Académique Documents
Professionnel Documents
Culture Documents
Abstract: Fechner's conception of psychophysics included both "outer psychophysics," the relation between stimulus intensity and
the response reflecting sensation strength, and "inner psychophysics,' the relation between neurelectric responses and sensation
strength. In his own time outer psychophysics focussed on the form of the psychophysical law, with Fechner espousing a logarithmic
law, Delboeuf a variant of the logarithmic law incorporating a resting level of neural activity, and Plateau a power law. One of the
issues on which the dispute was focussed concerned the appearance of contrasts if the overall illumination was increased or decreased;
another issue was the question of whether a sensation of a "just noticeable difference" established for one value of a sensory dimension
appeared the same for a value elsewhere on the dimension. The development of "inner psychophysics" led through the works of
Delboeuf, Solomons, Jastrow, and Thurstone to modern signal detection theory. A third line of research, devoted to the question of
what was meant by the "measurement" of sensation strength, stemmed from the criticism of Fechner's work by von Kries (1882) and
others. Although a valid body of science could be built up without the intervening variable called "sensation strength," such a science
might be a cumbersome representation of reality. When an optical contrast is set up, and its overall illumination is increased or
decreased, subjective contrasts involving medium levels of lightness vary little as illumination varies (as a power law based on
sensation ratios or a logarithmic law based on sensation differences predict), but subjective contrasts involving extreme levels of
lightness might be subject to the effects of other variables.
Keywords: Delboeuf; distinctiveness; Fechner; history of psychology; measurement; philosophy of science; psychophysics;
sensation; Stevens
(constant) subjective magnitude of a just noticeable differ- showing sensation magnitudes as a logarithmic function of
ence. In the Elements, Fechner went on to apply calculus stimulus intensity. It can also be claimed that the calcula-
to Equation 1, as follows. Dividing both sides of Equation tion of sensation magnitudes was unnecessary given the
2 by A/ gives presentation of a formula for calculating them. Falmagne
(1986, pp. 1-10) has suggested that the actual construc-
AS/A/ = C(l/1) (3)
tion of a scale of sensation magnitudes based on estimates
Fechner (Elements, vol. 2, p. 9) then makes the as- of jnds would necessarily be approximate because errors
sumption that, just as a physically small quantity A/ can made in the determination of an initial jnd would have a
be reduced without limit to the differential dl, so a small cumulative effect in a scale constructed of concatenated
"quantity" of sensation can be expressed, in the limit, as jnds.
dS. It may be noted that this assumption has been Fechner also devoted a great deal of space to discussing
challenged by Luce and Edwards (1958) on formal mathe- criticisms of Weber's Law: These criticisms, and Fech-
matical grounds. 2 Fechner then proposes the differential ner's rebuttals, were reviewed in the nineteenth century
equation (taking limiting values of A/ and AS) that follows by Kohler (1886) and more recently by Murray and Ross
from Equation 3: (1988), who also assess Fechner's use of the law in a
variety of contexts. The criticisms resulted from work by
dS/dl = C(l/I)
Aubert on brightness discrimination, by Hering on
The standard solution is S = C In / + constant of weight discrimination, and other work which in general
integration. Fechner showed that this in turn reduced to led Fechner to conclude by the end of his life that
Weber's Law was valid, except for extreme values, for
S = C In (///) (4) prothetic (intensitive) dimensions like brightness and
where / 0 is the "absolute threshold" value of/,- S = 0 (no loudness, as well as length and time discrimination; it
sensation) where / = l0. In order to apply this logarithmic might not be valid for metathetic dimensions like hue and
psychophysical law to real life, Fechner suggested that we pitch, in which sensory quality as well as "quantity"
could estimate both Zo, and A/ in the laboratory by various changes.
psychophysical methods. These include the method of
adjustment and the method of "right" and "wrong" cases:
We shall return later to a discussion of how these fit into a 1.2. Fechner's "Inner psychophysics"
more general scheme concerning methods. For Fechner, Fechner made an important claim about the logarithmic
the analytical arguments could only take on life if the function which was apparently shrugged off by Hering
psychophysical methods could be applied in the labora- and other investigators of his time on the grounds that it
tory; the study of the advantages and drawbacks of the was too metaphysical (Scheerer 1992); he proposed that
various methods came to dominate psychophysics partic- the transformation of physical intensity into sensation
ularly in the period between 1890 and 1920 - Titchener's magnitude, which had to be compressive in view of the
monumental Experimental psychology (1901-1905) is an logarithmic nature of the transformation, took place at the
exhaustive account of the early literature on Fechner's time physical events in the brain were translated into
methods. mental events of sensation. In order to understand this
In developing his theory, Fechner considered that view, we have to understand Fechner's distinction be-
since Weber's Law formed a cornerstone from which he tween "outer" and "inner" psychophysics. (Here, "inner"
could move forward to derive his logarithmic psycho- is a translation of the German word "inner" but "encapsu-
physical law by a logical procedure, it was of the greatest lated" might be an alternative translation.) Outer psycho-
importance to establish the validity of Weber's Law. To physics studied the relationship between the observable
do so required accurate measurements of the differential physical stimulus and the reported psychological re-
threshold, and this was the incentive for Fechner to sponse; for example, the psychophysical law relates / to S,
develop his psychophysical methods; furthermore, one of both of which are based on observable data. It was
his ways of demonstrating that Weber's Law might be Fechner's view that the most reliable estimate of S was
valid was to take various values of I and show that the obtained from a summated jnd scale or a bisection scale
percentage variability ofjudgments concerning whether I and he had little respect for what we would now call
+ A/ was different from 7 remained roughly constant for "direct magnitude" estimates of S. He felt that these had
all values of I (Fechner 1860/1966, pp. 85-93). This no more value than ratings of "beauty" or other subjective
treatment of variability involved an analysis of properties impressions. Nowadays we adopt the distinction between
of the normal distribution. (Later, in his Kollektiv- "confusion scales" (such as summated jnd scales or paired-
masslehre (1897), Fechner developed a number of other comparison data), "partition scales" (such as bisection data
statistical ideas, including the evaluation of data whose or category judgments), and ratio scales (such as are
distribution was skewed and of whether in a series of obtained from direct magnitude estimates or production).
measurements, each measurement was independent of Fechner might have found this nomenclature inappropri-
the previous one; see also Heidelberger 1987). Fechner's ate: For him, more "confusion" would have been associ-
own experimentation, as reported in the Elements, was ated with magnitude estimates. Moreover, if one insisted
largely restricted to the determination of thresholds using on making a dichotomy between "direct" and "indirect"
weights and line lengths; perhaps because of the expense measures (where the reality may be that we are dealing
of engraving figures, nowhere in any of his three books on with a continuum) some individuals may consider parti-
psychophysics, the Elements (1860), In Sachen der Psy- tion scales to be direct rather than indirect. But, since the
chophysik (1877), or the Revision der Hauptpuncte der data of outer psychophysics in all these cases were based
Psychophysik (1882), does Fechner give a graph or a table on measurement, even though nominally indirect as in
the case of a summated jnd scale, outer psychophysics was that this view anticipated signal detection theory; how-
the type of psychophysics most easily carried out. ever, signal detection theory had a separate history,
Inner psychophysics, however, was just as important starting with Neyman and Pearson's (1928) analysis of the
for Fechner. Inner psychophysics concerned the relation- process of statistical decision making. Moreover, signal
ship between what we would now call neurelectric as- detection theory is more explicit than was Fechner on the
pects of the stimulus and the mental experience thereby need to distinguish "internal noise" in the observer from
evoked: At the heart of inner psychophysics was the "external noise" present in the stimulus (for example, the
"mind/brain" problem - Scheerer (1992) has remarked of quantum or particle nature of light insures that the
Fechner's opinion: ". . . to use terms not employed by stimulus has intrinsic variability). It can also be claimed
Fechner himself, the mind is an emergent property of the that the interpretation of the word "noise" as neurological
entire psychophysical system and cannot be reduced to noise is only one way to handle it within the framework of
one of its (functional or spatial) components" (p. 8). the signal detection model of decision making. However,
Essentially Fechner thought of a physical stimulus as since signal detection theory is an intrinsic part of the
influencing the nervous system by imposing a set of history of psychophysics it is included in Figure 1 in the
ripple-like oscillations on a resting state of high- context of its logical relationship to Fechner's inner psy-
amplitude low-frequency oscillations: The oscillations im- chophysics, even though a historical connection is diffi-
posed by the physical stimulus had to be discriminable cult to trace.
from the resting state and furthermore had to be of such a It may also be noted here that, in assuming that a
degree of discriminability as to cross a particular "thresh- nervous excitation had to exceed a particular nervous
old" if the subject was to become conscious of the stim- threshold before a conscious perception concerning a
ulus. It was quite possible, in his view, for activity from a difference in sensation was possible, Fechner came close
sensory receptor to influence brain activity without influ- to asserting something like Stevens's later neural quan-
encing the activity of consciousness, and as early as the tum theory: Corso (1967) explicitly stated that Fechner
Zend-Avesta of 1851, Fechner speculated that the and others "argued for the noncontinuity notion which
amount of "work" necessary for a stimulation to become included the threshold concept" (p. 423). This raises the
conscious could be reflected in a number. A conscious difficult question of how Fechner would have viewed
sensation could have a number ascribed to its magnitude subsequent work according to which the mere existence
(a number predicted by Fechner's Law): If nervous activ- of variability in data entailed that there was a variability in
ity was aroused by a stimulus / that was below the 70 the nervous system that would make the concept of a fixed
needed to produce a sensation, the nonexistent sensation threshold nonviable. Later, we shall review evidence on
that could not be aroused was nevertheless given a nega- the line of research stemming from inner psychophysics
tive number. These negative numbers had no meaning in according to which Thurstone and the signal detection
reality except as an index of how subthreshold a level of theorists made variability a datum in the derivation of
activity was; in later years Fechner fretted at the way ideas about difference sensations; and, as mentioned, we
people assumed he was talking about something called know that Fechner himself was interested in the mea-
"negative sensations" as if they had an ontological status surement of variability. So this is perhaps the place to say
equivalent to real sensations, when he never meant to that, although Fechner foreshadowed signal detection
imply this at all (this issue is discussed at greater length by theory in the important respect that he considered signals
Murray 1990). to have "distinctiveness" only in the context of a back-
Thus, the activity induced by a physical stimulus in the ground of noise, he may well have disapproved of any
nervous system had to cross several "thresholds" of ner- reliance on the variability of data in order to measure that
vous activity before a particular threshold was reached at "distinctiveness." Stout (1989) has shown how even in the
which it was translated into conscious experience; it was 1920s psychologists did not always see eye-to-eye in the
at that point that Fechner localized the transition from meaning to be ascribed to the variability of data.
many to one, from a broad range of physical intensities to To summarize Fechner's work so far, we may say that
a narrow range of discriminably different sensations. This he intuited that sensation magnitude was a logarithmic
transformation was nonlinear - it was at this junction, function of physical intensity; an appropriate demonstra-
rather than at the periphery, that Fechner localized the tion would show that, once Weber's Law was validated, a
logarithmic transformation. Three other aspects of Fech- scale of sensation magnitude could be based on a knowl-
ner's writings on inner psychophysics may be mentioned edge ofjust noticeable differences. Fechner (1987b) him-
in passing: (1) Once a sensation had become conscious, it self realized, however, how the construction of a scale of
was possible by "selective attention" to concentrate on concatenated jnds could be carried out independent of
something else and thereby be temporarily unconscious Weber's Law. He noted that "measurement of difference
of that sensation; (2) as mentioned earlier, the nervous sensations is by no means restricted to the validity of
activity referred to was held to occupy the whole brain, Weber's Law. Indeed . . . in principle it has nothing to
with memory activity being aroused by a sort of reso- do with it" (p. 215). Moreover, Luce and Edwards (1958)
nance; (3) since the nervous activity evoked by a stimulus have argued that since in the construction of a scale of
was added to the preexisting level of activity, this implied magnitude jnds are concatenated, this could be carried
that for Fechner all stimulus-related nervous activity had out even if Weber's Law were invalid, a point also made
to be measured in relation to other nervous activity. It by Krueger (1989a). And Parker and Schneider (1980)
would then follow, because of the property that conscious claimed that although Weber's Law was indeed invalid for
experience depended upon nervous activity, that individ- judgments of loudness, a scale of subjective loudness
ual conscious experiences also had to be discriminated could nevertheless be constructed on the basis of jnds.
from a background level. Scheerer (1987) briefly indicates Fechner also suggested that the locus of the logarithmic
transformation was the junction of the brain/mind inter- S = log [(c + l)lc] (6)
face. He disliked the idea of measuring sensory magni- where c is a constant, to be evaluated by experiment, that
tude by ratings. He felt that a full discussion of sensory reflects the ongoing state of activity in the nervous system
magnitude would involve a consideration of preexisting at the time the stimulation measured by Z is presented.
levels of neural activity, and he maintained that sensory One important fact to note about Delboeuf's formulation
magnitude was a measurable entity. Our main contention is that, like Fechner's, it is logarithmic. Another is that
will be that the history of psychophysics since Fechner's Delboeuf started using a new method for estimating
time can be seen as having two broad strands, one sensation strength, a method based on estimating what
resulting from his activity on outer psychophysics - a stimulus intensity should be chosen so that it yielded a
strand that included unexpectedly strong criticism both of sensation that looked halfway between two other sensa-
his belief that the psychological law was logarithmic and of tions. The originator of the idea that sensation strength
his belief in the measurability of sensory magnitude and could be measured by finding a sensation midway be-
a second strand, logically connected with his inner psy- tween two others was J. A. F. Plateau (1801-1883) of
chophysics, that ultimately led to signal detection theory. Brussels; we shall describe his findings below. Delboeuf,
The chart of Figure 1, however, does not imply that those however, only contacted Plateau about his own experi-
who pioneered the research leading to signal detection ments after he had started them; they were designed
theory necessarily knew of Fechner's contributions: The explicitly to test Fechner's theory.
chart is in terms of commonalities of interest rather than
Delboeuf, then, set out to examine the validity of
of direct historical interaction. It must be understood that
Equation 6 by carrying out a systematic series of experi-
Fechner himself saw inner psychophysics as the central
ments in which he obtained any desired grey by revolving
core of his theory and said that a psychophysical law
a black disc that contained a white sector of a given
applied particularly to inner psychophysics; it would not
angular width: The greater the area of white, the lighter
be expected to be applicable to outer psychophysics and
the resulting grey. His visual display was presented
to fit the data perfectly, because there are many sources of
against a black background, with the whole set-up being
variability determining the data of outer psychophysics.
viewed in a variety of daylight illuminations or by candle-
We can now trace the development of these two strands,
light. The display was composed of three concentric
reviewing first the nineteenth-century evidence concern-
rings, a dark outer ring, an intermediate middle ring, and
ing alternative psychophysical laws.
a light inner ring (see Figure 2). In all his experiments,
the greyness of the outer and middle rings was prear-
ranged and the subject had to set the greyness of the inner
1.3. The line of research leading ring so that the middle ring looked intermediate between
from "outer psychophysics" the outer (dark) and the inner one (light, set by the
subject). If the three greynesses thus represented are
1.3.1. Retaining the logarithm: Delboeuf's equation.
labelled as 7X (outer darker ring), 72 (middle ring), and 73
Apart from the logical rigour of its derivation, the most
(inner ring), each with its corresponding sensation S j , S2,
powerful evidence in favour of a logarithmic law relating
and S3, then from Equation 6, we surmise that
sensation intensity to physical intensity was believed to
be that obtained from a series of experiments carried out S, = log[(c + /,)/<:]; S2 = log [(c + Qlc\,
by J. R. L. Delboeuf (1831-1896), a scholar associated S3 = log [(c + !3)/c)] (7)
with the University of Liege in Belgium. Delboeuf wrote and moreover that
a monograph of 115 pages in 1873 in which he first
reviewed Fechner's derivation of Fechner's Law and then Sz-S^Sz- S2 (8)
he also reviewed a modification of the law that had been From these identities we can express c in terms of Iu Z2
made by Helmholtz (1856-66/1962). According to Helm- and Z3; in fact, Delboeuf deliberately picked values oflu
holtz, any new sensation of brightness is added to a Z2, and Z3 so that Z2 looked intermediate between lx and Z3
"natural light of the retina," and this is in turn a function of for a particular placement of a candle, and he solved for c;
the total ambient light in a perceptual situation. For he then used this value of c to calculate what Z3 ought to
example, if a contrast is set up between a light object and be, given various other values of 7X and Z2. This calculation
another object in the surround of nighttime illumination, of an estimated Z3 was based on the following relationship,
the contrast looks less striking if the two objects are then derived from Equations 7 and 8:
viewed under daytime illumination. Helmholtz sug-
gested a modified version of Fechner's fundamental for- 73 = (/| - c/, + 2cZ2)/(c + /,) (9)
mula, one which made an important qualitative change. The question of course was whether the estimated Z3 as
In our notation, Helmholtz suggested that calculated in this way matched the Z3 that was actually set
I')]
by the subjects. In Delboeuf's first table, 13 different
(5)
values of Zj and Z2 are investigated and shown along with
where /' is a function of the natural light of the retina and the subject's settings of Z3; each setting of Z3is the average
A is a c6nstant. Possibly inspired by Helmholtz's intro- of five settings and Delboeuf stressed that the subject
duction of a factor indicative of a resting state of activity in could often take a long time (minutes) in choosing Z3.
the nervous system, Delboeuf went on to derive an Moreover, the overall level of daylight illumination, gen-
equation of his own; in so doing he extensively discussed erally resulting from a cloudy sky, could not be con-
such factors as the degree of adaptation or fatigue in trolled. Delboeuf quoted two measures of central ten-
receptor systems, but the end result was that Delboeuf's dency, the halfway mark between the highest and lowest
equation, in its simplest form, was: scores of the five settings of Z3, and the arithmetic mean of
BLACK SURROUND
ooo
Background i Background)
preset preset
1 preset \ / variable \
Table 2. A numerical demonstration of predictions generated by Fechner, Plateau, and Delboeuf on whether a contrast set up
under one illumination (in which stimulus ratios are 2:1) will change in appearance if the overall illumination is brightened
(making the stimulus ratios 10:5 respectively)
Fechner
s=yiiog/) Plateau Delboeuf
S = log[(c + l)lc]
Take logs
to base 10
s =m Let c = .5
Stimulus Let n = .5
Stimuli intensity I Sensation Sensation Sensation
boeuf himself made a simple demonstration that a given 1.3.3. Rejecting the logarithm: Plateau's equation. As
contrast could be reduced if overall illumination was mentioned earlier, Plateau (1872a) devised the method
darkened, in line with the prediction from his law on the of equisection, now known as the method of equal-
basis of sensation differences. On a white disc he drew a appearing intervals. He did it extremely informally:
portion of a narrow black sector. When the disc was When he asked eight painters to mix a grey that appeared
rotated, a grey zone appeared on the white surface of the equidistant between white and black he found that the
disc. If this zone was just visible by the light of a candle greys thus produced were very similar. He also noted that
placed nearby, it became invisible if the candle was an etching, consisting of strong contrasts of black and
moved away from the display to a considerable distance: white, seemed perceptually identical in terms of contrast
To make the zone visible again, the angular width of the whether viewed under daylight, strong sunlight, or can-
sector of which a portion was drawn had to be increased. dlelight. According to our Table 2, this would be expected
Two points of general interest may be noted here. if apparent differences between adjacent areas are what
First, consider the special case of two stimuli A and B remain constant as overall illumination changes, and
where B is just noticeably different from A. According to Plateau (1872a, p. 9) explicitly stated: "Fechner's formula
Fechner's Law, you should be able to darken or brighten has the consequence that, if the overall illumination
the scene, and the contrast between A and B will be such varies, it is differences between sensations that remain
that B continues to look just noticeably different from A. constant." Nevertheless, Plateau felt that it was not nec-
Fechner's own demonstration was to ask the observer to essarily differences between adjacent areas that we would
pick out a wisp of cloud that was just noticeably different expect to remain constant, but ratios, or relations, be-
from the background of the sky and then look at the cloud tween adjacent areas; he went on to remark: "It appears to
through a dark glass. He predicted that the cloud would be more reasonable, that if we are to understand the
continue to be just noticeably different from the sky. If persistence of the general appearance of the etching, to
Delboeuf was right, there should be a limit to how far this accept a priori the constancy of the ratios, not the differ-
is true; there should be a particular shade of darkness in ences, between sensations. "As Table 2 shows, if ratios are
the glass where the contrast is so reduced that the cloud to be preserved under different illuminations then the
becomes indiscriminable from the sky. Second, it is appropriate psychophysical law is a power function.
possible that in contrast situations white areas have a Plateau proved this algebraically; Brentano (1874/1973)
darkening effect on dark areas (so that increasing illu- commented that he felt that equal stimulus ratios implied
mination could make dark areas even blacker), whereas equal sensation ratios; and it was Fechner himself (1877,
dark areas may have no effect on white areas. This matter p. 24) who proved that a modification of the fundamental
will be taken up later, but we would note that this kind of formula (Equation 2) would lead to a power function.4 The
asymmetrical effect would be consistent with Delboeuf's modification consisted of assuming that it was not AS
results. alone that was incremented if the starting sensation /
were increased by A/, but the relation between AS and subjects are asked to judge which is the larger of two
the sensation S aroused by I differences in loudness (a - b) or (c - d), a scale can be
constructed in which estimated loudness is not a log-
&S/S = C(M/I) (10) arithmic but a power function of stimulus intensity
Solving this equation by the same chain of argument that (Schneider 1980a); and (c) evidence that some sensory
led to Equation 4, the logarithmic law, leads to the power dimensions may be associated with relatively easy differ-
law ence judgments (such as loudness, as just noted), whereas
on other sensory dimensions subjects more naturally
S = Klc (11) make ratio judgments. Among the latter dimensions are
where K is a constant. 5 length (Parker et al. 1975), area, volume (Schneider &
Yet despite this intense interest in the possibility of a Bissett 1988), and possibly pitch (Schneider et al. 1982).
power law, it was Plate' a (1872b) who, after reviewing Schneider and Bissett (1988) have shown that this distinc-
Delboeuf's paper, came to the conclusion that Delboeuf's tion may be unrelated to the prothetic/metathetic dichot-
equation was superior both to Fechner's equation and to a omy; instead, it seems that dimensions such as length,
power law when it came to predicting the data. More- area, or volume, along which stimuli are easily "decom-
over, Delboeuf's equation predicts that contrasts will posed" into a number of smaller perceptual units, are
change in appearance if overall illumination changes, more likely to be associated with ratio judgments,
whereas Plateau's equation had predicted that if apparent whereas dimensions such as loudness or heaviness (see
ratios were preserved they would not. Hence Plateau Rule et al. 1981), where such decomposition is subjec-
(1872b) retracted his views in the following words: tively unfeasible, are more likely to be judged in terms of
As for my formula, from the mere fact that it differs difference judgments.
from that of M. Delboeuf, it is clear that it is inexact. It Returning to the historical argument: Despite Plateau's
rests on the hypothesis that the level of illumination retraction, the power law did not vanish. Even Fechner
influences neither the ratios of sensations nor the (1887), in his last article, admitted that if the fundamental
equality of apparent contrast; but it can be easily formula of Equation 10 were valid then a power law could
demonstrated using M. Delboeuf's formula as a start- be a reasonable contender for the psychophysical law. He
ing point, that the ratios of the sensations do depend on claimed, however, that the logarithmic law worked better
in practice; moreover, he stuck tenaciously to the opinion
the level of illumination, and it has been seen that the
that the power law suffered from a serious drawback,
equality of the contrasts also depends on this, though in
namely, its assertion that if / were nonzero, the subject
a less pronounced fashion, (p. 261)
would necessarily have a sensation S of some strength,
At the risk of interrupting the historical flow, we also
since the constants K and C in Equation 11 were both
note a problem here that did not receive sufficient atten-
positive. For Fechner, building on the inner psycho-
tion in earlier years, at least until the work of Torgerson physics described earlier it was reasonable to suppose
(1961); in Plateau's bisection task, the subject is not asked that an intensity 7 could be positive yet unable to arouse a
to choose 72 so that S2/S1 = S3/S2, he is asked to choose 72 sensation because it elicited excitation below the thresh-
so that Sz S1 = S3 S2. The question may be raised old required for a conscious experience. As noted above,
whether a subject asked to do the former task - preserv- his opponents misinterpreted his use of negative signs to
ing ratios - would produce contrasts that are preserved refer to such subthreshold excitations as implying that
under different illuminations, a finding that would favour subjects had "negative sensations," an incomprehensible
a power law. Over the past two decades, however, it has concept; Fechner (1887) wearily remarked that "it is
become increasingly apparent that task performance is a incontestably to be desired that the controversy over
complicated function of whether subjects are asked to negative sensations should come to an end once and for
judge differences or ratios. Put overbriefly: Torgerson all; but if my experience up to now is my guide, even my
(1961) suggested that subjects made all their judgments ghost will have no peace because of it" (p. 224). For
on the basis of a sense of "difference' or "distinctiveness" Fechner, it was a fault, not a virtue, of a power law that it
and that it might be misleading to assert that a given implied there could be no subthreshold excitation arising
contrast is arrived at by a subjective judgment of ratio. from a positive intensity. Modern writers on psycho-
Birnbaum and Veit (1974) showed that subjects could say, physics, however, have suggested several ways in which
by giving a number, what the difference was between two subthreshold excitation arising from a positive intensity
weights (one in each hand) and what the ratio was be- can be reconciled with a power law (see e.g., Atkinson
tween the weights. However, following a complicated 1982; Dawson & Miller 1978).
literature, Birnbaum (1980) came to maintain that a single
mental operation underlay both difference judgments
and ratio judgments and that this single operation re- 1.3.4. Evidence for the power law based on the inequality
flected a subtractive procedure, so that a given difference of difference sensations. Another reason the power law
sensation Sab was equivalent to a sensation difference (Sa continued to hold sway arose initially from an observation
Sj,). There is a current of modern opinion to the effect by Hering (1875) attacking the view that, throughout a
that when a person sets up a "contrast" between two range of/ values, AS is constant. According to Hering, a
stimuli, A and B, he is judging their sensation difference difference sensation arising from an 7X (say a weight of 2
rather than their sensation ratio. To be weighed against units) and an 72 (say a weight of 1 unit) is not perceived to
this claim are (a) evidence that some subjects sponta- be the same difference sensation as arises from two other
neously give ratio judgments even when the task is weights (say, a weight of 10 units and a weight of 5 units)
devised to yield difference judgments (Popper et al. 1986; that bear the same ratio as 7j to 72. Hering actually did the
see also Schneider 1980a; 1980b); (b) evidence that when experiment, placing weights of 500 and 1,000 grams on
one hand and weights of 1,000 and 2,000 grains on the point where 7/70 equals e (2.71828) is passed (see Fechner
other; the latter difference, he claimed, seemed larger. 1860/1964, vol. 2, p. 49), a feature that is often over-
Fechner simply denied that the experiment was intro- looked; a logarithmic law cannot handle a positively
spectively valid: While the weights on the second hand accelerated curve such as is apparently obtained with
were certainly heavier in toto, the task of judging differ- rated heaviness, tactile roughness, and subjective
ences was too difficult using this paradigm (in fact Fech- warmth associated with a metal object touching the arm.
ner was always worried about the difficulty of evaluating The exponent of 3.5 quoted above for electric shock has to
either AS or Sl S2, where the two stimuli involved were be balanced against a claim of Swartz (1953): For three
separate in space or time, a worry shared by Krueger different ranges of current applied to the tooth pulp to
1989r, sect. 2.3). yield sensations of pain, if the subject set a variable
Nevertheless, Stevens (1957) has made it a cornerstone resistor to yield sensations half-way between the thresh-
of his belief in a power law that if a jnd (S1 S2) is old level and the pain tolerance level, the resulting graph
established for a pair of low intensities Ix and 72, the jnd of bisection points against actual fractions of the range of
(S3 S4) that is established for a pair of higher intensities stimuli was linear. Power laws can certainly be fitted to
73 and 74 on the same sensory dimension is not necessarily scales of "sensation strength" based on magnitude esti-
equal in apparent "size" to (Sv - S2). This assertion can be mates and cross-modal matching, and Stevens (1970)
validated by establishing (Sl - S2) and (S3 - S4) by a jnd claimed they could also be fitted to neurelectric data. In
method or alternatively showing that the difference be- fact, here Stevens seemed to argue that the nonlinear
tween sensory magnitude estimates of Ix and 72 is less transformation of an increase in stimulus intensity into an
than the difference between sensory magnitude estimates increase in sensation strength took place in the receptor
of/3 and 74. Stevens proved that if a power law holds, and rather than in the brain, as Fechner had claimed. The
if Weber's Law is true, then the subjective size of the jnd present situation, to judge by the responses to Krueger's
increases as an exponential function of the number ofjnds attempt to show that a modified power law can also be
by which the stimulus is above threshold (Stevens 1957, applied to summated jnd scales and category judgments,
p. 172). Later, Ekinan (1956; 1959) formulated "Ekman's is that the controversy sparked in the nineteenth century
Law," which stated that the psychological size of the jnd is by Plateau is still with us. In fact, we may say that it is the
a linear function of the sensation magnitude it is based on; major issue that arose from Fechner's outer psycho-
Teghtsoonian (1971; 1973) amended Ekman's Law to take physics.
account of the range of stimulus magnitudes used in an Figure 1 suggests that modern psychophysics stems
experiment - subsequent experimentation, such as that from two main lines, the line just described, leading from
of Marks et al. (1986) on cross-modality difference estima- outer psychophysics, and a second line arising from Fech-
tion, has shown the influence of the range of stimulus ner's definition of inner psychophysics. We now turn to
magnitudes on subjects' responding. The persuasiveness this second line of research.
of Ekman's argument was well illustrated by Gescheider
(1976, p. 160), who gave a proof by numerical demonstra-
1.4. The line of research leading
tion showing that if both Weber's and Ekman's laws are
from "inner psychophysics"
true then a power function of the type favored by Stevens
would be the resulting psychophysical law connecting S As noted, Fechner assumed that the excitation arising
with /. Special note should be made of a claim concerning from a stimulus was added to a resting level of excitation
loudness by Parker and Schneider (1980) according to and indeed that this excitation might have ensuing excita-
which, if Weber's Law is false for a sensory dimension, it tions that had to cross several "thresholds" before the final
is possible to have a psychophysical law that is a power threshold leading to a sensation, a conscious experience,
function as well as subjective jnds that do appear constant was breached. We also noted Helmholtz's assumption
in size across a range of loudness, as Fechner postulated. that at the level of the measurement of sensation whether
That Weber's Law might not hold for the loudness of or not an increment in visual sensation was detected
pulse sinusoid tones is suggested by data obtained by depended upon the increment's being large enough to
Jesteadt et al. (1977). In this study, log (A7/7) did not exceed "natural light in the retina," whereas Delboeuf
remain constant, but decreased linearly, as a function of assumed that a variable he called c had to be taken into
the stimulus intensity measured in decibels SL. The account in the psychophysical law. According to Krueger
tones varied in frequency between 200 and 8,000 Hz. (personal communication) it may even be possible to
For the moment we rest content with pointing to the reconcile Delboeuf's model with Fechner's if a nonzero
enormous popularity of a power law in twentieth-century (but subthreshold) value of 7 could induce a nonzero S that
research following Stevens's espousal of it. As an example was nevertheless not reflected in conscious experience.
of its general applicability, we can refer to Table 5.1 in Krueger has also pointed out that according to Fechner's
Gescheider (1976, p. 129), which shows that in 33 experi- Law, when 7 = I o , S must equal zero, yet it is at that point,
ments dealing with various stimulus dimensions, power the absolute threshold, that'S becomes consciously per-
laws could be fit to the data relating S to 7 with exponents ceptible. Fechner wrote that this was proof that a thresh-
varying from .33 (brightness as a function of the lumi- old must exist, for at a finite value of 7, namely 70,
nance of a 5 target in the dark) to 1.0 (apparent length of a sensation value S must be zero (Fechner 1860/1964, vol.
projected line) to 3.5 (the apparent "shock" value of an 2, p. 14). This does not rest easily, however, with those
electric current passed through the fingers). The fact that who would make the sensation at the absolute threshold a
some of these exponents can be greater than one illus- conscious experience.
trates the generality of the power law. A logarithmic law is The notion that a stimulus is never processed in isola-
associated with a negatively accelerated curve once the tion, but only against a background, holds both for physi-
ological discussions of sensory excitation and psychologi- this difference has by some been taken to be the
cal discussions of increments in sensations. Later we will differential threshold. What is most usually denoted by
examine the relationship between these two types of the threshold is the smallest difference that can be
discussion in more detail, but as a historical note we perceived. It is the threshold of conscious-
would remark that Delboeuf s psychophysical equation ness . . . either there is a threshold - be it a point or a
(Equation 5) was derived from an assumption about neur- more or less variable line - below which is homoge-
electric events. The same can be said of an argument neous unconsciousness; or from the region in which the
produced by our next figure, L. M. Solomons, who in sensed difference has its maximum of clearness down to
1900 offered an explanation of why Weber's Law took its the point where it utterly vanishes because this differ-
particular form in terms of a discussion of random activity ence between the stimuli vanishes, there is a contin-
in the brain. Solomons assumed that any brain activity uous series of intermediate degrees of clearness, and
underlying a sensation was not fixed but variable, and that there is no point on the curve with characteristics
sensation in turn was itself marked by variability. He peculiar to itself, no threshold in any true sense, (pp.
presumed that the degree of variability in brain activity B 276-77)
associated with stimulation from a physical activity / But this skepticism did not stop others from assuming that
varied between B AB and B + AB, with AB itself being the threshold could be defined as a particular point in the
a function of B. For an increment in intensity 7 + A7 to psychometric function: Urban (1910) gave details for
yield a sensation just noticeably different from that estimating the midpoint of such a function from raw data,
aroused by 7, 7 + A7 had to set up brain activity not merely and Boring (1917) gave an account of the psychometric
exceeding B but also exceeding B + AB. Since AB was function showing how it could arise if there were random
held to be a constant fraction of B, AZ would have to be a fluctuations in responding caused by a number of factors
constant fraction of 7 if / + A/ were to be discriminated that could not be experimentally controlled.
from 7. This, of course, is Weber's Law. But Solomons's
The theory that the psychometric function was an
name is mentioned not merely to add to the list of
artefact of chance variables disguising the true sensation
theorists who argued that stimulations have to exceed
of a difference between A and B was called the phi-gamma
resting levels, but because he was one of the first to stress
hypothesis by Urban and Boring.6 In Figure 1 we have
that variability in the sensory system is associated with
indicated that Fechner's inner psychophysics formed a
variability in sensation and that any theory of psycho-
physics has to take variability not merely as a source of historical backdrop to research that ultimately culmi-
error but as a given quantity to be evaluated. We have nated in signal detection theory only because Fechner's
already noted that Fechner measured variability. The inner psychophysics stressed that sensations were the
idea that variability is a useful measure per se in the outcome of stimulus-produced oscillations superimposed
estimation of sensation strength was criticised by Stevens on resting-level oscillations; it was the doubt raised by
(1957). observers of the psychometric function as to whether a
sensation arose discontinuously from these superimposed
oscillations that instigated subsequent research on vari-
1.4.1. The psychometric function. The point of contact ability, and thence signal detection theory. By the time
between the line of research concerned with the psycho- signal detection theory had arrived, doubts were being
physical law (shown in Figure 1) and the line of research expressed as to the existence of a discontinuous "thresh-
concerned with variability and "noise" lies in what is old," and the line in Figure 1 connecting Fechner's inner
known as the psychometric function. In establishing just psychophysics with signal detection theory should be
noticeable differences, the method of right and wrong interpreted as indicating a common thread (signal against
cases involves presenting two stimuli, A and B (with A and noise) but also a contradiction (concerning whether or not
B being close to each other on a physical continuum of there was a discontinuous threshold). By the time signal
weight, length, intensity, etc.) and having the subject detection theory appeared in the context of psycho-
decide whether A > B, "yes" or "no." If we plot the physics (Swets et al. 1961) the problem of "variability" was
percentage of yes judgments (p) against the size of the being related to the problem of "noise" of various kinds:
physical difference (A - B), we typically find an approx- (1) noise intrinsic to the stimulus, as exemplified by the
imately ogival curve rising from p = 0 for small values of quantum nature of light; (2) noise as a physical back-
(A - B) to p = 1.00 at large values. The term "psycho- ground to the stimulus; (3) noise in the sensory receptor;
metric function" was given common currency by Urban (4) noise in the brain; and (5) noise in the sense that
(1910), but prior to Urban's time a number of authors had subjects' responses were variable because of biases,
claimed that the mere fact that the rise in the curve from p order-of-presentation effects, fatigue, and so on (but
= 0 to p = 1 is not discontinuous, but continuous, see, e.g., Wertheimer 1953, for the claim that "order-
suggested (a) that the definition of what constituted a of-presentation effects should not be considered as ran-
threshold value on the A B continuum was necessarily dom").
arbitrary and (b) that the continuous nature of the curve A continuous psychometric function might be influ-
indicated that in the nervous system itself there was not enced by noise of any of the above types, as Urban and
necessarily a strong continuity between "no sensation" Boring had suggested; but it was also claimed, partic-
and "some sensation." As Corso (1963) outlines in detail, ularly by Stevens et al. (1941) and Corso (1967, pp. 423-
suspicions as to the value of the concept of a neurelectric 45), that a continuously increasing psychometric func-
threshold were expressed by G. E. Muller(1878), Jastrow tion, particularly if it was linear as opposed to ogival, was
(1888), and Cattell (1893); Jastrow wrote: consistent with the hypothesis that the sensory receptor
The conception [of the threshold] grew directly out responded in a quantal fashion, and that if a receptor was
of the method of the just observable difference; in fact already firing to a given stimulus intensity, there were
ranges of stimulus intensities over which no changes in cesses were often modified to make them probabilistic
firing would take place until a particular value of stimulus (see Falmagne 1986, for several examples of this trend in
intensity (below or above the original) had been reached. psychophysical theory).
This limit on receptor sensitivity can be described in Thurstone's model can be diagramed by showing a
terms of noise of type (3), and Cornsweet (1970, Ch. 4) has range of sensory states along an axis and drawing two
a particularly detailed analysis of receptor noise in the normal distributions on top of the axis, one representing
context of neural quantum theory. Thresholds for firing in the range of states excited by stimulus A (with its mean at
receptors were later distinguished from thresholds con- a) and the other representing the range of states excited
tingent upon brain and response variables ("observer" by stimulus B (with its mean at b). This diagram is shown
thresholds), and Luce (1963) and Krantz (1969) showed in Gigerenzer and Murray (1987, p. 37) but they go on to
how, if observer thresholds existed, it could be predicted argue that Thurstone for some reason did not add a
that curves of receiver-operating characteristics could be decision rule, that is, a rule determining which response,
rising discontinuously as opposed to continuously. yes or no, will be given if a particular state is experienced.
Discontinuities in sensory functioning are also illus- The addition of such a rule was provided by the signal
trated by the phenomenon of "categorical perception," in detection theorists in the 1950s; Gigerenzer and Murray
which only a few stimulus values within a range of possi- show in detail how the typical signal detection model
ble stimulus values along a dimension can be discrimi- showing two normal distributions of sensory states, one
nated; a variety of theories as to why this should be the associated with noise and the other with signal-plus-
case are discussed in the book Categorical perception noise, along with a movable criterion, is an analogue of
edited by Hamad (1987): A particular problem raised by the Neyman-Pearson model of inference-making con-
Macmillan (1987), Massaro (1987), and Pastore (1987) cerning two hypotheses. J. A. Swets (personal communi-
concerns the nature of the relationship between the cation) has suggested that the main innovation of the
discrimination of a difference between two stimuli and researchers on signal detection theory was to postulate a
the facility with which the two stimuli can be assigned to a movable criterion. In modern terms, this means that
category. There is agreement that context effects, anchor when we are attempting to establish thresholds, the task
effects, and memory demands affect both tasks, but par- involves a certain level of "cognitive penetrability," as
ticularly the latter, whereas within a "category" discrimi- Pylyshyn (1984) expresses it [see also Pylyshyn: "Com-
nations between individual stimuli are often possible putation and Cognition: Issues in the Foundations of
given appropriate psychophysical tasks. Because the as- Cognitive Science" BBS 3(1) 1980].
signment of a category reflects a dimensional process,
both Macmillan (1987) and Rozsypol et al. (1985) have
suggested that signal detection theory can be applied to 1.5. Some current Issues
categorical perception tasks. Wilson (1987) has argued It is not my purpose here to review the enormous amount
that the various contextual effects that influence a deci- of modern work either on signal detection theory (see
sion concerning which category a stimulus belongs to can e.g., Falmagne 1985, Chapter 10; Macmillan & Creelman
be modelled by adaptation level theory. 1991; McNicol 1972; Swets et al. 1961) or on the psycho-
From a historical point of view, the link between the physical law (e.g., Falmagne 1985, Chapters 3, 14;
phi-gamma hypothesis of the 1920s (with its generalized Krueger 1989a; 1989r). If we approach the signal detec-
treatment of variability) and the signal detection theory of tion model as a model of brain activity, the following
the 1960s (with its analytical treatment of various kinds of matters remain open:
noise) was provided by Thurstone (1927a; 1927b; 1927c;
1927d). Instead of trying to establish a "threshold" from 1.5.1. On signal detection theory. It has been frequently
the psychometric function, Thurstone argued that each claimed (see e.g., Green & Swets 1966) that as signal
stimulus, A and B, set up a range of internal states (a1, strength increases, so does the variability of the underly-
o 2 . . . a j a n d ^ j , b2 bn). Each of these states had a ing signal-plus-noise distribution. If so, this would be
possible frequency, so that the states (alt a2 an), for consistent with Solomons's suggestion that the effect of
example, could be taken as normally distributed with a range of brain states increases as stimulus intensity in-
particular state a representing a mean subjective state creases. Yet we have found little experimental work using
associated with stimulus A, and similarly, another state b signal detection theory, and a variety of signal strengths,
representing a mean subjective state associated with apart from that of Markowitz and Swets (1967), devoted to
stimulus B. Using what is now known as the method of establishing whether this claim is true. Moreover, Swets
paired comparisons, Thurstone established the "law of (1986a; 1986b) examined both theoretical and empirical
comparative judgment" which took various forms de- receiver operating characteristic (ROC) curves derived
pending on the degree to which A and B were considered from a variety of indices relating hit rates to false alarm
independent: For complete independence of A and B, rates and came to the conclusion that there was no
common explanation for cases of unequal variances of
Pnb = N(a-b) (12) signal and noise distributions as estimated from the slopes
where p a b is the probability that A is judged greater than of empirical ROC curves. Indeed, one of the studies
B and N is the cumulative normal distribution function. surveyed related to recognition memory, where it was
This is a rephrasing of the equation for the psychometric shown that the linear ROC curves relating to memory for
function in which a and b represent mean sensory states words studied once was nearer the chance line than was
out of the many that can be aroused by each of A and B. that relating to words studied twice; but the slopes of the
Incidentally, Thurstone's work started a trend in which two curves were both about 1.00 (Egan 1958). Neverthe-
deterministic mathematical models of psychological pro- less, research by Tolhurst et al. (1983) indicates that an
increase in visual signal strength causes not only an from tasks when the subject must detect an increment
increase in the mean number of impulses in single cells in added to a stimulus ("increment detection"). This finding
the visual cortex but also an increase in the variance would hearten Fechner, who always claimed that the
associated with that number. Furthermore, Thurstone's measurement of differential thresholds was extremely
Case V assumes that if we have two mean states of difficult because of the inability to control all aspects of
sensation strength a and b, with b > a, the distributions the display.
associated with b and a are supposed to have equal
variability. But if the variability associated with b is 1.5.3. Von Kries. We suggest that the scheme shown in
greater than that associated with a, and this variability is Figure 1 can facilitate thinking about psychophysics in
held to be a function of sensation magnitude, then scales general, but before turning to the second aspect of our
obtained from the Thurstonian paired comparisons "perspective," we must briefly mention that Fechner's
method would be linearly related to scales obtained from work was criticised by von Kries (1882) on the grounds
ratio methods. that Fechner's whole endeavour rested on quicksand:
Other questions on which considerable literatures now Sensations are not the sort of thing that can be measured
exist concern how the criterion location is related to a in the way that length or time can. Von Kries claimed that
variety of stimulus parameters, including the proportion all the measurements routinely used in physics, including
of signals in the experiment (Parks 1966), the effect of a measures of mass, temperature, and various aspects of
signal presented prior to a critical signal (Treisman electricity and magnetism, rest ultimately on measures of
1964b), and the effect of varying the signal-to-noise ratio space (lengths) and time (duration). Sensations cannot be
while the response bias is kept constant (Dusoir 1975; measured, he argued, because there are no agreed-upon
1983; see also Curry et al. 1977). Much of this empirical units: I cannot say that a sensation of strength "4" is
work on signal detection has concerned signals varying compared with 4 units of strength " 1 , " because the latter
along one dimension only, but the theory can be extended is impossible to define, and moreover I cannot show
to multidimensional signals; according to Macmillan and ostensively how a sensation of strength "4" can be com-
Creelman (1991, p. 240), Tanner (1956) first proposed posed of 4 units of strength " 1 . " This is to be contrasted
measuring how a stimulus of one frequency, Su could be with the case for length, where I can prove that a line is 4
discriminated from noise, measuring how a stimulus of a metres long by laying 4 one-metre-long sticks alongside it
different frequency, S2, could be discriminated from in a chain. For von Kries, a number assigned to a sensa-
noise, and then measuring how discriminable Sj was from tion magnitude had a status different from that of a
S2. These questions have all arisen within the context of number assigned to length or time: Whereas Fechner had
the signal detection model, however; recent attempts to considered direct magnitude estimates as poor measures
relate signal detection parameters to those obtained from of sensation strength, von Kries argued that sensation
other models have focused on relationships with Luce's strength could not be measured at all, either by sum-
choice theory rather than with Thurstone's theory (see
mated jnds or by magnitude estimates. Any number
e.g., Swets 1986a; Macmillan & Creelman 1991).
assigned to a sensation strength, for von Kries, was no
more informative than a word like "bright" or "dazzling"
1.5.2. On the psychophysical law. The question has been or "glaring." True, these words indicated physiological
raised whether experimental data can be used to decide byproducts associated with particular stimulus inten-
between a logarithmic and a power law of psychophysical sities, but number ratings were believed by von Kries to
functioning. Krueger (1989a; 1989r) has pointed out that if have little extra value compared with such words, and
a power law has a very small exponent, its predictions should certainly not be used as a basis for a "scale" of
cannot be distinguished from those of a logarithmic law sensation strength that could have the same scientific
(see also footnote 3). reliability as a scale of length.
As Krueger's 1989 article and its accompanying com- Fechner replied to this argument in his Revision der
mentaries indicate, at least some members of the psycho- Hauptpuncte der Psychophysik (1882, pp. 321-24). He
physics community would like to integrate findings from clarified a point about von Kries's claim: The reason that
scales based on jnds, bisection methods, category judg- length and time can be "measured" so easily is that we are
ments, and magnitude estimates within one broad theo- using measuring devices of the same kind as the magni-
retical framework (Krueger himself has suggested a mod- tude to be measured, lengths to measure lengths, seconds
ified power law as one possible framework). As a side to measure seconds, and so on. Von Kries objected that
issue, the question has been raised as to the nature of the psychophysics offered no similar advantage - sensory
internal "scale" representing the natural numbers, a scale magnitudes cannot be measured in terms of sensory
presumably used in the assignment of category judg- magnitudes. Fechner replied that this claim runs counter
ments and magnitude estimates (Baird & Noma 1975). to the evidence: Astronomers can measure star-
The nineteenth-century dream of starting with as- brightnesses, by comparing them with other star-
sumptions about neurelectric phenomena and from these brightnesses, and Delboeuf's subjects could indicate a
deriving, by as rigorous a deductive chain as possible, a greyness intermediate between other greynesses. Fech-
psychophysical law, or failing this, an account of receiver ner therefore felt that von Kries's argument could be
operating characteristics, is very much alive in the writ- dismissed, but it is clear that he considered it an attack on
ings of D. Laming (1985; 1986; see also Laming 1989; the very heart of psychophysics. If von Kries was right, he
1991). One of Laming's discoveries is that tasks involving maintained, then we would have to abandon psycho-
judgments comparing two adjacent stimuli (the "detec- physics, with the serious result that we would be back in a
tion of differences") can yield estimates of differential wasteland where psychology had no scientific status and
thresholds or ROC curves that differ from those obtained no body of measurements to contemplate. We should
remember that Fechner wrote this before Ebbinghaus, In the twentieth century it was of course Stevens (1951)
Pavlov, or Freud had made their discoveries. who introduced the distinction between ordinal, interval,
It is perhaps appropriate here to mention that Stevens and ratio scales. In the 1980s Laming (1984; see also
(1957) also attacked Fechner's principles of sensory mea- Laming 1991) argued that the difficulty we have in ascrib-
surement, but on grounds different from those of von ing absolute category values to individual stimuli suggests
Kries. For Stevens, it was Fechner's choice of the "just that we do not have an internal representation of those
noticeable difference" as a unit of sensation strength that stimuli which is of such a form that category values can be
could not be tolerated. The just noticeable difference is an read off the internal scale: However, Laming does show
index of the smallest sensation increment that can be that if on trial n we have a stimulus of a given intensity,
detected, but no good system of measurement, according then on trial n + 1 we can judge whether the stimulus is
to Stevens, can be based on a unit that simply reflects the equal to, less than, or greater than the stimulus on trial n,
"resolving power" of the measuring instrument. The units with the predictability of the resulting assignations of a
of length in physics are not dictated by the smallest length category to the stimulus on trial n + 1 increasing system-
that can be seen in a microscope, and only a bizarre atically with the size of the difference between the stim-
physics would result if we did choose such a unit as a basis ulus used on trial n and the stimulus used on trial n + 1. In
for length measurement. The fact that the subjective size my opinion, Laming's argument also supports the notion
of the jnd, in Stevens's opinion, did not remain constant is that limitations of memory determine the ascription of
a further argument against the use of the jnd as a unit of absolute category values and that scales linking stimulus
sensation strength. Stevens's argument against Fechner's intensities with estimates of sensation magnitude may be
choice of unit should be strictly distinguished, however, subject to variability resulting from memory limitations.
from von Kries's argument that sensations cannot be Luce (1972) has proposed that the observer himself is a
measured at all in the sense that length and time can be measuring device measuring the effects of external stim-
measured. ulus magnitudes upon the timing of impulses in the
The claim that sensations cannot be measured when nervous system, a view that has an affinity with Mach's
"measure" is defined in terms that have been found theory of measurement. Luce and Narens (1987) have
suitable for the physical sciences has led to the develop- developed a new view of measurement theory according
ment of a branch of science called "measurement theory." to which items on a continuum can be related to items on a
In the nineteenth century reactions to von Kries's argu- derived continuum by a relationship that can be de-
ment ranged from the supportive (it was not only von scribed as having a certain degree of "homogeneity"; Luce
Kries who objected to the concept of "sensation inten- and Narens show that the measurement of sensation has
sity") to the far-sighted, such as the views of Ernst Mach, properties characteristic of certain degrees of homoge-
who saw that the problem raised by Fechner had implica- neity. In their concluding words: "Processes that allow
tions not only for psychology but for measurement theory behavioral attributes to have strong forms of measure-
in general. It was Mach who claimed that measurements ment have been developed and measurements of such
themselves fundamentally depended upon sensations: attributes, if they exist, will act in much the same way as
The measurement of length or time is only possible physical units. Moreover, it is mathematically feasible for
because the physicist can see that a distance of three them to be combined among themselves and with physi-
metres is three times as long as a distance of one metre; cal units in just the same way as physical units combine"
the measurement of temperature is chosen to be based on (p. 1552). This contemporary optimism is in stark contrast
length (the length of a column of mercury) because that is to the conclusions of the committee chaired by Ferguson
a sensation conveniently related to other sensations. The (1940), who were unable to come to an agreement on
measurement of temperature could have been based whether sensations were measurable, with some mem-
upon time (e.g., the time to boil a certain volume of water) bers upholding the pessimism of the von Kries tradition.
but length was found more convenient for practical pur- Falmagne (1985; 1986) has made an effort to show how the
poses. A measure of a magnitude, then, is a description of psychometric function, the psychophysical law, signal
a relationship between the magnitude and a measuring detection theory, and conjoint measurement theory can
device; a measure of length is a description of a relation- all be related to fundamental propositions in measure-
ship between a length and a ruler. ment theory.
Following the same argument, the measurement of a It is clear, then, that many issues of interest to contem-
sensation might be a statement of a relationship between porary ps_ychophysicists were first raised and discussed
a sensation and an indicator of sensation, which might be a with considerable acumen by their nineteenth-century
stimulus intensity corresponding to that sensation. How- predecessors. The problems of the psychophysical law,
ever, Mach actually wrote, "there can be no question of an the general problem of the measurement of sensation,
actual measurement of the sensations; all that can be done and the problems associated with both neural and experi-
is to characterize them exactly and make an inventory of mental noise were all first approached more than 100
them by numerical means" (Mach 1886/1914, p. 81). On years ago. During the intervening years some questions
the other hand, Mach's recognition that measurement is persisted in an unresolved state and we now turn to an
about relationships between sense-data raised the ques- examination of these,
tion of whether measurements necessarily concerned
"objective" relationships in the spatiotemporal domain.
The story of how Fechner's views gave rise to consider- 2. Contemporary solutions to nineteenth-century
able controversy in the nineteenth century, culminating problems
in the expression of Mach's opinions, which in turn The plan shown in Figure 1 is based on a taxonomy of the
influenced Einstein, is told by Heidelberger (1993a). key questions of psychophysics. Other issues were raised
One Background G
Two A on G 1. Adjust C on G
(A barely discrim- to match A on G
inable) 2. Adjust A to be
just noticeable
A on G Pick out C from 2 Pick out C from n 1. Adjust C on G
(A clearly discrim- alternatives that alternatives that to match A on G
inable) match A match A ("magnitude pro-
duction")
2. Adjust C on G
to be a multiple
or fraction of A
("ratio produc-
tion")
Three A and B on G Adjust A to be
(A on B; A barely just noticeably dif-
discriminable from ferent from B
B)
(A near B; A
barely discrimin-
able from B)
A and B on G 1. Select C on G
(A, B clearly dis- that is "in be-
criminable) tween"
A and B on G
("bisection")
2. Select C and D
on G to match A
and B on G either
directly, or in
terms of an equal
difference, or in
terms of an equal
ratto
Four or more A, B . . . on G Arrange stimuli
(A, B . . . clearly on a scale of
discriminable) equal-appearing
sense distances
Note: Modern psychophysical methods arranged (a) according to how many stimuli there are in the display
(rows) and (b) according to the range of possible responses that can be given (columns). Nonverbal responses
consist of either pointing to a stimulus or adjusting a stimulus directly. Empty cells reflect cases where it is
impracticable to obtain responses of the type desired (e.g., "adjust a stimulus to look like this Ganzfeld") or
where no data have been found using that particular combination of stimuli and responses.
ceived distinctiveness (subjective contrast) and the physi- classification scheme is the following: We can take re-
cal difference between target stimulus and background sponses generated to the stimulus arrays shown in the
will be discussed in the next section. We should also be rows of Tables 3 and 4 and this constitutes a body of factual
reserved in any welcome for the term "sensation dis- information. This same body of information might then be
tinctiveness" as opposed to "sensation intensity" because related on rational grounds to the internal scalable magni-
Schmidt (1991) has shown that loose applications of the tudes of sensation strength or distinctiveness. Engineers
word "distinctiveness" have generated many problems in also make use of the concept of subjective sensation
the theoretical interpretation of memory data. strength (Indow 1989). Whether such an expanded body
Another question raised but not answered by this of information is of more value in a scientific psychology
Table 4. Modern psychophysical methods arranged as in Table 3 but for verbal responses
Note: Verbal responses can range from yes/no judgments, through category judgments where perhaps seven
possible answers are available, to direct estimates where there is no restriction on possible responses other
than the stipulation they may be ordered. The special case where yes/no judgments are also assigned confi-
dence ratings is called "canonical ratings" following Sakitt (1974).
() = neurelectric function; (S) = subjective sensation.
than is the original body of information has been a contro- strength or distinctiveness in the foreseeable future.
versial question. For example, Savage (1970) claimed that Moreover, the question of how measures of sensation
the expanded body of information lacked scientific value strength are related to stimulus variables and other cogni-
because it depended on introspective judgment. My tive variables will continue to be a major one in psycho-
feeling is that if we want a science in which assertions physics.
about individual psychological experiences (e.g., "I don't In the past several years, attempts have been made to
like stimulus X . . .") are to be meaningfully related to tease apart the effects of the physical stimulus / on the
others (". . . because stimulus X feels too intense"), it is subjective sensation S from the effects of the subjective
difficult to see how we could do without the expanded sensation on some other overt response. A stimulus of a
body of information. Roskam (1991) expresses a similar given physical intensity gives rise to a neurelectric func-
view. But in a behaviorist world where subjective experi- tion E [let E =_/(/)], which in turn gives rise to a sensation
ence is considered unimportant, the body of information [let S = g(E)], which in turn gives rise to a nonverbal or
produced from experiments such as those classified in verbal response R [let R = h(S)]. That is, R = h(g(f{I))). As
Tables 3 and 4 would suffice. I think most psychophysi- Shepard (1981) put it, there is no way whereby, from
cists do want to talk about psychological experience and knowing an equation connecting R and /, we can sep-
that therefore responses in psychophysical tasks will con- arately estimate / , g, and h, although through experi-
tinue to be related to conjectures about sensation ments we may try to evaluate these functions in isolation.
ate the far-sighted sophistication of many of Fechner's Using Fechner's equation (Equation 14), the line relating ob-
speculations. tained to predicted data / 3 data had as its equation
obtained data = .997 (predicted data) + 2.24 (r2 = .990)
ACKNOWLEDGMENTS
This work was supported by Social Sciences and Humanities Delboeuf's equation predicted the data slightly better than did
Research Council of Canada Grant 410-89-1040 and a grant from Fechner's equation, but both predictions were remarkably
the Queen's University Advisory Research Council. The author good.
is particularly grateful to Maureen Freedman for research If sensation ratios are being judged, then we expect S3/Sz =
assistance; to Peter Dodwell for a critical reading of the first draft SJSi, from which it follows that S 3 = SySy. US = Klc (Equation
of the paper; and to all the BBS referees, who gave invaluable 10 in text), then K/ = (K/|)2/(K/f), which simplifies to
suggestions for improvements. Requests for reprints should be
/ 3 = antilog [(2Clog/2 - Clog/,)/c] (15)
sent to David J. Murray, Department of Psychology, Queens
University, Kingston, Ontario, Canada K7L 3N6. Unlike the case for Delboeuf's equation or for Fechner's equa-
tion, C and K in this case have to be estimated from data. Taking
NOTES the first entries in Delboeuf's table and setting S = 1 where / =
1. Krueger (1991, p. 197) has raised the question whether 9, S = 2 where / = 47, and S = 3 where / = 232, the obtained
amplitude or power should serve as a measure of the intensity of data can be fitted by the power function
the physical stimulus. Fechner (1860/1964, vol. 2) wrote of an
intensity as being caused by a body of mass m which in a given S = .49Z34
time period moves with a velocity v so that its energy is mo 2 . Let The best-fitting line relating the obtained to the predicted / 3
b be the velocity associated with no sensation (i.e., the threshold data using Equation 15 was
velocity). In our notation, Fechner proved that if S is the
sensation strength and Fechner's Law is correct then obtained data = .98 (predicted data) + 2.58 (r2 = .99)
where it and K are constants and dS and do are differentials. He K/ = 2K/ - K/f
went on to remark that "it is a matter of indifference as to
But this equality only holds if K = 1 and C = 0; otherwise there
whether one uses energy or simply speed in these equations,
will always be a slight difference between the left and right
because in the latter case, one merely doubles the value of the
sides. If a very low exponent is chosen (e.g., C = .01), the
constants k and K" (vol. 2, p. 30). He also noted that the mass of
resulting values of S will be indistinguishable from those pre-
the body did not enter into either equation.
dicted by a logarithmic equation (Krueger 1989a, Figure 1). It
2. Luce and Edwards distinguished between a "sensation
should be noted that K is determined by C and a scaling factor,
jnd" (AS) and a "stimulus jnd" (A/) and defined a Weber function
because if/A is the stimulus intensity corresponding to a sensa-
as a "function relating stimulus to sensation jnd's." If Weber's
tion intensity of 1, then K = l^-
Law is true, then since all sensation jnds are equal by definition,
It is of interest that the exponent for the power function fitting
there is a constant P, so that AS/A/ = P/I. Fechner's "mathe-
Delboeuf's equisection data was .34, which is almost identical
matical auxiliary principle" allowed him to replace AS/A/ by
with an exponent of .33 reported by Teghtsoonian (1971) for
dS/dl - the essence of the principle, according to Luce and
magnitude estimates of the brightness of areas subtending an arc
Edwards, was to claim that "what is true for differences as small
of at least 5. Since Ward (1972) has shown that the exponent for
as jnd's ought also to be true for all smaller differences and so
magnitude estimates is often twice that derived for the same
true in the limit as they approach zero (differentials)" (p. 225).
sensory dimension by categorization methods, it would seem
But Luce and Edwards stressed that for many Weber functions,
unwise to think of equisection methods and categorization
for example (A/// 2 ) = k, AS was not constant for all values of/.
methods as somehow equivalent just because both yield parti-
The Weber function (A///) = k, however, is a case in which the
tion scales.
values of AS do remain equal.
Figure 3 shows that the log / data of Delboeuf can be
3. The question may be asked how well Delboeuf's predic-
transformed by the Naka-Rushton equation to yield S-values
tions, as contrasted with Fechner's predictions, fit Delboeuf's
that fall along a curve that has been found useful for fitting log /
data, and whether a power function can be fitted to those data.
data to neurelectric measures of receptor response (Lipetz
Given Delboeuf's stipulation that, on a scale of sensation S,
1971). However, Delboeuf did not use the extreme values of
lightness that would have better permitted us to evaluate the
goodness of fit of the Naka-Rushton function.
then Fechner's Law would predict that A logarithmic function predicting equal sensation differences,
Clog/3 - Clogi2 = Clog/2 - Clog/, a power function predicting equal sensation ratios, and a trans-
formation yielding a curve similar to one found for receptor
if sensation differences are being judged, or, cancelling out C, potentials could all be applied to Delboeuf's data. This fact
that indicates that curve-fitting alone is unlikely to decide between
psychophysical functions as applied to real data in certain cases.
/ 3 = antilog(21og/2 - log/,) (14)
4. S. S. Stevens was aware of this early research and briefly
An indication of how well an equation predicts the obtained data mentioned it in his 1957 Psychological Review article.
can be given by plotting predicted values against obtained 5. It is perhaps appropriate to mention here that several
values (here, we used Delboeuf's medians); a perfect match will authors have pointed out that a power law can be derived
give a straight line of slope 1, a zero intercept, and r 2 = 1 (r is the without reference to the jnd, including Stevens (1957) and
correlation coefficient relating the predicted to the obtained Yilmaz (1967). For example, according to Yilmaz, if we assume
data). For Delboeuf's equation (Equation 9), the best-fitting that to physical variables i, and t2 there correspond sensory
line relating the obtained data to the predicted Z3 data had as its variables sl and s 2 respectively and that the percept relating s,
own equation and s 2 is given by
obtained data =1.006 (predicted data) + .237 (r2 = .991) (S2/S]) = ji^h) (^)
Nonconscious sensation
and inner psychophysics
Norman H. Anderson
log I log Io
Department of Psychology, University of California, La Jolla, CA
Figure 3. A demonstration that a curve that has the shape of 92093-0109
Electronic mail: nanderson@ucsd.edu
the Naka-Rushton function can be generated from Delboeuf's
data. In the original formulation, which was about the way the 1. Nonconscious sensation can be defined and measured with
receptor potential Vo grew as the logarithmic of the stimulus the functional measurement methodology of integration psy-
intensity (u>) increased, Vo = .5 + .5 tanh (w w0), where w0 chophysics. Integration psychophysics thus involves a way of
was the log stimulus intensity corresponding to the half- thinking rather different from most versions of traditional psy-
saturation intensity, of Vo. Here we have simply relabelled the chophysics, which almost define sensation as conscious.
abscissa showing (w w0) as (log / - log / 0 ) and the Vo ordinate Nonconscious sensation is common. What reaches conscious-
as sensation strength. The abscissa was arranged so that Vo = 1 ness is often, perhaps nearly always, an integrated resultant of
corresponded to log 360, the highest possible value of lightness multiple stimulus determinants. The multiple effects, or sensa-
in Delboeuf's experiment, and Vo = 0 corresponded to a tions, evoked by these separate stimulus determinants will be
Delboeuf value of 0. Following the instructions of Lipetz (1971), inaccessible to consciousness in many cases. Traditional psycho-
a rough curve was drawn through a few data points to estimate physics, having largely ignored the integration problem, has
the value of / (corresponding to w0); then the rest of the data remained largely unconscious of nonconscious sensation. Inte-
points were calculated from this by transforming (log / log / 0 ) gration psychophysics can dissect conscious sensation to deter-
according to the above equation. There are two matters of mine - measure - its determinants, both conscious and
interest. First, Delboeuf did not explore extremely dark light- nonconscious.
nesses or extremely light lightnesses, so there are few data at the One example is the size-weight illusion, in which the effect of
ends of the curve. Second, all the points above the level the visual appearance on felt heaviness is beyond the reach of
corresponding to V() = .8 are data for subjects' responses; a few consciousness. Because the two determinants are integrated by
response data lie below this level, but most of the points below an addition rule, both can be measured on true linear scales with
Vo = .8 reflect lightnesses of stimuli preset by Delboeuf. functional measurement.
These separate sensations, it may be emphasized, can have
substantial effects on the conscious sensation. They are not
nonconscious because they are marginal, at or below "thresh-
where / m e a n s "is a function of," then it can be shown that old," but because they operate at a stage of information process-
s = Bf (17)
ing prior to consciousness.
2. Murray errs in his description of the functional measure-
where B is a constant and x can be specified in terms of/. As ment methodology of the theory of information integration.
Yilmaz points out, "the derivation of this first-order law makes There are three quite separate issues, it is true, but only one is
no use of the concept of the just noticeable difference (jnd) or correctly given. Contrary to what Murray suggests, measure-
resolving power. Nor is Weber's Law required" (p. 534). Equa- ment of the physical stimulus and determination of the psycho-
tion 16 implies that "equal stimulus ratios imply equal sensation physical law have only secondary importance in integration
ratios"; later evidence related to this assertion will be consid- psychophysics.
ered in section 2.2. Luce and Edwards (1958) note that since Instead, the three issues refer to: (a) sensation prior to
jnds are simply cumulated it does not necessarily matter for integration, often nonconscious; (b) the integration function
practical purposes in constructing a scale whether Weber's Law itself, also called the psychological law; and (c) the integrated
is actually valid for the sensory continuum being investigated. resultant, prototypically conscious. All three issues can be
6. Corso (1963) has pointed out that for Fechner (and, later, resolved jointly when the psychological law has an algebraic
Stevens), any change in the point at which the threshold is form. Numerous applications (Anderson 1982; 1990; 1992;
measured (e.g., the point at which the subject detects the McBride & Anderson 1991) demonstrate the empirical effec-
stimulus 50% of the time) will change the zero point on the tiveness of this functional measurement methodology.
psychological continuum that is supposed to reflect the physical The conceptual shift from psychophysical law to psychological
continuum. Some theoretical accounts of the psychometric law is essential to integration psychophysics. The psychological
function, such as Falmagne's (1986), make this point a variable law, being a function of two or more variables, embodies
The head and tail of psychophysical algebra anatomicae et physiologicae in Wagner's Handworterbuch des
Physiologie, Band III. One must emphasize that Fechner's
Robert A. M. Gregson mathematics has all its roots in the eighteenth century, and the
Department of Psychology, Australian National University, Canberra ACT contemporary developments in the mid-nineteenth century,
2601, Australia associated with such workers as Clerk Maxwell, were in a
Electronic mall: rag655@cscgpo.anu.edu.au physics which had moved on (Bell 1939). Late twentieth-
century developments - analytically intractable but computable
Murray's scholarly and timely target article does much to put ones - were obviously not even imagined and yet to come; all
things back into perspective and reinstates nineteenth-century Fechner had was pencil and paper (and presumably log tables).
contributions later either lost or misattributed. He is to be The surprising and indeed powerfully insightful features of
warmly thanked. Herbart's mathematical model are:
There are two ways of looking at psychophysics, however; (i) The process is in real time.
either as the roots of experimental psychology, which Murray (ii) The process is bounded, with limits and with internal
does meticulously, or as applied mathematics. From the latter inhibition as feedback.
viewpoint, Murray's account is like an artist trying to draw a still (iii) Both the output of the process and its time course follow
life of a fish from which, unfortunately, his cat has already from the same model.
chewed off the head and the tail. No matter how accurately he (iv) The concern is with the metric properties of psychologi-
delineates what remains (which admittedly are the guts of the cal variables.
fish) essential fish qualities are lost. Although Herbart modestly (at the beginning) and less mod-
The mathematics of Fechner are not sophisticated, even by estly (in the middle, half way through the algebra) makes
the standards of his time. Precursors had used linear differential reference to his mathematical competence, and although the
equations on psychological topics before him, and he knew of exercise ha.s no experimental data to complement it, the formu-
some. Setting aside some sensation ratio model equations writ- lation is coherent and ties together what later fell apart.
ten in the mid 1700s from Tartu (Dorpat) in Estonia, the best If you look at the literature of contemporary psychophysics
starting point is still Herbart's (1812) paper on the strength of a and its applications in cognitive psychology, you will find dis-
given impression as a function of its observed duration; this was tinct and separate treatments of
obviously well ahead of his contemporaries, both in the detailed (i) the psychometric function, with excessively massaged
consideration of sensory processes and the use of differential data plots of stimulus-response mappings,
equations in model construction. The radical step in the argu- (ii) response latency frequency distributions, with even more
ment, which was apparently dropped from psychophysics for massaging into skew normal or Pearson type III form, and
some time, or rather partialled out to become the separate study (iii) formal treatment of scaling theory, as in the well-known
.of reaction times, comes when Herbart states "Both the abso- Foundations of measurement, vols. 1 to 3, 1971-1990; (see
lute, and the relative strengths are Functions of Time. The word Smith 1991).
'function' is here taken in the mathematical sense, as it means a We know that Euler published some bizarre mathematical
changing magnitude which itself depends on another thing "proofs" that the soul is not a material substance (Bell 1939) and
changing." The great interest for the modern mathematical this could have created the intellectual licence for Herbart to try
psychologist is that Herbart used no stochastic theory; the what he did. But Herbart's attempt to solve his algebra with a
equations antedate the partition into deterministic and proba- series expansion and an approximation to a few terms is Eule-
bilistic parts which Gauss introduced a bit later in his theory of rian. Today we would do it by simulation and numerical
errors. At the same time, many technical devices used by methods on a computer if there were no ready known analytic
Herbart are familiar, and the emphasis on conceptualising solution, but that was not really practical until the 1960s.
processes as extending in real time, and converging on stability, Herbart was 150 years ahead of the hardware he needed. The
is a precursor to aspects of dynamical systems theory that surface important point in the exercise is the date, 1812. (I don't refer to
again with Lyapunov's habilitation thesis in 1892 (Goldner Napoleon's contemporary retreat from Moscow!) Though Gauss
1983). has reworked de Moivre's 1732 analysis of error in 1809, to get
Contemporary mathematical developments in nonlinear sys- what we call a normal distribution, it was only in 1827 that Gauss
tems theory (Wiggins 1990) make it possible to reexamine the used the theory of errors (1821, 1823, Theoria combinationis
fundamental questions which Herbart insightfully addressed; observationum erroribus minimis obnoxiae) to facilitate the
the computational difficulties that blocked him, and perhaps resurveying of the city of Hannover, having developed it for his
dissuaded his successors from exploring the same algebraic astronomical work. The idea that we might treat the psycho-
structures, no longer impede or intimidate us. physical problem as a linear law with superimposed residual and
Obviously Fechner's mathematics were not unique in his own normally distributed error is nowhere in Herbart; the 1812
day. Sprung and Sprung (1987) note that a tradition of "Physik algebra is deterministic but involves something like a closed-
der inneren Erfahrung" tracing from Euler and Herbart to loop recursion.
Drobisch was alive; indeed Drobisch (who is misspelt as Fro- Now, Gauss coedited a geomagnetism journal with Wilhelm
bisch in Adler's translation of Fechner's Elemente) had pub- Weber and collaborated with him up to 1837 (Biihler 1987).
lished two texts before Fechner. They were Empirische Psycho- Wilhelm was a physicist, insofar as any polymath of the day can
logie in 1842 and Grundlehren der mathematischen Psychologie be so labelled, and his brother Ernst Heinrich is the physiologist
in 1850. Fechner was aware of both, and cites Drobisch. who started Fechner on the path to posterity. So the mathemat-
Sometime professor of mathematics at Leipzig, Drobisch (1861) ics available to Fechner came from Gauss through the Webers,
promptly amended the derivation of Fechner's formulae. Kiilpe and uses directly the estimation methods we call least-squares.
(1893) refers explicitly to this tradition; it is in opposition to the But at what price? The Fechnerian approach to psychophysical
Kantian stance denying the possibility of psychological mea- scaling is outside real time, a mistake that Herbart didn't make.
surement which Murray properly associates with von Kries. It is The computations are now tractable, and the implicit model
not even the case that Fechner's extensive treatment of the matches some of the invariances of observable data.
Method of Limits is original; the method goes back at least to Fechner weakened the theory seriously in order to be able to
1827 with Delezenne, and the early evidence for Weber's Law match some data properties to theory. In doing so he gave an
rests on data obtained by that method; indeed the two stand or impetus to psychology as an experimental activity, for which we
fall together. Weber used the method in 1831 in Anotationes should be eternally grateful, but not as a science in the way
(black) interact with each other, leading to the following equa- ours to relate the phenomena of visual sensation and perception
tion for the sensation of a "whiteness' (Hering 1874; 1920): to stimulus intensity and to deduce quantitative rules ("laws")
from the experimental data, one is led to a revised view of the
E = W/(S + W) [units of sensation] (4) discoveries of such laws. The power law of psychophysics,
When only the perceived brightness of a spot of light is consid- usually attributed to Plateau or Stevens, was first described
ered, Hering assumed that Equation 4 could be transformed more than two generations before Plateau by Tobias Mayer in
into a function relating the brightness of the light and the 1754/55. The same is true for the hyperbolic function, which
stimulus luminance /, which would result in the notation used in relates stimulus intensity to perceived brightness (or darkness)
Murray's target article as: and corresponds to the Michaelis-Menten equation. This non-
linear function was first applied to modem neurophysiological
E = a//(l + kl) [units of sensation] (5) data by Naka and Rushton in 1966. The law itself was discovered
by Ewald Hering to interpret his observations in visual psycho-
with a and it as constants. This hyperbolic function indeed gives physics three generations before its application in modern
an s-shaped curve when / is plotted on a logarithmic scale neuroscience.
(Figure 2; Hering 1920). Equation 5 is identical to the formula-
tion of the Michaelis-Menten kinetics and the equation applied
by Naka and Rushton to describe the horizontal cell responses of
the fish retina as a function of the stimulus luminance /. Naka
and Rushton (1974) also demonstrated that for certain stimulus Psychophysics, its history and ontology
conditions an exponent a < 1 has to be added to describe the
experimental data reliably (cf. also Foerster et al. 1977 for cat Horst Gundlach
horizontal cells): Institut tar Geschichte der Neueren Psychologie, UniversitSt Passau, 8390
Passau, Germany
E = al" I (1 + kl") [mV] (6)
Psychophysics started out as one individual's invention. It
Hering (1920) discussed extensively how Equations 4 and 5, changed into a scientific community enterprise nearly two
which should be called Hering's Law (Grusser 1979), depend on decades after its cardinal components were elaborated. It is
different levels of adaptation and discovered that simultaneous therefore appropriate that attention turns back to Fechner
contrast phenomena also follow the same rule. For the spatial whenever the muddle that is psychophysics appears to need
summation in retinal ganglion cells an analogous rule was elucidation. Murray's initiation of a fresh look at the early stages
described (Buttner & Grusser 1968; Grusser et al. 1970), which of psychophysics is, in my opinion, praiseworthy and promising
was generally called "shunting inhibition" in models of neuronal for present-day psychophysics. I assume, however, that his
networks like those applied by Reichardt and others (cf. Furman "perspective for viewing" needs the support of a few more facts,
1965). and that his observations on contemporary psychophysics could
In summary, when one considers the history of early endeav- profit from a critique cutting somewhat deeper.
Murray's historical narrative shows the characteristic of most
narratives of psychophysics - solid anachronism. That is, early
stages are accounted for from the perspective of later concerns.
Yet Murray fares better than most, as he directs attention to the
often-neglected "inner psychophysics." He misses, however,
the true nature of the relationship between inner and outer
psychophysics.
When Fechner had hisflashon Fechner Day 1850, he had in
mind nothing but what he afterwards called inner psycho-
physics. He developed outer psychophysics later as a subordi-
nate, subservient enterprise to corroborate the fundamental
inspiration.
This inspiration consisted in the conception of the universe as
an undivided, non-Cartesian entity. In technical terms, Fech-
ner espoused the metaphysics of a neutral ontological monism,
that is, while rejecting both materialism and spiritualism, he
attempted to unify their positive assumptions. The surmised
fundamental entity nevertheless exhibits two forms of appear-
ance, physical and mental, or, to use another word, spiritual. In
short, a phenomenalist dualism went with the cardinal ontologi-
cal monism. This was, of course, in Fechner's times not a new
vision of the universe, but a rather unoriginal one, usually
termed Spinozism by its opponents.
- 3 - 2 - 1 0 1 2 3 The novelty Fechner introduced was the notion that the two
log 10 stimulus luminance [arbitrary units] appearances of an underlying primary entity were coupled in a
definite way, namely, the numerical values of counterpart
Figure 2 (Grusser). Relationship between the perceived appearances behave as a logarithmic function. This implies that,
brightness of a spot of light seen on a grey background (ordinate) viewed from the other side, they behave as an exponential
as a function of stimulus luminance (abscissa, arbitrary log- function. In simple words, multiplication in the physical realm
arithmic scale). Hering plotted these curves corresponding to is matched with addition in the mental realm. So far, this was
Equations 4 and 5. The two different curves relate to different bare and barren a priori fabrication and reproached as such by
brightness functions at different background luminances lead- Wilhelm Weber.
ing to two different "Eigenfarben" of the background (0.5 and To wit, any part of the primary entity has the faculty of double
0.3 on the black-white scale of the ordinate). On the abscissa a appearance: people, animals, plants, minerals, stars, and so on,
base-ten logarithmic scale is added to the original figure (from not only brains. The mind-brain problem is not at the heart of
Hering 1920, Figure 17). inner psychophysics.
The auxiliary science, outer psychophysics, created to gener- Inner psychophysics, neurelectric function
ate empirical support for Fechner's quaint metaphysics, intro- and perceptual theories
duces a new ingredient: causality. Inner psychophysics assumes
only the relation of identity between matching appearances. Stephen Handel
Outer psychophysics, however, presupposes a causal chain from Department of Psychology, University of Tennessee, Knoxville, TN
outer stimulus to activity modulation at the sense organ to 37996-0900
nervous excitation to central activity. The brain activity caused Electronic mall: handel@utkvx.utk.edu
by stimulation is ontologically identical to sensation, and the
numerical relation between their matching values is expressed What is the appropriate role for psychophysics in a theory of
by the basic psychophysical law. Outer psychophysics needs perception? From one point of view it is pivotal; the perception
identity plus causality. It assumes proportionality between of the complex overlapping coordinate worlds within seeing and
stimulus and brain values, and logarithmicity between brain and hearing can be and will be explained by simple psychophysical
sensation values. functions. From a different point of view, psychophysics is
This is what Fechner tells us outer psychophysics is to be largely irrelevant. Seeing and hearing involve the pickup of
about. This tale fits his desire to substantiate the claims of inner complex invariants within a context, and simple psychophysical
psychophysics. And it has been repeated over and over again, as functions can have little predictive power to explain that pickup.
Murray has now repeated it. Fechner's narrative is a fable, Although a middle ground is possible based on the ability to
however, and one that fails to correspond to what he and other abstract attributes or dimensions of a complex stimulus and
psychophysicists actually do when they turn from reasoning to thereby explain quantitative aspects of the precept (but see
experimenting. This fable has prevented generations from ask- Garner 1974, and Melara & Marks 1990, for constraints on this
ing the obvious, because the answer was seemingly already ability), this has proven to be difficult. I have come to believe
given. The obvious question is: What is the realm of reference of that the usefulness of psychophysics is limited. Any attempt to
outer psychophysics as a research domain? What do statements obtain a unified law will fail because there is as yet no unified
in outer psychophysics refer to? neurelectric function as a consequence of perceptual systems
A negative answer first: They certainly do not refer to sensa- evolving to group organized structures.
tions or other mental phenomena. Consider the weight experi- Many different attributes have been scaled. For example,
ments by E. H. Weber (1834; 1846; 1851; 1978). They are about Gescheider (1984, p. 230) lists 33 attributes along with their
objects, namely, weights. They explore human capacity to judge power law exponent. These are truly a mixed bag. At one end,
differences between weights under specified conditions of expo- there are dimensions in which the magnitudes could be signaled
sure without the assistance of instruments. This is then com- by the rate of firing of a single type of receptor (e.g., brightness,
pared with judgements succoured by precision balances. pressure on palm, and angular acceleration). At the other end,
Consider the case of the so-called extensive sensations. Fech- there are dimensions in which the magnitudes could not be
ner, like others, judges line lengths. He talks about the lines and signaled by any single type of receptor (e.g., visual length or
produces unaided judgements about their lengths. These he area). In the middle, there are attributes that could be signaled
compares with finer judgements produced by using a ruler. The by some neural property, but not necessarily the firing rate of
case is more complicated in sound or light perception. Here the one type of fiber (e.g., sound duration or length of silent
judgements pertain to either objects producing sounds or lights interval). It is the very diversity of these attributes that suggests
(e.g., stars) or to the events produced. This depends on the there is no common psychophysical law in spite of the fact that it
instruction to the subjects. Each is feasible, but in neither case is possible to fit a logarithmic or power function to all of these
do any sensations need to be talked about. As sensations do not attributes.
belong to the realm of reference of outer psychophysics, there is If we look more closely at the possible neurelectric functions
a fortiori no place for sensation strength and no necessity to underlying inner psychophysics, it is clear that transformations
discuss whether sensations are measurable. between intensity and firing rate at the receptor level are only
Empirical work in outer psychophysics, today "psycho- the first step in an intricate set of transformations at intermedi-
physics" for short, assesses the fineness of calibration of various ate nuclei. The neural firing at the cortical sensory areas repre-
senses under various circumstances in determining the values of sents the signal by the pattern of firing across fibers. There is no
diverse attributes of objects, or, depending on closer ontological subset of fibers that faithfully represents the firing pattern at the
analysis, events. The way out of the muddle of psychophysics is receptor level. Instead of considering the perception of loud-
the analysis of what is done in psychophysical experiments and ness that has been extensively reviewed, I would like to use
the construction of a sober ontology freed from Fechner's sound duration as an illustrative example. A recent review of
strange ambition of demonstrating empirically a metaphysical work on frogs (Feng et al. 1990) demonstrates the immense
system of the universe (Gundlach 1992). Plain (outer) psycho- transformations that take place between the auditory periphery
physics simply refers to objects or events and humans (and and the thalamus with respect to the neural representation of
animals). There is no place or need for sensations and similar natural sounds such as mating calls. At the periphery, regardless
ontologically dubious characters. of the temporal characteristics of the sound, all auditory fibers
We must forget Fechner's fable, originating from the desire to will respond to any signal as long as there is sufficient energy
corroborate his metaphysics, that smuggles in the contraband of within that fiber's excitatory frequency range. Nearly all fibers
sensations. Sensations may help in obtaining judgements, but increase their firing rate to longer signals (i.e., high pass).
they would play a similar role in obtaining instrument-aided However, at higher levels, neurons become more sensitive to
judgements, for example in meter reading or scale inspection. temporal characteristics of the signal. In contrast to peripheral
Whatever they might be, they are none of psychophysics's fibers, the majority of neurons in the thalamus respond maxi-
business. mally to signals below a certain duration (i.e., low pass) or to
signals within a certain range of durations (i.e., band pass). In
addition to those neurons coding duration, there are other
groups of neurons that code amplitude modulation irrespective
of amplitude. The discrimination of duration and modulation
becomes based on the patterning of the neurons. The firing
pattern does not mimic the time-intensity pattern of the periph-
eral fibers.
Each sensory system is organized into parallel information
channels for different spectral and temporal attributes. The Fechner's correlational theory of measurement. Fechner
perception of real events is therefore based on the firing pattern (1858; 1860, vol. 2, Ch. 7; 1882, Ch. 26; 1887; for a more detailed
both within and across channels. This is a common strategy for description see Heidelberger 1986; 1993a, Ch. 5) argued that
all sensory systems of vertebrates. Each channel can be innately the measurement of any attribute Q generally presupposes a
tuned for specific values that may signify important environ- second, directly observable attribute R and a measurement
mental events such as conspecific mating calls for the frog and/or apparatus A that can represent variable values of R in correlation
each channel can be experientially tuned by the naturally to values of Q. The correlation is such that when the states of A
occurring environmental events (Kaas 1982; Merzenich et al. are arranged in the order of Q they are also arranged in the order
1984), as might be the case for speech (Kuhl et al. 1992). of R. The different values of R are defined by an intersubjective,
What this means for Fechner's concept of inner psycho- determinate, and repeatable calibration of A. They do not have
physics is that the neurelectric function must be derivative. As to be measured on their part. The function that describes the
Fechner himself believed, in no sense is it a direct measure of correlation between Q and R relative to A (underlying the
peripheral sensory function. The parallel sensory channels pro- measurement of Q by R in A) is precisely what Fechner called
vide the mechanism for the abstraction and judgment of differ- the measurement formula. Normally, we try to construct (or
ent sensory attributes, but that very process of abstraction find) a measurement apparatus which realizes a 1:1 correlation
would appear to change the neural code so radically that a between the values of Q and the values of R so that we can take
straightforward neurelectric function is impossible. What it also the value of R as a direct representation of the value of Q.
means is that the typical experimental situation is unrepresenta- In developing his principles of sensory measurement Fech-
tive of the natural functioning of perceptual systems (Gibson ner several times referred to time measurement as an analogy
1979). Sinusoidal tones presented in isolation are not a "pure" (Fechner 1858, p. 2; 1887, p. 217). His argument can be
signal; the tones are presented in a context, but that context is so reconstructed as follows: The attribute "time' is measured by
fundamentally unrepresentative that the resulting processing the attribute "distance" as realized in a clock or in the movement
may be distorted. It may be for this reason that many instances of the sun. Time values cannot be measured abstractly without
of categorical perception where Weber's Law apparently is using their relation to distance values. Normally, clocks are
violated (Pastore 1987) are found with complex time-varying constructed so that we obtain an invertible 1:1 map from the set
stimuli (Rosen & Howell 1987). of time values {t} that can be possessed by Q into the set of
From this perspective, it is remarkable that the results from distance values {d} that can be possessed by R. If we know the
psychophysical experiments are so consistent. Although the value of some d (the distance covered by a clock's hand) we can
concepts inherent in signal detection theory can be used with pick out directly the value of the associated t (the time value). So
success in physiological investigations of discrimination based in the case of time measurement, the measurement formula that
on firing rate (Kim et al. 1990), the basic issue seems to me to be underlies the construction of a clock is a linear function d = kt,
the way discrimination emerges out of firing patterns across the for a constant k. Note that this measurement formula does not
neurons. The question is not how to dig the signal out of a noisy state that a certain empirical law holds between time and
background or whether a logarithmic or power law is more distance. Rather, it describes how time and distance are related
correct. The similarities in neural organizations for abstracting to each other by the clock that was chosen for time measure-
different attributes coupled with the commonalities in psycho- ment.
physical functions relating intensity to sensation strength for Consider now Fechner's transposition of this reasoning
each attribute argue strongly that there is common neurological (Fechner calls it his "measurement principle") to the problem of
strategy. This strategy in turn yields a common neurelectric measuring sensation. To measure sensations we first have to find
function. The objective is to understand the nature of that an instrument for which we know the measurement formula
strategy and resulting function; then I think it will be possible to involved. Suppose that S refers to sensation strength. The only
ascertain the relevance of Fechner's inner and outer psycho- attribute from which we know that it varies with S is the physical
physics for theories of perception. intensity / of the stimulus. The only device which could possibly
serve as a measurement apparatus relating variable values of/ to
values of S is the living human body. We know from experience
that sensation grows as a function of stimulus intensity such that
S is isotonically related to / for some specified range. (That is, for
Fechner's impact for measurement theory all pairs sm, sn of sensation values there exist correlated stimulus
intensity values im, in so that: if sm is greater than s n , then im is
Michael Heidelberger greater than tn.)
Philosophisches Seminar, University Freiburg, D-7800 Freiburg, Germany
It is obvious that the measurement formula realized by the
Electronic mall: engelhar@sun1.ruf.uni-freiburg.de
human body as a measurement device is much more compli-
None of the presentations of Fechner's original conception of cated than in the case of a constructed clock. So the first problem
the psychophysical law known to me (and Murray's presentation to be solved in order to measure sensation is to find out the
is no exception) give an explanation of the fact that Fechner measurement formula by which the outer stimulus is correlated
called his function a "measurement formula" (Massformet) and with the inner sensation.
not a "law." This might seem an unimportant detail without any For the deduction of his logarithmic measurement formula
further relevance at all. My claim, however, is that the answer to Fechner proceeded as described by Murray in sect. 1.1. There
this innocent question reveals that Fechner's reasoning is based are three empirical cornerstones on which Fechner builds his
on a highly original conception of measurement in general; one argument: (1) the existence of the absolute threshold, (2) the
that is at variance both with the predominant conception of equality of all jnds, and (3) the "fundamental formula" (Murray's
measurement in Fechner's own time and with the most popular Equation 2), as deduced from Weber's Law.
conception of today, the representational/operational theory of So what we call the "psychophysical law" today was first of all a
Suppes and Stevens (Suppes & Zinnes 1963; Stevens 1946; on "measurement formula" for Fechner. Although any measure-
the history of this conception see Newman 1974; Michell 1986). ment formula has important empirical import it does not have
I think that Murray's historical treatment of the psychophysical the status of a law of nature. It is more akin to a definition or to
law must be supplemented by this measurement-theoretical what philosophers call an analytic statement. A natural law,
dimension before we can fully understand the force and scope of however, is a synthetic, factual statement about the way vari-
Fechner's original ideas in the line of research leading to ables that are measured independently of the law itself are
modern measurement theory. related to each other. For each of the variables in a law there
largely reenacts it. This is most apparent in section 1.5, where of the Anomalous or Search for the Soul" and Rao & Palmer:
Murray (in the tradition of many of his psychophysicist col- "The Anomaly Called Psi: Recent Research and Criticism' BBS
leagues) carefully takes note of von Kries's original critique - 10[4] 1987.) Murray, like the scores of researchers he cites,
which discounted even the possibility of psychophysical mea- might well ask: Why would psychophysical research continue to
surement - only to act in subsequent sections of the article as if be done if its assumptions are so problematic? Why, indeed?
this argument had no force. But force it has, even after a "Fechner's whole endeavour" may have "rested on quicksand,"
hundred years, and Murray's dismissal of this critique, known in as von Kries put it (sect. 1.5.3, para. 1), but psychophysicists
the nineteenth century as "the quantity objection," reveals a have not let this fact slow their labors one jnd. Even in the
surprising presentism in his approach. 1940s, when the quantitative status of psychophysical data was
Von Kries argued that numerical values assigned in psycho- the subject of formal challenge (Ferguson et al. 1940), re-
physical judgments are not quantitative. Rather, they are labels searchers persisted. Stevens, following Fechner's lead, help-
for qualitative distinctions, arrayed on what Stevens later called fully redefined measurement in a way that obscured its link to
ordinal scales. Just as relative degrees of loudness can be quantification. If even ordinal scales qualified as a form of
differentiated by terms like "noisy," "piercing," and "deafen- measurement, how could there be any doubt about whether
ing," we can call tones "2," "5," and "9" on a scale of loudness. psychophysical judgments were scientific? The introduction of
But such judgments are subjective and qualitative, and von magnitude estimates as a "direct" way to assess a subject's
Kries felt that using numbers as labels gave a misleading sense of perception was also useful. With the experimenter's own sub-
precision where none really existed. jective judgments now reified as "the standard," psychophysi-
According to Murray, Fechner dismissed von Kries's argu- cists could enter an Alice-in-Wonderland world in which num-
ment because to have taken it seriously would have meant bers meant whatever they wanted them to mean. Perhaps this is
abandoning psychophysics and returning to "a wasteland where why they took to calling certain measures "confusion scales"
psychology had no scientific status and no body of measure- (sect. 1.2, para. 1).
ments to contemplate" (sect. 1.5.3, para. 2). Precisely. A field We can certainly agree with Murray that "responses gener-
which justified its existence on the grounds that it constituted ated to stimulus arrays . . . constitute a body of factual informa-
the first scientific psychology could hardly afford to doubt the tion" (sect. 2.1, para. 4). The question is whether these re-
underlying quantitative foundation of its work. But there was no sponses provide any evidence to support psychophysical "laws."
way to prove von Kries wrong. (If anything, the subsequent Murray and his colleagues keep claiming that they do, and since
century of debate, with its plethora of psychophysical "laws" and science is as much an exercise in persuasion as anything else,
constant bickering about definitions of measurement, may have these repetitions have the effect of making psychophysical data
proved him right.) Fechner's response to von Kries's critique, appear robust. But if, as Murray insists, "what can be said clearly
while not exactly a model of scientific open-mindedness, was can be said briefly" (sect. 3, para. 7), then perhaps it is time it be
nevertheless brilliant sociology. Instead of entering into an said: Psychophysics is to a science of psychological measure-
argument about the substance of von Kries's claims, Fechner ment what the Red Queen's words to Alice were to the truth.
simply pointed to the fact that subjects (in experiments such as
Delboeuf's) seemed perfectly able to perform psychophysical
tasks (sect. 1.5.3, para. 2). This fact is, of course, beside the
point: Subjects will attempt to perform any task they are given,
and their ability to choose among degrees of greyness could
History of psychophysics: Some
easily be seen as just the sort of qualitative judgment von Kries unanswered questions
described.
Later psychophysicists refined Fechner's strategy. One fre- Lester E. Krueger
quent tactic was to create mathematical rules that enable one Department of Psychology, Ohio State University, Columbus, OH
psychophysical scale to be related systematically to another. 43210-1222
Electronic mail: krueger.2@osu.edu
Such transformations: (a) obscure the fact that methods that
ought to yield the same data produce different results, while (b) This commentary will discuss in turn four issues raised by
making it appear as if such differences are merely mathematical Murray's History: (1) contrast effects, (2) psychophysical func-
variants of one another. That the "rule" created to permute one tion, (3) the nature of variability, and (4) the physical scale.
set of values to another is itself mathematical is also helpful, 1. Contrast effects. In order to reveal the true underlying
because it serves to reinforce the basic assumption that percep- psychophysical function, it is necessary to pare away bias and
tual processes are in fact capable of being quantified. context effects not only at the cognitive level, but also at the
Attempts to triangulate perceptual phenomena by using dif- sensory or neural level (Krueger 1991; Wasserman 1991). Inhib-
ferent scales have the effect of masking the uncertainty that is an itory sensory or neural interactions between stimuli are evident
unavoidable part of the psychophysical enterprise. We can in the contrast data Murray cites. He notes (sect. 2.2) that when
understand these attempts as analogous to those of the British greys are viewed against a white background, lightness con-
localizationists that Star (1985; 1986) analyzes. Unable to recon- stancy (which is consistent with both the power function and the
cile findings from animal experiments with data from clinical logarithmic function) is found only for medium greys. Darker
observations of brain-damaged patients, advocates of localiza- greys look darker as illumination is increased, whereas lighter
tion of function relied on whichever set of data best supported a greys look lighter. Murray (sect. 1.3.2) also mentions the asym-
particular hypothesis. This triangulation among different lines metrical effect of contrast: "White areas have a darkening effect
of work both deflected attention from the uncertainties in any on dark areas . . . whereas dark areas may have no effect on
particular set of data, and made all findings seem more robust white areas."
(see Star 1989 for a fuller account). According to Murray, Delboeuf's function, S = log [(c + /) /
Murray tells us to "rest content [as he simply points] to the c], can deal with these contrast effects, whereas "neither a
enormous popularity of a power law in twentieth-century re- power law nor a logarithmic law can predict how all contrasts
search following Stevens's espousal of it" (sect. 1.3.4, para. 3). In will look under increasing, or decreasing, illumination" (sect.
so doing, he joins a well-established tradition in which the 2.2, see also sect. 1.3.3). It does not trouble me, however, that
existence of psychophysical research is taken as a sign of its firm the power and logarithmic functions cannot handle all contrast
foundation. (Such a criterion might please the followers of J. B. effects, because contrast effects represent distortions that must
Rhine, who yearn to have their hundreds of ESP experiments be gotten rid of, not clasped to one's bosom, when devising a
taken seriously as science; see Alcock: "Parapsychology: Science psychophysical law. Dual-stimulus tasks, which introduce new
factors (such as contrast effects) into the perceptual and judg- ministic ones. Chaos theory (Gregson 1991; Krueger 1991)
mental process, should be avoided as far as possible (Krueger shows that even simple processes may produce complex behav-
1989r, "Keep it simple," sect. 2.3), not given center stage, as is ior and that deterministic factors may produce unpredictable
the case with Murray. results, owing to extreme sensitivity to initial conditions. Thus,
2. Psychophyslcal function. Equal stimulus ratios (which de- not all that appears to be random may be truly random
fine lightness constancy) may produce equal subjective differ- (nondeterministic).
ences (Fechner's logarithmic function), equal subjective ratios The Weber fraction, k, in A/ = kl (see Murray's Equation 1),
(Plateau's power function), or neither (Delboeuf's function). is larger in some modalities, indicating increased variability in
Delboeuf's function produces nearly equal subjective differ- observers' behavior. The reciprocal of this factor, Ilk, is a
ences and nearly equal subjective ratios, however, as shown in component of the measure constant, C, for Fechner's log-
Murray's Table 2; and thus it represents a (potentially very arithmic function (see Murray's Equations 2-4). Thus, for Fech-
useful) compromise between a logarithmic function and a power ner, the subjective magnitude, S, ought to rise faster with
function. The power function and logarithmic function are objective magnitude, /, when discriminability on the dimension
related. The power function very closely approximates the is poor (i.e., large k and thus small C). However, as far as I know,
logarithmic function when the power-function exponent ap- Fechner never made anything out of this fact; Murray (sect. 1.1)
proaches zero (see Krueger 1989a, Fig. 1). The Delboeuf func- said that nowhere in his three books "does Fechner give a graph
tion evidently bridges the gap between the power function and or a table showing sensation magnitudes as a logarithmic func-
the logarithmic function. When c = .5, as in Murray's Table 2, tion of stimulus intensity."
the Delboeuf function better approximates the power function, Fechner gave more attention to the absolute threshold, /,
producing more nearly equal subjective ratios than subjective which anchors his scale (see Murray's Equation 4), than to the
differences. As c approaches zero, however, Delboeuf's func- difference threshold or just noticeable difference (jnd), whose
tion better approximates the logarithmic function, as Fechner size determines k and thus C. Thus, it would appear that
(1877) pointed out (target article, sect. 1.3.1). Fechner was not very interested in certain aspects or forms of
The extra degree of freedom provided by c enables Del- variability, perhaps regarding them as due to second-order
boeuf's function to obtain some of the greater expansivity found nuisance factors. This indifference may be well founded. Ac-
in the power function, as compared with the logarithmic func- cording to the proportional jnd theory (Houtsma et al. 1980; Lim
tion, and thus to equal or slightly surpass the logarithmic et al. 1977; Riesz 1933), the subjective size of the jnd depends on
function in predicting Delboeuf's data (see Murray's Note 3). m, the total number ofjnds in a particular modality, and thus the
Because it is estimated from the data rather than based on an \lk component of the measure constant, C, ought to be replaced
independent measure of the "natural light of the retina" (Helm- by l/mk (see Krueger 1989a, sect. 8). If, in addition, k and mare
holtz) or the degree of adaptation or fatigue in the receptor reciprocally related or constrained, that is, ink = In (dynamic
system (Delboeuf), however, c is a free parameter. range) (Norwich 1987), then "resolving power per se may have
Two other intermediate functions (see Krueger 1989a, sect. 1) nothing to do with the psychophysical power function" (Krueger
are the Naka-Rushton function (Lipetz 1971), which Murray 1989a, sect. 9.2, p. 264).
(see his Figure 3) also fitted to Delboeuf's data, and Norwich's Fechner anchors his scale at what is really a statistical fiction;
(1987) information-theoretic entropy reduction measure. No his anchor, the absolute threshold, / 0 , is operationally defined as
doubt there are several other intermediate functions lurking the level of/ producing 50% detection, and thus it is more an
about as well. Thus, I don't see anything particularly special experimenter's "criterion" than an observer's "threshold." In
about Delboeuf's function, and Murray concedes "that curve- fact, owing to variability in the observer, positive sensations
fitting alone is unlikely to decide between psychophysical func- may be experienced well below the absolute threshold, contrary
tions as applied to real data in certain cases" (Note 3). to Fechner's Law, which holds that when / = /, S must equal
Murray nevertheless seems caught up in what I regard zero. Fechner faulted the power law because in its unmodified
(Krueger 1989a, sect. 9.2) as the vain search for "the" psycho- form (target article, sect. 1.3.3), "its assertion that if I were
physical function. The sad fact is that we do not yet know (and nonzero, the subject would necessarily have a sensation S of
may never know) which trends (linear, quadratic, cubic, etc.) of some strength." Given the intrinsic noise of the receptive
the underlying function are truly significant and ought to be system (sect. 2.1), as evident in the "natural light of the retina"
captured by the psychophysical equation. In the interim, scales (Helmholtz), however, there may generally be a positive sensa-
ought to be dealt with on a holistic basis, rather than in terms of tion S even when / is zero, not to speak of when / is above zero
the parameters of a particular function. If Fechner had been less (e.g., / = U
committed to the logarithmic function, and had dealt simply 4. Physical scale. Fechner's lack of concern regarding the size
with summated jnds (as he himself realized was possible; see of the measure constant, C, is shown again in his indifference
target article, sect. 1.2), he would presumably have accepted regarding the choice between energy and velocity for the
the more expansive scale that the near miss to Weber's Law so physical scale. He stated that the measure constant is simply
strongly implies (Krueger 1989a). doubled in the case of velocity (see Fechner quote in Murray's
3. Nature of variability. Although variability is a pillar of "inner Note 1). Most psychophysicists seem to share Fechner's indif-
psychophysics' according to Murray's Figure; 1, it is not clear ference about the choice of physical scale, judging by the scant
from Murray's History whether the founders of psychophysics treatment accorded this issue in Murray's History. It is a serious
had a deep understanding of the nature and meaning of vari- issue nevertheless (Krueger 1991, Note 2; Myers 1982; Weiss
ability. Variability is reducible insofar as it represents the effect 1981; 1989), and may be just as important as the issue of rating
of variation on uncontrolled factors (see Murray's citation of method because the exponent of the power function can be
Boring in sect. 1.4.1). Thus, Stevens et al. (1941) obtained a doubled or halved depending on the choice of physical scale
distinct type of variability - which they associated with the (amplitude, power) as well as on the choice of rating method
operation of discrete neural quanta - after they removed or held (magnitude estimation, category rating).
constant many such factors. Even their remaining variability
might have been eliminated if further controls had been used,
such as locking stimulus presentation to particular brain states.
A key question is whether any portion of the variability in
psychophysics, like the uncertainty associated with quantum
mechanics, is truly irreducible, that is, based on probabilistic
(static) or stochastic (dynamic) "noise" factors rather than deter-
The antecedents of signal detection theory appear to have had no essential sources other than their own
earlier papers (e.g., Neyman & Pearson 1928), except that the
Donald Laming idea of a ratio of likelihoods came from Fisher (1922, p. 326).
Department of Experimental Psychology, University of Cambridge,
These antecedents of signal detection theory are summarised
Cambridge CB2 3EB, England in Figure 1. I emphasise that they are all extraneous to experi-
Electronic mail: drjl@phx.cam.ac.uk mental psychology. In particular, although Thurstone (1927a;
1927b) is referenced by Swets et al. (1961), that reference
Murray's Figure 1 links some of the more important studies in signifies no more than a superficial similarity. It was not a source
psychophysics according to their "commonalities of interest" of ideas. Thurstone did not add a yes/no decision rule to his
rather than any "direct historical interaction. " A casual reader model (target article, sect. 1.4.1) because that particular idea
might expect each column to be addressing a common problem, did not arrive until six years later (Neyman & Pearson 1933). He
each study building on the work of its predecessor; he might did not develop his "law of comparative judgment" into a signal
take the historical span of Murray's diagram to indicate the detection theory because his eyes were set on scaling attitudes
profundity of Fechner's work, shaping the development of and other social entities (Thurstone 1929; Thurstone & Chave
psychophysics for 130 years. But the column headed "Vari- 1929).
ability," to take one example, simply lists studies that have 3. The problem of "variability" -1. Murray's approach to signal
hypothesised about the internal variability of sensory processes, detection theory focusses on its wrapping and entirely passes
irrespective of any relation between them or of where their over its essential contribution to psychophysics. To make this
respective ideas originated from. point Figure 2 reproduces data from Swets et al. (1961, Figure
1. Tracing antecedents. This commentary shows by example 13, Observer 1). The empirical reality of a signal detection
that tracing the sources of the essential ideas in each article leads experiment is a set of data points (here, five of them) exhibiting
to quite a different diagram (see my Figure 1). By "essential," I an essential relation between a, the probability of wrongly
mean those ideas that enable theory tofitdata - which make the reporting a signal when noise alone has been presented (corre-
sums add up - without which there would be no theory. Now, in sponding to the significance level of a statistical test), and B, the
psychology at least, the presentation of theory is usually accom- probability of reporting a signal when one is present (the power
panied by images or examples that, although not themselves of the test). In measuring detectability there are two proba-
part of the theory, provide a vehicle for the reader's thoughts - a bilities to be estimated. If one of them (a) is left uncontrolled, as
kind of wrapping paper for the psychological goods. Signal in most experiments prior to 1954, the data are confounded.
detection theory will provide an instructive example. I am Signal detection theory was a revolutionary idea. It showed
enquiring here about the origin of the goods, not of the wrap- psychophysicists, for the first time, how experiments on detec-
ping paper, no matter how fashionable that might be. tion and discrimination should be designed and analysed, and
1.1. Three principles. I set out three principles for tracing the why. There was, indeed, a profound problem of "variability" (a
origin of ideas: uncontrolled), quite distinct from the problem Murray (sect.
(i) The trace goes backwards in time because references to 1.4.1) has in mind.
other publications are necessarily retrospective. I shall begin 3.7. The problem of "variability" 2. Murray (sect. 1.5.1) exam-
from the principal publications in signal detection theory, not ines the possible relation of the normal distributions of signal
from the sources of the ideas. detection theory to underlying brain activity. That falls into the
(ii) Essential ideas are most easily identified in the earliest category of wrapping paper for this reason:
publications. Later publications (e.g., Swets et al. 1961) are Any theoretical operating characteristic can be succinctly
more elaborately wrapped, with references to other analogous written in parametric form,
ideas included to make the theory easier to assimilate. I shall P("Yes" | Noise) = o(i)
begin my inquiry from Tanner and Swets (1954). P("Yes" | Signal + Noise) = |}(x). (1)
(iii) Attention must be focussed on those ideas actually used in
calculation. Quantitative theories are therefore easier to trace. In terms of a conventional signal detection model, x is the
2. Signal detection theory. Tanner and Swets (1954) list only decision variable and l-ot(x) and l-B(x) are the cumulative
two references. One (Blackwell et al. 1954) concerns only the distribution functions. The point to be emphasised is that
apparatus used in their experiments. The other, a technical Equation 1 continues to describe the same operating charac-
report subsequently published as part of Peterson et al. (1954),
"The theory of signal detectability," is the immediate source of
Likelihood ratio
all the theoretical ideas. (Fisher 1922)
2.1. The theory of signal detectability. Peterson et al. (1954)
addressed the specifically military problem of optimum detec-
Random signals and noise
tion of radar signals. They list 25 references, of which I judge (Rice 1944/45)
three to be essential (to theory rather than practice). Sampling theorem
{Shannon 1949)
Neyman-Pearson Lemma
[Neyman Pearson 1933)
A radar signal is picked up as a continuous waveform over
some interval of time, confounded with background noise.
Some mathematical theory is needed to support a discussion of
the probability attaching to one sample waveform rather than Theory of signal detectability
(Peterson, Birdsall Fox 2954)
another. That theory comes from two sources: the analysis of
random signals and noise by Rice (1944/45) and the sampling
theorem of Shannon (1949).
Once probabilities have been defined for sample waveforms, Signal detection theory
an optimal basis is required for deciding whether or not a true {Tanner & Swets 1954)
(Pieron 1914) with an exponent the absolute value of which is extent, if any, of this ability one is asked to judge which of two
approximately equal to the exponent in Stevens's Law. This objects weighs more. Fechner assumed that each object, when
relates to brightness and loudness (Bonne 1989). lifted, produced sensations of heaviness that were mentally
Equal perceptual delays correspond to equal sensations. One combined. In this way the average of two heavinesses became a
can therefore think that the range of subjective sensation standard against which an individual heaviness was judged
strength (S) evoked by different intensities (/) is equal to the (Fechner 1860/1966, pp. 85-89). In making a judgment of
range of changeable perceptual delays: which of two stimuli weighed more, the heaviness produced by
one object was compared against the average. When an object's
2P _ Tmax _ Smax _ / /max \" heaviness was perceived to be greater than the average, that
r2 Tmin Smin V /min / (1)
object was judged to be the heavier of the two. When less, the
where n is Stevens's exponent and max and min are indices of object was judged to be the lighter.
maximum and minimum values. Fechner, once professor of physics at Leipzig, supposed that
As a sequence, the sensory system evoking the sensation of heaviness obeys the
same law of error that affects any physical measuring device.
n = (2*log(t/r) + Iog2)/log(/max/7min). (2) According to Gauss's (1809; 1821/1880) Law of Error, there
exists a true value of the object to be measured and an obtained
This formula reflects the known inverse relationship between value that is the sum of the true value and a randomly deter-
stimulus range and Stevens's exponent (Sarris 1967; Teghtsoo- mined amount of error. Thus, the sensation of heaviness evoked
nian 1971). The value of the exponent is determined by typical by a constant physical stimulus varies depending upon the
neurophysiological parameters. The maximum range of per- amount of error added to the true heaviness. Whenever the
ceived intensities is limited by the maximum diversity of all lighter of the two weights was judged to be the heavier, or vice
possible neural codes versa, an incorrect judgment occurred that was attributed to the
/max __ tt_ _ i role of error.
/min \r (3) Measuring the size of this invisible sensory error in units of
the visible physical stimuli was one of the most outstanding
where the base is an alphabet span of groups of neural impulses scientific achievements in the nineteenth century. Fechner
and the exponent is the number of such overlapping groups argued that the difference between two stimuli, say weights, is
appearing sequentially during a single cycle of neural activity. measured in grams. These two stimuli also produce a difference
Jarvilehto (1982), working with single tactile receptors, has between sensations measured in units of sensory error. Thus,
shown that the magnitude estimate actually grows in linear the number of sensory error units separating the two sensations
relation to the number of impulses within a single group. Jointly equals the number of grams separating the two weights. Divid-
solving equations (2) and (3) yields n Ss 0.3 which accords with ing the number of sensory error units into the number of grams
the experimental data of many authors. determines the unit of sensory error in grams. Thus the differ-
The right half of formula (3) simultaneously limits the diver- ence between the two physical stimuli is measured with respect
sity of all possible units of perception and memory. It has been to sensory error units, that is, with respect to the resolving
shown that short-term memory span, visual and memory search power of the mind. In this way Fechner and his psychophysics
speed, and latency of simple and choice reaction time depend on exposed to scientific scrutiny differences between sensations
the above-mentioned neurophysiological parameters (Lebedev that were previously hidden from public view and observed only
1990; Lebedev & Myshkin 1989). Hence, psychophysical regu- by personal awareness.
larities closely interrelate with cognitive ones through common The theory of this measurement, based upon derivations
neurophysiological information-processing parameters. The corroborated by the well-known topologist A. F. Mobius (b.
core problem of "sensation strength" underlined by Murray will 1790 - d. 1868), will appear obvious to students of classical
be exhaustively solved if one keeps in mind the neuro- statistics. It is illustrated in Figure 1, where the abscissa repre-
physiological regularities of cyclic brain processes which indeed sents heaviness. On the abscissa are the true heaviness values, a
belong to inner psychophysics.
Distribution of
Fechner's theory of mental measurement Heaviness for S A
RESPOND "HEAVIER"
Stephen Link
Department of Psychology, McMaster University, Hamilton, Ontario,
Canada L8S 4K1 ^^S a Errors \ ^
Electronic mall: link@mcmaster.ca
/
r\ \ \ Distribution of
there are viewers, for example, Scheerer's (1992) view of Fech- >. Heaviness for S_
b
ner's inner psychophysics. From my vantage point there is no ^ ^ Errors
and b, for two weights corresponding to stimuli SA and SB. The deviations from the true heaviness occurred and the measure of
average of these two values (a + b)l2 Fechner referred to as die sensitivity increased.
schwelle, or the threshold. Fechner postulated that when the The measurement of the probable error, based on Herbart's
lighter of the two weights SA was judged to be the heavier of the and Weber's idea of a threshold, allowed psychological scientists
two, it was because the true heaviness of the lighter weight had to determine whether one judge could discriminate a smaller
added to it enough error to make its apparent heaviness greater difference between two stimuli than another, or whether, for
than the average heaviness of the two weights. Because the the same judge at different times, the size of the physical
lighter weight was perceived to be heavier than the average, it difference needed for 75% correct discrimination varied. In this
was judged, incorrectly, to be the heavier. way Fechner showed how to investigate individual differences
By hypothesis, when stimulus SA was lifted the sensory and also how to examine variations within an individual to such
system added a Gaussian error having mean value zero to the causes as fatigue, time of day, and order of stimulus presen-
true heaviness, a. When an error occurred, the heaviness of SA tation.
necessarily exceeded the rerjrent located at (a + b)l2. Because The major point of Fechner's psychophysics is that the wholly
(a + b)l2 also equals a + (o - a)/2, Fechner reasoned that the invisible error of the sensory measuring device was made
amount of error was larger than (b a)/2. Therefore, the measurable through an extension of Gauss's Theory of Error.
probability of an error ofjudgment must equal the probability of The theory was later recreated by engineers who were entirely
errors larger than (b a)l2. The shaded area under the Gaussian unaware of Fechner's contribution to the founding of psychol-
probability distribution to the right of the midpoint, (a + b)/2, ogy. This new rendition is a form of signal detection theory
represents the probability that SA has sufficient error added to known as ideal observer theory.
its true heaviness to exceed the average heaviness of the two Somehow this quite stunning idea became lost to psychology.
weights and produce an error of judgment. Many students trained by Wundt in Leipzig took up residence
When the larger weight, SB, was judged to be the lighter of in the United States. Yet their textbooks did not present
the two, the heaviness was judged to be smaller than (o + b)l2, Fechner's theory. Perhaps the loss of Fechner's theory is a
which equals b (b o)/2. The total heaviness was less than b consequence of the loss of many young European students in
by an amount at least as big as (b d)/2. Thus the error was of two World Wars, compounded by the zealous grip on experi-
the same magnitude for SB as for SA. Due to the symmetry of the mental psychology of the antiquantitative, behavioristic forces
Gaussian distribution and the use of the midpoint as a threshold of the twentieth century. The antiquantitative bias is revealed in
for judgment, the probability of an error in judging SB equals a telling comment found in the introduction to Experimental
the probability of an error in judging SA. psychology: A manual of laboratory practice by E. B. Titchener
To estimate the probability of an error, Fechner determined (1901), scion of Wundt's line in American experimental psychol-
the relative frequency of an error by computing the ratio of the ogy: "Knowledge of elementary mathematics is a part of a man's
observed number of such errors divided by the total number of general scientific outfit: but one may work a lifetime, and with
judgments where such errors could occur. Therefore, an error success, in psychology, without needing the knowledge" (p.
larger than (b a)l2 occurred with a probability that was xxiii).
determined through mental judgments - self-reports about the Fechner's great insight did not entirely disappear, but the
status of a sensation. Fechner suggested that by repeated band of psychological scientists dedicated to the measurement
stimulus presentations enough judgments could be gathered to of psychological phenomena grew smaller as the statisti-
provide a statistically reliable estimate of this unknown proba- cal/mathematical models of additive effects came to dominate
bility. For this reason the probability of an error in judgment many psychologists' methodology. The story of the theorists and
became an important element of the theory of measurement. experimentalists who continued beyond Fechner's pioneering
For Fechner, estimating the probability of an error of judg- work is told in Part I of Link (1992, "Psychophysical origins of
ment was only a step along the path toward mental measure- difference and similarity"). Part II presents the theory that
generates feeling from sensation and resolves many previous
ment. His major concern was to measure the sensitivity of the
psychophysical paradoxes by showing how Weber's constant,
sensory system in much the same way as determining the Fechner's Law, and Stevens's exponents are derived and
sensitivity for any physical measuring device. In the mid- related.
nineteenth century the statistical quantity called the probable
error served this purpose. The probable error equals the dis-
tance from the mean of a Gaussian probability distribution to a
point that yields an area under the Gaussian distribution equal A parallel view of the history
to 0.25. Plus or minus one probable error from the mean defines
the interquartile range containing 50% of the Gaussian proba- of psychophysics
bility distribution. Beyond one probable error above the mean
Gregory R. Lockhead
is 25% of the distribution which corresponds to large positive
Department of Experimental Psychology, Duke University, Durham, NC
deviations from the mean. Another 25% of the distribution
27706
consists of large negative deviations that exceed one probable
Electronic mail: dgreg@dukemvs.bitnet
error below the mean.
Suppose two weights of 300 g and 318 g are correctly discrimi- Murray has outlined a history of psychophysics in terms of
nated with probability 0.75. The mean heavinesses of these Fechner's "outer" and "inner" psychophysics. To make that
stimuli are shown as points a and b on the abscissa of Figure 1. In history more complete, I would add a column to his Figure 1 and
terms of decision making, the average heaviness is a decision entitle it "Contextual Psychophysics," to reflect Egon
criterion that lies midway between the heavinesses correspond- Brunswik's history of psychophysics (1956, Figure 7, p. 32).
ing to the two stimuli. The average heaviness corresponds to an Like Murray, Brunswik began his history of psychophysics
interpolated hypothetical stimulus of 309 g. The probable error with the classical approaches of the early and middle nineteenth
equals 9 g because 25% of the probability distribution of heav- century. Brunswik's history described relations between exper-
inesses exists between 300 g and 309 g. imental design and data interpretation. As an example of the
To measure the sensitivity of the sensory system, Fechner beginnings of psychophysics, Brunswik reviewed studies of the
used the reciprocal of the probable error. As the probable error Galton Bar that provided threshold measures in terms of differ-
increased, and larger errors became more likely to be added to ence limens and PSEs, examined generalizations of the Weber-
the true heaviness, this reciprocal decreased, indicating a re- Fechner law, and introduced new statistical measures.
duction in sensitivity. For a small probable error, few large Brunswik called this "structuralism." It was alleged then by
investigators that individual stimulus variables could be isolated Brunswik suggested using the modern procedure plus the
and measured. "omission technique' (Brunswik, p. 25) of gradually eliminating
It gradually became clear that factors other than the variable cues as information accumulates. Because we do not know a
of interest determine the perception of that variable. To make priori what determines judgment, this method allows poten-
these demonstrations, that attribute was systematically isolated tially essential cues to be preserved and thus possibly to be
from neighboring variables. As examples, Brunswik reviewed discovered.
studies of the Miiller-Lyer figure which showed that the appar- Classical experimentalists bypass this process because they
ent length of a line depends on what lines surround it. Such already "know" that attribute intensity determines perception.
research was an important part of the beginnings of Gestalt When the assumption is correct, this is the more efficient
psychology. method. When it is wrong, one might expect irregular data,
It further came to be understood that such multidimensional so the erroneous idea would quickly be rejected. Unfortunately,
determiners of perception were related to thing-constancies. the data are sometimes confounded with what actually causes
This was shown in studies that systematically isolated distal cues the performance so that face validity gives encouraging results
from other mediating cues. The results are that a distal stimulus with no motive to look further.
appears constant even though the proximal stimulus varies Consider typical brightness studies. It is not possible to vary
widely. This fact requires relaxing the idea that the intensity of only the intensity of a spot. Then the difference between the
an attribute at the receptors determines its perception and that spot and its surround also changes. Furthermore, brightness is
the way to measure appearances is to hold constant everything determined by this difference and by the overall illumination,
in the situation except that attribute. not by spot intensity (Lockhead 1992). Using only the classic
Brunswik next extended his analysis to emotions and attitudes approach and assumptions, this remains unknown because
and reported studies showing how these are also integral to there is no apparent need to seek a causal factor beyond spot
judgment. His consistent emphasis was on the biological value intensity.
of the situation rather than on the intensity of an attribute at the Because Brunswik recognized this as a potentially general
receptors. difficulty, his history of psychophysics applauds what he saw as
This does not mean that his approach could not determine progress in moving from the classic view of independent attri-
thresholds or that he considered such measures unimportant. butes to a Gestalt psychology and then to a functional approach
Neither would be true. But it does mean that Fechner-Iike laws where constancy is the important fact and the study of percep-
cannot be correct beyond being points in a complex function tion calls for an overt, distal, and multidimensional psycho-
where performance depends on many factors which sometimes physics based on the idea that what is most important is "the
interact in complex ways (Lockhead 1992). relationship between ecological validity and utilization"
Not everyone was persuaded by Brunswik and separate (p. 141).
literatures resulted. As examples, Brunswik did not reference
Delboeuf or Plateau, who are important to Murray's review, and
Murray did not reference Brunswik or most of the other re-
searchers who are important to Brunswik's review. Let's not promulgate either Fechner's
In addition to different methods, different dependent vari-
ables are also sometimes used. In a sample "outer psycho-
erroneous algorithm or his
physics" study, the brightness of a disc is measured when its unidimensional approach
intensity is varied and the surround is held constant. The result
is that disc brightness increases with 'intensity. In a sample
R. Duncan Luce
"inner psychophysics' study, as reviewed by Murray, the over- Department of Cognitive Sciences and Institute for Mathematical
Behavioral Sciences, University of California, Irvine, CA 92717
all illumination of the entire display is varied and the perceived
Electronic mail: rdluce@uci.bitnet
contrast between the disc and surround is evaluated. Brightness
is not reported here, and perceived contrast is not noted in disc Murray several times cites Luce and Edwards's (1958) critique
brightness studies. of Fechner's derivation of Fechner's Law from Weber's Law
The constancy literature shows that disc brightness changes (sects. 1.1, 1.2, and Note 2). Unfortunately, he seems not to
little if at all when the overall illumination changes a lot, even have absorbed our message and he makes the same mathemati-
though disc intensity changes a lot (Brunswik 1956). These cal error where it matters some. Fechner correctly stated
conflicting conclusions - brightness increases with intensity Equation 2, AS(I)/A7 = Cll, as the functional equation embody-
(when disc intensity increases) and brightness is independent of ing both his hypothesis of equal subjective jnds and the assump-
intensity (when disc and surround intensity both increase) - tion of Weber's Law. To solve for S(/), he proposed the algo-
argue against the classical assumption that individual variables rithm - calling it a "mathematical auxiliary principle" - of
can be isolated and measured. Something beyond the models replacing AS(/)/AZ by the differential dS(l)/dl. Thus, Equation 2
reviewed by Murray is required for completion. is transformed into a (well-understood) differential equation
Because of such considerations, Brunswik concluded that our whose solution is Equation 4. Our point was that while this
understanding of perception must move away from the classical algorithm happens on a correct solution when either Weber's
considerations, which examined effects of the proximal stim- Law or the generalization implicit in Equation 5 holds, it fails to
ulus, to considerations of constancy, which examined relations arrive at the correct solution for any other Weber function.
between perception and the distal stimulus. This was a move Falmagne (1985) provides a good discussion of a correct
from a structuralism that emphasizes effects of stimulus inten- algorithm.
sity at receptors to a functionalism that emphasizes the biolog- This may seem a bit of esoterica, but the issue reappears in a
ical importance of perception. more significant way in section 1.3.3 where Murray notes that
In relating method to theory, Brunswik contrasted proce- Fechner later considered a reformulation (see Equation 10) of
dures used by classical (mid-nineteenth century) researchers the initial problem. To "solve" Equation 10 using Fechner's
who were interested in direct intensity effects with procedures algorithm, rewrite it as AS(/)/A7 = CS(1)/1. Replace the ratio of
used by then modern (1890s) experimentalists who were inter- differences by a differential to get dS(l)/dI = CS(I)/I. This is
ested in appearances. The classical methods held everything equivalent to d[\og S(I)]/dI = Cd(log l)ldl. Integrating and
constant except the attribute of interest. The "moderns,' at- taking exponentials yields Equation 11. Murray appears to
tempting to mimic the ordinary world, held essentially nothing accept this not only as historically accurate, but also as mathe-
constant. matically correct. About the latter he is wrong. One disconcert-
ing feature of the "solution" is that it is independent of the form respond. Here the models are much more process oriented and
of the Weber function. Another worse feature is that Equation have little in common with the measurement-theoretic models.
11 simply does not satisfy Equation 10 except when C = 1. To The measures of sensation are parameters of the model and are
show this, let/' = / + A/, andsoAS(/) = S(J') - S(I). Assuming S estimated from behavioral data. Again, trade-offs are the name
is given by Equation 11, of the game, in these cases often between two or more sensory
variables such as signal duration and signal intensity but also
=m _1= [AI
1
I* _ between a sensory variable and some sort of motivational crite-
rion. The models simultaneously develop measurement and
s(i) sy)
L/ J theory. Summaries of various of these models can be found in
which agrees with Equation 10 when and only when C = 1. Luce (1986) and Townsend and Ashby (1983).
Considering that 34 years have passed since it was noted that My only point is that one is probably misguided to continue to
substituting a differential for a ratio of differences in general fuss at the one-dimensional measurement case; unless a relevant
leads to an incorrect solution, it is sad that this not very subtle concatenation operation or some other rich internal structure
mathematical issue continues to mislead. can be found, the situation is simply too underdetermined to be
A conceptual issue permeates Murray's target article. Is it of much theoretical interest.
plausible to expect successful measurement of an attribute when
just one independent variable is manipulated? To my knowl- NOTE
edge, the only case where this approach has succeeded is for the 1. At the time, the only cases that were understood involved opera-
extensive measures of physics, of which mass, length, charge, tions with additive representations. Later work (see Chs. 19 and 20 of
and time are prime examples. Such dimensions include a Luce et al., 1990, for a summary) yields a whole family of inherently
concatenation operation of combining two entities that each nonadditive operations. Homogeneity, described in Luce and Narens
(1987), underlies this result; this paper, which is purely expository,
exhibit the attribute in question to form a third entity that also gives references to the original contributions.
exhibits the attribute. Classically, such operations were repre-
sented numerically as addition, although that choice is highly
conventional - multiplication is equally good, as are a contin-
uum of other (associative and commutative) mathematical oper-
ations. No other purely one-dimensional example of fundamen- Quantifying, valuing, choosing
tal measurement is known, and that was the reason N. R.
Lawrence E. Marks
Campbell (1920/1957) and the Ferguson (1940) committee,
largely a creature of Campbell's concerns and on which Camp- John S. Pierce Laboratory and Yale University, New Haven, CT 06519
Electronic mail: marks@yalevm.bitnet
bell played a major role, concluded that psychology was incapa-
ble of fundamental measurement: It has no empirical concatena- In his thoughtful historical review, Murray recounts several
tion operations of its own. issues that were raised to prominence by Fechner and others in
If one accepts that measurement is a one-dimensional matter, the nineteenth century and that remain topical today. Of all
the committee was right. x For a careful contemporary treatment these issues, perhaps none is more central to the enterprise of
of the one-dimensional approach and its ambiguities, see psychophysics than the matter of quantification. In a nutshell,
Narens (submitted). Falmagne (personal communication) points Murray's target article poses once again those thorny questions
out that if one is willing to deal with choice probabilities rather that have bedeviled psychophysicists: Can sensory/perceptual
than orderings, it is possible in principle to construct a binary magnitudes be measured (quantified)? If so, how? And if not,
operation over the probability space. The difficulty is that why not?
because most choice probabilities are 0 or 1, one is forced to It may seem to those who are familiar with psychophysical
piece together the global scale from highly local data. scaling, with its history and its ongoing controversies, that the
What the Ferguson committee failed to acknowledge, and matter at hand constitutes another example of the proverbial
many psychophysical sealers seem to continue to ignore, is that dead horse. Today, as in Fechner's time more than a century
something very like one-dimensional measurement becomes ago, some in the scientific tribe eschew the measurement or
feasible when two or more independent variables affect the deny the measurability of perceptual magnitudes. And this
same attribute. One can use the resulting trade-off between the stance may seem justified, for, unfortunately, psychophysics has
independent variables as a source of measurement of the attri- failed so far to come up with any "measurements" with sufficient
bute and how the factors combine. Indeed, trade-offs typically theoretical underpinning to allay skepticism and engage the
induce mathematical concatenation operations on the compo- scientific community at large into accepting them. Even so, the
nents. This was completely familiar in physical measurement of matter of quantification does not go away.
such quantities as momentum, energy, density, and so on, but it Nor should it. Quantification of the mental is significant not
had not been axiomatized in a fashion analogous to the turn-of- just in the domain of sensory-perceptual processing but
the-century axiomatizations of extensive measurement. The throughout psychology: in what Kurt Lewin (1936) called "val-
lacuna was corrected in the early 1960s (see Chapter 6 of Krantz ence," in what economists and decision-theorists call "utility," in
et al. 1971 and Chapter 19 of Luce et al. 1990 for historical what I'll loosely refer to as "valuation." In brief, valuation is the
details). domain of objects of desire; of the goals underlying decision-
Thoroughgoing behavioral examples of the trade-off approach making, including moral acts and judgments; and, ultimately, of
are Luce's (1977) axiomatization of power and logarithmic func- happiness and unhappiness - all of which have quantitative
tions when the data are orderings of stimulus pairs, such as aspects.
loudness to binaural tones; the functional measurement proce- The treatment of mental qualities as quantities - sometimes
dures advocated and applied to psychophysical as well as other explicitly, often implicitly - did not appear de novo in the
psychological problems by Anderson (1981; 1982); and the nineteenth century, but has much older and deeper roots.
entire complex literature on axiomatizations of preferences Plato, for example, calculated a number to represent the ratio
and/or judgments of uncertain alternatives. In each case, one between the happiness of an ethical life and an unethical one;
uses the trade-off between variables to establish simultaneously specifically, he suggested (Republic, IX, 587) that a just ruler
the measures involved and the law relating them. (king) is 729 times as happy as a tyrant (the number three
If axiomatics are not to one's taste, another approach is to squared, or nine, raised to the third power): "And if a per-
increase the dependent variables from just choice and/or judg- son . . . tells the measure of the interval which separates the
ments about psychophysical stimuli to include the time taken to king from the tyrant in truth of pleasure, he will find him, when
in psychophysics. Yet he accepts this as representing the "slip- leagues seemed to imply a simple linearity between magnitude
pery ground" that is avoided through the study of "empirical estimates and sensation strength, it has been convincingly
questions," as these represent a "terra firma." This is a false argued by many commentators that this is simply an assump-
geology. tion, with little or no supporting evidence. Critics of magnitude
estimation have rightly pointed out the inconsistent and even
idiosyncratic ways in which subjects use numbers, but even an
"ideal" subject able to use numbers on a ratio scale consistently
would leave uncertainty about how those numbers relate to
Sensation strength: Another point of view sensation magnitude. But is a solution to this puzzle really
needed?
Robert Teghtsoonian
The method of cross-modal matching (CMM), mentioned
Department of Psychology, Smith College, Northampton, MA 01063
only briefly by Murray (and apparently omitted from Table 3),
Electronic mall: rtex@smith.bitnet
seems to me to have a unique place in the context of debate
Murray deserves much praise for his wide-ranging analysis of about the form of the psychophysical law. (Because matches that
the nineteenth-century antecedents of contemporary issues in involve numbers so often involve assumed relations between
psychophysics; we owe him a debt of gratitude for helping us those numbers and sensation strength, it will simplify the
better understand the important work done by the pioneers in argument to exclude such procedures from this treatment of
our field. Of the many interesting topics he has raised I have CMM.) Since both the target stimulus intensity (set by the
selected just two for comment - Delboeuf's classic study of experimenter) and the matching response stimulus intensity
judged greyness, and the status of the problem of measuring (selected on another continuum by the subject) are physical
sensation strength. values that involve no assumptions other than that of subjective
Murray notes that Delboeuf compared the results of his equality, the result certainly qualifies as an answer to Fechner's
version of the psychophysical law with the obtained data and question about "outer psychophysics," and for CM M the answer
took comfort from the good agreement. In an extended footnote has been repeatedly shown to be a power law. Indeed, it is
Murray shows that Delboeuf had little grounds for pleasure with notable that (so far as I know) no one has ever suggested that the
these data since it can readily be shown that they are compatible power law is found for CMM only because the experimenter
both with Fechner's version of a logarithmic law and with a induced a particular bias in his subject. Nor (so far as I know) has
power law (for which Plateau was the contemporary advocate). it ever been reported that some subjects in CMM experiments
The point I wish to make about Murray's analysis is that it is not show power functions while others show logarithmic functions.
really necessary to undertake the calculations made by Del- The results are clear, consistent, and make possible a systematic
boeuf or by Murray to show that the data are consistent with description of a very large number of perceptual continua in
either a logarithmic or a power law, and that this dual con- terms of the operating characteristic relating perceived magni-
sistency can be understood in terms of the uncertainty about tudes under the instruction of subjective matching.
what it is that the subject is actually doing. Why, it may be asked, is this important discovery given such
The important feature of all 13 sets of values for/,, J2, and/ 3 is short shrift not only by Murray but by many others concerned
that they define geometric progressions. Were it not for the with psychophysical scaling? I think the answer lies in the fact
fact that Delboeuf was concerned to distinguish his own model that CMM data make no claims to measure sensation strengths.
from that of Fechner, he could have tested the adequacy of a There is nothing in such data to address the ancient questions
simple logarithmic law by using the ratio / 2 //i as a multiplier for about ratios or differences among sensations, and it is probably
72 to obtain a predicted value for 73. (In fact, such "predictions" for that reason that they are of such little interest to those who
are in excellent agreement with Delboeuf's data, perhaps better continue to see the main problem in Fechner's terms, that is,
than the fit to his own model.) It only remains to argue that the the relation between stimulus intensity and sensation strength.
three corresponding sensations define equal intervals on a scale But after nearly a century and a half in which it has been
of sensation strength to allow the conclusion that there is a impossible to agree on defining operations for "sensation
logarithmic relation between sensation and degree of greyness. strength" it may be time to recognize that the goal of measuring
The critical question, of course, is whether one can indeed it is a will-o'-the-wisp.
assume equal intervals on a sensation scale under the instruction What seems to me important about CMM is that although it
to make 72 appear "intermediate" between 7, and 73. As Murray makes no assumption stronger than subjective equality, it pro-
notes, a number of writers have drawn attention to the fact that vides data that tell so much about the nature of so many
there need be no simple or universal relation between what the perceptual continua: The virtually exhaustive matching that
subject is told to do and what he in fact does. Assuming that the obtains for most (if not all) pairs of continua; the fact that the
subject in Table 1 of the target article adjusts 73 so that S3IS2 = matching functions generally follow a power function; the fact
S2/Slt that is, creates a geometric progression of sensations, his that there is transitivity for the exponents of those functions; and
results can be said to conform with a power law. (It should be the likely fact that those exponents are inversely related to the
noted that this line of argument is rather different from that dynamic ranges for the corresponding continua (Teghtsoonian
proposed by Murray in Note 3 where he shows that a power law 1971). Furthermore (pace Laming 1989), if there is cross-modal
fits Delboeuf's data even if it is assumed that the sensations in subjective equality for jnd, and the jnd is even roughly approxi-
question define equal intervals; that is the familiar demonstra- mated by Weber's Law, then power law exponents must be
tion that, with a low exponent, a power relation is difficult to systematically related to measures of resolving power (Teght-
distinguish from one that is logarithmic.) soonian 1974). This is a great deal to know, or even to speculate
The problem is: How are we to know the relations among S,, about, in a completely testable context, without ever asking
S2, and S 3 when these are not directly observable? This brings about the relation among sensation strengths.
me to the second issue on which I wish to comment, and that is But if this last concept has become so compelling that few are
the difficulty of working with the theoretical construct "sensa- willing to give it up, it remains simply a matter for the Bureau of
tion strength" when there is no agreement about the defining Standards to convene a committee whose task would be to
operations. Although everyone accepts that stimulus intensity announce a standard reference continuum, be it sound pressure
constitutes a major defining antecedent, there has been contin- level, luminance, or our old friend the number scale, and that
uing dispute about an appropriate outcome measure and would provide the operational definition of sensation strength.
whether it is possible to specify the relation of any outcome So far there seems to be no experimental or theoretical solution
measure to sensation strength. Although Stevens and his col- that seems preferable to such a resolution by fiat. Although it is
on an arbitrarily chosen scaling convention. We can scale the good at using our senses to make guesses at the objective
mechanisms determining our perception of constant contrast so properties of the world. For example, he argues that brightness
as to give either answer, just as the intensity of a tone can be tends to be judged in terms of proportions (ratios), because it is
measured in decibels or power units. Within the scope of this proportional to background illumination, changing atmospheric
kind of experiment, the information does not exist to give one or absorption (in the case of stars), the size of the pupils, and so on.
other scale a special status (Treisman 1961; 1962; 1963a; 1963b; Nevertheless, von Kries is unsympathetic to psychophysics in
1964a; 1964c; 1985; 1989). the sense that sensations themselves cannot be measured in the
same way as physical length, time, and mass. However, the
significance of the difference (in particular, its negative appeal)
only makes sense if the priority of physics is taken for granted. It
is here that Maeh offers an alternative. (I am not endorsing this
The analysis of sensations as the foundation alternative - just suggesting that it might throw light on what
of all sciences the purpose of psychophysics could be.) His view can be
summarized as follows:
J. van Brakel (1) No physical (or any other) measurement is possible with-
Department of Philosophy, University of Utrecht, 3508 TC Utrecht, out measurement of sensations. The measurement of sensations
Netherlands provides the basis for the unity of knowledge, grounded in the
Electronic mail: brakel@phil.ruu.nl
manifest (daily, folk, common sense) experience of the world.
In 1932 a committee of the British Association for the Advance- There is therefore no ground to worry about psychophysics
ment of Science was appointed to report on "the possibility of losing out on physics.
Quantitative Estimates of Sensory Events." Eight years later (2) The sense of measurement intended, however, is that of
(Ferguson 1940): ordinal measurement. Interval or ratio scales for sensations (or
The Committee . . . found it possible . . . to state their conclusions judgments for that matter), as well as correlations that look like
in terms that all of them could accept; but in doing so they . . . had to psychophysical laws, are artefacts of data-gathering procedures
ignore the wording of their terms of reference. . . . After examining or the result of entrenched conventions.
[the various] points of view [concerning the possibility, etc.] the How should we understand Mach's view that the most basic
Committee [were] satisfied that no practicable amount of discussion form of measurement is the measurement of sensations?
would enable them to express an agreed opinion. . . . (Though not a literal quote, I think it captures Mach's spirit.)
The same might be said of the discussion 50 years earlier The first thing to do is to forget about the difference between the
between Fechner and von Kries. And again the same could be things mentioned in the left and right column of the table below:
said of the commentaries on Krueger (1989a) and on Murray's
target article. sense perception of time clock time
My comments on Murray's target article will be restricted to sense of space (seen length, area, volume) length, area, volume
felt force/pressure/pull/push physical force
section 1.5.3 and will deal with the views of von Kries and Mach. felt warmth temperature
I will suggest that Mach's views might provide a "novel" per- (musical) pitch frequency (of longitudinal waves)
spective on psychophysics. experience of brightness luminosity, quantity of light
Von Kries's (1882) argument is that on a practical level
Fechner's jnd scales (or any other quantitative scale for sensa- According to Mach (in the interpretation I'm offering) the true
tions) may be useful for some purpose or other, but on a nature of the world consists of elements that correspond to the
theoretical level the idea of a sensation scale in the same sense as rows in the table above. But there's only one column. Making
a scale for length, mass, and time is meaningless, because the the distinction between, say, felt warmth and temperature is a
very idea of equality of sensation difference (at different points later (human) addition. Having made the distinction it turns out
along the scale) is empty: "die Gleichsetzung zweier ver- that the two corresponding entities in any row are in ordinal
schiedener Empfindungszuwuchse . . . hat gar keinen Sinn" correspondence. This fact is the basis of all measurement. If
(p. 274). For any particular type of sensation, not only does a jnd there were no prior ordinal experience (on the level of the
depend on "surrounding" sensations, the part of the sense organ elements), there would be no metrical science. This ordinal
used, the history of exposure, and so on, but the whole idea of space of (pre)sensation elements is, as it were, the metaphysics
looking for units of sensations is meaningless. That people are on which both psychology and physics are built (cf. Bradley
often good at "producing' ratio or interval scales in experiments, 1971, p. 68). Failure of transitivity and so on should be under-
just means that they have been trained through survival or stood in terms of more than one row being actual - an experi-
otherwise to be good measurers of physical magnitudes. (I have ment "about one row" always abstracts, if only from a
been told that there are people who are extremely good at background.
estimating the average water content of large piles of fresh When the physical column has been constructed, empirical
natural rubber. Do they have "average water content of large laws can be found that warrant the use of ratio scales. This is not
piles of fresh natural rubber-sensations"?) the case for the sensation column (here Mach agrees with von
Basically, von Kries's message is: Use any scale that serves Kries). However, it's easy to overestimate, as von Kries does,
your needs, but don't think there's a reality behind it (such as the metaphysical significance of this (as distinct from the practi-
the true metric structure of sensation space). Note that this gives cal relevance). Mach's advice to psychophysics is, I suggest:
a somewhat different emphasis from Murray's account. He Analyse sensations - it's the most fundamental science. But
seems to suggest that if von Kries is right, there's no room for don't worry about scaling (or the "true" form of the psychophysi-
psychophysics. However, the only thing von Kries opposes is cal law if you think that is something different): Any perspicuous
Fechner's appeal to the correct unit of sensations. He concludes and insightful presentation of your research will do.
(p. 277) "dass wir es mit einer willkiirlichen Festzetzung von As I do not expect everyone to be convinced by Mach's lack of
Grossenbezeichnungen zu thun haben, welche zweckmassig deference for physics, I therefore add two examples which may
oder unzweckmassig sein kann, aber nicht richtig oder falsch." help to realise "the danger of overestimating the instruments,
Similarly, I couldn't place Murray's suggestion, on Fechner's with which we are so constantly employed, or even of regarding
behalf but against von Kries, that "astronomers can measure them as the objective point of science" (Mach 1960, p. 609).
star-brightnesses" and that "Delboeuf's subjects could indicate Temperature is often used to illustrate interval scales or
[an intermediate] greyness." Von Kries (1882) discusses both difference measurement. Stevens and others use it as an exam-
examples in detail, but concludes they merely show that we are ple to illustrate the "evolutionary" development from nominal to
the stimulus "looks"), and objective instructions (which ask distinguish among contending functions which make essentially
subjects to take into account distance to determine the objective equivalent predictions. A different, nonempirical, argument
size of the stimulus as measured by a ruler). The different against a universal law focuses on the arbitrary measure of the
psychophysical functions arise because the instructions induce physical stimulus (Myers 1982; Weiss 1981; 1989).
subjects to look at their subjective experience in different ways. Another nineteenth-century issue brought out by Murray is
These different perspectives each give a slightly different mean- also still with us. Fechner criticized the introspective validity of
ing to the word "size." Hering's difference sensations. Validity is the vexing concern in
Similarly, the different psychophysical methods subtly ask empirical attempts to determine sensation values. Despite a
subjects to look at their subjective experience in ways that give fairly sophisticated typology of the concept (e.g., Nunnally
slightly different meanings to the word "magnitude." Some 1970), validity ultimately seems to be in the eye of the beholder.
instructions ask subjects either explicitly or implicitly to pre- Progress might be made by viewing a response as a judgment
serve differences, and subjects dutifully construct a scale, a way rather than as the observer's reading a number from an internal
of looking at their experience, which more or less preserves meter. Then, by modeling the judgment, the researcher can
differences. Some instructions ask subjects to preserve ratios, attempt to validate a view of the judgmental process.
and subjects dutifully construct a scale, a way of looking at their This cognitive emphasis characterizes the functional mea-
experience, which more or less preserves ratios. Undoubtedly, surement approach to the issues of psychophysics (Anderson
subjects are cognitively flexible enough to produce other psy- 1970). Sensation scale values fall out as a byproduct of successful
chophysical functions if an experimenter cleverly presents the modeling. A characteristic of functional measurement analysis is
right instructions. that it calls for complex judgments, in which two or more stimuli
Finally, I would like to discuss briefly Murray's third issue, are integrated by the observer, presumably in accord with the
the possibility of "inner psychophysics." I believe inner psycho- hypothesized model. A major advantage of this approach is that
physics died the death it deserved for attempting the impos- context effects are a positive force. They enrich the model,
sible: resolving the mind-body problem empirically. (It's a providing additional leverage for analysis (Anderson 1975; Birn-
stretch to say signal detection theory derived from inner psycho- baum 1974) rather than obscuring pure estimates of sensation as
physics. Noise, in SDT, refers to the mathematical construct of the earlier tradition held.
variability and is only peripherally related to neural activity.) The nineteenth-century technique of equisection offers a
Although one could trace neural firing patterns as a function of beautiful illustration of an integrative judgmental task. Func-
stimulus intensity upon the nervous system, how could one ever tional measurement affords a validation scheme for equisection
know which collection of neurons represents the crossover point and allows the derivation of psychophysical scales (Weiss 1975)
from brain to mind? Is it the brain as a whole, some section of the in a way I like to think would have pleased Fechner. By allowing
cortex, the thalamus and something else, or maybe Descartes' for unequal weighting of the integrated components, the anal-
pineal gland? Higher in the nervous system, neural activity will ysis avoids the assumption (also made by Garner 1954) that the
be a mixture of thoughts and feelings, as well as sensations. At subject divides the interval in half. Finally, Fechner would
the critical point, it may be difficult to know what neural activity surely have been pleased at the way methodological and analytic
produces sensations of magnitude apart from other activity. procedures developed using psychophysical tasks that have
Finally, as in outer psychophysics, if we can't know with cer-
been found applicable in other domains (Anderson 1982).
tainty what the "true" magnitude of a sensation is, we can't
derive the "true" relationship between neural activity and
sensation either.
sinusoidal gratings) that are summarised by Graham (1989). the target article. I shall attempt to do this here by
Psychophysical "laws" or the measurability of sensation (or presenting the framework of the target article again but
lightness constancy) are nowhere discussed in that book. This this time rounded out by discussion of the contributions
implicit acceptance of von Kries's position has not, as Fechner in the commentaries and by some of my own addenda.
feared, led to the abandonment of psychophysics, indeed quite Actual criticisms of my remarks in the target article will be
the opposite. It has produced a most impressive "body of dealt with in the appropriate place. The plan of this
measurements to contemplate."
response will first involve some integration of various
We can see in this work transitions from discovery to method
that are not only a recurrent feature of science but a sign of its commentaries concerned with the antecedents and back-
health. Psychophysical methods are prominent, but they are ground of Fechner's Elements of psychophysics (1860/
used only as methods. Questions of the reality of thresholds, or 1964). Then we shall cover the same ground as the target
of the constancy of variance of the signal-plus-noise distribution, article but accentuating certain points, particularly the
which once seemed to be substantive generalisations worth issue of the logarithmic law and that of sensation measure-
fighting over, are now seen to be local questions to be answered ment. We will conclude by elaborating on two matters
in particular situations not for their own sake, but en route to raised by several commentators: the need for psycho-
something of more current interest. We now have a body of physicists to take context effects into account and the
psychophysical method and theory that provides the analytical suggestion that they transfer their focus from measure-
tools to do this (e.g., Macmillan & Creelman 1991). ments of sensation magnitude to measurements concern-
I have recently argued (Whittle, in press) that "Fechner's
integration" - the idea that we can measure sensation strength ing the duration of psychophysical decisions.
by summing jnds - should also be seen as the basis for a method
rather than as a substantive hypothesis. In the situation I R1. Historical and philosophical precursors
studied there was good agreement between jnds as measured by of Fechner's psychophysics
a two-alternative forced-choice task and step size in an equal-
subjective-interval series. I do not think, however, that we R1.1. Fechner's predecessors. Exactly when mathema-
should take this simply as vindicating Fechner, and wait with tics first entered psychology is a matter of debate. Marks
bated breath for someone to produce a counterexample. We informs us that Plato tried to estimate how much happier
should rather take it as showing that the idea holds in one ethical people were compared with unethical people, a
situation (and note the methods needed to show this), although
it may well not hold in others. This would not be a stalemate but question itself open to question ever since Freud taught
a pointer to interesting differences between the situations. For us about the dangers of a dominating superego. Marks
example, it would suggest that in the second case different cues also points out that both Plato and Pascal described
were used for discrimination and for scaling. We need more situations in which the subject basically carries out "intu-
techniques for studying perception of suprathreshold stimuli; itive statistics" to make decisions as to what to do or
Fechner's idea points to one that has the very desirable property whether to believe in God. In the latter case, I again refer
of allowing us to evaluate the degree of convergence between to depth psychology, which could claim that no matter
different methods. how rational you may wish to be in deciding whether or
The same argument can be made for another topic Murray not to believe in God, if the idea of "God" was put into
discusses, the constancy of lightness if luminance ratios are your head as a child you will probably never really escape
constant. This holds quite precisely in some situations but not at
all in others. The difference can be seen not as the failure of a "believing" in God no matter how hard you try. Despite
promising generalisation but as an important pointer to differ- the fact that many of our decisions are hopelessly biassed
ences between subjects' interpretations of the two situations in by our emotional needs and by our childhood learning,
terms of surface colour and lighting (Whittle 1991). intellectuals since Plato and Pascal have persisted in
In these instances the old questions are not settled so much as trying to bring numbers into psychology. Thomasius in
put into perspective. We see with hindsight that their impor- 1691 asked his readers to rate personality characteristics
tance was to spur us to develop methods for doing this. As (e.g., greed for money) on a scale of 1 to 60 (see
generalisations, they become of "merely academic" interest. McReynolds & Ludwig 1977; Ramul 1963, p. 657);
Hutcheson in 1725 invented a "moral algebra" in which he
basically elaborated on Pascal's notion that moral good
could be ascribed a numerical rating (see Brooks & Aalto
Author's Response 1981); Wolff (1731/1968) first used the word "psycho-
metrics" as a name for a proposed branch of science in
which numbers could be ascribed to degrees of "pleasure"
and "displeasure" (see Metraux 1983; Ramul 1960, p.
The place of psychophysics in the history 256); and, as Gregson stresses, the first full-scale attempt
to bring mathematics into psychology was made by Her-
of sensory science
bart in individual articles (e.g., Herbart 1812/1889), in his
David J. Murray book Textbook in psychology (1816/1891), and in his very
Department of Psychology, Queen's University, Kingston, Ontario, Canada large work whose title translates as Psychology as a
K7L3N6 science newly founded on experience, metaphysics, and
Electronic mall: murrayd@qucdn.queensu.ca mathematics (1824/1892). The historical background to
When I first read the assembled commentaries, I was Herbart's mathematization of psychology is described in
struck by how many added to and amplified the argument detail by Leary (1980), who notes direct influences on
in the target article, so that a reader who started with the Herbart's thinking of the writings of Leibniz, Kant, and
target article and then read the commentaries in order others.
might appreciate an opportunity to have the gist of the I am grateful to Gregson for finding Herbart's article of
commentaries reorganized within the "perspective" of 1812, whose title translates as "Psychological investiga-
responding can lead experimenters to mistakenly believe and Norwich's own psychophysical equation
that power functions are suitable descriptions of psycho-
physical functions. S = yl")
First, we must describe what is meant by a "logarithmic where k", y, and n are constants. This equation takes the
response bias." In magnitude estimation or category rat- form of Fechner's Law when yln > 1 and Stevens's Law
ing tasks, if subjects are given a large range of numbers as when yln 1 (Norwich 1991, pp. 160-63).
possible responses, they will use 1, 2, 3 . . . n-digit
numbers about equally often, particularly when judg- R2.2. Signal detection theory. I am grateful to Laming for
ments are being1 made about sensory dimensions that do arguing explicitly what was only implicit in the target
not have familiar units (e.g., brightness, loudness). Thus, article, namely, that signal detection theory arose inde-
on any scale on which there is a step change in the range of pendently of psychology in the contexts of engineering
numbers available (e.g., 1, 10, 100), the effect will be to and statistical decision making; Laming's Figure 1 would
make subjects give as responses numbers between 0 and suggest that I should have elaborated my Figure 1 to
9, numbers between 10 and 99, and numbers between shown an extra branch, a line reading from Rice, Shan-
100 and 999 about equally often, even though the range of non, and Fisher to the point labelled "signal detection
numbers comprised within each of these limits increases theory: Green & Swets 1966." Wagner also thought I
in the order mentioned. Other circumstances in which stretched things too far by relating signal detection theory
there is a logarithmic response bias include descending to inner psychophysics; the "noise" in the former theory is
judgments and cases in which instructions are given to a mathematical construct reflecting variability that is
judge ratios; as Poulton says (p. 173), "these instructions "only peripherally" related to neural activity. These quali-
encourage the undergraduates to use for responses a scale fications to my remarks in section 1.5 of the target article
of numbers calibrated logarithmically instead of linearly." are noted with appreciation.
Because this bias is so pervasive, Poulton recommends However, I had not intended readers to believe I
using linear ascending judgments as the only sure proce- thought signal detection theory was in any essential way
dure for avoiding the logarithmic response bias. based on the assumption of normal distributions, and I
The effect of this bias, according to Poulton, is to give welcome Laming's demonstration, in his Figure 2, of the
experimenters the impression that a power function is good fit of a distribution that is not normal (in this case, it
operative, because when log estimated magnitude is is a noncentral x 2 distribution) to some data of Swets et al.
plotted against log stimulus intensity, a straight line often (1961). In fact, the generalizability of the signal detection
results. This relationship, however, particularly in the model to nonnormal distributions has been amply at-
case when stimulus magnitudes cannot be judged in tested to in the book by Egan (1975). I remain with my
familiar units, can be produced by the logarithmic re- claim, however, that we do not have enough data to
sponse bias or a range-determined bias. Poulton shows supplement the large theoretical edifice that has been
how straight lines on log-log plots can be produced by erected around signal detection theory in works such as
experimenters if they choose the right range of numbers those by Green and Swets (1966), Egan (1975), and
to be used as responses. Similarly, cross-modal matches Macmillan and Creelman (1991).
between two stimulus dimensions can be made consistent I put Laming's Sensory analysis in the neurelectric
by appropriate choices of the range of responses to be column of Figure 1 because the book (and also the
used for each dimension. Since the size of the exponent multiple review, BBS 11(2) 1988) deals in detail with
depends on the range of responses the experimenter probabilistic events that occur in the eye, ear, and other
uses, exponents cannot be said to be unique to particular sense organs. I am happy to acknowledge that its rightful
sensory dimensions. As for the case of loudness, which lineage is signal detection theory, however, and in fact
was historically the dimension for which Stevens claimed this is where I originally placed it in earlier drafts of the
that a logarithmic psychophysical function did not hold, target article.
Poulton (1969) indicated that when range effects were
controlled, the discriminability of loudness for noise in- R2.3. Measurement theory. When I wrote the target
creased linearly with log sound pressure level, whereas article, I thought I was relatively isolated in worrying
for the loudness of tones, this psychophysical function for about whether sensation can be measured, but section
various methods tended to be intermediate between 1.5.3 elicited considerable commentary. In the article, I
linear and logarithmic (Poulton 1989, Fig. 4.2). The mentioned von Kries as being at the centre of the move-
reader finishes Poulton's book with a strong scepticism ment whose aim was to discredit the notion that sensa-
about the validity of a power law and a renewed feeling tions could be measured, but I would stress again that von
that perhaps the logarithmic law has been rejected too Kries was relatively late in the line of Fechner's detractors
hastily; in Figure 1 of the target article the question mark on this issue: The neo-Kantians, for example, objected to
under the line heading from the logarithmic law might be the idea that sensation could be isolated at all from the
supplemented by "Poulton 1989." apperception process; and in France certain mathemati-
Krueger notes that Delboeuf's Law is more suitable for cians had argued before von Kries that sensations were
dealing with certain contrast effects than is a power law not measurable in the way that lines were (Heidelberger
(though see sect. R2.5 below) or a logarithmic law. I 1993a). In this context the distinction should be made
would also note that Norwich (personal communication) between "extensive" measurement, where a value of a
points out a resemblance between Delboeuf's Law, magnitude on a numerical scale (say 6 units) can be
which can be written derived from adding together two shorter magnitudes
(say 3 units each), and "intensive" measurement, where
S = log[l + lie] this is not possible. Length, for example, is an extensive
magnitude: I can lay two 3-metre sticks end-to-end to van Brakel indicate, his sentences have to be understood
produce a 6-metre length. Temperature is an "intensive" in their broad context.
magnitude: I cannot add a beaker of water at 10C to However, my second point is more worrying: If Mach
another at 10C and produce a new volume of water at said that sensations can only be categorized on an ordinal
20C. Von Kries stressed that length was subject to the scale, is there any hope that a psychophysical function will
concatenation criterion, as just outlined, and claimed that have scientific validity? Temperature can certainly be
sensation strength, being intensive, could not be mea- predicted from calculations about movements of mole-
sured given a concatenation criterion. Ramul (1960, p. cules; so it should be possible to predict sensation magni-
258) shows that Ploucquet in 1764 had also claimed that tude from a knowledge of stimulus intensity. The doubt
psychological magnitudes, being intensive, could not be Mach raises is whether any inference about whether one
concatenated. sensation magnitude is "twice" another is valid; and, of
In my target article, I gave a quotation from Mach course, as Wagner reminds us, this is what William James
which seemed to support von Kries because it said "there had in mind when he said that one red is not a sum of
can be no question of an actual measurement of the pinks. Similarly, if two liquids of different temperatures
sensations; all that can be done is to characterize them are combined, we can predict the resulting temperature
exactly and make an inventory of them by numerical at different intervals after the mixture is made, knowing
means." This quotation is from Mach's most widely read the laws of cooling; but how do we mix two sensations?
book, and even Stevens (1975, p. 59) stressed the first We can compare them, as we move a finger from a vessel
sentence, although he read the second sentence as imply- of hot water to a vessel of cold water; but these are "be-
ing that sensation could be categorized by means of fore" and "after" sensations, not a "mixture" of sensations.
numbers. However, both Heidelberger and van Brakel Luce's contribution to a resolution of the dilemma is to
have hastened to correct me: By "measure" Mach meant argue that the mistake psychophysicists make is to try to
"measure according to the concatenation criterion" and measure just one dependent variable, sensation, when
by "make an inventory" Mach meant "assign numbers to only one independent variable, physical intensity, is
different degrees of sensory magnitude in the way that manipulated. He says that inferences made about the
numbers can be assigned to different degrees of the effects of single independent variables are only successful
intensive magnitude of temperatures." Or, rephrasing it, when the dependent variables are extensive measures of
Mach was sceptical about whether sensation could be physics such as mass, length, charge, and time, all of
measured on a ratio scale (whose attractiveness should not which can be concatenated. Luce cites his work on sensa-
be seen as limited to the fact that concatenation is an tion measurement when two or more independent vari-
analogue of the arithmetical operation of addition, since ables are manipulated, an approach also espoused by
other arithmetical operations can also give rise to mea- Anderson. Mach says that sensation magnitudes cannot
sures concordant with concatenation, as both Luce and be concatenated; Luce writes that "unless a relevant
van Brakel note). concatenation operation or some other rich internal struc-
I have two comments. First, Mach has a tendency to ture can be found, the situation [the one-dimensional
use words idiosyncratically. Consider this passage from measurement case] is simply too underdetermined to be
his article entitled "Some questions of psycho-physics" of much theoretical interest." If underdetermination, like
(Mach 1891) controversy, leads to overspeculation, William James
Now I say, that if I see a tree with green leaves (A), with (1890/1950, vol. 1, p. 549) was vindicated in his worrying
a hard (B), gray (C) trunk, that A B C are elements of the about the "dreadful literature" of psychophysics.
world. I say elements - and not sensations, also not One way in which the problem of sensation measure-
motions - because it is not my purpose at this place to ment of a single sensory magnitude can be bypassed is by
arrive at either a psychological or a physiological or a studying the behaviour of subjects confronted by two
physical theory, but to proceed descriptively. . . . If, kinds of sensation, as in the cross-modal matching (CMM)
however, I close my eye (K), withdraw my feeling hand paradigm. Teghtsoonian in fact claims that CMM is the
(L),ABC . . . disappear. If I contemplate A B C in f/iis answer to the riddle of how to measure sensation strength
dependence they are my sensations. This is but a when it is so hard to define it on a one-dimensional scale.
special point of view within the first. CMM avoids the "ancient questions" about ratios or
According to my conception, therefore, the same A B differences between sensations, questions that, as we saw
C is both element of the world (the "outer" world in the target article, have caused unusual difficulties. My
namely) and element of feeling, (pp. 394-95) main problem with Teghtsoonians argument is that, as far
This is a difficult passage to read at first because there is a as he knows, "no one has ever suggested that the power
natural tendency to read Mach's definition of "sensation" law is found for CMM only because the experimenter
as referring to a memory image; but I think he is saying induced a particular bias in his subject." Poulton (1989,
that whether or not the green leaves of the tree are objects pp. 271-75), however, indicates how biasses operating in
or sensations depends on your point of view, and that one Stevens's experiments could have induced subjects to
can only tease apart a "sensation" of an object from the respond in such a way as to yield power functions for the
object itself by the artificial experiment of closing one's individual modalities investigated.
eyes. In the same paper, Mach objected to Fechner's Hornstein, in suggesting that von Kries's argument was
comparison of matter and mind as being "as inseparable as unfairly ignored for over a century, says that my dismissal
the concave and convex sides of the same circle"; Mach of von Kries's critique reveals a "surprising presentism" in
says this makes a duality where in fact unity exists, a my approach. First, I don't think I ever "dismissed" it; as I
sophisticated form of monism. Mach's ideas were not wrote the target article, which was meant to be a "per-
easily understood in his time, and, as Heidelberger and spective," that is, a mental framework for encapsulating
the history of psychophysics in the reader's mind, I gave on psychophysical methods, yielded few comments con-
von Kries a prominent place in the history of measure- cerning the classification of the methods shown in Tables
ment theory. Now, I may freely say that I think the 3 and 4. Teghtsoonian notes that I had not referred to
commentaries I have just reviewed have been extraordi- cross-modal matching, which would presumably fit in
narily helpful in improving that "perspective": We now Table 3 (nonverbal responses) with two stimuli (A on G, A
know, from Heidelberger and van Brakel, that von Kries clearly discriminable). There would be a continuous
did not properly discuss the idea that sensation strength range of stimuli in another modality, B on a background
could be measured on an ordinal or interval scale, and H, which have to be adjusted to match A. Borg stresses
restricted his idea of measurement to one permitting a that for purposes of generalizing beyond measures of
concatenation criterion. We have also learned from Luce sensation strength to such measures as perceived exer-
that von Kries's views fit into a particular place in the tion, ratio scales or category ratings are useful, partic-
context of the whole of science. Nevertheless it must be ularly when it comes to establishing a range from "lowest"
acknowledged that there are still persons who object to to "highest." Another classification of psychophysical
the idea of measuring sensations, either because they methods, which does include cross-modal matching, will
believe the word "measure" has to be fitted in a Procrus- be found on pp. 39-40 of Poulton (1989). Here the
tean way to the meaning of the word "sensation" (Horn- methods are classified into five broad headings: discrimi-
stein likens this to the Red Queen's twisting of meaning) nating and matching judgments; interval or partition
or because sensations themselves are "ontologically du- judgments; category ratings; arithmetic judgments; and
bious characters" that have no place in outer psycho- cross-modal judgments. Choosing between classifications
physics (as Cundlach claims). will be a matter of personal taste; in the history of
Second, I am becoming worried by the growing use of psychology, a similar question about classifications was
"presentism" as a negative term meaning "biassed his- extensively discussed in the nineteenth century, namely,
tory. " It is impossible to write a history of a large topic the question of the classification of the emotions. Many
which is unbiassed by one's knowledge of the present writers from early times, including Aquinas, Descartes,
state of the topic, because presumably in a history of a Hume, and Thomas Brown (1820/1830), had classified the
topic as large as psychophysics, I cannot simply convey emotions in different ways; it was perhaps Bain
information about every paper without making some (1859/1880) who laid the fashion to rest when, in his The
evaluation of its usefulness from a present-day point of emotions and the will, he summarised previous classifica-
view. There are indeed histories of other large topics in tions and rejected them all as armchair exercises. Since
psychology that rigorously avoid any kind of presentism. then, with the important exception of J. B. Watson (with
Jackson's splendid history of ideas about depression down his three basic emotions of rage, fear, and love; Watson
the ages (Jackson 1986), for example, takes us through the 1930, p. 138), classifying emotions has become a lost art.
ancient and medieval periods with little reference to Yet I feel that my armchair exercise in classifying the
modern theories, and this objectivity persists in his dis- psychophysical methods has the important spinoff of
cussions of nineteenth- and twentieth-century theories of forcing us to realize that all psychophysical judgments are
depression. The end result is that the reader feels a strong made against a sensory background and that the more
sense of the sheer weight of theorizing about depression flexibility given to the subject, as in n-choice responses,
and yet also feels a sense of incompleteness (one, I hasten the more information the experimenter will obtain,
to add, that is justified) by the lack of any feeling that we though possibly at the cost of giving the subject more
are close to a simple explanation of depression. The opportunity for bias.
history of psychophysics also leaves one with a sense of On the other hand, the middle section of section 1.2
incompleteness. However, I think there is room for a aroused more comment than any other passage in the
particular kind of presentism, one which takes a current target article. Borg wishes I had elaborated on my remark
opinion and looks back through history for evidence or that "if we want a science in which assertions about
theories that support it (as well as evidence against it). individual psychological experiences (e.g., "I don't like
The first half of Link's (1992) book is like this, and it leaves stimulus X . . ,")are to be meaningfully related toothers
the reader with a sense of progress rather than the lack of (". . . because stimulus X feels too intense"), it is difficult
it. But modern psychophysics, alas, is in such a state of to see how we could do without the expanded body of
flux that instead of saying "this is the truth: Here is the information [involving the concept of sensation
evidence that supports it," I have been obliged to say: strength]." Teghtsoonian points out that since the choice
"These are the issues: Here is the evidence adduced on of a reference continuum for describing sensation
these issues," leaving readers, of course, with the sense strengths is entirely arbitrary, the relations that are ob-
that psychophysics is still too controversial for them to tained among "sensation strengths" will depend on the
base any strong reliance on any one of its theories. choice made. Whittle points out that hundreds of studies
However, if the word "psychophysics" is extended to of visual science are carried out with no mention of
include any studies in which subjects make responses "sensation strength" or "psychophysical laws"; he gives
concerning what they sense (perceive), then psycho- convincing examples from his own work indicating that
physics is not such a disappointment as the target article "local" phenomena of sensation (such as brightness con-
makes out. As Whittle writes, you can do psychophysics stancy) are often not generalizable into comprehensive
without some prior commitment to an ideology about assertions concerning other situations with ostensibly
sensation strength or a particular psychophysical law. similar characteristics. Lebedev, on the other hand, is
optimistic that we shall be able to relate assertions about
"sensation strength" to brain processes that show regular
R2.4. Do we need "sensation strength" in a valid psycho- cycles in time. Lockhead confirmed me in my suspicion
physics? The first part of section 2.1 of the target article,
rounded by green differed from that of a patch of identical function was obtained but the exponents changed with
grey surrounded by red; they disagreed on the explana- the background. Then he took the same stimulus on a
tion, however, with Helmholtz favouring an account fixed background Bl and changed the overall illumina-
based on perceptual learning and Hering favouring an tion: The effect of the overall illumination was to change
account in terms of neural interactions at a peripheral both the apparent lightness of the stimulus and the
level. Mach (1866) showed that a new sensation (a "Mach apparent lightness of the background. He found that the
band") could arise when a piece of paper divided into phenomenal appearance of the stimulus was predicted
equal black and white areas, but with a border showing a from the earlier power functions relating / to R, given
sharp discontinuity in its contour, was rotated. The paper each value of background. In particular, he found that
looked mainly grey except for a darker strip or band at the with a dark grey, the grey looked blacker as overall
locus of the discontinuity. Some time later, von Ehrenfels illumination increased.
(1890) pointed out that the entity known as a "tune" is Since psychophysical power functions hold for each
based on relationships between adjacent items indepen- stimulus on a given background, and these could be used
dent of the individual frequencies of the tones. Evidence to predict appearances of the stimulus when both stim-
like this persuaded the Gestalt psychologists of the early ulus and background changed because of the effect of
twentieth century to stress that the "clear" percepts of overall illumination, Stevens would claim that one can
everyday experience are the end results of brain pro- take contrast effects into account and still make use of the
cesses that act automatically (probably innately) when- psychophysical laws applying to single stimuli. The point,
ever the brain receives sensory information, in organizing however, is that the exponent of the law in each case is a
that information so that it becomes ecologically useful to function of the contrast between the stimulus and the
the percipient. In fact, almost the last sentence of Koh- background, a point elaborated by du Buf (1987).
ler's (1969) last book summarizes the Gestalt question as When is a stimulus ever judged in isolation, then? A
succinctly as any other: He asked "why do brain processes point of light in an otherwise pitch-dark room? A Ganz-
tend to produce perceptual organizations of remarkable feld? I think many people would claim that experimental
clearness of structure?" This, of course, is almost identical data obtained under such circumstances are artificial in
with Leibniz's question about clear and obscure ideas: In that these circumstances rarely correspond with real-life
this long-range historical perspective, Fechner's focus on situations. Psychophysical experiments, in a word, are
individual sensations can almost be seen as a diversion or claimed to lack ecological validity. The same claim has
digression. been made for the experiments on letter or word identi-
The importance of global factors in perception is fication using tachistoscopic presentation that seem to
stressed by several commentators including Weiss and support the notion of iconic memory (Haber 1983); we
Anderson as noted above; Gigerenzer and Lockhead, never run into this situation in ordinary life (how often do
both of whom support Brunswik's attempt to redefine the we read newspapers by lightning flashes?). And the claim
perceptual process in terms of correlations between stim- has also been made for experiments on memory using
ulus events and mental events; Gregson, who claims that nonsense syllables or isolated words (Bartlett 1932;
psychophysical equations that might be valid for a single Neisser 1982).
stimulus-on-a-background, cannot simply be somehow The solution of rejecting all experiments involving
"put together" to yield equations valid for more than one isolated stimuli as ecologically invalid is in my opinion too
stimulus on the same background; and Handel, who harsh. Instead we need a comprehensive answer in which
stresses how neurelectric events at the receptor level variables that are undoubtedly influential in determining
cannot be simply tracked through the nervous system in the perception of isolated stimuli can be argued to be less
order to predict a neurelectric event at the brain level. important in determining the perception of multistimulus
However, there seem to be two camps here: The above displays for a reason. In the case of perception, explana-
commentators tend to be sceptical of any attempt to base a tions like this are in the offing, I think, if we consider such
theory of holistic perception on some kind of elaboration work on contrast as that by Whittle (1992) and others. For
of psychophysical laws applied to relationships between / example, there is the phenomenon known as the "crisp-
ening effect," which is seen for isolated stimuli but is not
and R found for single stimuli; on the other hand,
always apparent if the display is made more complicated.
Krueger feels that holistic effects such as contrast effects
If a circle of light is set up in a surround and the lumi-
should be avoided as far as possible in making determina-
nances of either are changed, there is a steep change in
tions of psychophysical laws; and Olsson et al. stress how
the phenomenal brightness of the circle when its lumi-
estimates of the size ofjnds can be different because of the
nance passes through that of the surround. This crispen-
variability induced by other psychological factors. Which ing effect (Takasaki 1966) is possibly related to the fact that
view is correct? sensation intensity at borders may not necessarily in-
Stevens himself, I think, would agree with Krueger crease with the logarithm of/ but with log A/, a finding I
that there is a true psychophysical law relating the inten- should have mentioned in section 2.2 of the target article.
sity of a single stimulus (on a background) to the strength Whittle, however, has shown that the crispening effect
of the associated sensation. For example, in order to can be abolished by surrounding the circle with a thin
evaluate the effect of a change of overall illumination on a outline or by introducing hue differences between circles
single grey stimulus against a white background, Stevens and background. In his commentary, Whittle notes that
(1961) first took the stimulus on a background Bl and lightness constancy can be obtained if luminance ratios
showed that a power law described the relationship be- are held constant in some situations but not in others; this
tween the lightness of the stimulus and the magnitude does not mean that the generalization has failed; it points
rating. But then he did the same again for the stimulus on instead to the need for a more general interpretation in
new backgrounds (B2, B3 . . . ). In each case a power
the analysis done by Luce (1986, p. 446) of Chocholle's speaking one cannot represent the results of the two
data: As stimulus intensity increased, and therefore mean subjects on the same graph. Krueger notes that in select-
response time decreased, the standard deviation of the ing an anchoring point on a Fechnerian sensation scale,
response time also decreased. But if we make the decision the point named 70 is really an experimenter's "criterion"
harder by presenting one of two tones (of equal intensity), rather than an observer's "threshold," while the choice of
each tone requiring a different response, and vary the units on the physical scale has been neglected in many
intensity of the two tones, it is found that response times accounts of psychophysics even though it determines the
change in a U-shaped pattern as stimulus intensity in- value of the exponent in any power function. Teghtsoo-
creases (van der Molen & Keuss 1979). nian is so impressed with the arbitrary elements involved
Other problems with response times include the scep- in scaling sensation that he says the goal of measuring
ticism that can be engendered when we read the worries sensation strength is a "will-o'-the-wisp," which my dic-
by Luce (1986, p. 421) about our lack of understanding of tionary defines as a light which misleads travellers, often
Hick's (1952) Law concerning the increase in choice seen over marshy places (!). Treisman notes that once you
response time as the number of alternative stimuli in- have defined the function relating S to I, whether or not
creases; or when we read his criticisms of the previously sensation differences or ratios are judged by subjects falls
acclaimed suggestion by Sternberg (1969) that analysis of out of the definition. Van Brakel points out that arbitrari-
variance be used to estimate whether processes take place ness applies not only to sensation scales, but also to
at simultaneous or at successive stages (Luce 1986, p. physical scales such as length; as he writes, "although,
481); or when we read his criticisms of Donders's subtrac- allegedly, the only arbitrary choice for a ratio scale is its
tion methods (Luce 1986, p. 213). I do not deny that the unit, . . . the same unit can 'support' two different
study of response times might add rigour to our under- scales, which are theoretically equally adequate." Wag-
standing of how sensations are processed; response times ner claims that, just as instructions can determine how
are probably more reliable as data than are judgments perceived size seems to vary with physical size, so in-
such as magnitude estimates; and one of the merits of the structions can determine how "sensation strength" seems
recent wave theory of difference and similarity offered by to vary with "stimulus strength." Stout and Wagner note
Link (1992) is his unification of judgmental and response that William James rejected Fechner's psychophysics on
time measures within a single theoretical framework. But grounds related to the difficulty of measuring sensations.
response time data have to be approached with caution In the light of these comments, so-called measure-
because response times are so sensitive to any change in ments of sensation strength seem to be among the less
experimental procedure. valuable accomplishments in the history of psychophysics
On the other hand I think it is essential to note the so far. Another group of commentators, however, claims
similarity of the ideas expressed by Lebedev to those that the exercise of measuring sensation strength might
recently expressed by Newell in his Unified theories of also be of limited usefulness in the overall perspective of
cognition (1990; see also multiple book review, BBS 15(3) studies of sensation and perception. Anderson claims that
1992). Both authors indicate that different neurological Fechner did not properly appreciate the importance of
processes (reflecting different levels of cognitive activity) the way the nervous system integrates information from
tend to operate in time-quanta of approximately .01, . 1 , 1 multisensory displays even though, as I noted, previous
or more seconds. That time units, rather than internal eighteenth- and nineteenth-century German work on
scalable magnitudes, are the appropriate numbers to apperception can be seen to anticipate Anderson's views
include in cognitive science is an argument that is given on unconscious processing. Brigner finds Fechner's work
considerable credibility by the work of these writers. only incidental to the development of global theories of
inner psychophysics - the brain itself seems to provide a
context for some sensory experiences such as phantom
R3. Concluding remarks limb pain (and, I would add, dreams). Gregson (1988) has
offered a new perspective altogether for viewing psycho-
When I had finished the main part of the target article, I physics, one in which the dynamics of large networks
had to write a concluding paragraph and found myself in depend very much on local connections and not on the
conflict between my desire to say that psychophysics had sheer size of the networks; when the behaviour of such
made progress since Fechner and my desire to say, more networks is very simple, it yields behaviour almost look-
reluctantly, that in writing a perspective which took ing like traditional psychophysics, but it is in this context
Fechner as its starting point, I was forced to admit that I that the psychophysical laws of Plateau and Fechner have
frequently ran into stumbling blocks that made me less limited value. In his commentary, Gregson criticizes
positive in my evaluation of how useful psychophysics has these for not being "dissipative." Krueger.also notes how
been. The latter sentiment won out (at the last minute) extreme sensitivity to initial conditions can give rise, by
and the final paragraph of my target article mentioned two deterministic processes, to unpredictable and complex
stumbling blocks in particular, the problem of whether behaviour. Handel shows neatly how the nervous system
sensation strength can be measured and the problem of is too complex for a simple "neurelectric law" that is
whether we use sensation differences or ratios in judging parallel to a simple "psychophysical law"; stimulus dura-
contrasts. These issues have also been discussed by a tion, for an example, is a variable ignored in psychophysi-
number of commentators and several focused on the cal laws which actually determines the behaviour of
matter of defining the units of sensation strength and neurons at higher levels in the frog's nervous system.
physical strength. It is worth summarizing these. Laming argues that Fechner "directed attention to some
Borg notes that in a single experiment two subjects may inappropriate questions (to which answers are never
use different units of sensation strength, so that strictly found); and his successors have become fixated on those
particular questions, unable to turn away, even when content to see it as that, a device for acquainting young
experimental evidence so dictates." Laming quotes the persons who wish to do psychophysics with a framework
literature on fixed thresholds. for thinking about this topic. It is clearer to me now,
Lockhead claims that there are two literatures on however, than it was when I first wrote the article that it is
psychophysics, one which sees the role of the stimulus a framework limited to attempts to answer Fechner's
environment as a nuisance contaminating the measure- questions in Fechner's terms. It was he who stressed
ment of the sensation corresponding to that stimulus (a thresholds, just noticeable differences, sensation differ-
view that Brunswik called "structuralism"); and another, ences and ratios, and the difference between inner and
Brunswik's "contextual psychophysics" as Lockhead calls outer psychophysics. The commentators have made it
it, which would presumably include adaptation level quite clear that in the future psychophysics may see
theory and Gestalt psychology as well as Brunswik's own Fechner's psychophysics as one (historically important)
theory. In this second literature, Lockhead says that aspect of a more comprehensive psychophysics in which
"Fechner-like laws cannot be correct beyond being points context effects are central, biasses to respond are properly
in a complex function where performance depends on integrated into the topic, and neurelectric events affect-
many factors which sometimes interact in complex ways." ing cognition are not thought of as simple monotonic
Presumably Krueger's claim that "in order to reveal the transformations of neurelectric events at the periphery. A
true underlying psychophysical function, it is necessary psychophysicist of the future who wished to write his own
to pare away bias and context effects not only at the "perspective" would probably see my perspective as
cognitive level, but also at the sensory or neural level" is a
Gregson sees it, the middle of a fish whose head (in the
dogma of Lockhead's first kind of literature, not the past) lay in eighteenth-century theories of apperception
second. Weiss, like Luce and Anderson, sees a psycho- and whose tail (in the future) will incorporate the
physics based on measurements of unisensory stimuli to twentieth-century writings of Helson, Brunswik, the Ge-
stalt writers, and Anderson, as well as new developments
be a sort of "special case" which can fall out of a more
in neurophysiology and in mathematical modelling.
general system of functional measurement - "sensation
scale values fall out as a byproduct of successful model-
ling. A characteristic of functional measurement analysis
is that it calls for complex judgments, in which two or References
more stimuli are integrated by the observer, presumably Letters a and r appearing before authors' initials refer to target article and
in accord with the hypothesized model . . . Context ef- response respectively.
fects . . . enrich the model . . . rather than [obscure] Anderson, N. H. (1970) Functional measurement and psychophysical
pure estimates of sensation as the earlier tradition held." judgment. Psychological Review 77:153-70. [aDJM, DJW]
(1972) Cross-talk validation of functional measurement. Perception and
Finally, Whittle sees Fechnerian theory as having pro-
Psychophysics 12:389-95. [rDJM]
vided methods and questions which once "seemed to be (1975) On the role of context effects in psychophysical judgment.
substantive generalisations worth fighting over, [but] are Psychological Review 6:462-82. [DJW]
now seen to be local questions to be answered . . . en (1981) Foundations of information integration theory. Academic
route to something of a more current interest." Press. [RDL]
(1982) Methods of information integration theory. Academic Press. (NHA,
In the above, I have deliberately used quotations from
RDL, DJW]
the commentators because they clearly indicate that (1990) Integration psychophysics. In: Psychophysical explorations of mental
many psychophysicists now writing think of Fechner's structures, ed. H.-G. Geissler. Hogrefe & Huber. [NHA]
work as limited in scope. In the last 15 years or so, there (1992) Integration psychophysics and cognition. In: Psychophysical
has been a revival of Gestalt-type thinking among stu- approaches to cognition, ed. D. Algom. Elsevier. [NHA]
Atkinson, R. C. (1963) A variable sensitivity theory of signal detection.
dents of perception, and psychophysicists now - using Psychological Review 70:91-106. [DL]
the word in Whittle's broad sense - wish to delve into Atkinson, W. H. (1982) A general equation for sensory magnitude. Perception
neuropsychology and bring in cognitive psychology in a and Pstjchophysics 31:26-40. [aDJM]
way that is foreign to traditional psychophysics. As Whit- Bain, A. (1859/1880) The emotions and the will, 3rd ed. Longmans,
tle says, you can know a great deal about visual science Green. [rDJM]
Baird, J. C. (1970) A cognitive theory of psychophysics: II. Fechner's law and
without worrying too much about whether Fechner's Stevens'law. Scandinavian Journal of Psychology 11:89-102. [MO]
questions are answered by that science. At the same time, Baird, J. C. & Noma, E. (1975) Psychophysical study of numbers. 1.
caution must be applied when drawing conclusions that Generation of numerical responses. Psychological Research 37:281-
"holistic" or "global" properties dominate the perceptual 97. [aDJM]
processing; Kimchi (1992) distinguishes between "global" (1978) Fundamentals of scaling and psychophysics. Wiley. [MO]
Baird, J. C. & Wagner, M. (1991) Transformation theory of size judgment.
properties of a stimulus, defined by their position in the Journal of Experimental Psychology: Human Perception b Performance
hierarchical structure of the stimulus, and "holistic" prop- 17:852-64. [MW]
erties, defined as a function of interrelations between the Baker, K. E. & Dudek, F. J. (1955) Weight scales from ratio judgment? nd
component parts of the stimulus. Her review is a good comparisons of existent weight scales. Journal of Experimental Psychology
starting point for a more careful examination of the Ge- 50:291-308. [HER]
Balmer, H. (1977) Albrecht von Holler. Haupt. [O-JG]
stalt revival. Hoffman and Dodwell (1985) show how Bartlett, F. C. (1932) Remembering: A study in experimental and social
global Gestalt laws can be derived from a "geometric psychology. Cambridge University Press (Cambridge). [rDJM]
psychology" that postulates how the brain processes par- Bartley, H. S. (1970) The homeostatic and comfort perceptual systems.
ticular local responses. Journal of Psychology 75:157-62. [GB1
Baxt, N. (1871/1982) On the time necessary for a visual impression to come
When I submitted the manuscript of the target article, into consciousness. Psychological Research 44:1-12. (Translated by D. J.
one referee said that it might be useful as a teaching Murray.) [aDJM]
device. The more I think about this, the more I am Beck, J. (1972) Surface color perception. Cornell University Press. [aDJM]
Eisler, H. & Montgomery, H. (1974) On theoretical and realizable ideal Foss, C. E., Nickerson, D. & Cranville, W. C. (1944) Analysis of the Oswald
conditions in psychophysics: Magnitude and category scales and their color system. Journal of the Optical Society of America 34:361-
relation. Perception and Psychophysics 16:157-68. [MO] 81. [aDJM]
Ekman, C. (1956) Discriminal sensitivity on the subjective continuum. Ada Furman, G. G. (1965) Comparison of models for subtractive and shunting
Psychologica 12:233-43. [arDJM, MO] lateral inhibition in receptor-neuron fields. Kybernetik 2:257-74. [O-JG]
(1959) Weber's law and related functions. Journal of Psychology 47:343-- Gamau, G. (1811) Erinnerungen an Lichtenbergs Vorlesungen Ober Erxlcbens
52. [aDJM, MO] Anfangsgrunde der Saturlehre, vol. 2. Wien: Geistinger. [O-JG]
Ekman, G. & Kiinnapas, T. M. (1957) Subjective dispersion and the Weber Garner, W. R. (1954) A technique and a scale for loudness measurement.
fraction. Reports from the Psychological Laboratory, University of Journal of the Acoustical Society of America 26:73-88. [DJW]
Stockholm, No. 41. [rDJM, MO] (1962) Uncertainty and structure as psychological concepts.
Ellis, B. (1967) Measurement. In: Encyclopedia of philosophy, vol. 5, ed. P. Wiley. [aDJM]
Edwards. Macmillan. [MH] (1974) The processing of information and structure. Erlbaum. [SH]
(1968) Basic concepts of measurement. Cambridge University Press Gauss, C. F. (1809) Theoria motus corporum coelestium. Hamburg: Certhes &
(Cambridge). |JVB] Besser. [SL]
Engen, T. & Tulunay, U. (1957) Some sources of error in half-heaviness (1821/1880) Theoria combinationis observationum erroribus minimis
judgments. Journal of Experimental Psychology 54:208-12. [rDJM] obnoxiae, pars prior. In: Carl Friedrich Causs Werke, vol. 4. KSniglichen
Eriksen, C. W. (1966) Temporal luminance summation effects in backward Gesellschaft der Wissenschaften zu Gottingen. [SL]
and forward masking. Perception and Psychophysics 1:87-92. [aDJM] Geissler, H.-G. (1990) Foundations of quantized processing. In:
Falmagne, J.-C. (1985) Elements of psychophysical theory. Oxford University Psychophysical explorations of mental structures, ed. H.-G. Geissler.
Press. [aDJM] Hogrefe & Huber. (ANL)
(1986) Psychophysical measurement and theory. In: Handbook of perception Geldard, F. A. (1972) The human senses, 2nd ed. Wiley. [HER]
and human performance, vol. 1, ed. K. R. Boff, L. Kaufman & J. P. Gescheider, G. A. (1976) Psychophysics: Method and theory.
Thomas. Wiley. [aDJM] Erlbaum. [aDJM, HER]
Fechner, G. T. (1848) Nanna oder iiber das Seelenleben der Pflanzen. (1984) Psychophysics: Method, theory and application, 2nd ed.
Leipzig: Voss. [GG] Erlbaum. [SH, MO]
(1851) Zend-Avesta: Oder iiber die Dinge des Himmels und des Jenseits. Gibson, J. J. (1979) The ecological approach to visual perception. Houghton
Leipzig: Voss. [aDJM, GG, ANL] Mifflin. [SH]
(1858) Das psychische Mass. Zeitschrift fur Philosophic und philosophische Gigerenzer, G. & Murray, D. J. (1987) Cognition as intuitive statistics.
Kritik 32:1-24. [MH] Erlbaum. [aDJM, GG]
(1860/1964) Elemente der Psychophysik, vols. 1 and 2. Leipzig: Breitkopf & Gilchrist, A. L. (1980) When does perceived brightness depend on perceived
HSrtel. (Reprinted: [1964] E. J. Bonset.) [aDJM, GB, GG, O-JG, MH, spatial arrangement? Perception and Psychophysics 28:527-38. [arDJM]
SL] Goldner, K. (1983) Mathematische Grundlagen der Systemanalyse, vol. 2.
(1860/1966) Elements of psychophysics, vol. 1, ed. E. G. Boring & D. H. Verlag Harri Deutsch. [RAMG]
Howes. (Translated by H. Adler.) Holt, Rinehart & Winston. [arDJM, Graham, N. (1977) Visual detection of aperiodic spatial stimuli by probability
SL, DAS] summation among narrowband channels. Vision Research 17:637
(1866) Das Buchlein vom Leben nach dem Tode, 2nd ed. Leipzig: Voss. 52. [DL]
(Originally published in 1836 using the pseudonym Dr. Mises.) [GG] (1989) Visual pattern analyzers. Oxford University Press. [PW]
(1876) Vorschule der Aesthetik. Leipzig: Breitkopf & Hartel. [rDJM] Green, D. M. & Swets, J. A. (1966) Signal detection theory and
(1877) In Sachen der Psychophysik. Leipzig: Breitkopf & Hartel. [aDJM, psychophysics. Wiley. [arDJM, DL]
MH, LEK] Gregson, R. A. M. (1988) Nonlinear psychophysical dynamics.
(1882) Revision der Hauptpuncte der Psychophysik. Leipzig: Breitkopf & Erlbaum. [RAMG]
Hartel. [aDJM, MH] (1991) Dead men tell odd simple tales! Behavioral and Brain Sciences
(1887) Ueber die psychischen Massprincipien und das Weber'sche Gesetz.
14:187-88. [aDJM, LEK]
Philosophische Studien 4:161-230. [arDJM, MH] (1992) n-Dimensional nonlinear psychophysics. Erlbaum. [RAMG]
(1897) Kollektivmasslehre, ed. G. F. Lipps. Leipzig; W.
Grusser, O.-J. (1979) Cat ganglion-cell receptive fields and the role of
Engelmann. [aDJM, GG] horizontal cells in their generation. In: The neurosciences. Fourth Study
(1987a) Outline of a new principle of mathematical psychology (1851). Program, ed. F. O. Schmitt & F. G. Worden. MIT Press. [O-JG]
(Translated and edited by Eckart Scheerer.) Psychological Research (1989) Quantitative visual psychophysics during the period of European
49:203-7. [aDJM]
enlightenment. The studies of the astronomer and mathematician Tobias
(1987b) My own viewpoint on mental measurement (1887). (Translated and Mayer (1723-1762) on visual acuity and colour perception. Docunwnta
edited by Eckart Scheerer.) Psychological Research 49:213-19. [aDJM] Ophthalmologica 71:93-111. [O-JG]
Feng, A. S., Hall, J. C. & Gooler, D. M. (1990) Neural basis of sound pattern
Grusser, O.-J., Schaible, D. & Vierkant-Glathe, J. (1970) A quantitative
recognition in anurans. Progress in Neurobiology 34:313-29. [SH]
analysis of the spatial summation of excitation within the receptive field
Ferguson, A., Meyers, C. S., Bartlett, R. J., Banister, H., Bartlett, F. C. et
centers of retinal neurons. Pfliigers Achiv der gesamten Physiologic
al. (1940) Quantitative estimates of sensory events. (Final report of the
319;101-21. [O-JG]
committee appointed to consider and report upon the possibility of
Guilford, J. P. (1936) Psychometric methods. McGraw-Hill. [GB]
quantitative estimates of sensory events.) British Association for the
Gundlach, H. (1988a) Fechner scholarship. In: C. T. Fechner and psychology,
Advancement of Science 2:331-49. [aDJM, GAH, JVB]
ed. J. Brozek & H. Gundlach. Passavia University Press. [HG]
Fisher, R. A. (1922) On the mathematical foundations of theoretical statistics.
(1988b) Index psychophysicus. Passavia University Press. [HG]
Philosophical Transactions of the Royal Society of London 222A:309-
(1992) Entstehung und Cegenstand der Psychophysik. (Formation and object
68. [DL]
of psychophysics.) Springer-Verlag, in press. [HG]
(1935) The design of experiments. Oliver & Boyd. [GG] Haber, R. N. (1983) The impending demise of the icon: A critique of the
Fodor, J. A. (1968) Psychological explanation. Random House. [MH] concept of iconic storage in visual information processing. Behavioral and
Foerster, M., Grind, W. van de & Grusser, O.-J. (1977) The response of cat
Brain Sciences 6:1-11. [rDJM]
horizontal cells to flicker stimuli of different area, intensity and
Haller, A. von (1759) Elementa physiologiae corporis humani. Tomus quintus.
frequency. Experimental Brain Research 29:367-85. [O-JG]
Sensus externi interni. Lausanne: Grasset. [O-JG]
Foley, J. M. & Legge, G. E. (1981) Contrast detection and near-threshold
(1768) Anfangsgrunde der Phisiologie des menschlichcn Korpers, vol. 5.
discrimination in human vision. Vision Research 21:1041-53. [DL] (Translated by J. S. Hallen.) Berlin: Voss. [O-JG]
Forbes, E. G. (1971) Tobias Mayer's opera inedita. The first translation of the Harnad, S., ed. (1987) Categorical perception: The groundwork of cognition.
Lichtenberg edition of 1775. Macmillan. [O-JG] Cambridge University Press. [aDJM]
(1972) The unpublished writings of Tobias Mayer. 3 vols. Vandenhoeck & Harper, R. & Stevens, S. S. (1948) A psychological scale of weight and a
Ruprecht. [OJG] formula for its derivation. American Journal of Psychology 61:343-
(1980) Tobias Mayer (1723-1762): Pioneer of enlightened sciences in 51. [MO]
Cermany. Vandenhoeck & Ruprecht. [O-JG] Hartley, D. (1967) Thoughts about the human being, his duty and trust in
Forrin, B. & Cunningham, K. (1973) Recognition time and serial position of Good. In: English materialistic scientists of the eighteenth century.
probed item in short-term memory. Journal of Experimental Psychology Academy of the Pedagogical Sciences of USSR (in Russian). [ANL]
99:272-79. [rDJM] Harvey, O. J. & Campbell, D. T. (1963) Judgments of weight as affected by
adaptation range, adaptation duration, magnitude of unlabelled anchor, Indow, T. (1989) Psychophysics: On the possibility of another approach.
and judgmental language. Journal of Experimental Psychology 65:12- Behavioral and Brain Sciences 12:276-77. [aDJM]
21. [rDJM] Jackson, S. W. (1986) Melancholia and depression from Hippocratic times to
Hecht, S. (1928) The relation between visual acuity and illumination. Journal modern times. Yale University Press. [rDJM]
o/Ceneral Physiology 11:255-81. [O-JG] James, W. (1890) Principles of psychology. Holt, Rinehart & Winston.
Heidelberger, M. (1986) Zur Philosophic der Messung im 19. Jahrhundert. (Reprint 1950, Dover) [rDJM, DAS, MW]
In: Die historische Metrologie in den Wissenschaften, ed. H. Witthoft et (1911) The will to believe and other essays in popular philosophy. Longman,
al. Scripta Mercaturae Verlag. [MH] Green. [LEM]
(1987) Fechner's indeterminism: From freedom to laws of chance. In: The Jameson, D. & Hurvich, L. M. (1961) Complexities of perceived brightness.
probabilistic revolution, vol. 1, Ideas in History, ed. L. KrOger, L. J. Science 133:174-79. [aDJM]
Daston & M. Heidelberger. MIT Press. [arDJM, GG] Jarvilehto, T. (1981) A new combination of methods for the study of neural
(1988) Fechner's Leib-Seele-Theorie. In: G. T. Fechner and psychology, ed. mechanisms of mental activity. In: Soviet-Finnish symposium on
J. Brozek & H. Gundlach. Passavia Universitatsverlag. (MH] psychophysiology, Moscow. Finnish-Soviet Committee on Scientific
(1993a) Die innere Seite der Natur: Custav Theodor Fechners Technical Cooperation. [ANL]
wissenschaftlich-philosophische Weltauffassung. Vittorio Klostermann (in Jastrow, J. (1888) A critique of psycho-physic methods. American Journal of
press). [arDJM, MH] Psychology 1:271-309. [aDJM]
(1993b) Gustav Theodor Fechner's conception of the unity of nature and Jesteadt, W., Wier, C. C. & Green, D. M. (1977) Intensity discrimination as
mind. In: Romantic sciences and unity of knowledge: Science 1790-1840, a function of frequency and sensation level. Journal of the Acoustical
ed. M. Bossi & S. Poggi. Kluwer (in press). [MH] Society of America 61:169-77. [aDJM]
Heinemann, E. C. (1955) Simultaneous brightness induction as a function of Jones, L. A. (1986) Perception of force and weight: Theory and research.
inducing- and test-field luminances. Journal of Experimental Psychology Psychological Bulletin 100:29-42. [HER]
50:89-96. [aDJM] Kaas, J. (1989) The segregation of function in the nervous system: Why do
Heller, O. (1990) Scaling and orientation. In: Fechner Day 90 - Proceedings sensory systems have so many subdivisions? In: Contributions to sensory
of the Sixth Annual Meeting of the International Society for physiology, vol. 7, ed. W. D. Neff. Academic Press. [SH]
Psychophysics, ed. F. Miiller & F. R. G. Wuraburg. [GB] Kaestner, A. G. (1762) Elogium Tobiae Mayeri. Gottingen: Schulz &
Helmholtz, H. von (1856-66/1962) Treatise on physiological optics. Rosenbusch. [O-JG]
(Translated by J. P. C. Southall.) Dover. [aDJM] Katz, D. (1925/1989) The world of touch. (Translated and ed. by L. E.
(1866) Handbuch der physiologischen Optik. Leipzig: Voss. [O-JG] Krueger.) Erlbaum. [aDJM]
(1887/1921) Zfihlen und Messen, erkenntnistheoretisch betrachtet. In: Kim, D. O., Chang, S. O. & Sirianni, J. G. (1980) A population study of
Schriften zur Erkenntnistheorie, ed. P. Hertz & M. Schlick. Julius auditory nerve fibers in unanesthetized decerebrate cats: Response to
Springer. [MH] pure tones. Journal of the Acoustical Society of America 87(4):1648-
Henmon, V. A. C. (1906) The time of perception as a measure of differences 55. [SH]
in sensation. Archives of Philosophy, Psychology and Scientific Methods Kimchi, R. (1992) Primacy of holistic processing and global/local paradigm: A
8:1-75. [WMP] critical review. Psychological Bulletin 112:24-38. [rDJM]
Herbart, J. F. (1812) Psychologische Untersuchung ueber die Staerke einer Knibestol, M. & Vallbo, A. B. (1980) Intensity of sensation related to activity
gegebenen Vorstellung, als Function ihrer Dauer betrachtet. In: Johann of slowly adapting mechanoreceptive units in the human hand. Journal of
Friderich Herbart's Schriften zur Psychologic, Drifter Theil, ed. G. Physiology 300:251-67. [aDJM]
Hartenstein (1989). Leipzig: Voss. [rDJM, RAMC] Kohler, A. (1886) Ueber die hauptsachlichsten Versuche einer
(1816/1891) Lehrbuch der Psychologie. (Translated by M. K. Smith as: A mathematischen Formulirung des psychophysischen Gesetzes von Weber.
textbook in psychology.) D. Appleton. [rDJM] Philosophische Studien 3:572-642. [aDJM]
(1824/1892) Psychologie als Wissenschaft neu gegriindet auf Erfahrung, Kohler, W. (1969) The task of Cestalt psychology. Princeton University
Metaphysik und Mathematik. In: J. F. Herbart's samtliche Werke Part 1 Press. [rDJM]
(vol. 5) & Part 2 (vol. 6) ed. K. Kehrbach. Langensalza: H. Beyer & Konig, A. (1897) Die Abhangigkeit der Sehscharfe von der
S6hne. [rDJM] Beleuchtungsintensitat. Berichte der Koniglich Preussischen Akademie
Hering, E. (1874) Zur Lehre vom Lichtsinn V. Grundziige einer Theorie des der Wissenschaften 1897:559-75. [O-JG]
Lichtsinnes. Sitzungsberichte der Kaiserlichen Akademie der Krantz, D. H. (1969) Threshold theories of signal detection. Psychological
Wissenschaften in Wien. Mathematisch-Noturwissenschaftliche Classe, Review 76:308-24. [aDJM, DL]
Abth. Ill 69:85-104. [O-JG] (1972) Visual scaling. In: Visual psychophysics, vol. 7/4, Handbook of
(1875) Zur Lehre von der Beziehung zwischen Leib und Seele. I. sensory physiology, ed. D. Jamson & L. E. Hurwich. Springer. [O-JG]
Mittheilung. Ober Fechner's psychophysisches Gesetz. Sitzungsberichte Krantz, D. H., Luce, R. D., Suppes, P. &Tversky, A. (1971) Foundations of
der Kaiserlichen Akademie der Wissenschaften, Wien. Mathematisch- measurement, vol. 1. Academic Press. [RDL, WMP]
Natunvisscnschaftliche Klasse. Abth. Ill 72:310-48. [aDJM, MH] Kries, J. von (1882) Ober die Messung intensiver Grossen und fiber das
(1920) Crundziige der Lehre com Lichtsinn. Springer. [O-JG] sogenannte psychophysische Gesetz. Vierteljahrsschrift fur
(1931) Wissenschaftliche Abhandlungen, 2 vols. Thiem. [O-JG] wissenschaftliche Philosophic 6:257-94. [arDJM, MH, JVB]
Hess, C. & Pretori, H. (1894) Messende Untersuchungen uber die (1886) Prinzipien der Wahrscheinlichkeitsrechnung. Mohr. [MH]
Gesetzmassingkeit des simultanen Helligkeits-Contrastes. Archiv fur Krueger, L. E. (1989a) Reconciling Fechner and Stevens: Toward a unified
Ophthalmologie 40:1-27. (Translated [1969] by H. R. Flock & J. H. psychophysical law. Behavioral and Brain Sciences 12:251-320. [aDJM,
Tenney). Technical Report FLP.-l, York University. [aDJM] JVB, LEK, DJW]
Hick, VV. E. (1952) On the rate of gain of information. Quarterly Journal of (1989r) Psychophysical law: Keep it simple. Behavioral and Brain Sciences
Experimental Psychology 4:11-26. [rDJM] 12:299-320. [aDJM]
Hoffman, W. C. & Dodwell, P. C. (1985) Geometric psychology generates the (1991) Psychophysical law: Taming the cognitive and chaotic aspects.
visual Cestalt. Canadian Journal of Psychology 39:491-528. [rDJM] Behavioral and Brain Sciences 14:193-204. [aDJM, LEK)
Hooke, R. (1705) The posthumous works. London: Smith & Walford. [O-JG] Kruskal, J. B. (1964) Nonmetric multidimensional scaling: A numerical
Hornstein, G. A. (1988) Quantifying psychological phenomena: Debates, method. Psychometrika 29:115-29. [WMP]
dilemmas, and implications. In: The rise of experimentation in American Kuhl, P. K., Williams, K. S., Lacerda, F., Stevens, K. N. & Lindblom, B.
psychology, ed. J. G. Morawski. Yale University Press. [GAH] (1992) Linguistic experience alters phonetic perception in infants by 6
Houtsma, A. J. M., Durlach, N. I. & Braida, L. D. (1980) Intensity months of age. Science 255(5044):606-08. [SH]
perception XI. Experimental results on the relation of intensity resolution Kiilpe, O. (1893) Grvndriss der Psychologie. Herzil. [RAMC]
to loudness matching. Journal of the Acoustical Society of America LaMettrie, J. O. (1751) L'homme machine. English translation in: Man a
68:807-13. [LEK] machine [1912], ed. C. C. Bussey. Open Court. [O-JG]
Hueck, A. (1840) Ober die Grenzen des Sehvermogens. Miillers Archie fur Laming, D. (1984) The relativity of "absolute' judgments. British Journal of
Anatomic, Physiologic und Wissenschaftliche Medizin 1840:82-97. Mathematical and Statistical Psychology 37:152-83. [aDJM]
[O-JG] (1985) Some principles of sensory analysis. Psychological Review 92:462
Hurvich, L. & Jameson, D. (1966) The perception of brightness and darkness. 85. [aDJM]
Allyn & Bacon. [aDJM] (1986) Sensory analysis. Academic Press. [aDJM, DL]
lllich, 1. (1985) H2O and the toaters of forgetfulness. Heyday Books. (1988) A reexamination of Sensory analysis. Behavioral and Brain Sciences
[DAS] 11:316-39. [rDJM]
(1989) Experimental evidence for Fechner's and Stevens's Laws. Behavioral Macmillan, N. A. (1987) Beyond the categorical/continuous distinction: A
and Brain Sciences 12:277-81. [aDJM, RTl psychophysical approach to processing modes. In: Categorical Perception,
(1991) Reconciling Fechner and Stevens? Behavioral and Brain Sciences ed. S. Hamad. Cambridge University Press. [aDJM]
14:188-91. [aDJM, MO] Macmillan, N. A. & Creelman, C. D. (1991) Detection theory: A user's guide.
Leary, D. E. (1980) The historical foundation of Herbart's mathematization of Cambridge University Press. [arDJM, PW]
psychology. Journal of the Behavioral Sciences 16:150-63. [rDJM] Marcel, A. J. (1983) Conscious and unconscious perception: Experiments on
Lebedev, A. N. (1990) Cyclic neural codes of human memory and some visual masking and word recognition. Cognitive Psychology 15:197
quantitative regularities in experimental psychology. In: Psychophysical 237. [aDJM]
explorations of mental structures, ed. H.-G. Geissler. Hogrefe & Markowitz, J. & Swets, J. A. (1967) Factors affecting the slope of empirical
Huber. [ANL] ROC curves: Comparison of binary and rating responses. Perception and
Lebedev, A. N. & Myshkin, I. Yu. (1989) Contemporary problems of inner Psychophysics 2:91-150. [aDJM]
psychophysics. In: Fechner Day 90. Proceedings of the Sixth Annual Marks, L. E. (1974) On scales of sensation: Prolegomena to any future
Meeting of the International Society of Psychophysics, ed. F. Muller. psychophysics that will be able to come forth as science. Perception and
Universitat Wuerzburg. [ANL] Psychophysics 16:358-76. [aDJM]
Legge, G. E. (1978) Sustained and transient mechanisms in human vision: (1979) A theory of loudness and loudness judgments. Psychological Review
Temporal and spatial properties. Vision Research 18:69-81. [DL] 86:256-85. [aDJM]
Legge, G. E. & Foley, J. M. (1980) Contrast masking in human vision. Marks, L. E., Borg, G. & Ljunggren, G. (1983) Individual differences in
Journal of the Optical Society of America 70:1458-71. [aDJM] perceived exertion assessed by two new methods. Perception and
Lehmann, A. (1886) Ueber die Anwendung der Methode der mittleren Psychophysics 34:280-88. [GB]
Abstufungen auf den Lichtsinn. Philosophische Studien 3:497 Marks, L. E., Szczesiul, R. & Ohlott, P. (1986) On the cross-modal
533. [aDJM] perception of intensity. Journal of Experimental Psychology: Human
Lewin, K. (1936) Principles of topological psychology. McGraw-Hill. [LEM] Perception and Performance 12:517-34. [aDJM]
Libet. B.. Alberts, W. W., Wright. E. W. Jr. & Feinstein, B. (1967) Marshall, M. E. (1982) Physics, metaphysics and Fechner's psychophysics. In:
Responses of human somatosensory cortex to stimuli below threshold for The problematic science: Psychology in nineteenth century thought, ed.
consciousness. Science 158:1597-1600. [aDJM] W. R. Woodward & M. G. Ash. Praeger. [rDJM]
Lichtenberg, G. C. (1775) Tobiae Mayeri opera inedita I. Gottingen: (1990) The theme of quantification and the hidden Weber in the early work
Vandenhoeck. [O-JG] of Gustav Theodor Fechner. Canadian Psychology 31:45-53. [DAS]
Lim, J. S., Rabinowitz, W. M., Braida, L. D. & Durlach, N. I. (1977) Massaro, D. W. (1987) Categorical partition: A fuzzy-logical model of
Intensity perception VIII. Loudness comparisons between different types categorization behavior. In: Categorical perception, ed. S. Hamad.
of stimuli. Journal of the Acoustical Society of America 62:1256-67. Cambridge University Press. [aDJM]
Erratum: (1978) 63:1233. [LEK] Matthews, B. H. C. (1931) The response of a muscle spindle during active
Link, S. W. (1992) The wave theory of difference and similarity. contraction of a muscle. Journal of Physiology 72:153-74. [rDJM]
Erlbaum. [rDJM, SL] Mayer, T. (1741) Neue und allgemeine Art, alle Aufgaben aus der Geometrie
Lipetz, L. E. (1971) The relation of physiological and psychological aspects of vermittelst der geometrischen Linien leichter aufzulosen; insbesondere
sensory intensity. In: Handbook of sensory physiology, vol. 1, ed. W. R. wie alle regulare und irregulare Vielecke, davon ein Verhaltnis ihrer
Loewenstein. Springer-Verlag. [arDJM, LEK] Seiten gegeben in den Circul geometrisch sollen eingeschrieben werden,
Lockhead, G. R. (1992) Psychophysical scaling: Judgments of attributes or samt einer kurzen hierzu notigen Buchstaben-Rechenkunst und
objects? Behavioral and Brain Sciences 15:543-601. [GRL] Geometrie. Als Erstlinge an das Licht gestellet von Tobias Mayern.
Lotze, H. (1887) Metaphysic, vol. 2 (Translated by B. Bosanquet). Clarendon Esslingen: Mantlern. [O-JG]
Press. [DAS] (1745) Mathematischer Atlas, in welchem auf 60 Tabellen alle der
Lowrie, W., trans, and ed. (1946) Religion of a Scientist; Selections from Mathematik vorgestellt, und nicht allein iiberhaupt zu bequemer
Gustao Th. Fechner. Pantheon Books. [rDJM] Wiederholung, sondern auch den Anfangern besonders zur
Luce, R. D. (1959) On the possible psychophysical laws. Psychological Review Aufmunterung durch deutliche Beschreibung und Figuren entworfen
66:81-95. [DJW] werden. Augsburg: Pfeffel. [O-JG]
(1963) A threshold theory for simple detection experiments. Psychological (1754a) Observationes astronomicae A. 1753 Gottingae habitae. Commentarii
Review 70:61-79. [aDJM, DL] Societatis Regiae Scientiarium Cottingensis 3:441-51. [O-JG]
(1972) What sort of measurement is psychophysical measurement? American (1754b) Experimente uber die Scharfe des Gesichtsinnes. Cottingische
Psychologist 27:96-106. [aDJM] Anzeigen von gelehrten Sachen (20):401-2. [rDJM, O-JG]
(1977) A note on sums of power functions. Journal of Mathematical (1755) Experimenta circa visus aciem. Commentarii Societatis Regiae
Psychology 16:91-93. [RDL] Scientiarium Cottingensis 4:97-112. [rDJM, O-JG]
(1986) Response times: Their role in inferring elementary mental (undated, ca. 1757) Von den Farben. Unpublished manuscript, Gottingen
organization. Oxford University Press. [rDJM, RDL] University Library, Mayer 15, 13. [O-JG]
Luce, R. D. & Edwards, W. (1958) The derivation of subjective scales from (1758) De affinitate colorum (Uber die Verwandtschaft der Farben).
just noticeable differences. Psychological Review 65:222-37. [aDJM] Cottingische Anzeigen von gelehrten Sachen, 147. Stuck (9
Luce, R. D., Krantz, D. H., Suppes, P. & Tversky, A. (1990) Foundations of Dezember): 1385-89. [O-JG]
measurement, vol. 3. Academic Press. [RDL] (1759) Die verschiedenen Grade des Schattens zu bestimmen, wenn man
Luce, R. D. & Narens, L. (1987) Measurement scales on the continuum. eine einzige Farbe zugleich zum ganzen und halben Schatten nehmen
Science 235:1527-32. [aDJM, DJW] will. Unpublished manuscript January 22, 1759, Gottingen University
Mach, E. (1866) Uber den psychologischen Effect raumlich vertheilter Library, Mayer 13. [O-JG]
Lichtreize. Sitzungsberichte der Kaiserlichen Akademie der (1775) De affinitate colorum. In: Tobiae Mayeri opera inedita /, ed. G. C.
Wissenschaften in Wien. Mathematisch-Naturwissenschafliche Klasse, Lichtenberg. Gottingen: Vandenhoeck. (English translation in Forbes
Abth II. 54:131-44. [rDJM] 1971.) [O-JG]
(1868) Uber die physiologische Wirkung raumlich vertheilter Lichtreize. McBride, R. L. & Anderson, N. H. (1991) Integration psychophysics in the
Sitzungsberichte der Kaiserlichen Akademie der Wissenschaften, in Wien. chemical senses. In: Contributions to information integration theory, vol.
Mathematisch-Naturwissenschoftliche Klasse Abth U 57:11-19. [O-JG, 1, Cognition, ed. N. H. Anderson. Erlbaum. [NHA]
MH] McElree, B. & Dosher, B. A. (1989) Serial position and set-size in short-term
(1883/1960) The science of mechanics: A critical b historical account of its memory: The time course of recognition. Journal of Experimental
development, 6th ed. (after the 9th German edition). Open Court. [JVB] Psychology: General 118:346-73. [rDJM]
(1886/1914) The analysis of sensations and the relation of the physical to the McKenna, F. P. (1985) Another look at the "new psychophysics." British
psychical. (Translated by C. M. Williams & S. Waterlow.) Open Journal of Psychology 76:97-109. [HER]
Court. [aDJM] McNicol, D. (1972) A primer of signal detection theory. George Allen &
(1886/1919) Die Analyse der Empflndungen und des Verhaltnis des Unwin. [aDJM)
Physischen zum Psychischen, 8th ed. Gustav Fischer. [JVB] McNicol, D. & Stewart, G. W. (1980) Reaction time and the study of
(1886/1922) Die Analyse der Empfindungen und das Verhaltnis des memory. In: Reaction times, ed. A. T. Welford. Academic
Physischen zum Psychischen, 9th ed. Gustav Fischer. [MH] Press. [rDJM]
(1891) Some questions of psycho-physics. The Monist 1:393-400. [rDJM] McReynolds, P. & Ludwig, K. (1977) Psychometrics in the seventeenth
(1896/1919) Die Principled der Warmelehre. Historisch-kritisch entwickelt, century: The personology of Christian Thomasius. Paper presented at the
3rd ed. Johann Ambrosius Barth. [MH] 9th Annual Meeting of Cheiron, the International Society for the History
Woodworth, R. S. & Schlosberg, H. (1954) Experimental psychology. Wyszecki, G. (1986) Color appearance. In: Handbook of perception and
Methuen. [aDJM] human performance, vol. 1, ed. K. R. Boff, L. Kaufman & J. R. Thomas.
Wundt, W. (1862) Beitrdge zur Theorie der Sinneswahrnehmung. Leipzig: Wiley. [aDJM]
C. F. Winter. (The introduction, titled "On the methods of psychology," Yilmaz, H. (1967) Perceptual invariance and the psychophysical law.
is translated in: Classics in psychology ed. T. Shipley. Philosophical Perception and Psychophysics 2:533-38. [arDJM]
Library 1961.) [rDJM] Zwislocki, J. J. & Goodman, D. A. (1980) Absolute scaling of sensory
(1874) Cnindziige der physiologischen Psychologie, 1st ed. W. magnitudes: A validation. Perception and Psychophysics 28:28-
Engelmann. [rDJM] 38. [aDJM]
PSY1 eta
PSYchologistsforthe
Ethical Treatment of Animals