54 THE BEHAVIORAL DATA LANGUAGE
CONCLDSIO:N
Behaviorist conceptions of cmpiricalness and objectivity in observation carr
over to behaviorist views on the behavioral data language. However, the a:
teiupt to insure objectivity and empiricalness by restricting the behaviora1 data
language to physical descriptions fails for a number of reasons. This rejection
I of a purely physical language opens the way for psychological properties, es-
' pecially those distilled by a functional approach to behavioral analysis. Func-
tional properties n1ust still be objective, and one way to measure their objec-
tivity is through the intersubjective agreenient they achieve.
Intersubjective agreen1ent is also a criterion for judging other candidates
for acceptance into the behavioral data language. Descriptions that are infer-
ential, intensional, purposive, molar, or in action language are suspect, partly
because of this criterion. Because these qualities, as well as the degree of in-
tersubjective agreement that they achieve, vary and are not all-or-11one, many
decisions about acceptability into the behavioral data language are not clear-
cut and often depend on a context of tradeoffs between intersubjective agree-
ment and other desirable qualities.
'.'.
1'
'
.. Ii,I:i'
"'(t:
''
II
CHAPTER 4
Theoretical Concepts
Be/iavioris,n admits a 11aricry ,,[ rlzcorcriml co11uprs i1110 rlic sdc11cc of /,cliauioz-. These
collcepts diffe1· in how closely tlicy arc liuked to rlic he/wi,iora/ dara la11gua.f!e. Trnns-
fanuations within the behavioral d,1/a la11g1w.\!l' :7ri· scmrcly linked mid pose 110 prob-
lem. State, dispositional, mid operationally de.fined co11cepts, liowePer, pro11idc 011/)'
partial defi11itiolls since they arc illdetem1i11ate and, i11 t/ic case of t/ie _first two, opcu
as well. !11 practice, eve11 operatio11ally de{i11cd w11cepts arc also open.
hiterve11i11g variables tie togr:tha a sci <:f imcrrclared i11depc11dc1U uarial,/es witb a
set of iflterrelated depend cm varia/,/es. BdiaPiorisrs d({fer as to wbetlier tlie hcmjits <:f
;11tervelli11g Pariah/es outwe(f!h their da11.~ers. This deliatc ccmcrs 011 the taaical iss11e
of ivliether concepts facilitate or in/ii bit research a11d 1·xpla11atio11 rat ha than 011 the
plii/osop/iical respectability of in£apr:11ili.i! 11ari<1l1/es.
A similar debate arises OPer tlic admission of /iypot/ietical co11stmcts, tlieoretical
coucepts which explicitly refer to wwbsenn·d emities. Propo11e111s m:~11e tbar hypo-
t/ietical constrncts are necessary ([ the scie11cc •:f /1diaPior is to be red11cib/c, explmia-
lor}', and fertile. Opponents see these constrncts as di/11till.i! the <'l,jcaiuc and empirical
11al11re of the science. A/tlwu_R/, 11011c of these mxumellts is co11c/11siPe and many de-
pe11d 011 indiPidual intuitio11s abom liow research is best pe1:for111ed, most be/iaPiorists
do incorporate hypothetical constructs in their theorics. Neucrtlieless, they implicitly
impose restrictions 011 the posw/atcd featrn·cs of these cons/mets so rlrnr the dan.f!crs arc
reduced.
For the reasons discussed in the preceding chapter, the boundaries of the be-
havioral data language are not sharply defined. Yet, wherever they are drawn,
the study_ of behavior must exceed them to establish a science. It n1ust tran-
scend the immediate mon1entary observation to impose or discover coher-
ence in its subject matter. Therefore, concepts must be introduced which do
not q~alify as terms in the behavioral data language.
This requiren1ent, however. raises serious problems for the behaviorist
program. B_y tying the behavioral data language to observation and intersub-
 I!. '
!

'· 56 THEORETICAL CONCEPTS

jective agreeinent, behaviorists ensure that .tenns within this language 111:et
the twin criteria of empiricalness and obJecttvtty. To mtroduce tenns outside
this language is to risk including subjective and speculative notions. There-
fore behaviorists insist that theoretical concepts be linked securely to the be-
havi,oral data language. They disagree, however, on the nature of this link-
age, differing on how loose a linkage is permissibl':. 1 This issue forn1s the
next n1ajor node of the behaviorist conceptual tree diagram.
FlHST-OHDEH CONCEPTS 57
two sti1nuli, S1 and S2, and only two responses, R 1 and R 2. Suppose it is found
'.hat at tunes S1 f'roduc~s R 1, but on other occasions it produces R . Likewise
2
1t 1s found that S2 occas1onally_produces R 1 and at other tin1cs it produces R •
2
At first glance, no lawfulness 1s apparent. However, assume further that it is
found that A's response to a stimulus depends on its state, T(A), which can
take on two possible values, T 1(A) or T 2(A). Lawfulness n1ight be introduced
as follows:

T,(A): s, R,
FIRST-ORDER CONCEPTS T,(A): s R,
T 2 (A): s, R,
Transformations Within the Behavioral Data Language T,(A): S2 R,
Many theoretical concepts are unproblcn1atic because they repn:sent.. mere
Laws of tl~is fonn _arc c~mmon. Often Tis referred to as the "parameter"
transforn1ations within the behavioral data language. "Rate of response 1s an
for the equation rdatmg stunulus and response. Note, however, that the tern,
exan1ple of a concept which integrates over multiple observations but is_ not
considered the paratncter is n1erely a matter of convention. and there is noth-
the equivalent of the set of then1. Similarly, the "average nun1ber of subjects
ing inherently "paran1etric" about a state. A stimulus, too, can function as a
turning right," when not a whole number, represents a datu1n that cannot para1neter in the above model:
I possibly be observed. . . .
Because none of these concepts refers to any entity or process existing m
.
S,: T 1 (A R,
son1e unobserved realm and because they can all be defined solely by terms S,: T,(A R,
within the behavioral data language, they pose no methodological problems. 2 S2: T1(A) R,
Indeed, by some definitions of the behavioral data language, th_ese concepts S2: T 2 (A R,
are included in it. Under the other definitions, their unproblen1at1C nature and
State variables thus serve the function of restoring lawfulness to stimulus-re-
their proximity to the data language should adn1it then1 to what might be
sponse connections when a sin1ple one-to-one correspondence between stim-
labeled the "extended behavioral data language." ulus and response is lacking. 4
' li,

States and Dispositions
In the simplest case, behavioral laws permit the inference of statctncnts about
responses from staten1ents about the external environn1ent. The form of s~ch
laws would be: "Given stin1t11i a, b, c, . . . then responses x, y, z. . . . It
is clear from elen1entary considerations about behavior, however, that laws
of this form are inadequate. More often than not, the response .is not uniq~ely
detennined by external environmental conditions alone. Consider a few sim-
ple examples: a dog's response when food is placed before 1t depends on
whether it is hungry; a fe1nale rat's response to a n1ale rat depends on its es-
trous cycle; whether a tap to the knee elicits a jerk depends on the subject's
health. In all these illustrations, the response depends on factors other than ~n
external stimulus. These factors arc "states" of the organisn1 and include its
age, health, species, hunger, wakefulness, and emotion. . havioral data language.
A state then is a variable determining which stimulus-response laws are ll1
effect. 3 The functions served by state variables can be demonstrated by an
idealized model. Assume a universe consisting of only one organism, A. only
States arc not observable in the san1e way as stinn1li and responses. For
example, the dog is thought of as hungry even when no food-related stimuli
or resp01_1scs are present. Thus an organism may be reported to be in state T
at a particular t1n1e even though nothing observed at that time indicates the
state. Hence, terms denoting states do not qualify for the behavioral data lan-
guage.
A common approach to states is to adn1it them to the science but to define
them ~olely by ter111s in the behavioral data language. The organis 1n is said
to be m the state 1f and only if certain stimuli produce certain responses. In
the model above this can be fonnulated:
T,(A) iff[(S,--------,R,) or (S, 7_------.R,)l
That is, A is in state TI if and only if either S 1 produces R 1 , or S, produces
Rz. _The definition contains only sti1nulus and response terms fro-m the be-
b Und~rstood this way, states are dispositions, concepts commonly used in
0
th sc'.ence and everyday speech. 5 To say that an object is soluble, for ex-
ample, 1s not to say that it is in the process of dissolving, but that when placed
 I ,
58 THEORETICAL CONCEPTS
in water it will dissolve. In general, to attribute a disposition to a object is to
state that under certain conditions, called "test conditions," certain outcomes
will result. Thus states (e.g .• hunger) can be construed as dispositions in which
the sti1nuli are the test conditions (e.g., food). and the responses are the out-
I con1es (e.g., eating).
I, Note that for both dispositions and states, the conditional is assmned to
I .
1'I! hold only when certain other conditions obtain: the soluble substance is not
expected to dissolve if the water is frozen. These other conditions are covered
I operanomst n1ethodology to concepts as yet unquantified. thereby ensuring
;: by the rctcris paril,11s clause i1nplicitly included in the definition of a disposi-
tional term, although '
these conditions cannot be exhaustivdv, listed. Since both
test conditions (stinrnli) and outcomes (responses) are public and observable,
the attribution of a state n1eets behaviorist criteria of objectivity and empiri-
I' ' calness,
Operationally Defined Concepts
A 1nethodological procedure ensuring that all introduced concepts possess a
siniilar linkage to observations is offered by operationism. Bridgman (1927)
suggests that a concept be identified with the operations used to n1casure it:
To find the length of an object, we have to perform certain physical operations. The
concept of length is therefore fixed when the operations by which length is measured
arc fixed: that is, the concept of length involves as much as and nothing more than a
set of operations; tl,c co,uept is sy11011ymc11s with tire corresp"'1di11_l? s,·t ,:f opaarit>ns. (p, 5)
In adapting operational principles to psychology Stevens (1939) asserts:
A term denotes something only when there are concrete criteria for its applicability;
and a proposition has empirical meaning only when the criteria of its truth or falsity
consist of concrete operations which can be performed upon demand. . . . When we
attempt to reduce complex operations to simpler and simpler ones, we find in the end
that discrimination, or differential response, is the fundamental operation. (p. 228)
Operationist principles arc neatly congruent with behaviorist aspirations. If
it is stipulated that operations 1nust be publicly observable (i.e., introspection
is excluded), repeatable. and specified precisely enough for others to carry
then1 out, then there will be intersubjective agreement and verifiability for
6
the operations, their results, and consequently the application of the concept.
As a prescription, opcrationisn1 stipulates that concepts be introduced only
by operational dcfinitions-i.e., in terms of operations used to nlcasurc the
concept. As a test, operationisn1 states that concepts that cannot be given an
operational definition arc scientifically unacceptable.
For application in behaviorist psychology, operationis1n requires certain
modifications. In Bridgn1an's original forn1tdation, operationisn1 concerns the
relationship between concept and n1easurcment in physics. As cmnpared with
I' physics, psychology has developed very few concepts which can be 1neasured
,I
"ii
F[RST-OHDEll CONCEPTS 59
in any precise ~ay. Therefore, operational definitions in psychology often give
only ,the ..operanons used to d:tcct ..,md apply the concept. For example: the
con~cpt_ h~vmg a sensation of red may be operationally defined by the op-
'.:rat1.~ns, us.:d to set up a perceptual experiment in which the subject reports
I sec red. Herc the operations define when the concept "'having a sensation
of red" may .be applied
• . to th·c su b.~cct
, I1ut d o not me
· Ju d c measunng ' , opera-
7
tions .. By this broadenmg of operationisn1, behaviorists extend the use of
that a lar~cr range of concepts meet behaviorist standards of acceptability.
Although_ m prmciple opera~1omsm seems straightforward, in practice its
apphcanon is not without a111b1gmty. ()ne source of indeterminateness is the
sense m _winch _the ~oncep: is "synonymous with" the operations. Taken lit-
erally
" this 1s. unmtdhg1bk.
.. . It would mean ~ for insta 11 c,·.... tl1..at "d.st
1 ancc " 111ca11s
the operatton of choosmg a standard unit, the operation of markina off a
rod with these umts, the operation of placing the rod next to an objec~ etc. ..
A more ac.ccptablc_ mtcrpretat1on of operationisn1 would be that the con-
cept (c. g .• distance) is_ defined as the result of a set of specified operations
(e.g., n1easuren1cnt with a ruler). In particular. to say that a certain object,
A, possesses a property, T, at a particular value, N (e.g., "The distance to
the table 1s 90 nn") 1s to specify the results ("pointer readings") of those op-
eranons. Tln1s,
(T(A)=Nl iff ((operations on A\------->NI
When T i~ a propeny which is not measurable, then N represents a two-val-
ued function 111d1catmg whether or not A possesses T.
Fonnulatcd in this way, operationally defined concepts arc dispositions. The
operations ar~ equivalent to the test conditions, and the results of the opcra-
twns arc eqtuvalent to_ the test outcomes.'' What operationis1n adds to the
analysis of dispositions 1s the principle that all concepts are dispositional, even
ones hke weight and distance.
Partial Definitions
States, dispositions, and operationally defined concepts appear, therefore, to
have a con1n1on logical form:
T(A) iff (S(Al----->RI
A serious problc1n with all three types of concepts is that the definition is in
the for~1 of a _conditional: "S(Al------➔ R." By truth functional logic,
a cond1t1onal _is true !f the antecedent of the conditional is false. Thcref~re,
T(A) is true 1f S~A) 1s false. For example, any rock not placed in water is
soluble by this kmd of definition of a disposition. 111 Therefore, Carnap sucr-
gests an alternate formulation: "'
 60 THEORETICAL CONCEPTS
FIRST-ORDER CONCEPTS 61

S(Al----➔IR iff T(A)I ad111ission of concepts tied to the behavioral data language by a link weaker
than explicit definition.
This says that if the test condition is in effect, then an object may be said to
have the disposition if and only if the required outcome occurs, In the ex-
ample, "soluble" would be defined as follo':"s: If an obj.ect is .placed in water Partial Definitions and Operational Definitions
I,
then that object may be said to be soluble 1f and only 1f It dissolves. Under Operational definitions arc also indeterminate since they assign no value when
p
''
this forn1ulation the rock not in water ceases to be a problem, for it docs not the operation is not in effect. However, in contrast to reductive chains, op-
'' satisfy the definition. Carnap calls his formula a "bilateral reduction sen- erational definitions are meant to be closed definitions. A concept is defined
!1 11
: tence," and he rccomn1cnds its use for introducing dispositional terms. by only one operation, and different operations define different concepts. If
Bilateral reduction sentences have logical properties with in1portant i111pli- a particular concept is defined by a certain 1neasuring operation, and a second
cations for introducing terms into behaviorist psychology. First, the state or operation is discovered that apparently measures the concept as well. strict
disposition T is defined only if S(A) obtains, and the application of the con- opcrationism demands that the second operation be considered as defining a
cept is therefore indeterminate for all those instances in which S(A) docs not second concept rather than as an alternate measure of the first. Thus, strict
obtain. This area of indeterminacy leaves open the possibility that test con- operational definitions do not share the open texture of Carnap's reductive
ditions other than S(A) can be used when S(A} docs not obtain. Thus a sec- chains and therefore cannot function in concept formation as reductive chains
ond characteristic of introduction by bilateral reduction sentences is that it is do. 13
"open" in the sense that further bilateral reduction sentences-i.e., test con- Behaviorists treat this issue in several ways. The heroic response is to sac-
ditions and outcomes for the disposition-can be added as they arc discov- rifice the integrative role of concepts in favor of the protection offered by
ered, thereby reducing the area of indeterminacy. strict operationism. This decision means that for each operation, a separate
' As additional bilateral reduction sentences accrue, the result is what Carnap set of empirical laws is required to relate a concept to other variables. Strict
'! calls a "reductive chain" in the form: operationism demands that a law relating two nonoperationalized variables,
!
iff T(A)I
X, which can be measured in m ways, and Y, which can by measured in n
S1 (A IR,
S2 (A {Ro iff T(A)I ways. must be construed not as a single general law but as 111 x 11 separate
S,(A IR., iff T(A)I laws. Furthermore, according to strict operationism, the 111 methods thought
• to measure X, cannot be viewed as different ways of disclosing the same
• magnitude. Instead, each method defines a different variable. The resulting 111
• variables arc then considered to be related by empirical laws over an observed
. range rather than by the fact that they all reveal some common aspect of na-
• ture. 14 Although such a conservative approach creates a science less vulnera-
S,.(A) (Rn iff T(A)I ble to the need for major revision with new discoveries, it clearly pays a price
where S, represents a test condition, and R, represents the associated out- in loss of simplicity and integration. 15 If the convergence of concepts and laws
come. Indeed, most dispositions in everyday speech are n1anifcst in more than is a prerequisite and consequence of theory development then operationism
one way. For example, to say that Jones knows algebra is to attribute to hin1 may shield science only by inhibiting its growth.
an ability to do many things. Moreover, the attribution is open in that other Other behaviorists, unwilling to sacrifice either integration or caution, try
manifestations may be added as, say, he learns to use a c01nputer and can to fuse the two. This feat is accomplished by suggesting a method for inte-
therefore demonstrate his knowledge of algebra in new ways. This open na- grating operations into a single concept. According to this approach, con-
ture of dispositional concepts permits them to play an integrative role in con- cepts are initially to be defined by only one operation. If continued experi-
mental research reveals that two operations consistently yield the same results,
cept formation. . . .
i The open texture of reductive chains and the1r mdetermmacy mean that
12
then the two operations may be considered as measuring and defining the
same concept. 1" Additional operations can be added to the concept by the same
the reductive chain provides only a "partial" definition of a dispositional term.
:I Ii The reductive chain is therefore neither an explicit operational definition of means, as further research warrants. This approach is appealing because it ap-
pears to retain the virtues of operationism without sacrificing theoretical con-
i· I the term nor a "translation" of the term into the behavioral data language.
vergence. However, this approach raises difficulties of its own.
Acceptance of dispositional terms into the behaviorist science thus means the
,i I
I .

1 I
1li ~!
 ,.' 62 THEORETICAL CONCEPTS
First, the suggested n1cthod for integrating operations undermines the very
rationale for Bridgman's opcrationism. Relativity theory had shown that physics
was led to a crisis by the assumption that its basic concepts, such as tin1e and
space, can be measured in various ways all revealing the same underlying
physical property. This asstm1ption had to be abandoned with the discovery
that the equivalence of different measurement procedures breaks down at the
extremes of the range and that measurement is dependent on the conditions
of the observation. Opcrationism is intended as a replace1nent. Thus, the be-
haviorist suggestion that two operations n1ay be said to measure and define
the san1e concept if they yield the san1e results is in effect to make the induc-
tive leap that operationism was incant to forbid. 17 The observed congruence
,I of two measurement operations over a restricted range and under a li1nited
t
set of circumstances docs not permit the identification of the two, according
i
''
I to the strict operational approach.
d A second problem with the behaviorist suggestion arises in determining when
'!
I l the two operations give the same results. If initially each measurement op-
eration defines a distinct concept, then a third operation is needed to define
the concept of "sameness of results." 18 However, the nature of this third op-
eration is never precisely specified. This problem is especially acute when there
is little overlap in area of application for the two operations. For example,
the method of triangulation is commonly used to measure cxtraterrestial dis-
tances where there is no possibility for a comparison with distance as mea-
sured by a rod. Although the problem of defining the third operation may
not prove to be insurmountable, it has yet to be resolved in ways that are
consistent both with operationism and with the exigencies of concept for-
mation in scientific theory.
Underlying these questions about opcrationism and concept formation is a
more fundamental problen1 in opcrationisn1. Opcrationists have not success-
fully provided criteria for the individuation of opcrations. 19 That is, there is
no formal way of determining whether or not two activities represent the
same operation. In Bridgman's classic example, measuring distance by use of
a rod is not the sa111c operation as measurc1nent by use of light rays; therefore
they define two concepts. But what about the operation of measuring with a
ruler nurked in meters compared with one marked in feet? Are Smith's
movements in using a ruler the same operation as Jones'? Is n1easuring with
a rod in the n1orning the same operation as measuring in the evening? There
i arc an infinite number of ways in which any two activities are si1nilar and
different. Without a criterion of individuation, there is no way to decide when
to ignore the differences and classify the two activities as the same operation
and when to ignore the similarities and consider the two as defining different
concepts.
In practice it is only a posteriol'i that such distinctions are drawn. Only after
Ii
I
it is known that temperature affects length is measurement at extreme tern-
' I
'·'
I NT ERV ENING V ARIA BL ES 63
peraturcs considered to be a different operation from measurement at middle-
range temperatures. It cannot be known in advance which similarities and
differences are to be selected or ignored in defining operations. If the initial
intuition is incorrect in that certain ignored differences are found to be im-
portant and the class is too broad, then the resulting empirical laws and cor-
relations will not have the same simplicity and orderliness as a classification
which distinguishes on the basis of those differences. Conversely, if the con-
cept is defined too narrowly-i.e., important similarities arc ignored-then
the resulting laws will display redundancy. Expansion of the classification re-
sults in the convergence of many laws into fewer but more general ones. 2"
There are 110 formal rules for constructing categories that n1aximize both
sin1plicity and co111prehensiveness. 21 Therefore, operationism is not an algo-
rithm for concept formation. 22
From these considerations it follows that although an operational definition
may be formulated at any given stage, it is subject to revision in so far as its
criteria for individuation of operations are vulnerable to change. Viewed this
way, operational definitions acquire the open-textured quality of reductive
chains. The only major difference is that with the reductive chain, the con-
cept is viewed as open at all stages, and new reduction sentences arc merely
added to the existing scheme. From the operationist point of view, the con-
cept is given a fixed closed definition, and every revision of individuation cri-
teria is viewed as introducing a new concept, albeit with the same name. 23
Whether concept development is visualized as the gradual evolution of a sin-
gle concept or as series of discrete substitutions of improved concepts for less
useful ones of the same name seems to be more a matter of taste rather than
fundamental episten1ological principle. Thus the differences between con-
cepts introduced by reductive chains and those operationally defined all but
vanish. 24
INTERVENING VARIABLES
Dispositional concepts and operationally defined concepts share important
logical features with the "intervening variable," a type of concept introduced
by Tolman which plays a central role in behaviorist matathcory. 25 Tolman
(1938a) describes his use of the intervening variable in this way:
A theory, as I shall conceive it, is a set of "intervening variables." These to-be-in-
serted intervening variables are 'constructs,' which we . . . evolve as a useful way of
breaking down into more manageable form the original complete [empirical] func-
tion. . . . And I have conceived a set of. . . functions to connect these intervening
variables severally to the independent variables, on the one hand, and a function . . .
to combine them together and connect them to the final dependent variable, on the
other. {p. 9)
 ;p_,
I
I

64 THEORETICAL CONCEPTS
INT EH VEN ING VAR I A 1l LES 65

An intervening variable thus represents a relationship between a set of inter- there are four different sequences in which this can be carried out. Because
i related independent variables and a set of interrelated dependent variables. all four achieve the same purpose, only one will be illustrated.
I When behavioral measures covary, this correlation can be named and con- In a second experiment, the relationship between 1) 1 and 12 is detennined:
I ceptualized as a comn1on factor among all the measures. For example, a rat's
(4) D, = H(I,)
rate of lever pressing for water, its toleration for quinine in its drinking water,
I

;I and the amount it drinks when provided with free access to water arc three By the inductive logic of the intervening variable, it is assmncd that values
behavioral measures which tend to covary. This covariation can be said to of D1 when produced by changes in '2 reflect T in the same way as those
represent an intervening variable, termed "thirst," reflected by each of the produced by changes in 11 • Therefore from equations (2) and (4):
three behavioral n1easures. (5) T=G(H(l 2 ))
ii'' The intervening variable is also securely linked to the independent vari-
20
ables. It is, as behaviorists often describe it, "anchored on both sides. " To In a third experi1nent, the relationship of D 2 to I I is detcrn1incd:
be anchored on the other side, the state must be shown to correlate with a (6)
II j set of interrelated independent variables. Thus, for example, the number of
hours the rat is deprived of water, the amount of salt in its diet, and the tem- By the inductive logic of the intervening variable it is assun1ed that D 2 also
perature of its cage correlate, in that each one affects behavior in similar ways. reflects T. From equation (3) it follows that:
Therefore, it can be said that these independent variables all affect a common (7) I,= F*(G*(T))
I '
state called "thirst." Anchored on both sides, the intervening variable repre-
sents the relationship between a set of dependent variables, which covary as Where F* _and G* arc the inverse functions of F and G, respectively; i.e., for
a function of each of a set of independent variables, and a set of independent any function, P, there is an inverse function, P*, such that for all x,
:( variables each of which produces the same covariation in the set of dependent P*(P(x)) = x. Therefore, fron1 the assmnptions of the intervening variable and
equations (6) and (7):
l variables.
Ideally the anchoring on the two sides should be quantitative and given by (8) I ) 2 = K(F*(G*(T)))
' defining equations. In practice this kind of definition is rarely achieved, but = K(F*(G*(G(H(l 2))))) from equation (5)
<Jj,:
. '.:.
·I
a model of the ideal introduction of an intervening variable in the simplest
case helps to illustrate its conceptual status. At the minimum, the introduc-
(9) = K(F*(H(I,))).
d: tion of an intervening variable, T, requires two independent variables, 11 and Equation (9) is the key equation for the intervening variable. It is an em-
:1: 12 , two dependent variables, D 1 and D 2 , and four empirically derived equa- pirical law relating an e1npirically defined dependent variable D 2 to an em-
'., pirically defined independent variable '2. However, this law is derived by in-
tions. In an initial experi1nent, often termed the "defining experiment," the
!i I
ductive logic, prior to c1npirical observation, from three empirically derived
I' ,I relationship between D 1 and 11 is determined while all other relevant variables
'I equations and the assumptions of the intervening variable. The derivation of
! :! are held constant. 27 Fr01n the defining cxperin1ent:
i' I

(1) D, = F(l 1 )
equation (9) can thus be viewed as a prediction prior to the fourth experi-
ment, or as an explanation of equation (9) if it is indeed found to be the result
I The intervening variable T is introduced by stipulation either in terms of D1 of the fourth experiment. In either case, the derivation makes use of T. If the
ii!' r~sults ~f t_he fourth ~·xperiment conform to equation (9) then the assump-
or 11. It is of no importance which is chosen since the choice of one logically
iI entails the relationship of T to the other. Thus:
T = G(D 1 ) by stipulation, therefore
tions gam 111 degree of confirmation, and the intervening variable T proves
to be a useful concept.
' (2) Additional variables can be added to the intervening variable by the same
(3) T = G(F(lt)) by the stipulation in conjunction with equation (I). metl'.ods. _A depen_dent variable is added by first c1npirically detcnnining its
Equations (2) and (3) define T in terms of dependent and independent vari- re~at1onsh1p to an mdepcndent variable already known to belong to T. From
ables and thus securely anchor the intervening variable on both sides. Three tlus result, the assmned relationship of this dependent variable to T may be

,,
Ii 1nore empirical relationships between dependent and independent variables must deduced. [ cf. equations (6) and (8) ]. Fro1n this, the relationship of the depen-
be determined. Since there arc two dependent and two independent variables, dent vanable to another independent variable in T may be deduced [ cf. equa-
"! . l
 66 T H E O H. E T I C A L C O N C E P T S
í T H E I N T E R P R E T A T l <.) N t) F C () N (: E P T S 67

tion {9) ¡. I f thc dcductíon is con firmcd by cxperimcnt , t hcn thc � a1� didatc ís
alt h o:1gh it is not currcn tl y undcr test cond itions for thc intcrvcnin g va riable
acccptc d into thc íntervcning variabk; i f not , it is cxdudcd . A _s11111br tcch­
of tb 1 rst bcca usr..� the intcrveníng variable is also di:fined in tcnns of indcpen­
niquc can be uscd for tcsting indcpcndcnt varia bles for p ros p ccn vc m c � 1: bcr­ _
d ent. va � 1ables , such a s hou rs o f water dcprivation, roo1n tcn1p1..� ratun:. and
shíp in T. Thus the intcrvening va r iable is an o pen conccp t m that add1nonal .
,. sal t _ 111 d 1ct, i.� h1 ch c � n be . 1� 1casured . Thcrcforc, a wdl formulatcd intcrvcning
variables may bt.> addcd w ith ncw d iscovcrii:s. :?H
!!
. 1·¡ ¡¡ vanablc, unhkc a d 1spos1t1onal concept, is determ ínate c vcn whcn test con­
As equations (2 ) and (J) indicatc, intcrvcn ing variables are � cfo�ed b y b� th
d ttions do not p rcvai i . and í t is not introduced b y rcd tt ct ivc chains . · 1
-,,
1 ¡r .
i ! d epcn dcn t and indcpen dcnt v ariables. To say that thc valuc ot a11 m : crvcnmg
variable T for an organisn1 has thc va.luc N is thcrcforc to say somethmg about
1 ; !
t1
¡: thc beha vior of tha t organism . H o wcvcr , it is not ncccssaril y to sa.y so m e­ THE INTERPRETAT ION OF DE
1. !i thing about its currcnt beha vier. Thc va luc o f thc th írst-intcrvcning-variablc FIN ED CO N CEP TS
for ; particular rat as mcasurcd by t � e índe� cn dcn � variabl_ cs m a � be v cr y They Are Mere Lab els
high , for exa111pk� but bchaviors assoc1atcd wtth the mtcrvemng vanablc may A com mo n bch a viorist i nte rpre ta tion
o f thco rct ica l con ccp ts is tha t they are
11 b� absc-nt beca use othcr fac tors intcrfcn.'. Without availabl e water, thc dcpcn­ n1e rcl y lahc ls fer obs crv cd rcl atio nsh ips
i( bct wec n bch a. v ior and the env iron ­
d dcn t va ria bl e a1nount-drunk cannot be m casurcd . Thus. th(.· valu c of an in ­ nu.·n t. _ Mo � t i m po rtant, thcy do not sta
n d for cn tüies , cv cn ts, or pro ccss t.."S
q tcrvcning variable is rcflcctcd in b chavior onl y un d cr � pprop riatc círcum­
_
stanccs. Thcrcfon:·. to say th at T = N fer a n organ1sn1 1s t o say that undcr
occu rnn g m an uno b c rvcd rcah n in an o
�
tc mporal l y lo cata bl c, m thm rda tion sh íps
rga 11is1 n's bod y or min d. Thc y a re
ma y hol d o vcr lint i ted i n tc rvals o f
circun1stances C 1 dc p enden t variable D 1 will hav e valuc x, undcr ci rcum­ tin1 e. but l ike aH rela cion al con cep ts, thc y
pos scss no spa tial l oca tion .
stanccs C 2 dcpcndcnt varia ble D 2 will have valuc y . and so forth throughout Althoug h a bstr act con ccp ts are adm
ittc d into bch avior íst s ci cn ce thc y a re
¡ thc class of dcpcn de11t variables in T. not t � bc rci �1cd, 31 or assi gnc d a cau sal r
_ . . , ole. :n A n intcrve n ing var iable " inte r­
Whcn v icwcd ín this w ay. thc in terv cning varia ble rescmblcs d1spos1 t1on al : cnc s on � y m the s cnsc tha.t thc tenn rep resc ntin g th c in terv cn
i'J· ¡•
: \
·. ! con ccpts and opcrationaHy ddincd conccpts di scus sed � bov c. 29 _ Thc circu �11- m a bch av1o ral bw ma. y be p bcr..•d 1 11 a pos i ng var iable
ition ín tc rnH�dia tc bct wccn th c dc­
' J1 •' s tances under which an intcrvcnin g variable is rcflccted m bcha v1or are cqu1v­ p cn den t an d thc ind cpcn dcn t vari a.ble s. Thu
} ¡ s fron1 cqu atio ns ( 1 ) and {2) it
t · alent to th c test cond1tions under which a disposition is mani fcs t and the de­ nuy be sta ted that :
, . l1·1F
fining opcrations for an opera tionall y defincd conccpt . Note that it is the
. r! _
ci rcu mstanccs. C m and not tbc indcpcndcnt varia bles , that a re c q m valent to D1 = G*(T) = F(1 1 }
1 \
.. . l �

thc test condi tions . Similarlv, the values of thc dcpcndcnt varia bles far the
in tcrvcning variable are thc � quivalent of the 1nan ifost outco1ncs for thc dís­ �nd far cxa m p k, that d rínkíng is a func tion of thírst, and thirs t . in turn, is a
tunction of deprivation o f water. Howcvcr , i nt crvcnin g varia bles are not to
p osition a 1; d the resul ts of thc mcasu ring opcra tions for thc opcrationally de-
be conccivcd as cv cn ts or p roccsscs o ccurring a ftr..· r an ..._ cn víron m cntal cvcnt
finc d conccpt .
an d befa re a bchavioral cvcnt.
O n the othcr ha.nd. Íl1tcrvening variables di ffer in significant ways iro111
dis positiona l and opcra tionally dc fincd conccpts . In contras t with strictly � p­
.
era tionally dcfincd concc p ts, intcrvcning variables are op en. 30 An mtcrvcnmg They Are Eliminable
_
variable is rcal izcd by a progrcssivdy incr casing numbcr of bchav1oral dcpcn­
This vi cw of � on ccpts lcads quite naturally to a second . m ore radical, opinion
dent variables whi lc a strictly opcrationally defincd conccpt ís defincd by only .
among b cha v 1ons ts. lf conccp ts are truly nothing more tha n namcs of ob­
, ¡\
o ne o pcra tion . N c vcrthdess. operational dcfuütions serve an ünportant ro �e
. ¡ '
. '

servcd rclat ionshíps, thcn why not d iminatc thcn1 altogethcr and dcal di­
wi thin che forn1 t1lation of a n intcrv cn ín g variable bccause cach of the va ri­
re � tly w i th those obscrv � d rdationships ? A wcll dcfincd conccpt is casily di­
ables in clu dcd in thc i ntcrvening va riable can be opera t ionally dcfincd.
muu � lc, a �1 � a � a guel y ton11 ulated onc docs n ot bclong in a scic11ce any way.
lntcrvening va riables also d'iffer fro m dispositional conc�pts . As � qna �ion
1 f a d1spo : 1t1011 1s attnbuted on the basis of s0111e observable prop erty, then
(3) in dicates, intervcning variables are dcfined from the st1nu1lus s1de (1. e . •
thc beha v1or shoutd be cxplained as a function of that observable propert y
indepcndent va riables) . Althou gh response n1easures occur only undcr ª ? p ro­ rathcr than the dis positi on . 34 If thc disposition i s as cribcd on th e basis o f th c
p riatc circumstances, the in depcn dent vari ables ca n be 111casurcd a11 d ass1gne d
_ �ehavi o r to b e cxplaincd (c. g . • catin g is cxplaincd as d uc- to hungcr which is
a determínate value even whcn the circumstances are not appropnatc for ob­
111f: rrcd fro ! 11 the ca ting bch avior) then thc con ccp t is ci rcular. and thc cx pla­
serving ch e bcha vioral mcasures . For cxa m plc, a rat n1ay be said to be thirsty
nat 1on s p unous .

·[, : i ,;
· : l wra
 68 T H E O R ET I C A L C O N C E P T S TH E l N T E R P R ETATI O N OF CO N C E PTS 69
n­
f In tcrvcn ing va ria bles ca n be dimin ated i n a sinülar m a nncr . 1 f an i nterve re cted toward the understanding of thc inncr processes and cntitics rathcr than
ing vari able is dcfincd by cqu a tí on s of thc form of cqua tions (2) an d 3p), t hen
j

l' at changing thc indcpen dent cx ternal variables which truly control beha vio r.
! wherev er T appca rs . ít n1ay b e rcp laccd by cithcr G ( D 1 ) � r G (F ( I i ) ) . :-, 'I_"' here­ M o rco vcr, b cca u sc thc concept is not sccurdy ancho red to measurable obs er­
!; fore , equati on (9) dcri ved with thc hdp of T 1na y be den ved as wdl
w 1 th ou t vati on s , i t is often assig ncd din1cnsions difforcnt fro m thosc uscd in rhc other
l
ít: s cien ces, thereby i solating psy cholog y from the rcst of scicn cc.

!H.i ( 1 0) 1 1 = F* (D 1 ) from cquat ion ( 1 )

Thus thc in tro duction of díspositional conccpts and intcrvening variabl es is
b elie ved by son1c bcha vio rists t o kad to severa! evcntualítics whi ch are at
ji (1 1) D 2 = K (F*(D 1 )) from cquat ions (6) and (1 0) . variancc wit h fundamental bchavi ori st doctrine. At thc samc tinte, howcvcr,
¡l ¡'
! From cqua tíons (4) and ( 1 1 ) , cqua tion (9) can be dcrived : it is i m portan t to note that thesc arguments a re d i rectcd a t the p ractica( con­
; :
seq_tu'nces of introduci n g certain concepts i n to bcha viorist psychology rather
¡ :
l , D 1 = K ( F*(H(l2)))

¡·¡:
; ¡,
witb ou t using T .
¡ '
¡
¡ They Are Undesirable
l
l.
¡ Not on ly are con ccp ts clintina blc. bu t, opponents argue, their clinlina t � on is
a lso desira b k . Many behavi oris ts fear that a l though a wcll fonnu latcd mter­
vcning variable or disposi tional concep t may be logicaU ):' i m pcccabl c, in prac­
tice the in t rodu ct i on of con ccpts inevitab ly lcads to n11s use . Such concepts,
thc; claim. ten d to b e hy postatí zcd and are easil y rcgardc d as s yn1bo lizing
unobserv cd even ts . AH too oftcn, T con1cs to be vícwcd as an unobsc rved
inner state of thc organis m-ci thcr a physiol ogical onc, or worse. from a be­
haviori st pcrs pcctívc, a n1en tal one . Mentali sm thus foll o ws in thc wake of
thesc conccp ts , cs pccial l y whcn thcy a re i dcnti ficd with statcs known only
.
t l1roug l1 mtros . . ·Jh
pect10n
Bcside s thc dangcrs o f misint crprcta tíon of wd l formub tcd concep ts, t �e�e
1-
is the fu rthcr risk , accordi ng to thcsc cri tics , that on ce con ccp ts are adn11ss
37
ble. poorly formu latcd oncs will be devclop cd . This is Jik_cly to o � cu r :' hen
an intcrvc ning variabl e is not sccurd y an chorcd on the stun ulus (u: . , m de­
pcnden t variable) sidc. In this case, thc int ervcn í n g variabl e is thc cq ui valent
as
o f a disp ositional con ccpt . Thc pcríl hen: is that behavio r will be cx plain � d
,
brough t on by thc d i s position , but th c disposí tion itsd � is kft uncxpla mc�
with no atte1np t to lin k thc dísposi tion wíth its causes m thc ex ternal . en '_' 1-
1t­
r0 11 1nent. Thus bchavio r is cxplain cd b y a conccp t infern:d fr0111 beha v10r
self. Althou gh such correlat ions within bcha vio r n1a y be o ccasíon ally u seful,
te­
they are n1 ore likely to inhi bit thc sca rch for thc en viron men tal ca1_1sal an
cedents o f both the b ehavior and thc corrdation s . Becaus e thcsc env1 ron me11-
tal anteced ents a re direct1 y manipu l able while disposi tions are not, 8their dis-
cov ery is far n1o re importa nt for thc practi ca! cont rol of b ch_avior. :'
1s
The situatio n is wo rse whcn thc concept is an chorcd on 11e1thcr s t de. In th
l y are
case the concept is often in vented ad ltoc to c x plain a behavio r. N ot on
¡· 1 such explanations illegitin1ate. but they are also n1islcad ing. On t�1e onc �an �
'1
than a t the con ccp ts thcmsclves . Propcrl y formulatcd dispositíonal con cep ts
and in tcr vening varia bks are not logícally suspect. Thc a rgumcnts in favo r o f
their cfün ination ín vol ve the possiblc mísin terp rctation o f r h c con ccpts o r the
íntrod uction o f i m p ropcrl y formulatcd oncs. Thcrcfo rc� the case agains t dis­
positi onal concepts and ín tcrvening va ri a bles n1a y be sai d to be a "strategi c"
one ra thcr than a .. phílosophica l " onc.
The Case for Concepts
1. Tlz ey Mediate Ca11sality. Prop on cnts of con ccp ts a rguc that the allegcd
dangcrs are exaggeratcd and that conccpts offcr certaín in1portant bcncfits that
ought not to be forfeitcd. Y> O ne fun ction scrvcd by intcrvening v:.1riablcs i s
i'
that they bridge the temporal g a p bctwccn dcpcn dent and indcpcndcn t vari­
ables . Vcry o ftcn cons iderable t i m e may pass bct wcen the occurrcn ce of an
independent variable such as the rcin forccmcn t of a response and thc beha v­
ior (e. g. , thc latcr occu rrcncc o f tha t response) which is a fu n ction o f that
va ria ble. Thc introduction of an íntcrvcning variable, such as ha bit strength,
serves as a con tcn1porancous st11nn1ary of the organisn1•s prcvious history and
thus s erves to mcdía tc thc- t i m e intcrval. •to Whcn thc behavior finall y docs oc­
cur it n1a y be said to be caused by thc i m n1c-diatd y prcccdin g state, T. ra thcr
than by an indcpcndcnt variable í n thc pas t .
Although th i s is a com n1on justi fi catíon for thc introduction o f a n inter­
vening va riable. thcrc are two major proble111s wi th it. First, i f an intcrvcning
variable is t ruly j ust a mathcmatical tcrn1 fully defincd by thc equations in
which it ap pcars, thcn i t can not serve as a ca usal event disti nct from thc dc­
fining va riables . 4 1 Secon d , thc j usti fi ca tion assu n1c-s that all causal ít y is of thc
type ty pi fic-d by one billiard ball strik íng a sccon d one . Howcvcr, as Hume
defini tivdy argucd. causal necessity is not obscrvably prcscnt cven in this case.
Yet thc rcquircment that a case always be temporally an d spati aH y contig­
uo us with íts cffect , somctimes called the .. bea d thcory of caus a l i ty . " 42 as­
su mes causal ncccss ity.
Sorne behavio rists thcrcfore a d o pt a function al view of causal i ty. 43 Func­

¡:
d1-
expc ri n1ental research is diverted frmn thc true causes of b ehav1or and 1s
H
tional rela tions are substi tutcd for causal oncs, an d ' • x is a fun ction o f y re-

:¡
11

': I
.1

70 THEORETICAL CONCEPTS
places "x is caused by y." This approach to causality thus avoids any impli-
cation that unobserved causal necessity intervenes between cause and effect.
It also obviates the need for the contemporaneous occurrence of cause and
effect. Therefore, behavior 1nay be said to be a function of an event in the
ren1ote past, and there is no need for an intervening variable to mediate. This
staten1ent is not meant to deny that physiological events intervene between.
the two occurrences. Indeed, it is undeniable that such events do occur. The
point is that these physiological events do not have to be referred to in order
to state a functional relationship between behavior and the environment, and
a n1olar behavioris1n. therefore, does not have to sacrifice such relation-
ships. 44
2. Tltey Simplify. A second justification for the use of intervening variables
is that they simplify the formal expression of behavioral laws. Assmne that
an intervening relates n dependent variables to m independent variables. Without
the intervening variable, if each dependent variable is related to each indepen-
dent variable. m X 11 equations are required. With the introduction of the
intervening variable T, m equations are required to express T as a function of
each independent variable, and 11 equations for T as a function of each depen-
dent variable. for a total of m + n equations. Hence, for 111 or 11;;;,,2 and the
other ;;;,,3, the use ofT decreases the number of equations necessary. The sav-
ings are only in formal expression. There is no reduction of the number of
empirical experiments necessary to establish the behavioral laws, for a vari-
able cannot be assmned to belong to the intervening variable unless it is ex-
perimentally proven to be. 45
In reply to this claim, it can be argued that each dependent variable can be
expressed in one equation as a joint function of all the independent variables.
In this case, only 11 long equations are required to express the empirical con-
tent of the cxpcri111entally detern1ined laws, and this is fewer than the 11 + 1
equations necessary to express each of the dependent variables as a function
of T (n equations) as well as T as a joint function of all the independent vari-
ables.
However, even here, proponents of the intervening variable argue that it
serves an important role. They claim that the long equation relating a depen-
dent variable to all the independent variables is so complex that it cannot be
dealt with effectively. By subdividing the long equation into smaller groups
of independent variables and identifying each group with an intervening vari-
able, the entire equation becomes more intelligible and therefore manage-
able. 46 Relationships that would otherwise be missed arc clarified in an equa-
tion judiciously simplified by a chain of intervening variables. To be sure,
nothing in the simplified equation using a chain of intervening variables is
not captured by the long equations. The simplified equation has only a heu-
ristic advantage over the long equations, not a logical one .
.: : '
I i
• I
;I.,,\: ,.
+
THE INTERPRETATION OF CONCEPTS 71
3'. Tltey Facilit~te I11d11cti1 1 e Systematization. By systematizing various em-
p1ncally dctermmed relationships, an intervening variable facilitates predic-
tions and sug?ests new experiments. This is a third reason why some behav-
10nsts fa_vor_ mt~rvening variables. An important instance of this inductive
systemat1zat1011 ts to be found schematized in equations (1) through (9). This
schc_ma 1s, 111 effect, an algorithm for suggesting an experiment and predict-
mg its _results. A!though, as shown above, this san1e derivation can logically
be earned out ":ithout T,_ through equations (10) and (11). it is not as likely
that m _fact it will be earned out. Without T, equation (11) has no rationale.
There IS no reason _to expect D2 and D 1 to covary as suggested by equation
(11 ). _The mtroducnon of T along with the associated intuitions that T is a
f~nctton of I I ar~d 12 and that T is reflected by both D 1 and D 2 provide a ra-
tionale and motive for the derivation. As the number of variables increases
~nd n1ore than one intervening variable is introduced between dependent and
mdependent variables, the derivation becon1es more complex and T becon1es
a more important heuristic. Hence, T introd11ees a degree of inductive sys-
tematization that would be lost in its absence. 47
Thus the three. argun1ents in favor of the intervening variable, like the ar-
gun1ents agamst 1t, are strategic rather than philosophical. Neither side claims
that intervening variables are logically indispensible. The point of contention
revolves around the predicted consequences of their use. Proponents assert
that such concepts possess heuristic value to facilitate research while oppo-
ne~ts aver that concept". en~ourage abuses which impede scientific progress.
Evidence relevant to this d1sagreen1ent might be found in a historical study
of the actual consequences of the use of intervening variables or from a psy-
chology, yet to be achieved, which accounts for scientific thought.
Conventionalism
Th~ justifications offered for the use of intervening variables emphasize their
ut1hty and lead qmte naturally to a conventionalist view of theoretical terms
and theory. 48 Intervening variables are valued because they provide a short-
~and summary of observed relationships, mediate the derivation of predic-
?ons, and 1nake complex equations more manageable. To be sure, interven-
mg van_ables must be valid in that they must accurately summarize
observat10~1s, n1ed1ate confirmed predictions, and correctly subdivide com-
?lex equations. However, because they do not syn1bolize events or entities,
It is a matter of h un1an choice as to how they are formulated, and in this
s:nse they are conventions. Note, for example, the stipulative nature of equa-
'.1on (2): defini~g T .49 Consequently, there is 110 unique set of valid interven-
m~ variables, Just as there is no unique true summary of observations. For
tins reason, i'.1tervenin? variables are selected not merely for their validity but
more for their convemence, degree of economy, and effectiveness.
I
I
~
 I
'

72 THEORETICAL CONCEPTS HYPOTHETICAL CONSTRUCTS 73

From this conventionalist point of view, it also follows that intervening to cn1pir!Cal tern1s; they rcfC'r to processes or entities that arc not directly observed
variables have no unique representation, just as there is no unique shorthand. (although they need not be in principle unobservable); the mathematical expression of
Mathematical equations are a comn1on n1cthod of representation but not the them cannot be formed simply by a suitable grouping of terms in direct empirical
only one. Intervening variables can also be symbolized by diagrams, concrete equation; and the truth of the empirical laws involved is a necessary but not a suffi-
mechanical models, flow charts, or electrical fields. For exan1ple, Tolman's cient condition for the truth of these conceptions. (p. 104)
"dby spat1a·1 reprc:entat1ons.
. 50
intervening variables are often syn1bohze · H ov.:- The admission of hypothetical constructs into the science of bcha vior is a
ever, these fields are not intended to represent the subjects phenomenolog1- matter of controversy within behaviorist thought. It is obvious why this is
cal field, nor arc they maps of the subject's physical field. Instead, they _are so. Hypothctic:il constructs pose a far greater challenge to behaviorist ideals
economical, and, for Toln1an, convenient ways to represent the mteraction of empiricalncss and objectivity than do intervening variables. Hypothetical
of the variables affecting behavior and to derive spatial vectors representing constructs are associated with "existence propositions" which assign .. surplus
the subject's response. . . . meaning" over and above observed correlations. Admission of hypothetical
Opposed to a conventionalist interpretation of mtervenmg va~1ables and constructs thus means the inclusion of concepts referring to unobservables.
dispositional concepts is the realist view which argues, on_ philosophical
grounds, that these concepts must be more than mere conventions. If behav-
1. They Are Used by Other Sciences
ior shows the interrelationships symbolized by the concept then there must
be something about the organism by virtue of which those i?terrclationships Proponents of hypothetical constructs argue that science, in fact, does not lin1it
arc true. This "something" then is the true referent of the d1spos1t1onal con- itself cxclusi vcly to intervening variables. Physics has introduced a v:ist array
cept or the intervening variable. 51 Furthermore, this _"something"_ can play a of theoretical terms such as "electron" and "neutron" which refer to unob-
t causal role in behavior. If a disposition or intervemng vanable 1s an event served events and entities beyond the observations from which their exis-
l(! distinct from the observable variables, then it can truly "intervene" in the
sense that it is a state caused by the independent variables, and it, in turn, can
tence is inferred. Modern physics could not have progressed without the in-
troduction of these hypothetical constructs. 5~ Therefore, a methodological
.•,, I
. I 1 :I I cause behavior. 52 decision to disallow constructs of this sort in psychology is not reflective of
. '. . scientific method and might put psychology at a severe disadvantage.
HYPOTHETICAL CONSTRUCTS Empiricists in1pressed by the success of physics construct a meta theoretical
''
:i"!,
\ n1odcl of how such theoretical terms derive cmpiric:il significance although
I
A realist perspective leads quite naturally to the introduction of concepts ex-
they refer to unobservables. 55 According to this model, a theory is an axiom-
!'
I
I,
•I
plicitly intended to refer to unobserved events within the organis_m. It is l:clpful
i! :' to distinguish formally between these concepts and those wluch arc viewed atized uninterpreted calculus. Theoretical terms arc introduced by postulates
I
i1 as a mere sun1mary of observed relationships, and MacCorquodalc and Meehl within the theory that provide only an implicit definition of the theoretical
i 11 I I
(1948) provide the required differentiation. They tern1 the latter kind of con- terms by stating how they arc related to the other terms of the theory. 56 The

I; '
'
! cept an "intervening variable," and the forn1er a "hypothetical construct. "
They define the intervening variable as
53 theory, as a whole, is given empirical significance by a set of "correspon-
dence rules" which connect certain terms of the theory to en1pirical obser-
,. :1 vations. These correspondence rules may take, among others, the form of
!.
1'
I I 1
simply a quantity obtained by a specified manipulation of the values of empirical vari-
ables: it will involve no hypothesis as to the existence of nonobserved entitles or the
operational definitions or reduction sentences. However, only some of the
terms of the theory are included in the correspondence rules. Other terms,
i.
'I
I occurrence of unobserved processes; it will contain, in its complete statement for all
purposes of theory and prediction, no words which arc not definable cith_cr_ explicitly
not given definition by the correspondence rules, arc theoretical and have only
implicit definitions. They acquire empirical significance only by their role in
I :!:
::i·
or by reduction sentences in terms of the empirical variables; and the validity of em-
the entire set of postulates connected at various points to empirical observa-
'iI ;:1: pirical laws involving only observables will constitute both th~ necess~ry and_ suffi-
cient conditions for the validity of the laws involving these mtervcmng variables. tions by the correspondence rules.
1·
In response to this argument it can be claimed that there are cogent reasons
t·
'
:I (p. 103)
why psychology should not follow the example of physics. First, psycholog-
ii In contrast, hypothetical constructs are defined as
'
I
I I. concepts which do no/ meet the requirements for intervening variables in the st_rict
ical theory is nowhere near axiomatization. Therefore any model for the in-
troduction of theoretical terms which assumes an axiomatized theory is in-
'i sense. That is to say, these constructs involve terms which arc not wholly reducible applicable to psychology. Second, as compared to psychology, physics is a
l'/illl' " I I
.

HYPOTHETICAL CONSTRUCTS 75
74 THEORETICAL CONCEPTS
•
which prevent an 01acome or behavioral measure from defining the concept.
far more n1aturc and wdl developed science. Theoretical concepts arc more
Thus, when a bchav10ral outcome is not taken as a true measure of IQ or
highly quantified, postulates arc more explicit, and theoretical deductions are
,. ·II hunger, this 1s not because those intervening variables exist as entities over
:' stricter, so testing of theories is more rigorous. Therefore, the dangers asso-
I and above the measurements, but because the measurements were not pcr-
I
ciated with the introduction of theoretical concepts arc much less severe in
fonned under standard conditions.
'I
physics than in psychology. Third, unlike physics. psychology must contend
The interfering factors which preclude conditions from being standard can-
with the fact that researchers bring to their theorizing a wealth of prescicn-
".,!. tific concepts about human action derived from everyday speech, intuition, n~t be exhaustively listed, not only because they cannot all be anticipated a
! • pnon, but also because they arc infinite in number. Therefore, some intuitive
' and their own phenomenology. Inevitably, these subjective and mentalistic
understanding of the concept is necessary to determine when a factor is an
notions influence the formulation of theoretical concepts in psychology.
interfering one. Nevertheless, this intuitive grasp of the concept docs not en-
.i-''.' Therefore, hypothetical constructs arc 1nore vulnerable to misuse in psychol-
tail that an unobservable underlies the concept and is referred to in dctennin-
ogy than in physics. 07
i These arguments as to why psychology should not imitate physics arc not ing :vhen_ an outcome is or is not a true measure of the concept. Normally,
iiI' without irony. A central thrust behind the behaviorist movement is the at-
the mtmt1on results from experience with the observed correlations and the
ii correct_ unproblematic_ application of the concept. In fact, all concepts, whether
tempt to transform psychology into a science. Physics, in particular, is up-
d1spos1t1onal or not, mvolvc an intuitive understanding which cannot be ex-
if
!i'
held as the n1odel psychology should emulate. However, as physics pro-
haustively captured by a listing of rules of usage. To 'know an object is red
:;l I'j gressed, its associated philosophy of science evolved; but psychology, not
or triangular also mvolves knowledge about normal conditions and interfer-
blessed with corresponding progress, did not share in this development. 58
ing factors, knowlege which cannot be fi.1ll y explicated.''"
I 1
Ironically, behaviorist philosophy of science is often defended by appeal to
differences between psychology and physics .
I 3. They Are Important for Reduction
. '!i'.(·' 2. They Cannot Be Avoided in Practice Y ct a third argument in favor of hypothetical constructs is that they bridge
·1'. I the gap between psychology and physiology. " 1 Hypothetical constructs tend
:1 A second argument against the limitation of psychology to intervening vari-
to be characterized in physiological tcrn1s. They, therefore can serve a heu-
.. Ii Ii. ables and operationally defined concepts is that in practice psychologists do
ristic function in suggesting to physiologists the kind of mechanisms to look
,.I . not treat any concept as if it were one of those two kinds. Suppose, for ex-
for. At some point, the hypothetical constructs of psychology can be identi-
, I I an1plc, that IQ is operationally defined as what is measured by a particular
1
•I '. fied with the mcchamsms discovered by physiology, and the theoretical re-
I i test, and the intervening variable of hunger is measured by rate of lever pressing
I: I .I I
I l; for food. Suppose further that after the behavioral measures arc recorded, it
duction of psychology to physiology will be accomplished. By combining
~ I behavioral as well as physiological features, the hypothetical construct facili:.
i' is discovered that the subjects in both experiments were ill. Under these cir-
! •I I ! tatcs this reduction. If continuity with the other sciences is a goal of bcha v-
01 cumstances, it is c01nmonly concluded that the behavior was not a true mea-
. !1 sure of the concept. However, if the concept is in fact defined solely in terms mnst psy_chology then theoretical reduction of psychology to physiology via
I, hypothetical constructs 1s one way to accomplish this. Intervening variables
of the behavior, then whatever behavior is recorded is, by definition, a mea-
,I are unlikely to be helpful in this regard. Because of their conventional nature
sure of the concept. There should be no rationale for saying when the behav-
ior is or is not a "true" measure of the concept. To deny that the behavior is int~rvening va~iablcs can be defined and redefined arbitrarily, while physio~
,. logteal mcchamsms seem to be matters of fact, not convention.
a measure of the concept entails that the concept is something other than what
I :11
Ii is measured by the behavior. In practice, therefore, it appears that interven-
This argument in support of hypothetical constructs presupposes several
.1 ;,l questionable assu1nptions. First, it is not dear that hypothetical constructs arc
' ' ' ~.
rt' :1,, ing variables and operationally defined concepts arc treated as if they are hy-
more cffi_cacious for theoretical reduction than intervening variables. With a
I' I' pothetical constructs. 59
This argument docs not seem conclusive. Operational definitions as well hypothcncal construct, the possibility of attributing invalid properties is ever iI
'-I'. ,'
JI
f• I ◄ present. _In c01~trast, a ':"'"II formulated intervening variable, in not exceeding ,I
as definitions of intervening variables include implicit ceteris parilms clauses to
observat10ns, 1s 1nore likely to correspond to a physiological constuct than a
indicate that the outcome or response measure defines the concept only under
!' hypothetical construct with incorrect features. 62 Fu thermo re, the argunient
i. normal circumstances. Amount eaten is not assumed to 1ncasure hunger if
assumes that the subdivision of an entire organism into physiological systems i
the rat's mouth is sewn shut. It is assumed that interfering factors may occur
,'
I
 76 THEORETICAL CONCEPTS
is a given rather than the consequence of human conceptualization. When the
skull is opened, the parts of the brain do not appear already labeled with bor-
ders neatly laid out. Neural "mechanisms" are created on the basis of a num-
ber of considerations, including functional ties as well as anatomical connec-
tions. Therefore, physiological mechanisms are conventional in the same sense
intervening variables arc: both initially are human creations, but both must
I :
prove their worth.
' . :I : I Second, the argument assumes that theoretical reduction is the best method
I'.1·:' •' for assuring continuity between psychology and the other sciences. Yet, one
'',,. of the contentions of n1olar behaviorism is that the legitimacy of a behavioral
'!II
I
psychology does not rest on the possibility of its reduction to physiology. To
I!!:
I,,, the contrary, the aims of molar behaviorism are best achieved by a purely
1!1
r,I behavioral psychology, independent of physiology, which is effective for the
I, prediction and control of behavior and is unencumbered by premature spec-
Ii ulation about physiology. Continuity with the rest of science can be accom-
I:
plished by a com1nonality of methods and underlying assumptions."3
I with an intuitive understanding, and consequently the explanation is not ep-
4. They Are Necessary for Explaining Behavior
l'I.l'·.Lt
. I
' .' A fourth objection to theories limited to intervening variables and operation-
'I• .I ally defined concepts is that such theories arc not truly explanatory. Because
theories of this type refrain from making statcn1ents about events internal to
the behaving organism, they are often derogatorily referred to as "black-box"
theories. At best, it is often argued, black-box theories provide behavioral
laws relating input {stimuli) to output {responses), but they do not explain
these lawful relationships. Only by reference to events inside the black-box,
i.e., hypothetical constructs, can behavior be explained.''4
This objection to black-box theories makes certain tacit assumptions about
the nature of explanation that need to be examined. If by "explanation" the
non1ological-deductive pattern is meant, then clearly black-box theories are
explanatory. Particular instances of behavior are explained by deducing state-
ments describing them from general behavioral laws in conjunction with
statements about initial conditions. Instead, the objection may be that the be-
havioral laws used in black-box deductive explanations are simply descriptive
of observed regularities rather than explanatory. They are stated as givens rather
than as the results of the internal events mediating them.
It is true, as this argument notes, that black-box theories must have a few
fundamental behavioral laws from which other laws and statements about
particular responses are deduced. These fuuda1nental laws are indeed "prim-
itive" in the sense that they the1nselves are not deduced from other laws.
However, the argument is 1nistakcn in its implicit belief that the situation could
:• be otherwise. As Hume clearly showed, causal laws are not logically neces-
I I sary. Therefore even explanations referring to hypothetical constructs ulti-
,, l
' 'i
ii,,
,:/
· 11
;!
r
:j i
111
.1L:
HYPOTHETICAL CONSTRUCTS 77
mately appeal to fundamental laws, whether experimental or theoretical, which
are primitive in the same sense that the fundamental behavioral laws arc.
A proponent of hypothetical constructs might still argue that theories in-
corporating hypothetical constructs arc superior because primitive behavioral
laws can be deduced from such theories. Thus these theories explain what is
left unexplained in black-box theories. In response to this argument it should
be noted that it is always the case that the fundamental laws of one science
are possibly the derived theorems of another more fundan1ental science. This
holds as well for theories incorporating physiological hypothetical constructs.
Nevertheless, just because a science is possibly reducible to another science,
this is no reason to deny that the science is explanatory. If this were the case,
then no science, other than the most fundamental one, is explanatory, includ-
ing physiological pychology.
It appears that the claim that the black-box theories are not explanatory is
false when deductive explanation is meant. However, "explanation" may be
used in other senses. Often a deductive explanation does not provide the hearer
istemically satisfying. Tln1s, "explanation" may 1ncan "to provide a satisfac-
tory understanding." Those who object to black-box theories arc often say-
ing, in effect, that although such theories explain in one sense, they fail to
provide understanding. In contrast, theories appealing to physiological mech-
anisms are said to n1ect the hearer's epistemic needs and provide an intuitive
grasp of the explicandum."5 Thus, objections to black-box theories often re-
flect a preference for a certain kind of explanation. Therefore, the explanatory
adequacy of black-box theories depends on individual differences over what
constitutes subjectively satisfying understanding.
5. They Are Heuristics
A fifth and extremely common objection to a black-box science is that it would
be excessively "narrow," thereby impeding scientific progress."" Nun1erous
examples in the history of science can be adduced to demonstrate how unob-
servables were hypothesized to explain certain observations, and then later
these hypothetical constructs were found, through direct observation, to have
many of the hypothesized properties ascribed to them. Genes, molecules,
bacteria, and distant stars arc typical examples. In this way the hypothetical
construct guides and integrates research. Although the properties assigned to
the construct arc inferred frmn observation, they transcend the observations
to provide the construct's surplus meaning. This surplus meaning enables the
theory to integrate diverse observations which otherwise would not appear
to be related. Moreover, from statements about the hypothesized properties,
other statements are deduced which have observational implications. These,
in turn, suggest experiments which tend either to confirm or disconfirm the
 78
. THEORETICAL CONCEPTS
,,' I
1 statements about the construct. Thus the attempt to characterize fi.,lly the
1 I
:_._,:_I theoretical construct and to test this characterization organizes and directs re-
l'1 I
:I I scarch. 67
lil ·
•i
l'I!, The Case Against Hypothetical Constructs
','
l,f
These arguments in favor of the inclusion of hypothetical constructs appear
to be strategic ones, just as arc the arguments in favor of intervening vari-
ables. The strategic arguments-that psychology should emulate physics, that
hypothetical constructs facilitate theoretical reduction, that they provide more
satisfactory understanding, or that they are heuristically fortik~reduce to the
common theme that the introduction of hypothetical constructs will enhance
the progress of psychology, not that they are logically necessary. Similarly,
the case against hypothetical constructs is based largely on strategic consid-
erations. In addition to all the objections to intervening variables reviewed
above, it is feared that the use of hypothetical constructs will encourage un-
warranted speculation, resulting in premature "physiologizing" and the cre-
ation of theoretical °'fictions."
Hypothetical constructs seem less open to experimental tests because prop-
erties can be attributed to them ad hoc to accommodate any experimental re-
sult, and they therefore appear to provide only spurious explanations. Fur-
thermore, if they arc not logically required, then their introduction violates
the principle of parsimony by unnecesarily complicating a theory. Theorizing
devoted to determining the properties of the hypothetical construct further
diverts attention toward inner mechanisms not readily manipulated and away
from the environmental factors responsible for both behavior and its phys-
iological n1ediators. Again, the argument is strategic, not philosophical.'"'
Son1e behaviorists argue, furthermore, that hypothetical constructs are al-
ways dispensable. Constructs are said to intervene between independent and
dependent variables. If this relationship is lawful, then the insertion of the
construct does not add to the law and is unnecessary. Similarly, if the con-
struct is lawfully related to the independent variable, and the dependent vari-
I !J able is lawfully related to the construct, then again, the dependent variable is
t:i ! 1
:'.I , . lawfully related to the independent variable, and the construct is useless. If,
I .
·I on the other hand, there is no lawfulness, the construct cannot create it. In
' '
either case, the construct is dispensable, generating the behaviorist "paradox
:.! II1:' of theorizing." 69
,:;[ '1
Ii" . 1·
::i : '
1: :
1; :
Ii Hypothetical Constructs in Behaviorist Theories
In the absence of decisive arguments for or against the use of hypothetical
I. constructs, the decision to introduce them into bcha viorist science is left to
'I: t
the strategic intuition of the individual theorist. In point of fact and contrary
i ! to the popular image, the majority of behaviorist theories include hypothet-
HYPOTHETICAL CONSTRUCTS 79
ical constructs among their theoretical terms. Although there are a vast nu111-
ber of such constructs to be found throughout behaviorist theories, they can,
for the most part, be classified into four broad categories.
One category can be characterized as consisting of covert responses. These
are conceptualized as similar in all essentials to observable responses except
that they arc small in scale or else located inside the body. This class includes
the subvocal speech and visceral responses Watson and others use to explain
thought and emotion. 711 Hull's anticipatory goal reaction, Spence's emotional
responses, and Mowrer's "hope" and "fear" form another subclass. 71 Internal
perceptual responses, including the "coding response," arc also of this type. 72
Finally, one very large subclass of the covert response catcogry is the "me-
diating response." This covert response has been assigned manifold functions
within behaviorist theories, ranging from the mediation of generalization in
the absence of physical similarity to the acquisition of meaning in verbal be-
havior. 73
Covert stimuli form a second broad category of hypothetical constructs.
These are hypothesized to arise from at least three possible sources. First, cer-
tain stimuli arc caused by various physiological processes within the body.
Deprivation of food, for example, leads to physiological changes which con-
stitute potential stimuli for behavior. Or these stimuli may be caused by overt
behavior in the form of proprioceptive stimulation arising fron1 muscles and
tendons, such as the movement-produced stimuli of Guthrie's theory. 74 Third,
covert stimuli may be the result of covert responses, as in the stimulus con-
sequences of Hull's pure sti,nulus acts or anticipatory goal reaction, which
may in turn elicit overt behavior. 75
A third set of hypothetical constructs consists of long-term physiological
mechanism thought to underlie behavioral states or dispositions. State vari-
ables pose a proble111 for those behaviorists who prefer to conceive of re-
sponses and stimuli as brief and momentary events because states are long
lasting and continuous. However, tonus and postural reflexes provide a model
to resolve this difficulty. Therefore, a number of dispositional concepts, in-
cluding attention, determining tendencies, anticipatory set, attitude, and
readiness, arc interpreted as states of n1uscle tonus, postural reflexes, or their
conceptual ofTspring, the "motor set. " 7 <•
The fourth category of hypothetical constn1ct differs significantly from the
others. In this category belong overt stimuli and responses which arc not hid-
den inside the organisn1 and arc therefore not unobservable in the same sense
as internal events are. Nevertheless, these events are hypothetical in that they
arc postulated although not actually observed, identified, or recorded. Con-
sider, for example, a competing-response theory of extinction. According to
this theory, an unreinforced response stops occuring because incompatible re-
sponses are conditioned and interfere with it. These competing responses are
hypothesized in order to explain an observed decrease in a particular re-
sponse, but the competing responses are not actual! y identified. 77
 , I
'i 80 THEORETICAL CONCEPTS
I
5
1.
Characteristics of Hypothetical Constructs in Behaviorist Theories
1'.ri1''
'I :I., Although no restrictions arc logically imposed on the postulated properties CHAPTER
·I, I of the hypothetical construct, they nevertheless generally conform to four
I·,, I
'
conditions. First, the properties assigned to the hypothetical construct do not
differ substantially from those of observed stimuli and responses. They are
!l l
\ l conceived as material rather than n1cntal and as obeying the san1e laws as overt
' !

:, I
1 stin1uli and responses, with the possible exception that constants in the equa-
tions may differ somewhat because of the reduced size of covert forms. Sec-
78 Theorizing
ond, hypothetical constructs arc con1monly located peripherally, in the n1us-
:r:II: ··I. '
,-,1 cles and glands, rather than in the central nervous system_.,,, Third, hypothetical
'·.
I \ii• constructs are typically thought to operate under the functional control of en-

l 'I ..
),
vironmental variables, that is, their activities arc not "autonomous.""" Fourth,
although this condition is rarely achieved, it is preferable that the hypot_heti-
Ii'I'i . cal construct be linked by a functional relationship to an observable envtron-
mental variable, or an observable behavioral variable, or, ideally, to both. This
' '
,I
. ,1 linkage is different from the intervening variable. With the hypothet_ical con-
struct, the linkage relates three distinct entities, and _the hypothetical _con-
struct has a meaning and existence beyond that provided by the functional
relationship.
These four conditions con1bine neatly in a comn1on example of a hypo-
thetical construct. It is hypothesized to begin as an overt muscular response,
acquired through the normal laws of learning as a function of external_ vari-
ables. It gradually regresses in magnitude until it is unobservable. In its re-
duced state it yet remains a response, still n1uscular, and still under the func-
81
tional control of environment variables.
These four conditions help to guard against behaviorist fears of unwar-
ranted speculation and the introduction of "fictions." The first condition as-
sures that no properties will be attributed to the construct that are not prop-
I '
\i erties of observables. The second and third conditions allay behaviorist qualms
!1,,
about the taint of n1cntalism. Behaviorists often argue that because so little is
known about the central nervous syste1n, it serves as the last refuge of the
soul in psychology. Processes hypothetically located there often have quali-
ties of free will, agency, and the mental. 82 Therefore, by locating covert even~
in the periphery and keeping them under the functional control of the envi-
ronment, behaviorists prevent the contamination of their constructs by men-
talism. The fourth condition, in conjunction with the first, helps to ensure
, about the continuity o_f theory development aud tl1e relationship o_{ theory to data.
that the theory in which the hypothetical constructs are en1bedded is rigor-
ously testable.· Unconfirmed postulates can be discarded along with their hy-
pothetical constructs. Thus although behaviorist hypothetical constructs are
inferred rather than observed, they are closely related to observables by short
lines of inference, thus retaining their en1piricalness and objectivity.
Behm,iorist theorizill.rt occurs in many fonns. At the two extremes are the hypothet-
ico-deductivism of Hull and the descriptiFism of Ski1111cr.
The stat11s of H11/lian theoretical co11cepts is some11,Jwt ambiguous. Some arc clearly
hypothl'tical co11stn1cts. Others appear to be i11terveni11g i,ariables /mt operate as hy-
pothetical co11stmcts.
H11/l's hypotiletico-deductive method is 11ot like ded11ctii,e systems in lo,{!ic a11d
mathematics. I11stead, liis poswlates de.fine coucepts as hypothesized q11a11titative .fi111c-
tio11s of empirically defmed varia/,/es, and they state h)•potheses as to /zow these con-
cepts combine to determi11e behavioral varial,les.
The purpMe of a theory, according to Hull, is to mediate the ded11ctfon of theorems
about observable behavior. Theorems are experimentally tested and eitlier add confir-
matioll to a theory or force its revision. Theory develops thro11gh th is process of de-
duction, test, and 1-ci,ision. Explanatio11 is achieved whe11 the description of a phe;1om-
enon is dcd1.JCed as a tlieorem .fi'om a thco1y.
Skinner's theoretical concepts consist of intervening variables, s11ch as drive, and
private events, consisting of covert stimuli and responses. Althott,{!h the status of the
latter is not clear, they are best viewed as hypothetical co11stmcts.
Skinner objects to theories 011 the grou11ds that thq, i11hibit fi,1it/i1l research. In-
stead, he advocates a descriptiPism with an emphasis on discove,:in.f? ~rderliness in be-
havioral data. Theory, for Skinner, co11sists of eco11omical descriptions of functional
relations which subsume a 11umhcr of behavioral regularities.
Both Hullian and Skinneriau meta-theory are open to Cl'iticisms from a nonlinear
model of the de/Jelopment of science. This model q11estio11s l1ehaviorist assumptions
Despite behaviorist attempts to link concepts to the l,elwi,ioral data language, the
relationship of concepts to observation is 11nderdetermi11ed. This /inka.1:e is ultimately
psychological. Within this context, bchalliorist meta-theory can be i,iewed as a deci-
sion to specify concepts precisely enough for scientists to learn to achiei,e intersubjecti ve
agreement, prediction, and control while usin.<! tlzem.
 ..
.. .
.
.... ......
_
G. E. Zuriff

BEHAVIORISM:
A Conceptual
Reconstruction

COLUMBIA UNIVERSITY PRESS

(i j
NEW YORK

l
i
r

Preface

This book has been brewing since 1962. As an undergraduate philosophy n1ajor
at Columbia that year, I decided to learn son1ething about psychology. But
the introductory psychology course I took, the notorious one developed by
Keller and Schoenfeld, could have been better entitled "An Introduction to
Operant Conditioning," because that is what psychology was at Colutnbia
in those days.
Unlike most of my classmates, who grumbled about the narrowness of the
course, I found myself intrigued by the behaviorist approach to the mind.
My reading of Skinner's Science and H11ma11 Behavior affected me profoundly,
and I chose philosophy of psychology to be my field as it has been ever since.
Columbia was a particularly good place to begin, with behaviorism as an ag-
gressive and lively ideology in the psychology department and a renewed in-
terest in the smdy of mind in the philosophy department.
The next logical step seemed to be Harvard to work with Skinner. By the
time I arrived at Harvard. Skinner was retired from active laboratory re-
search, and most of my contact was with Dick Herrnstein. Dick had some-
how acquired the quaint notion that psychologists should devote themselves
to discovering the laws of behavior rather than fighting the ideological battles
of behaviorism. If you nagged him enough, he could talk philosophy of sci-
ence as well as anyone, but for the most part, my four years in his lab were
spent learning how to do good science. Eventually I emerged with several
completed experi111ents in operant conditioning, a couple of publications, a
thesis, and a Ph.D.
Concurrent with n1y work in the lab, I managed to establish some contact
with Fred Skinner and to learn about the exciting issues he was pursuing. I
also took advantage of the offerings of the Harvard philosophers, including
Van Quine and later, Hilary Putnan1 and Nelson Goodman.
In 1968 I assumed n1y current position in the Departn11.:nt of Psychology
at Wheaton College. Wheaton's liberal attitudes toward research and its sup-
l(

XII PREFACE PREFACE XIII

portive collegial atmosphere allowed me the two years of floundering it took continued encouragen1ent and Peggy Rosenthal who got me to stop research-
to define my intellectual goals. Also at this time, the founding of the journal ing and start writing. Finally, l want to share the joy of this accon1plish1nent
Behaviorism by Willard Day finally provided a ready outlet for my writing , with my parents whose love and support were always there for me. They
which, because of its philosophical orientation, did not find a welcon1c re- struggled so that I could have the opportunities denied them. The result is a
ception fro1n other journals. book they do not fully understand but which they fully appreciate.
During n1y first sabbatical, spent in Berkeley, California, I conceived the GERALD ZURIFF
idea for this book. Upon my return to Wheaton in 1975, I began developing
March 12, 1984
an outline and bibliography. Research began in earnest in 1977. I spent the 1
Cambridge, Massachusetts
next four years ensconced in the Psychology Research Library at Harvard
University, reading everything written about and by behaviorists. I am in-
debted to Miss Annelise Katz, Librarian, for her assistance and snacks during
that trying period. Bleary-eyed and enriched by hundreds of note cards, I be-
gan the actual writing of the book in 1980.
My writing accelerated during the academic year 1981-1982, n1y second
sabbatical, granted and financed by Wheaton College. I was also generously
supported by a National Endowment for the Humanities Fellowship for Col-
lege Teachers. My thanks also to the Department of Psychology and Social
Relations at Harvard University which provided n1e with an office, intellec-
tual stimulation, friendships, and a position as Visiting Scholar that year. Funds
to cover the costs of manuscript preparation were provided by a grant fron1
the Mellon Foundation and small grants from the Comn1ittee on Faculty Re-
search at Wheaton College.
Portions of the first draft, completed in Noven1ber, 1982, were read by a
number of colleagues: Jin1 Blight, Dick Herrnstcin, Pere Julia, Drazen Prekc,
Howie Rachlin, Fred Skinner, Larry Sn1ith, Mark Snydcrman, and Gladys
Topkis. My thanks to all of them for their comments and suggestions. A special
con1mendation to Barry Schwartz who was the first person, other than pos-
sibly myself, to read the entire manuscript and whose critique was particu-
larly helpful. Incorporating the suggestions of these readers, I completed the
final draft in August, 1983. Any deficiencies in the final product arc sold y
the responsibility of these readers, either for not noting the probk1ns or for
not persuading n1e to nuke the necessary changes.
I am grateful to a nmnber of people who were instrun1ental in preparing I
the manuscript: Nancy Shepardson and her staff for 1nuch of the early typ- '
ing, Sue Taylor for her wizardry at word processing, and Leslie Bialkr for
his skilled editorial pencil. Maggie Vaughan, assisted by her bottle of extra-
strength Tylenol, prepared the entire bibliography, a herculean task, per-
formed with dedication and care. My special appreciation goes to Susan Kos-
cielniak and Columbia University Press for their wisdom and courage in de-
ciding to publish this book at a tin1e when other university presses would not
even consider a book that is scholarly, behavioristic, and philosophical. I hope
Columbia makes a lot of money on it.
On the personal side, I thank Ken Prager and Elizabeth Van Ranst for their I,
j
I
l
''
1

CHAPTER 1
General Overview

Behaviorism is the co11cepmal _framework 1111derlyin,f? a particular science of behavior

ratlzff than that scieuce itse[f. This .framework consists of a philosophy of science, a
philosophy of mi11d, an empirical backgrotmd theory, and an ideolo,r;y. lvf y goal is to
present a co11ffptual reconstruction of behaviorism, l1e,t;i1111ing with a _few fimdamental
premises and then examining their logical dei>elopment. 111 followiu,(; a logical rather
than a chronological order, such a reco11strnction differs ./i'om a history of behat>iorism.
Yet, the reconstruction will take into acco1111f ideas proposed thro1t,(;lro11t the history of
behaviorism.
Because it is _fruitless to look far 11ecessary and s1![ficient conditions to identify be-
hai,iorist ideas, I shall portray a loose family resemblance which characterizes the va-
riety of positions that constitute behaviorism. I shall include both philosophical behav-
iorism and behavior therapy as members of the behaviorist conCl'plllal family.
In this reconstruction of behaviorism, the fimdamental premise is that psychology is
a natural science. Two corollaries are that psycholo,(;y is to be empirically based and
that it is to be objective. To a lm~(;e extent, the reconstruction of behaviorism is the
elaboration of what it means far psychology to be empirical and olyective.

After dominating American psychology for nearly a half century, behavior-

ism today finds itself on the defensive. New and competing approaches to
psychology have arisen in the wake of advances in psycholinguistics, cogni-
tive science, and philosophy. A reformulation of behaviorism is called for-
one that takes into account these critical challenges. Some of them may be
integrated into behaviorist thought to create a more sophisticated and viable
behaviorism. Others must be either discarded if found wanting or conceded
as damaging refutations of aspects of behaviorisn1.

BEHAVIORISM'S "ISM"
Behaviorisn1 is not the science of behavior developed by behaviorists since
the turn of the century. It is, rather, the conceptual fra1ncwork underlying
2 GENERAL OVERVIEW CONCEPTUAL RECONSTRUCTION 3

that science. The science itself, at times called "praxiology," "behavioris- philosophy of science and philosophy of mind extends to this empirical back-
tics," and "anthroponomy," consists of the findings, principles, laws, and ground theory.
1 Fourth, behaviorism also represents a certain set of values. It recommends
theories formulated through the study of behavior. Behaviorism, on the other ;
hand, represents the assumptions, values, and presuppositions implicit in this , goals for behavioral science and suggests standards for evaluating scientific
2 activity. Values are even more salient with respect to applied behavioral sci-
science.
To articulate the implicit conceptual framework of behaviorism requires a ence in which behaviorism promotes applications congruent with particular
special methodology. For the most part, behaviorists are, and have been, social aims. Behaviorism must therefore be seen as an ideology as well as a
dedicated scientists whose foren1ost concen1 is their experimental research and philosophy of psychology. Philosophy and ideology are tightly integrated in
theorizing. Relatively little of their writings is devoted to the explication of ' behaviorism, and one cannot be fully understood without the other.
the philosophy underlying their research. This was especially true during the ' These four interdependent components comprise the conceptual frame-
decades when American experimental psychology was synonymous with be- i work of behaviorism. 4 A major goal of this book is to explicate this frame-
haviorist psychology, and behaviorism was taken for granted. Therefore, the , work. It will not, therefore, be concerned with presenting particular theories
unearthing of behaviorism is an inductive task. First, an entire corpus of be- of learning or specific methods of behavior modification, except in so far as
haviorist research must be scrutinized to find its characteristic features. Then they illustrate conceptual points.
it is necessary to determine the sorts of presuppositions that must underlie '
the research in order for it to have manifested these features. CONCEPTUAL RECONSTRUCTION
Although behaviorism is an integrated framework, it is possible to discern
four components. First, behaviorism is a philosophy of science. Before any There are several ways in which this goal can be achieved. One might begin
empirical findings, behaviorism dictates canons concerning what sorts of by inspecting the best of contemporary behavioral research and constructing
psychological questions are worth pursuing and what methods are acceptable a philosophy to characterize and guide it. Although this approach may gen-
in searching for answers. Similarly, behaviorist doctrines define what kinds erate a sound philosophy, it is not true to behaviorism. First, it ignores the
of concepts are acceptable and what form a psychological theory should take. history of behaviorism over the past seventy years. More important, it misses
Behaviorism also advocates certain criteria for the adequacy of a scientific ex- the diversity of behaviorism. Although behaviorism is a "school" of psy-
planation. Thus, behaviorism provides a strict framework in which behav- chology, it encompasses a variety of philosophical positions, many of which
iorist scientific activity takes place, thereby detennining, a priori, important ,, still stand as powerful and interesting approaches to psychology. 5 Thus, by
1
characteristics of the resulting behavioral science. ignoring the past, de novo philosophizing adopts the term "behaviorism" only
Second, behaviorism is also a philosophy of mind with certain assumptions,, by expropriation while losing the conceptual substance of the term.
about human nature as well as about the essential workings of the mind. This A second method of explicating behaviorism is by analyzing its develop-
philosophy of mind is interdependent with behaviorist philosophy of science; ment since its inception and describing the changes in its fundamental ideas
each justifies the other. Given the assumptions of the behaviorist philosophy over successive periods. 6 Such a historical reconstruction would be true to
of mind, the kinds of methods, theories, and explanations favored by behav- both the history and diversity of behaviorism. However, this approach has
iorist philosophy of science appear most appropriate. Conversely, the behav- i its own disadvantages: it creates difficulties in the analysis of individual con-
iorist philosophy of science supports its philosophy of mind. A science re- ' ceptual issues, many of which cut across historical periods. A chronological
stricted to a limited set of methods and explanations will tend to confirm a approach misses the abstract continuity that transcends historical demarca-
3 tions. Furthermore, in a historical reconstruction the emphasis is often on
particular conception of mind.
Third, although the emphasis in the present work will be on the concep- , particular figures rather than conceptual themes. Interest shifts to the intellec-
tual, inextricably interwoven in the fabric of behaviorism are several impor- ,: tual development of prominent individuals, the influences acting on them, the
tant empirical assumptions. These very general e1npirical hypotheses consti- I consistency of their thought, and the correct interpretation of their writings.
tute a background theory for behavioral theories that n1ay otherwise differ : Conceptual issues are easily lost in the welter of these historical questions.
considerably from one another. These empirical presuppositions consist of a 1 This book adopts a third approach, that of conceptual reconstruction. The
few basic assumptions about behavior and its relationship to the environment .' · entire scope of behaviorism, roughly 1910 to the present, is considered, and
and also a set of beliefs about the relative effectiveness of various methods in, the reconstruction is organized around conceptual issues rather than historical
scientific research and theorizing. The interdependence between behaviorist periods or important individuals. The conceptual fra1nework of behavioris1n
 4 GENERAL OVERVIEW
is elaborated as a logical rather than as a chronological dcvclopn1ent. It begins
with a few fundan1cntal premises, but because the implications of these basic
underlying assun1ptions can be developed in more than one way, the recon-
structed conceptual framework is organized more like a branching tree dia-
gram than like an extended linear systen1. Each node of the tree represents a
conceptual choice point, often created by a criticism of behaviorism or by the
application of behaviorist analysis to a new question. Each branch growing
from a node syn1bolizcs a different behaviorist conceptual decision in re-
sponse to the choice point. Thus, from an initial set of sin1ple pren1ises, an
elaborate, rich, and complex network develops, with branches leading to nodes,
and nodes generating further branches. The branching tree image emphasizes \_
that bchaviorisn1 is a family of conceptual positions rather than a homage- ;
f neous school of psychology.
The justification for this conceptual reconstruction lies in its success in n1aking
I intelligible the scientific activities of a large number of behavioral psycholo-
gists. By explicating the conceptual framework underlying an enorn1ous cor-
pus of research and theory, it imposes a coherence on an important but diverse
body of American psychology. Most important, in reconstructing behavior-
ism in an abstract and coherent way, it exposes both strengths and weak-
nesses, leading to the formulation of a sound behaviorism to meet the chal-
lenges of contemporary psychology and philosophy.
A conceptual reconstruction differs from a history of beha vioristn in sev-
eral respects. First, chronology is not respected, and chronological anon1alies
abound. For example, a criticism of behaviorism raised in 1970 n1ay be ade-
quately answered by a behaviorist idea appearing in 1920. Second, there is no
concern for preserving consistency in the thought of individual behaviorists.
As practical scientists, behaviorists are not usually overly concerned with
philosophical nuances, nor are they averse to changing their views. There-
fore, their pronouncements often reveal ambiguities, contradictions, and gradual
changes. Third, the importance of an idea in a conceptual reconstruction n1ay
not correspond to its historical significance. Often a position that functions
critically within a conceptual reconstruction may have been proposed by
someone who l~ad little influence on other behaviorists. Fourth, in a concep- 1
tual reconstruction, ideas are examined independent of their original context.
Conceptual positions maintained to justify a particular research program or
proposed as a defense against a specific criticism are examined in isolation
and related to issues they were never intended to address.
THE BEHAVIORIST DIALOGUE
A convenient fiction for understanding the practice of conceptual reconstruc-
tion is the "behaviorist dialogue." In an imaginary forum, all behaviorists as
well as all their expositors, comtnentators, and critics, from the beginnings
THE BEHAVIORIST DIALOGUE 5
of the movement to the present, arc gathered to discuss the major conc.eptual
questions of psychology regardless of whether their lives, in fact, overlapped.
Early behaviorists respond to questions which may not have been raised until
after their deaths. Behaviorists with no following whatsoever speak with as
much right as the n1ost illustrious founders of behaviorist systetns. Also pre-
sent arc thinkers whose ideas are highly relevant to the task of conceptual
reconstruction although they do not discuss behaviorism directly.
Why include the early behaviorists in the dialogue? To be sure, the science
of behavior has made significant strides over the past seventy years, and the
behavioral scientists of 1920 have very little of interest to say to their coun-
terparts of today concerning behavioral theory. However, on the conceptual
level, bchaviorisn1 (and psychology in general) has not shown similar prog-
ress. The fundan1ental questions concerning the nature of psychology have
not been answered, nor is it obvious that progress is being nude. Instead,
certain recurrent themes are discussed, debated into a numbing stillness, and
dropped unsettled, only to reappear years later in a different guise and under
a new tern1inology. Therefore, the behaviorist of 1920 n1ay have n1uch of
relevance to say to the modern psychologist forn1ulating a conceptual frame-
work for psychology. Indeed, early behaviorist writings often are 1nore ex-
plicit on the reasons behind a particular position than later works, which may
adhere to a behaviorist tenet out of a sense of loyalty to a school even though
the original reasons have long been forgotten or may no longer be valid.
As n1odcrator of the dialogue, I structure the discussion, choosing the is-
sues to be exan1ined and developing their logical implications. As interpreter
of the dialogue, I 111akc the discussion accessible to the modern reader. Much
of early behaviorist writings appears in an idion1 foreign to conten1porary ears,
and I translate the early tern1inology and style into the language of today.
When necessary I also extrapolate from a behaviorist's position on one issue
to the topic under discussion.
As evaluator and arbiter of the dialogue, I attempt the task of constructing
a behaviorism that is sound by contemporary standards. Not all the positions
articulated in the behaviorist dialogue can be accepted as valid. Some have
internal flaws while others are discredited by recent research which has re-
futed aspects of their empirical background theory. Yet others have not ad-
equately met objections raised by their critics, including other behaviorists.
My goal is to prune the behaviorist tree diagran1 of these weak positions and
to note those paths through the diagram which I believe are critically sound.
Although there is more than one valid path through the behaviorist tree
diagram, none of them is complete. Behaviorisn1 has not responded fully to
the many objections raised to challenge it. In some cases behaviorists have
sin1ply been unaware of the challenges. In others, there has not been suffi-
cient time for assi1nilating and n1eeting the challenge. As yet, a sound n1od-
en1 behaviorism has not been fully developed. However, I believe that a sound
 6 GENERAL OVERVIEW
behaviorism generally consistent with the valid pos1nons of traditional be-_
haviorism can be developed. My fourth role, therefore, in addition to n1od-
erator, interpreter, and arbiter, is that of active participant in the behaviorist i
dialogue, filling the conceptual gaps or, at least, pointing in the direction of·
solutions. More often than not, pointing is all that is possible because the res- :
olution of a conceptual issue often depends on facts as yet unknown or on'
questions of values and personal preference.
WHO IS A BEHAVIORIST?
A conceptual reconstruction presupposes that behaviorist ideas can be iden- \
tified before the characterization of behaviorism. But how can we decide whom!
to ad111it to the behaviorist dialogue without first knowing the results of that \'
dialogue? Ideally what is needed is a set of necessary and sufficient conditions :
for an idea to qualify as behavioristic. However, given the diversity of be-,
ha viorist views, no such conditions exist. 7 The intellectual comn1itn1ents shared ·
by all behaviorists are so broad that they are held also by most nonbehavior- ·
ist and even antibehaviorist schools of experimental psychology. Conversely,
although it is possible to formulate a particular behaviorist position that is
clearly distinct from every nonbehaviorist psychology, this position would ,
not accurately describe the views of a large number of behaviorists. 8
A more fruitful approach is suggested by Wittgenstein's concept of"family
resen1blance." 9 In looking at a photograph of a large family, one may note
that although not every fa111ily n1en1ber has the same nose, or eyes, or mouth, ·
there is a similarity of features that constitutes the family resemblance. Each
family me1nber shares the rcse111blance even though there is no one feature .
i that characterizes each and every member. S0111e members show the fa1nily
' "look" very strongly; others, so weakly that they appear as belonging to the
family only in the context of the photograph.
So too behaviorists can be viewed as sharing a fa111ily rese111blance. Al-
though they cannot be identified by a strict set of necessary and sufficient
properties, they can be recognized by their com111011 family resemblance: a
set of overlapping features, some related by tics of similarity and son1e by'
historical association. Some of these features are more i111portant and relevant.
than others for defining behavioris111. An idea 1nay be said to be a "central" ,
family trait of behaviorism if: (1) it is shared by a large number ofbehavior-i
ists, and (2) it appears early in the lo.i:ical (not chronological) development of)
the behaviorist tree diagram and therefore plays a major role in deter111ining ':
the nature of subsequent choice points and branching positions. A position/
without these characteristics may be said to be "peripheral." Many peripheral'
positions lie on the border between behaviorism and another school, often'
resembling the latter more than the former.
To recognize the fan1ily resemblance the equivalent of a family photograph·•
I
.
)
WHO IS A BEHAVIORIST? 7
is needed. 10 The portrait of behaviorism as drawn by its critics cannot be used
for this purpose since the result would be merely a picture of a group of
scowling straw111en. Nor should the photograph include everyone who has
ever claimed to be a behaviorist; self-identification as a behaviorist is not a
sufficient condition for inclusion in the behaviorist family. 11 Conversely, re-
jection of behaviorism i: not _a su~cient condition for exclusion. ,;"1-ajor _be-
haviorists have on occasion d1ssoc1ated themselves from the label behavior-
ist" to distinguish their positions fro1n those of their behaviorist colleagues
while maintaining theories that, in retrospect, are clearly behavioristic. 12 In
addition, critics of one version of behaviorism have occasionally espoused points
of view congruent with other versions. Thus, an idea historically presented
as "antibehaviorist" 111ay find its way into the behaviorist family.
Even without any clear-cut criteria for identifying 111cmbers of the behav-
iorist family, a small number of psychologists are universally recognized as
behaviorists. Among the early behaviorists this group includes J. B. Watson
(1878-1958), the founder and popularizer of the movement, and A. P. Weiss
(1879-1931), a vigorous early chan1pion ofbehavioris111. As behavioris111 de-
veloped from a program1natic philosophy of psychology into a conceptual
basis for well-developed psychological theories, "neobehaviorists" appeared
with competing theories of behavior. Prominent among these are E. R. Guthrie
(1886-1959), E. C. Tolman (1886-1959), C. L. Hull (1884-1952), K. W.
Spence (1907-1967), and B. F. Skinner (1904- ).
From an examination of the works of this small prototype group, a family
resemblance emerges. If additional ideas which share the resemblance are added,
the structure is slightly altered, thereby allowing for the incorporation of still
other ideas. Eventually a broad behaviorist conceptual family develops, and
that resulting family resemblance is characterized in this book. This final
characterization of behaviorism n1ay not coincide perfectly with initial intui-
tions. For example, the theories of Tolman, one of the original "prototype"
behaviorists, appear on the periphery of the behaviorist conceptual family as
finally characterized, rather than at its center.
Philosophers and Clinicians
Two groups of behaviorists deserve special mention. First is a group of phi-
losophers whose approach to psychology is closely related to that of behav-
iorist psychologists. Some of them developed a philosophy of mind in which
mental concepts arc to be understood in terms of observable behavior rather
than private consciousness. Because this philosophy of mind, known as "an-
alytic behaviorism," is congenial to psychological behaviorism, some behav-
iorist psychologists borrow from analytic behaviorists, such as Ryle and
Wittgenstein, in formulating their conceptual positions. 13
Other philosophical behaviorists were more concerned with philosophy of
 .'
l
8 GENERAL OVERVIEW
science. These philosophers, known as "logical behaviorists," include logical '
positivists such as Carnap and Hcmpel. 14 They attempted to develop canons
to ensure the empiricalness and n1caningfulncss of scientific statements. Be-
cause these concerns arc shared by behaviorist psychologists, many behavior-
ists show the influence of logical behaviorism, especially during the period
when An1crican philosophy of science was dominated by logical positivism.
Philosophical behaviorists perceived no important alliance with psycholog-
ical behaviorism because they identified it with Watson's early brand of be-
haviorism. However, later and n1orc sophisticated versions of bchavioris111
have much in common with philosophical behaviorism. For this reason, the
various versions of philosophical bchaviorisn1 arc here included in the behav-
iorist conceptual fa111ily.
A second group of behaviorists meriting special attention consists of be-
havior therapists. In fact, today the term "behaviorist" is often identified with
behavior modification in the mind of the general public. In its beginnings,
behavior therapy, or behavior modification, grew out of behaviorist science.
The early therapy techniques were closely related, in name if not in fact, to
the 1nethods, findings, and theories of behaviorist learning theory. Similarly,
the philosophy of science underlying early behavior therapy had much in
con1mon with that of behaviorist expcrin1ental psychology. As behavior
therapy developed, however, its techniques, methods, and philosophy be-
came increasingly diverse. Today some of the 1nany versions of behavior
therapy continue to share a strong fa111ily resemblance with bchaviorisn1 while
others belong only on the periphery of the behaviorist family and have only
the most tenuous tics with behaviorism. 15 Therefore, in the present recon-
struction of behaviorism, some ideas associated with behavior therapy are ,
presented as behaviorist positions and others are considered criticisms of be-
haviorism.
PSYCHOLOGY IS A SCIENCE
The first premise in the conceptual reconstruction ofbehaviorisn1-thc open-
ing statement of the behaviorist dialogue-can be found in the first sentence
of Watson's f.1111ous article "Psychology as the Behaviorist Views It," the formal
beginning of behaviorisn1:
Psychology as the behaviorist views it is a purely objective experimental branch of
natural science. Its theoretical goal is . . . prediction and control. . . . (1913a, p. 158) ·
The premise is clearly enunciated: Psychology is a natural science. 16 All the
remaining conceptual reconstruction of beha viorisn1 is essentially an elabo-
ration of what it means for psychology to be a natural science.
In describing psychology as a "branch of natural science," the passage im-
plies some sort of unity among the various sciences. Although the sciences
PSYCHOLOGY IS A SCIENCE 9
may differ from one another in their subject niatter and their special tech-
niques, they have a ~on1monality .that n1akes it possibl_e to see the v~ri~us
sciences as "branches of natural sC1cncc. Thus, bchav1onsm secs a contmmty
among the sciences, and between psychology and the other biological sci-
ences in particular.
The passage also associates science with prediction and control. One way
of interpreting this is that behaviorists are suggesting a useful goal for their
science. A stronger interpretation is that prediction and control are proposed
as the defining characteristics of their science, or even of science in general.
That is, an activity is judged as "scientific" to the extent that it contributes
to the prediction and control of phenomena.
The last sentence of the quoted passage reads in full: "Its theoretical goal is
the prediction and control of /1ehavio1·." I omitted "of behavior" because the
belief that behavior is the su bjcct matter of psychology occurs later in the
conceptual reconstruction. From the mere premise that psychology is a sci-
ence, it does not follow that behavior is the only possible subject matter. Such
a conclusion follows from the initial premise only in conjunction with several
other corollaries, two of which arc especially noteworthy.
The first is that science, and psychology in particular, must be empirically
based. 17 This corollary is developed in great detail in the course of the con-
ceptual reconstruction. For now, the corollary can be interpreted in very gen-
eral terms. The cmpiricalness of psychology will be taken to mean that the
fundamental facts of psychology are to be derived through the senses, pref-
erably through careful perception, and ideally through experimentally con-
trolled observation. The corollary of empiricalness excludes the kinds of
knowledge derived from nonempirical sources such as mystical revelation.
empathy, intuition, religious dogma, innate ideas, and authority.
The second corollary is that science, and psychology in particular, must be
objective. 18 The meaning of objectivity for behavioris1n, like the meaning of
empiricalness. is apparent only in the context of the full reconstruction of be-
haviorism. At this point, objectivity will be taken to 111ean simply that psy-
chology should be independent of the individual prejudices, tastes, and pri-
vate opinions of the scientist. Statements of findings and theories should
therefore be as precise and unambiguous as possible so they are not subject
to diverse individual interpretations.
Many of the methods excluded by the empiricalness criterion, such as em-
pathy and intuition, are also excluded by the objectivity criterion. Methods
sanctioned by the former are generally also acceptable by the latter. Results
of empirical n1cthods are objective in that they are open to anyone's obser-
vation and do not depend on the subjective belief of an individual scientist.
Nevertheless, objectivity and empiricalness are not the san1c. For example, as
is discussed in the next chapter, introspection, as a 111ethod for acquiring
knowledge about the mind, has been considered by some to be e1npirical and
yet nonobjective. The two corollaries will therefore be kept distinct. To-
10 GENERAL OVERVIEW

gether the two support Watson's view of psychology as a "purely objective

experimental branch of natural science."

SYNOPSIS
PART I
To a striking degree, the conceptual reconstruction of behaviorism is the
elaboration of the fundamental premise that psychology is a branch of natural
science given the two corollaries of empiricalness and objectivity. Part I
(chapters 2-5) discusses how this premise and its corollaries develop into be- THE SCIENCE
haviorist views on the structure of science. Behaviorist science is founded on
observation, and chapter 2 shows how the canons of objectivity and empiri-
calness dictate what types of observation are acceptable by behaviorist stan-
dards and why introspection is rtjected. Chapter 3 examines behaviorist views .,:,
on bow observations are to be transforn1ed into data reports that retain em- :!
piricalness and objectivity. In chapters 4 and 5, behaviorist notions about the
nature of psychological theory are explored, concentrating on the question of
what sorts of theory can guide research and at the same time maintain em-
pirical links and objectivity.
Part II (chapters 6-8) looks at the major behaviorist assumptions about the
nature of behavior, that is, behaviorism's background theory. Chapter 6 dis-
cusses various n1eanings of "stimulus-response psychology" and rejects nearly
all of them in favor of a very broad definition compatible with behaviorist
premises. Chapter 7 defends this broad conception of stimulus-response psy-
chology against two major objections: (1) the contention that behavior is or- I
ganized by purposive principles incompatible with the principles of sti1nulus- I
response psychology and (2) the psycholinguistic challenge that the organi- I

zation of verbal behavior cannot be accounted for by stimulus-response prin- I

i'
ciples. Chapter 8 discusses how stimulus-response psychology, broadly con- ,I•!
ceived, approaches complex psychological processes and defends this approach
against the objections raised by cognitivism.
Part III (chapters 9-12) is concerned with behaviorist attempts to deal with
mental concepts in a scientific way. Chapter 9 explains why behavioris1n re-
jects the concept of a self and how this concept can be replaced in behaviorist
theory. Chapters 10 and 11 review behaviorist treatments of mentalistic lan-
guage when it is used to describe others (chapter 10) or when it is used to
talk about one's own mind (chapter 11). Finally, chapter 12 presents a behav-
iorist theory of knowledge loosely derived from behaviorist psychology. This
psychologically based epistemology closely links behaviorism to the intellec-
tual traditions of positivism and pragmatism. The thesis that behaviorism can
be understood as a psychological pragmatic positivism is then used to char-
acterize the behaviorist conceptual family resemblance and to make intelligi-
ble fundan1ental behaviorist assumptions about science, en1piricalness, and !
!
objectivity. i
I
1
j