Académique Documents
Professionnel Documents
Culture Documents
Predictors of Innovativeness
LYMAN E. OSTLUND"
Two studies are reported that apply diffusion theory to two new consumer
products. The research objective in both studies was to discriminate
buyers from nonbuyers using the personal characteristics of respondents
and their perceptions of each new product and thereby also to assess the
relative Importance of the predictor variables. In both studies perceptual
variables are found far more successful as predictors of the purchase
outcome than respondent personal characteristics.
or would not adopt the innovation, that is, the innovativeness of individuals. According to the Rogers
(1962:307) adoption model perceived innovation attributes are supposed to provide the framework for
the evaluation of an innovation by the potential adopter,
but Rogers did not refer to this kind of research in
his typology of "analyses completed or possible by
diffusion researchers" (Rogers, 1962:73), nor does the
expanded typology by Rogers and Shoemaker (1971:
72-73).
The importance of perceptual variables in determining purchase behavior is well established in marketing literature. The extensive application of multidimensional scaling to develop so-called perceptual maps
has been the most recent approach to such investigation (Green and Carmone, 1970: Johnson, 1971;
Lehmann, 1971; Neidell, 1969; Stefflre, 1969). It
seems therefore opportune that perceptual variables,
that is, perceived innovation attributes, assume greater
attention in diffusion studies.
It should be understood that predicting the rate of
adoption from perceived innovation attributes is a far
easier statistical task than to predict adoption or nonadoption by individuals. Rate of adoption is an aggregative measure resulting from many separate adoption
decisions. In effect it is a group average. In attempting
to predict individual adoption decision outcomes, the
sources of error are commensurately increased. Thus,
one must not expect too much accuracy from any set
of predictors.
The purpose of this paper is to present the results
of two studies wherein the Rogers set of perceived
innovation attribute dimensions were used to predict
the innovativeness of housewives and to contrast these
results with the predictive efficacy of commonly used
respondent personal characteristic dimensions. In both
studies it was hypothesized that the perceptual vari-
PREDICTING INNOVATIVENESS BY
PERCEPTIONS OF INNOVATION ATTRIBUTES:
TWO STUDIES
Diffusion theory within sociology concerns the
propagation of a new concept or object, termed an
innovation, among members of a defined social system.
Most diffusion studies have been conducted by rural
sociologists and have concerned new farming methods
or implements. The most common study has centered
on the innovativeness (incidence of early innovation
adoption) of members of a social system as predicted
by personal characteristics of members, the communications or influence flows as well as social or cultural
norms within the social system (Rogers and Shoemaker 1971:72-73). Only a few studies, regardless of
research context, have foeussed on the attributes of
innovations, as perceived by potential adopters. The
relevant studies which exist have been successful in
explaining from 49 percent to 87 percent of the variance in the rate of adoption (Rogers and Shoemaker,
1971:140-141), but not all of these studies actually
involved the perceptions of potential adopters. For
example, Kivlin (1960) used a panel of "dairying
experts" rather than dairying farmers, for his study
of 43 dairying innovations considered for adoption
among 229 Pennsylvania fanners.
It is curious, however, that not one study has apparently gone beyond using the rate of adoption as
the dependent variable to predict instead who would
* Lyman E. Ostlund is Associate Professor in the Marketing
Department of the Wharion School. University of Pennsylvania. The amhor wishes to acknowledge the assistance of
Professors Raymond Bauer of Harvard University and Thomas
S. Robertson of the University of Pennsylvania in designing
the first of the two studies. Appreciation is expressed for
research support from the Faculiy Research Fund of the
Columbia Graduate School of Business.
23
24
TABLE 1
VARIABLE DEFINITIONS AND HYPOTHESIZED RELATIONSHIPS
Variable Name
Personal Characteristics
Venturesomeness
Cosmopolitanism
Social Integration
Privilegedness
Interest Polymorphism
General Self-Confidence in
Problem-Solving
General Self-Confidence in
Psychosocial Matters
Family Income
Respondent Education
Occupational Status of Husband
Respondent Age
Perceived Innovation Attributes
Relative Advantage
Corhpatibility
Complexity
Trialability
Observability
Perceived Risk
Definition
Expected Relationship
with Innovativeness
+
+
-1+
+
+
+
+
+
+
Reiss, Albert J., Jr., Otis Dudley Duncan, Paul K. Hatt, and Cecil C. North, Occupations and Social Status. New York: Tbe
Free Press, 1961, Appendix.
25
finally introduced nationally. The product, a self-layering dessert mix, was introduced into the test area
(Boston) about one year after the laboratory study
was conducted. The identity and telephone numbers of
each respondent from the laboratory study had been
preserved in anticipation that one or more of the six
products would eventually be marketed. After the selflayering dessert mix had been on sale for two months,
telephone interviews were attempted. Because of relocations, changes to unlisted numbers and other obstacles, only 360 of the original 605 respondents were
reached. For control purposes, a random sample of
200 housewives was also interviewed by telephone using
the same questionnaire. No significant difference was
found between the mean purchase incidence score of
the control sample and that for the 360 respondents
of the test sample. This suggested that involvement
with the original laboratory test had not itself stimulated purchase of the dessert product.
The telephone questionnaire involved an ordered
series of questions concerning unaided and aided recall
of purchase attitudes and future purchase intentions
toward the product. Both unaided and aided recall of
purchase were sought from respondents so as to assess
the degree of yea-saying from only the aided recall of
purchase measure. These same questions were asked
for two other recent new products so as to disguise
the study's objective from respondents.
Data from the laboratory study were combined with
the telephone interviews in order to relate the predictor
variables to unaided and aided recall of purchase using
multiple discriminant analysis. The 360 respondent
sample was divided randomly into two groups of 100
and 260 respondents. The first subsample was used
to derive the discriminant function and the second subsample was used to evaluate its effectiveness in correctly classifying respondents as to buyers and nonbuyers. Additional details of the design are given
elsewhere (Ostlund, 1972).
Results
Four discriminant functions were derived: (1) unaided recall of purchase, using all predictor variables;
(2) unaided recall of purchase using only perceptual
variables as predictors; (3) aided recall of purchase
using all predictor variables; and (4) aided recall of
purchase using only perceptual variables as predictors.
All four functions were statistically significant (Mahalanobis D-). Two other criteria, Cmnx, and Cpr,,, have
been suggested by Morrison (1969) for use in evaluating the predictive value of multiple discriminant functions. Cm.x, the maximum chance criterion, refers to
the proportion of the sample which would be correctly
classified by merely classifying the entire sample according to the largest subgroup (buyers or nonbuyers).
This criterion is relevant only when the group of pri-
#4
#2
Aided
Unaided
#3
Aided
Recall
Recall
Perceptual
Recall
Perceptual
Variables
All
Variables
Predictors
Only
Only
Criteria:
mox.
pro.
68%
57%
Classification
Results:
Proportion of
buyers correctly
65%
classified
Proportion of
total validation
sample correctly
80%
classified
n =260
68%
57%
60%
52%
60%
52%
63%
74%
72%
79%
260
69%
260
68%
260
26
classified. One additional buyer and one additional nonbuyer were misclassified. Similar results were obtained
using the aided recall measure of purchase. Using all
of the predictor variables produced 74 percent correct
classification of buyers, considerably in excess of both
discriminant function criteria (function #3). The proportion of nonbuyers correctly classified dropped, such
that the proportion of the total validation sample correctly classified became 69 percent. Using only the
perceptual variables resulted in merely a two percent
drop in the proportion of buyers correctly classified
from aided recall data (function #4). Once again,
both Cpr... and Cmax. are easily exceeded.
The greater effectiveness of the discriminant functions on aided recall data, relative to that on unaided
recall data is thought to have resulted from two sources.
The first source is that of yea-saying which can result
from aided recall measures. A second explanation is
in order, however. According to the telephone interviewers, respondents tended to give the family brand
name (''Jello") instead of the exact brand name
("1.2,3") when responding to the unaided recall
question, and later corrected themselves after the
aided recall question was asked. This point of confusion had not been anticipated. In the author's opinion,
the "true" effectiveness of the unaided discriminant
function should be regarded as relatively close to that
of the aided recall function. In any case, the classification results from aided recall differed by only about
ten percent from those using unaided recall, which
does not suggest a major problem with yea-saying.
What about the relative importance of the predictor
variables? It should be clear from Table 2 that the
personal characteristic variables were unimportant predictors. That is, in removing the personal characteristic
variables from the multiple discriminant functions, the
reductions in classification acuracy are minor. In fact,
when examining the predictor variables individually
only venturesomeness related significantly (p < .05)
with innovativene.ss for both unaided and aided recall
data. Problem solving general self-confidence was
statistically significant (p < .05) in the aided recall
function and family income in the unaided recall function. Still, the absence of these three variables was
scarcely missed in functions 2 and 4.
The ranked order of importance among perceived
innovation attributes is given in Table 3 for both the
unaided and aided functions. As would be expected,
there is considerable consistency in the orderings from
the two measures. Relative advantage and compatibility
reverse position as first and second in the rankings. Similarly, complexity and observability exchange orders.
Clearly, however, the perceived innovation attributes
are far from equally important. As one might expect,
trialability for such an inexpensive product is of no
importance as a predictor. When moving from unaided
to aided recall, the greatly increased size of the stan-
Variable
Name
Standardized
Discriminant
Function
Coefficient
Relative
Advantage
Compatibility
Complexity
Perceived risk
Observability
Trialability
23.46
22.32
-14.72
-12.74
120
.14
Aided Recall
Standardized
Ranking Discrimiof
nant
Impor- Function
tance Coefficient
1
2
3
4
5
6
20.10
50.63
6.00
-12.60
14.21
1.60
Ranking
of
Importance
2
1
5
4
3
6
27
Compatibility
1. regarding self-concept
2. regarding family members
3. regarding existing habits
This expansion was considered appropriate for consumer products but not necessarily beyond that class.
The Data
The product selected for study was a brand of oven
roasting bag made of a special nylon-type material
which can sustain high temperatures. The meat to be
roasted is placed inside the bag on a roasting pan
within the oven. Benefits claimed by the manufacturer
from using such bags include self-basting of the meat,
reduced roasting time, less shrinkage of the meat, and
no oven mess from spattering juices. Several similar
brands of various forms and compositions have been
introduced into the market place since this test was
conducted. The product in question was the first of its
type sold in the test area and was at the titne of the
studv a new concept in food preparation for housewives.
The data were gathered by means of a representative housewife panel in the test area. Six months before
the product was introduced into the market a sample
of 300 respondents was selected from the on-going
panel to view a series of television commercials on
new products. The sequence of the commercials was
rotated so as to eliminate any order bias. Some of the
commercials concerned existing new products, others
concerned products still at some stage of pre-market
development. One such commercial concerned the oven
bags which gave a demonstration of their appropriate
use. For each of the featured products, the housewife
completed a perceived innovation attribute question-
Innovators
As Classified By Predictors
NonInnovators Innovators
Total
41
U
52
Actual
Non-Innovators
_24
J24
65
135
Proportion of Innovators Correctly Classified:
Proportion of Total Sample Correctly Classified:
C^ = 74%
US^
200
79%
83%
28
variables were of little importance. Only venturesomeness and socioeconomic status bore any relationship
(positive) to innovativeness. The levels of significance
were .05 and .10 respectively. For this reason, it is
not surprising that, as before, little change resulted in
the classification matrix by dropping the personal characteristic variables (Table 5). One additional innovator
was misclassified and, less importantly, seven additional
non-innovators were misclassified. In any case, the
function's performance regarding innovators is improved relative to functions 2 and 4 in the first study
(Table 2).
The relative importance of perceived innovation attributes in the discriminant function is given in Table
6. Separate coefficients are given for the components
of ^relative advantage and compatibility. Effort savings,
time savings and perceived risk take the first three
positions of importance followed by complexity and
monetary value. The importance of compatibility when
split into three components is not so prominent as was
true in the first study (Table 3). Nonetheless, it is
fair to say that the same general ordering is found as
was true in the first study.
CONCLUSIONS
The evidence from two studies of new consumer
packaged goods suggests that the perceptions of innovations by potential adopters can be very effective
predictors of innovativeness, more so than personal
characteristic variables. More work must be devoted
to the refinement and elaboration of perceived innovation attribute dimensions, however. Secondly, replication among differing types of potential adopters and
products is needed.
Diffusion researchers, particularly rural sociologists,
are urged to make the added effort to incorporate such
variables, where appropriate, in their studies, and devote relatively less effort to studying personal characTABLE 5
MULTIPLE DISCRIMINANT CLASSIFICATION TABLE
USING ONLY THE PERCEIVED INNOVATION
ATTRIBUTE VARIABLES
Study #2
Innovators
Actual
Non-Innovators
As Classified By Predictors
NonInnovators Innovators
Total
52
40
12
31
117
TT
T29
= 74%
148
200
77%
79%
TABLE 6
RELATIVE IMPORTANCE OF PERCEIVED
INNOVATION ATTRIBUTES IN THE
DISCRIMINANT FUNCTION
Study #2
Variable Name
Relative Advantage
a) time savings
b) effort savings
c) Monetary value
Compatibility
a) self-concept
b) family members
c) existing habits
Complexity
Perceived risk
Observability
Trialability
Standardized
Discriminant
Function
Coefficient
Ranking
of
Importance
21.7
27.8
6.4
2
1
5
5.6
2.2
4.7
-7.9
-9.2
3.4
1.7
6
9
7
4
3
8
10
29