Professional Documents
Culture Documents
REX S. T O H
Kirton (1976) suggested that the way an individual solves problems re-
flects the cognitive style of innovators, i.e., those who "do things better,"
and adaptors, i.e., those who "do things d~fferently."Innovators are said to
have a proliferation of ideas, think tangentially, and approach tasks from dd-
ferent angles. By contrast, adaptors have a cognitive style that emphasizes
efficiency, order and precision, conformity, production of a small set of solu-
tlons co problems, and adherence to rules (IGrton, 1976, 1978; Foxall, 1994;
Foxall & Hackett, 1994).
The 32-item Kirton Adaption-Innovation Inventory has been more wide-
ly used to measure the innovative personality and cognitive style of an indi-
vidual, called "innovativeness" (e.g., firton, 1976, 1978; Foxall, 1988; Gold-
smith & Hofacker, 1991). Although other multi-item scales to measure inno-
'We thank Kenneth Bollen for his help and guidance in the planning and writing of this paper
y d also Barry Bayus and the UNC Research Council for assistance in data collecuon.
Send correspondence to Dr. Rex S. Toh, Department of Management, Albers School of Busi-
ness and Economics, Seattle University, Seattle, WA 98122 or e-mail (rextoh@seattleu.edu).
884 S. IM, ET AL.
vative personahty and cognitive style have been proposed (e.g., Leavitt &
Walton, 1975; Jackson, 1976; Hurt, Joseph, & Cook, 1977; Goldsmith &
Hofacker, 1991; Manning, Bearden, & Madden, 1995), the inventory remains
the most popular measure of innovativeness and adaptiveness. This inven-
tory has been extensively tested for reliabhty, content vahdity, and general-
izability in numerous contexts, including organizational decision-making sit-
uations (Keller & Holland, 1978; Goldsmith, 1986; Foxall & Hackett, 1992;
Goldsmith, Freiden, & Eastman, 1995).
The Kirton Adaption-Innovation Lnvento~yhas also been shown to have
high convergent and discriminant vahdities ( K Irton, 1976, 1978; Keller &
Holland, 1978; Goldsmith, 1984, 1986; Beene & Zelhart, 1988; Taylor,
1989a, 1989b; Foxall & Hackett, 1992; Bagozzi & F o x d , 1995; Goldsmith,
et al., 1995). l r t o n (1976) reported a test-retest reliabhty of .82, and Foxall
and Hackett (1992) found Cronbach alphas ranging from .86 to .89. Also,
the stability of the loadings and correlations across samples from different
countries confirm the generahzabllity of the Krton Adaption-Innovation In-
ventory (Beene & Zelhart, 1988; Foxall & Hackett, 1992, 1994; Bagozzi &
Foxall, 1995, 1996; Goldsmith & Goldsmith, 1996).
Goldsmith (1984, 1986) reported that scores on the Kirton Adaption-
Innovation Inventory are highly correlated with those of other originality or
creativity scales, for example, the Open Processing Scale (r = .28, p < .01; Lea-
vitt & Walton, 1975) and the Jackson Personality Inventory ( r = .28, p < .01;
Jackson, 1976). Carne and Kirton (1982) further reported that the cognitive
style measured by the Kirton inventory is highly correlated with both the in-
tuition ( r = .44, p < ,051 and perception (r = .53, p < .01) personality types as
measured by the Myers-Briggs Type Indicator (Myers, 1962).
The l r t o n Adaption-Innovation Lnventory, as a measure of cognitive
style, is said often to be confounded with different cognitive levels, capaci-
ties of creativity, and attitudinal variables (Mudigan & Martin, 1980; Masten
& Caldwell-Colbert, 1987; F o x d & Hackett, 1992; Foxall, 1994). Further-
more, Foxall and Hackett (1992) and Taylor (1989a, 1989b) claimed that the
Sufficiency of Originality subscale is not homogeneous; instead, it comprises
two subdimensions, one related to "Idea Generation" and the other related
to "Preference for Change" (Taylor, 1989b, p. 922). When exploratory fac-
tor analysis was used to explore the underlying structure of the inventory,
significant amounts of test variance (above 60%) remained unexplained
(Hammond, 1986; Taylor, 1989a; Foxall & Hackett, 1992).
The main objective of this research was to determine whether the Suffi-
ciency of Originality subscale measures a homogeneous construct. Bollen
and Grandjean (1981) suggested that construct validity relies on the proper
identification of the dimension&ty of the construct, i.e., unidimensional or
multidimensional, otherwise we would get ambiguous results. In this paper,
DIMENSIONS O F ORIGINALITY SUBSCALE 885
'In our analysis, listwise deletion provides consistent paralncrcr estimators without influencing
the significance test for random missing values (Bollen, 1989) The sample covariance matrix
remains positive, and the reduction in sample size is nor cons~deredsevere for our analysis.
This process is necessary, since some respondents systemat~callyomitted a series of questions
related to the Kirton Adaption-Innovation Inventoty.
sponses on the key variables, 356 complete and usable responses remained,
representing a response rate of 78%. This sample had an average age of 56
(SD= 14), average income of $39,700 (SD= $19,200), and averaged 15 yr. of
education (SD= 2).
( Generation 1
FIG. 2. Respecified model with two double loaded items (x7 and x9) removed; Model 3:
= 1, Model 4:ol2= Free
The &ton Adaption-Lnnovation Inventory has 32 items and asks par-
ticipants to evaluate (on a 5-point scale) whether it is easy to describe their
self-images on each item. Of the 32 items, 11 items load Onto the Sufficiency
of Originality factor: has original ideas (xl), is stimulating (x2),has fresh per-
spectives on old problems (x,), likes to vary set routines at a moment's no-
tice (xJ, would sooner create than improve ( x ~ )copes, with several ideas at
the same time ( ~ , 5 ) can
, stand out in disagreement against a group (x7),pro-
Merates ideas (xg),often risks doing things differently (xg),w d always think
of something when stuck (xI0), and needs the stimulation of frequent change
(xI1).For our final sample of 356, the Cronbach for the 11 items is 3 7 ,
indicating very satisfactory internal consistency. As recommended by Mc-
Donald and H o (2002), the correlations of the variables as well as the means
and standard deviations are reported in Table 1.
TMLE 1
A N D DESCRIPTIVE
CORRELATIONS~ STATISI~CS
Item X2 X3 X4 X5 X6 X7 X8 x9 ~ 1 0X I ] M SD
x l : Has original ~deas. 1 3.67 1.06
xz : Is stimulating. .49 1 3.51 .90
x3: Has €resh perspectives on
old problems. .52 .60 1 3.61 .94
x4: Likes ro vary set routines at
a moment's notice. .29 .23 .27 1 2.86 1.13
x5; Would sooner create than
unprove. .36 .31 .30 .34 1 3.04 .94
xg: Copes with several new ideas
at the same time. .48 .46 .51 .31 .33 1 3.44 1.07
x,: Can stand out in disagree-
ment against a group. .47 .25 .32 .26 .26 .44 1 3.41 1.14
x8 : Proliferates ideas. .57 .55 .51 .23 .35 .50 .42 1 3.35 .86
x9: Ofren risks doing things
ddferendy. .48 .37 .42 .37 .45 .51 .49 .56 1 3.34 1.08
xlo: Will always thmk of some-
thing when stuck. .43 .39 .49 .26 .28 .51 .39 .46 .47 1 3.70 .95
xl . Needs the stimulation of
l;equenr change. .29 .29 .30 .35 .28 .32 .25 .33 .37 .31 1 3.11 1.04
tSlgn6cant a t .01.
previous findings (Taylor, 1989a, Foxall & Hackett, 19941, the results from
both exploratory factor analyses provided the same factor solutions (patterns)
and supported the proposition that within this subscale, there are indeed
two distinct subdimensions, Idea Generation and Preference for Change. We
also found that the two dimensions explain 52% and 55% of the total vari-
ance, respectively. Thus, our initial exploratory factor analyses and close ex-
amination of semantic contents of the measurement items provide support
for specdying the two-dimensional model a prior2 before testing it using con-
firmatory factor analysis. The variables X I , X Z , x3, xg, x,, X B , and xlo were
loaded into the Idea Generation subdimension, while the variables x,, x5,
x9, and xll were loaded into the Preference for Change subdimension.
Then we performed a series of confirmatory factor analyses as proposed
by Bollen and Grandjean (1981) to examine the homogeneity of the Suffi-
ciency of Originahty subscale. The confirmatory factor model is represented
by the following matrices (Bollen, 1989): x = Ax x k + 6, with E (6,)= 0, for all
j, and E (kiiSi')= COV (ti,
6;) = 0, for all i and j. Specifically, Bollen and
Grandjean's procedure (1981) examines the improvement in fit between two
nested measurement models, (1) a reduced model which is unidirnensional
when the correlation is constrained and equal to I , i.e., perfectly correlated,
and (2) a full model which is two-dimensional when the correlation between
the two subdimensions is unconstrained, i.e., freely estimated. A significant
xZ difference between the two models would indicate that the two-dirnen-
888 S. IM. ET AL.
'In addition, we examined the discriminant v&dity of the two subdimensions by correlating
them wich demographic variables such as income and the residence history, i.e., the length of
stay at the current residence. Income is correlated with scores on Idea Generation (r = .15, p <
.05) but not with Preference for Change ( r = . 0 4 , p > .5). Residence history is negatively corre-
lated with Idea Generation ( r = -.19, p < .01) but nor with Preference for Change (r = -.lo, p <
.07). These differential results from correlation analyses indicate the discriminant validity of the
two subdirnensions.
DZMENSIONS O F ORIGINALITY SUBSCALE 889
TABLE 2
S MUM LIKELIHOOD
ESTIMATION
RESULTSFORORIGINAL
CONFIRMATORY
FACTOR
MODEL
Model 1: Correlation (El, k2)= 1 Model 2: Correlation (5,. e2)= Free
Estimate SE SMC Esumate SE SM C
Coefficient
~ I I .76* .05 .52 .77* .05 .53
&?I .59* .05 .42 .60f .04 .44
%I .65* .05 .48 .67* .05 .51
?'6 1 .76" .05 .50 .76* .05 .50
'7 I .65* .06 .33 .64* .06 .32
ha1 .64* .04 .56 .64* .04 .56
~ I O I .61f .05 .42 .61* .05 .42
Ad2 .48* .06 .18 .55* .06 .24
5 2 .47* .05 .25 .54* .05 .32
%2 .77* .05 .51 .86* .06 .63
A1 12 .47* .06 .2 1 .52* .06 .25
Correlation
(51, 5 2 ) Fixed at 1.00. .84*
Fit Statistics
Chi-square 152.19, d/=44t 120.80, df= 43 t
NFI ,986 ,989
RFI ,980 ,983
IF1 .990 ,993
TLI 985 ,989
RMSEA 083 .07 1
* p < .05. t p < ,001.
the two dmensions. Bollen and Hoyle (1990) suggested that, if a concept is
one-dimensional, the correlation coefficient benveen the two latent variables
should be close to 1.0. Also, in terms of unexplained variance, substantial
amounts suggest that the two dimensions are not identical, favoring the two-
dimensional model (Model 2) over the one-dunensional model (Model 1).
Furthermore, as recommended by Tanaka (1993), multiple fit indices
are used to capture other dimensions of goodness of fit. From Table 2 , the
basehe comparison indices-Normed Fit Index (NFI), Lncremental Fit In-
dex (IFI), Relative Fit Lndex (RFI), and Tucker-Lewis Index (TL1)-are a l l
over .95, indcating that both models fit the data rather well5 Additionally,
based on Browne and Cudeck's cut-off criteria,' the Root Mean Square Er-
ror of Approximation (RMSEA) for Model 2 (.071) indicates a reasonably
good fit, while that for Model 1 (.083) indicates only an acceptable fit, thus
favoring the two-dimensional model over the one-dimensional model.
'AlthouEh .the:e m u l t s do not indicate a signhcant statistical difference between the two mod-
$,the t IS elther equal to or slightly better for Model 2 than Model 1.
Browne and Cudeck (1993) suggested the following criteria for the RMSEA value: .05 or less
for close fit, $08or less for reasonable fit, and .10 or less for acceptable fit.
890 S. IM.ET AL.
TABLE 3
~ I M U LIKELIHOOD
M ESTLMAT~ON FOR&SPECIFIED MODELWITH
ITEMS (X-I AND ~ q REMOVED
DOUBLE-LOADED )
The chi-square Merence test indicates that there are significant im-
provements in fit between the original and respecified models R AX^ between
Model 1 and Model 3 = 152, 44 - 79, 27 =73, Adf= 17, p < .001, and AX* be-
tween Model 2 and Model 4 = 121, 43 -51, 26=70, Adf=17, p=.OOl). With
model respecification, all the fit indces (NFI, RFI, IFI, and TLI) improved
slightly and are close to .99, indcating improved fit for the respecified mod-
els without the double-loaded items.
Next, the chi-square difference test was again applied to the two nested
DIMENSIONS OF ORIGINALITY SUBSCALE 89 1
respecified models without the double-loaded items. Results show that the
two-dimensional model is favored over the one-dimensional model (AX* be-
tween Model 3 and Model 4 =79, 27 -51, 26=28, Adf= 1, p < .0001). Also,
based on Browne and Cudeck's cut-off criteria, the smaller RMSEA for
Model 4 (.052) indicates a close fit, while the bigger RMSEA for Model 3
(.073) shows a reasonable fit, thus again favoring the two-dimensional model
(Model 4) over the one-dimensional model (Model 3 ) . Finally, for Model 4,
the correlation between the two latent constructs is .73 (down from .84 for
Model 2), meaning that only 53.3% of the variance is shared by the two &-
mensions, leaving 46.2% unshared, indcating that the two dimensions are
not identical but are, indeed, separate constructs.'
Addztional Analysis
In the previous section, we examined the homogeneity of the Suffi-
ciency of Originality subscale by comparing two alternative theoretical mod-
els, (1) a confirmatory factor model based on extant literature and initial ex-
ploratory factor analysis and (2) a confirmatory factor model respecified with-
out the double-loaded items identified in the exploratory factor analysis. Al-
though our results indicate that the respecified model without the double-
loaded items significantly improved the fit, all our models (Models 1 through
4) showed a lack of overall fit based on chi-square estimation, i.e., all p val-
ues for chi-square estimations < .01.' For example, the most favored Model
4 (respecifiedwithout double-loaded items and Lo-dlrnensional unconstrain-
ed) still lacks proper fit as reflected by the chi-square value of 51.25 with 26
df and associated p of <.01. Although the lack of fit as measured by the
chi-square test is common in social science (e.g., see Bollen & Grandjean,
1981), we decided to investigate further the source of the lack of fit.
Since our sample size (N=356) satisfactorily exceeds Boomsma's (1982)
cut-off criterion (n =200) to avoid erroneous Maximum LLkelihood estima-
tion, we suspected that the lack of overall fit may be attributed to the mis-
specification of the model. We then re-estimated the model after allowing
measurement errors belonging to the same subdimension to be correlated.'
When we respecified Model 4 with correlated measurement errors, the
chi-square fell to 27.68 with 19 df and an associated p value of .09, indcat-
ing a substantially improved and good fit. AU the other fit indices (NFI,
RFI, IFI, TLI) were above .99, and the RMSEA was only .03, indicating an
excellent fit.
'In addition, all the other fit indices (NFI, RFI, IFI, TLI) and SMC values for the nvo-dimen-
sional Model 4 are either che same or higher than chose for the one-dimensional model 3.
T h e chi-square test is used as an overall goodness o i Fit test in structural equation modeling.
Chi-s uare values which are not significant indicate a good fit of the model with the data.
W e l a s e d our decision to correlate measurement errors on extant literature (Taylor, 1989b;
Foxall & Hackett, 1992) and our initial exploratory factor analysis.
892 S. IM, ET AL.
DISCUSSION
The main objective of this paper was to investigate whether the Suffi-
ciency of Originality subscale of the Kirton Adaption-Innovation Inventory
should be subdivided into two separate subdunensions of Idea Generation
and Preference for Change. Using Bollen and Grandjean's (1981) approach
based on confirmatory factor analysis, we confirm that subdividing the Suffi-
ciency of Origin&ty subscale significantly improves the fit, inlcating Suffi-
ciency of Originhty r e d y consists of two distinct subdimensions. We fur-
ther found that an alternative model respecified without the double-loaded
items (found from exploratory factor analysis) improves the f ~ significantly
t
and that the amount of unexplained variance between the two subdimen-
sions increased from 29.46% to 46.21%. The final unexplained variance of
46.21% suggests that the Sufficiency of Originality subscale indeed has two
subdimensions. Addtional analysis further suggests that allowing correlations
for measurement errors that belong to the same subdimension significantly
improves the fit of the two-dimensional model with the data.
As with almost any study, our results may be idiosyncratic to the sam-
ple and methods used for this study. We suggest further studies using confir-
matory factor analysis but with different samples in different contexts. We
also suggest that further research be directed at other factors and mehods to
test the overall fit in chi-square estimations. One possible avenue is to inves-
tigate the robustness of the maximum likelihood estimation results to the vio-
lations of the multin~rrnalit~ assumptions because these tend to inflate the
overall chi-square test statistic.
We conclude that the construct vahdity of the l r t o n Adaption-Innova-
tion Lnventory related to the Sufficiency of Originality subscale can be
improved by subdividing the subscale into Idea Generation and Preference
for Change, supporting Taylor's original claim (1989a). The separate recog-
nition of these two subdimensions is more compelhg when the double-load-
ed items are dropped, and the model fits the data significantly better when
correlations of measurement errors are allowed.
REFERENCES
BAGOZZI,R. I?, & FOXALL,G. R. (1995) Construct validity and eneralizability of the Kirton
Adaption-Innovation Inventory. European lournal of ~ c r r o n a & i9.~ .227.1-227.23.
~ , P.. &FOXALL,G. R. (1996) Construct validation of a measure of adapdve-innova-
B ~ c o z z R.
tive cognitive styles in consumption. 111fernafionalJorrrnal of Research in Marketing, 13,
201-213.
BEENE,1. M., &ZELHART,I? F. (1988) Factor analysis of the Kirton Adapdon-Innovation Inven-
tory. Perceptual and Motor Skills, 66, 667-671.
BOLLEN,K. A. (1989) Strzrctural equafio~lswith lafenf variables. New York: Wiley.
BOLLEN,K. A,, & G W D J M N , B. D. (1981) T h e dimensions of democracy. American Sociologv
Review, 46, 65 1-659.
BOLLEN,K. A,, & H O W , R. H. (1990) Perceived cohesion: a conceptual and empirical exami-
nation. Social Forces, 69, 479-504.
DIMENSIONS O F ORIGINALITY SUBSCALE 893
BOOMSMA, A. (1982) The robustness of LISREL against small size in factor analysis models. In
K. G. Joresko & H Wold (Eds ), Systems zrnder directed observation: Part I . Amsterdam:
North ~ollan!. Pp. 149-173. '
BROWNE, M. W, &CUDECK, R. (1993) Alternative ways of assessing model €it. In K. A. Bollen
& J. S. Long (Eds.), Testing strzrctural eqzration models. Newbury Park, CA: Sage. Pp.
136-162.
CARNE. G. C., & KIRTON, M. J. (1982) Styles of creativity: test-score correlations benveen
Kirton Adaption-Innovacion Lnventory and Myers-Briggs Type Indicator. Prjchological Re-
ports, 50, 31-36.
FOXALL,G. R. (1988) Consumer innovativeness: novelty-seeking, creativity, and cognitive style.
Research in Consumer Behavior, 3, 79-113.
FOXALL,G. R. (1994) Consumer iniuators: adaptors and innovators. British Jotirnal ofManage-
ment, 5(Specid Issue), s3-sl2.
FOXALL.G. R., & HACKETT,P. M. W. (1992) The factor structure and conscruct validity of the
Kirton Adaption-Innovation Inventory. Persorzality and Individual Dzferences, 13, 967-
975.
FOXALL,G. R., & HACKET~, l? M. W. (1994) Styles of managerial creativity: a comparison of
adaption-innovation in the United Kingdom, Australia, and the United Scares. Bn'rish Jour-
nal of Management, 5 , 85-100.
GOLDSMITH, R. E. (1984) Personality characteristics associated with adaption-innovation. lour-
nal of Psychology, 117, 159-165.
GOLDSMITH, R. E. (1986) Convergent validity of four innovativeness scales. Educational and
Psychological Measzrrement, 46, 81-87.
GOLDSMITH, R. E., FREIDEN, J. B., & ~ T M A N J.
, K. (1995) The generality/specifcity issue in
consumer innovativeness research. Technovation, 15, 601-611.
GoLDSM~TH, R. E., &GOLDSMITH, E. B. (1996) An empirical study of overlap of innovativeness.
Psychological Reports, 79, 1113-1114.
GOLDSMITH, R. E., & HOFACKER, C. F. (1991) Measuring consumer innovauveness. ]ounzal of
Academy of Marketing Science, 19, 209-221.
HAMMOND, S. M. (1986) Some pitfalls in the use of factor scores: the case of the E n o n Adap-
tion-Innovation Inventory. Personality and Individzral Dzfferences, 7, 401-407.
HOLLAND. P. W. (1993) Which comes f rst, cause or effect? In G. Karen & C. Lewis (Eds.), A
handbook of data analysis Lz the behavioral sciences: methodological issues. Hillsdale, N J :
Erlbaum. Pp. 273-282.
HURT,H. T., JOSEPHK., &COOK,C. D. (1977) Scale For the measurement of innovativeness.
Hzrmmn Commtrnication Research, 4(Fall), 58-65.
JACKSON. D. N. (1976) Jackson Personality Inve~ztor~ manual. New York: Research Psycholo-
gists Press.
KELLER,R T., &HOLLAND, W. E. (1978) A cross-validation study of the b n o n Adapuon-Inno-
vation Inventory and three research and developmenc orgnnizations. Applied Psychological
Measzrrement, 2, 563-570.
KIRTON,M. J. (1976) Adaptors and innovators: a descripuon and measure. ]orrrnal of Applied
Psychology, 61, 622-629.
KIRTON,M. 1. (1978) Have adaptors and innovators equal levels of creativity? Prychulogrcol Re-
ports, 42, 695-698.
LEAVITT,C., & WALTON. J. R. (1975) Development of a scale for innovativeness. Advances in
Corzszrmer Research, 2, 545-554.
MANNING, K. C., BEARDEN, W. O., &MADDEN, T. J. (1995) Consumer innovativeness and adap-
tion process. Jozintal of Consumer Psychology, 4, 329-345.
MASTEN,W. G., & CALDWELL-COLBERT, T. (1987) Relationship of originality to Kinon's scale
for innovators and adaptors. Psychological Reports, 61, 411-416.
MCDONALD, R I?, &Ho, M. R. (2002) Principles and practice in reporting structural equation
analyses. Psychological Methods, 7, 64-82.
MULLIGAN, G., &MARTIN, W. (1980) Adaptors, innovators and the l r t o n Adapuon-Innovation
Inventory. Psychological Reports, 46, 883-892.
MYERS,I. B. (1962) The Myers-EMS Type Indicator. Palo Alto, CA: Consulting Psychologists
Press.
S. IM, ET AL.