duff-y
-
Upload
eliad-becker -
Category
Documents
-
view
216 -
download
0
Transcript of duff-y
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 1/17
Psychometric properties of Honey & Mumford’s Learning
Styles Questionnaire (LSQ)
Angus Duff*, Tim Duffy
Research into Learning Unit, University of Paisley, Ayr Campus, Ayr KA8 0SR, UK
Received 12 March 2001; received in revised form 27 June 2001; accepted 24 July 2001
Abstract
Honey and Mumford’s Learning Style Questionnaire (LSQ) has been proposed as an alternative for
Kolb’s Learning Style Inventory (LSI) and a later refined version (LSI-1985). The LSQ has been widely
applied in the fields of management training and education. Limited evidence exists concerning the psy-
chometric properties of the LSQ. Participants were 224 undergraduates enrolled in business courses and
164 undergraduates in health studies. Exploratory and confirmatory factor analysis failed to support the
existence of the two bipolar dimensions proposed by Kolb, and four learning styles hypothesised by Honey
and Mumford. An item analysis and pruning exercise failed to raise the internal consistency reliability to a
satisfactory level, or provide adequate model fit to the data. The results of a structural equation model
finds no consistent relationship between scores on the four learning style scales, two bipolar dimensions
and academic performance between the two samples. The tests of factorial invariance provide no support
for the stability or generalizability of the model. It is concluded: the LSQ is not a suitable alternative to the
LSI and LSI-1985; and its use in applied research considering higher education students is premature.
# 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Learning Style Questionnaire; Reliability; Validity; Academic performance
1. Introduction
A learning style is described as being:
. . .a description of the attitudes and behaviour which determine an individual’s preferred way
of learning. (Honey & Mumford, 1992, p. 1)
0191-8869/02/$ - see front matter # 2002 Elsevier Science Ltd. All rights reserved.P I I : S 0 1 9 1 - 8 8 6 9 ( 0 1 ) 0 0 1 4 1 - 6
Personality and Individual Differences 33 (2002) 147–163www.elsevier.com/locate/paid
* Corresponding author. Tel.:+44-01292-886296; fax:+44-01292-886250.
E-mail address: [email protected] (A. Duff).
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 2/17
Learning style is the composite of characteristic cognitive, affective, and psychological factors
that serves as an indicator of how an individual interacts with and responds to the learning
environment. The study of learning style involves the investigation of individual differences:
people perceive and gain knowledge differently, they form ideas and think differently, and theyact differently. Research on style as an individual trait has been of interest to psychologists for
many years (Jung, 1921; Myers & Briggs, 1962). Kolb’s (1976) Experiential Learning Model
(ELM) is a well-established model that has attracted much interest and application. The ELM
consists of a hypothesised four-stage learning cycle and is based on the work of Lewin (1936).
The hypothesised learning cycle can be entered at any stage but must be followed in sequence.
According to the theory, different individuals may cope better with, or prefer, some parts of the
learning cycle to others. In the learning cycle, or process, learners acquire information by con-
crete experience in the new experience. Second, a stage of reflective observation on the experience
occurs whereby the learner organises the experiential data from a number of perspectives. Third,
a stage of abstract conceptualisation occurs, whereby the learner develops generalisations fromwhich to assist them integrate their observations into sound theories or principles. Finally,
through active conceptualisation, learners use these generalisations as guides to action in new and
more complex situations. This process explains individual differences in learning style in terms of
relative abilities (i.e. level) for performing well (or less well) at various stages of the learning cycle.
That is, the ideal learner will possess maximum abilities for all four stages.
Since Kolb developed his concept of a four-stage process, the process has been developed fur-
ther as two orthogonal dimensions of learning derived from the Learning Style Inventory (LSI;
Kolb, 1976). These two dimensions are labelled prehension, grasping information from experi-
ence (Concrete Experience-Abstract Conceptualisation); and transformation, that is the proces-
sing of information grasped (Reflective Observation-Active Experimentation). This concept
explains differences in terms of two bipolar styles (i.e. the manner) by which each stage in thelearning process is approached and operationalised. These bipolar dimensions are sometimes
described as learning types.
Kolb’s 12-item LSI has been widely applied to measure learning style. Twelve short statements
concerning learning situations are presented and respondents are required to rank-order four
sentence endings that correspond to the four learning styles. Later research finds little factor
analytic support for the four styles and two independent dimensions (Freedman & Stumpf, 1978,
1980; Geller, 1979; Newstead, 1992; Stout & Ruble, 1991a; 1991b). Notably, the ipsative scoring
method guarantees that some scales must be negatively correlated. Similar psychometric pro-
blems exist with a refined version of the instrument, the LSI-1985. These problems are sum-
marised in: Geiger, Boyle, and Pinto (1992, 1993); Loo (1999); Ruble and Stout (1993); Willcoxonand Prosser (1996); Yahya (1998). Critics of the application of Kolb’s LSI maintain that its use
for education research purposes was premature in the sense that the instrument’s psychometric
properties had not been sufficiently assessed.
Honey and Mumford’s (1992) Learning Style Questionnaire (LSQ) has been proposed as an
alternative to Kolb’s LSI. The LSQ was developed to report management trainees’ learning style
preferences and has subsequently been applied to a wide range of subjects, including students in
higher education. Prudent scholarship requires that the LSQ be subjected to critical analysis
before it is used for applied research and correlation studies (Schwab, 1980). The LSQ is a self-
administered inventory consisting of 80 individually rated (1 or 0) items, differing in this respect
148 A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 3/17
from Kolb’s ipsative LSI (1976 version). The normative nature of the scale makes the instrument
a potentially attractive alternative to both the LSI and LSI-1985 to educational researchers given
the much-documented problems with ipsative measures (e.g. Cornwell & Dunlap, 1994; Dunlap
& Cornwell, 1994). The vast majority of LSQ items are behavioural, i.e. they describe an actionthat someone might or might not take. Occasionally, an item probes a preference rather than a
manifest behaviour. The LSQ is designed to probe the relative strengths of four different learning
styles: Activist, Reflector, Theorist and Pragmatist. These four styles correspond approximately
to those suggested by Kolb’s (1976) ELM: active experimentation (Activist), reflective observa-
tion (Reflector), abstract conceptualisation (Theorist), and concrete experience (Pragmatist).
Furthermore, the ELM reflects two independent dimensions: Pragmatist–Theorist (prehension)
and Activist–Reflector (transformation).
Table 1 reports the results of previous studies considering the internal consistency reliability of
scores produced by the LSQ. Most studies report alpha coefficients indicating scores produced by
the instrument of moderate internal consistency reliability.
1
Evidence regarding the factor analytic properties of the LSQ is mixed. Allinson and Hayes
(1988), using a sample of managers, factor analysed the scores of the four learning style scales to
yield the two hypothesised orthogonal factors Activist–Reflector and Pragmatist–Theorist. A
subsequent study of UK undergraduate students (Allinson & Hayes, 1990) confirmed the two
independent dimensions. With a sample of UK and Eire managers De Ciantis and Kirton (1996),
attempted to modify the LSQ by means of an item analysis and pruning exercise. As a result of
this exercise, DeCiantis and Kirton produced two bi-polar measures, Activist–Reflector and
Theorist–Pragmatist, containing 45 and 15 items, with internal consistency reliabilities of 0.90
and 0.69, respectively. The modified bi-polar measures of DeCiantis and Kirton were reported as
correlating negligibly (0.08) in accord with theory.
No previous work has utilised confirmatory factor analysis (CFA): Allinson and Hayes (1988,1990); De Ciantis and Kirton (1996); Sims, Veres, and Shake (1989) and Tepper, Tetrault, Braun,
and Romero (1993) all applied exploratory factor analytic techniques to their respective samples.
Table 1
Summary of internal consistency reliability estimates of previous research
Study Participants n= Coefficient
Activist Reflector Theorist Pragmatist
Allinson and Hayes (1988) UK managers 127 0.58 0.74African & Indian managers 40 0.71 0.63
Sims et al. (1989) US business students 270 0.68 0.68 0.78 0.75
Tepper et al. (1993) US undergraduate students 227 0.75 0.76 0.67 0.52
Jackson and Lawty-Jones (1996) UK psychology students 166 ‘‘all between 0.69 and 0.77 (p. 295)’’
De Ciantis and Kirton (1996) UK and Eire managers 185 0.76 0.76 0.67 0.64
1 Nunnally and Bernstein (1994) indicate an alpha coefficient cutoff of 0.7 is necessary for instruments to be used in
applied settings, although 0.8 is a more preferable and stringent cutoff criterion.
A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163 149
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 4/17
Mulaik (1987) identifies that exploratory techniques can never answer definitively questions
regarding the latent structure of a set of variables, arguing that exploratory factor analysis (EFA)
can only suggest structures and these require confirmation by CFA. Previous research consider-
ing the psychometric properties of scores produced by the LSQ is therefore limited.Rather more evidence exists of the criterion validity of scores produced by the instrument.
Furnham and Medhurst (1995) report high and consistent correlations’ between students’ scores
on the Pragmatist dimension and positive performance in university seminars. Other work has
considered the relationship between learning style as measured by the LSQ and personality
(Furnham, 1992, 1996; Jackson & Lawty-Jones, 1996). Such investigations have reported con-
siderable correlation between learning styles such as Activist and personality variables such as
extraversion. These studies conclude in general that learning style is a subset of personality. In a
recent study of telephone sales staff, Reflector and Pragmatist learning styles account for a small
but important amount of the variance in measures of work performance (Furnham, Jackson, &
Miller, 1999).In summary, rather mixed evidence exists as to the validity and utility of LSQ data. The
objectives of the present study were: first, to use EFA and CFA to determine if the four proposed
learning styles and two bipolar dimensions are clearly identified; second, to determine the gen-
eralizability and stability of LSQ responses across different student groups and third, to deter-
mine the predictive validity of the LSQ by relating scores on the instrument to examination marks
for the sample of health studies students using structural equation modelling (SEM).
2. Method
2.1. Participants
Scores on the LSQ were validated using participants (n=388) of undergraduate students
enrolled in two different faculties of a regional university in Scotland: business and health studies.
The instrument was administered to all participants at the start of the academic year. Those
enrolled in the business faculty completed the instrument in a classroom environment, the health
studies students completed the instrument at their home and returned the instrument by post. The
distribution of the sample across the faculties was as follows: business, n=224 (males=104,
females=120; average age=24.36 years, minimum age=17 years, maximum age=53 years); health
studies, n=164 (males=15; females=149; average age=34.82 years, minimum age=20 years,
maximum age=62 years). Students enrolled in the business faculty were undertaking taughtmodules requiring attendance on campus on a full-time or part-time basis. Those enrolled in the
health studies faculty were undertaking distance-learning modules requiring no attendance on cam-
pus and were located throughout the UK. As can be seen from the above description, the sample
was demographically diverse enough to generalize to other Western higher educational settings.
2.2. Statistical analyses
The appropriateness of the four styles was evaluated using CFA using the SPSS version of
AMOS v4.0 (Arbuckle, 1999). AMOS is a statistical program to perform structural equation
150 A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 5/17
modelling, a form of multivariate analysis. CFAs were performed using AMOS to test for the
goodness-of-fit between our obtained data and the hypothesised structure of Honey and Mum-
ford (1992). When undertaking CFA, there are ‘‘vague and sometimes contradictory guidelines
about the desirable amount of data’’ (Marsh & Hau, 1999, p. 252). Marsh, Balla, and Hau (1997)using a Monte Carlo study investigated the effect of varying numbers of indicators (items) per
factor (p/f ratio) on varying sample sizes. Their results support a ‘‘more is better’’ approach to
both sample size and p/f ratio. For a p/f ratio as small as six, a sample size of 50 was adequate.
Therefore, the combined sample size of n=388, is satisfactory for conducting CFAs at the item
level of the LSQ, where the number of items per factor (p/f) equals 20 (Marsh et al., 1997). In
evaluating goodness-of-fit, we present the w2 statistic, the Relative Noncentrality Index (RNI),
the Tucker-Lewis Index (TLI), the ratio of the discrepancy, w2, divided by the degrees of freedom
(w2 /df), the Adjusted-Goodness of Fit Index (AGFI) and an evaluation of parameter estimates to
ensure the solution is proper (Marsh, Balla, & Hau, 1996; McDonald & Marsh, 1990). Although
no precise standards exist to indicate what value of indices are needed for a satisfactory fit, typicalguidelines are that the RNI should exceed 0.9. Various rules-of-thumb ranging from 2 to 5 have
been suggested as cut-offs for CMIN/d.f.. The present study follows the recommendations of
Byrne (1989) that a w2/d.f. ratio of greater than 2.0 represents an inadequate fit.
In the results reported below, discussions as to the best model fit include each of the above mea-
sures. This is in keeping with the Hoyle and Panter’s (1995) recommendations that multiple indica-
tors of overall fit should be selected from ‘‘absolute-fit indexes’’ (such as w2, and the AGFI) and
‘‘incremental fit indexes,’’ which should be selected from ‘‘type-2’’ and ‘‘type-3’’ indexes, such as
the TLI (Tucker & Lewis, 1973) and RNI (McDonald & Marsh, 1990). A type-2 index compares
the lack of fit of a target model with the lack of fit of a baseline model, usually the independence
model. Value estimates the relative improvement per degrees of freedom of the target model over
a baseline model (Hoyle & Panter, 1995). A type-3 index ‘‘indexes the relative reduction in lack of fit asestimated by the noncentral (two of a target model versus a baseline model’’ (Hoyle & Panter, 1995).
The hypothesised bipolar structure is examining by first, an examination of the LSQ scale cor-
relation coefficient matrix; and second, by means of EFA using maximum-likelihood analysis
followed by oblimin rotation. The psychometric requirement for achieving two zero-correlated
bi-polar measures (see Kirton, 1994) would be for each of the four style scales to be negatively
correlated with one other scale and zero correlated with the two remaining scales.
Considering the EFA, principal components were used: first, because this method yields com-
ponent scores that have the same correlation coefficients as the rotated factors; and second, as
component analysis does not unduly capitalise on sampling error as the price for estimating
measuring error (Thompson & Daniel, 1996). A further consideration is to determine the numberof factors to be extracted. Thompson and Daniel (1996) recommend employing a number of dif-
ferent methods to select factors. Accordingly, the present study uses eigenvalues-greater-than-one
rule (Kaiser, 1960), scree tests (Cattell, 1978) and parallel analysis techniques (Horn, 1965).
Separate CFAs were conducted on responses by students from the two groups of business and
health studies. The generalizability and stability of the two-factor and four-factor model is eval-
uated—following Marsh’s (1994) recommendations—by comparing the fit of three models: total
non-invariance (where no parameter estimates were constrained to be equal across groups), factor
parameters invariant (where only factor parameters were constrained to be equal across groups)
and total invariance (where all parameter estimates were constrained to be equal across groups).
A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163 151
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 6/17
Educational attainment is generally believed to be a function of the learning environment, the
ability of the student and the individual’s learning style. Therefore it could be expected that stu-
dents with a preference for particular learning activities would outperform those with preferences
for other learning activities. To evaluate the relationship between learning style and performance,a structural equation model was constructed for both the health studies students and business
studies students using AMOS v4.0 (Arbuckle, 1999). Each model treats the four learning style
dimensions as observed exogenous predictor variables and academic performance as the observed
endogenous variable. Academic performance is proxied: for health studies students by a percen-
tage grade in a healthcare module (scores ranging from 19 to 79%; mean=58.0%; S.D.=9.4%),
and for business students by a percentage grade in an accounting module (scores ranging from 21
to 68%; mean=47.8%; S.D.=10.1%).2
3. Results
Alpha coefficients were calculated for the scores on the four LSQ scales. The coefficients indi-
cate the scores produced by the LSQ have only modest internal consistency reliability: Activist,
0.681; Reflector, 0.731; Theorist, 0.577; Pragmatist, 0.516. Item analysis of each scale removed 26
items, on a one-by-one basis, in total (four from Reflector, 11 from Theorist, 11 from Pragmatist)
that failed to improve the homogeneity of the scale to which they were assigned by the authors of
the LSQ. This item attrition failed to raise the scales to a level of minimum integrity of 0.70
suggested by Nunnally and Bernstein (1994): Activist, 0.671; Reflector, 0.750; Theorist, 0.661;
Pragmatist, 0.638. The results of the item attrition exercise are shown in the appendix and com-
pared to the results of De Ciantis and Kirton (1996).
Before item analysis, the combined 40-item Activist–Reflector scale yielded an alpha coefficientof 0.789. Item analysis of the combined Activist–Reflector scale created a 27-item scale (=0.817;
Table 2
Learning Style Questionnaire (LSQ) scale correlation matrix
Scale
Scale A P R T A–R T–P
Activist (A) – 0.038 0.464 * 0.105
Pragmatist (P) 0.219 * – 0.023 0.573 *
Reflector (R) 0.480 * 0.082 – 0.148 *
Theorist (T) 0.269 * 0.415 * 0.499 * –
A–R – 0.099
T–P 0.220 * –
The upper triangle represents scales after item analysis. The lower triangle represents original LSQ scale scores before
the scales were re-defined. *P<0.01
2 Like most measures of academic performance the data suffers from a restriction of range. Also, academic perfor-
mance is proxied by the results of only one module. However, the investigation benefits from an analysis of two distinct
samples.
152 A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 7/17
Appendix). Combining Theorist–Pragmatist into a single 40-item scale, produced an alpha of
0.689, and after item analysis this increased to 0.766 (16 items).
Table 2 reports the correlation coefficient matrix of the four learning styles of the LSQ, both
prior to and after item analysis. As hypothesised, Activist is highly negatively correlated withReflector (prior to item analysis, r=0.48; after item analysis, r=0.46). However, Activist is
also positively correlated with Pragmatist and Theorist.
Theorist is highly correlated with Pragmatist (prior to item analysis, r=0.42; after item analy-
sis, r=0.57). Not only is the correlation coefficient in the opposite direction to that expected in
theory, as Theorist and Pragmatist are considered opposites, it is of a high magnitude. Prior to
item analysis a stronger relationship exists between Theorist and Reflector (r=0.50). However,
after item analysis, the Theorist–Reflector relationship is attenuated (r=0.15). Prior to item
analysis, Activist-Pragmatist and Theorist Reflector are positively correlated (r=0.22); however
after item analysis this is negligible (r=0.10). We conclude, after the item attrition exercise, the
LSQ scale correlation matrix provides some evidence, albeit mixed, to support the two dimen-sions of prehension and transformation.
The results of the EFA are shown in Table 3. Principal component analysis using the eigenva-
lues-greater-than-one rule, extracted 27 factors (accounting for 62.51% of the total variance);
parallel analysis extracted nine factors (accounting for 34.59% of the total variance) and scree
tests extracted five factors (accounting for 25.77% of the total variance). Evidence exists to show
that the scree test is more accurate than the eigenvalue-greater-than-one criterion when both are
tested on artificially generated sample data (Zwick & Velicer, 1982, 1986). The results of the scree
test are shown in Fig. 1. Factor 1 consisted of 24 items with item coefficients greater than 0.25,
Fig. 1. EFA Scree Test.
A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163 153
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 8/17
Table 3
Factor structure matrix
Item Scale Factor
I II III IV V
15 Reflector 0.577
16 Reflector 0.549
46 Reflector 0.516 0.338
60 Reflector 0.496
25 Reflector 0.474
28 Reflector 0.472 0.322
2 Activist 0.457 0.322
56 Pragmatist 0.415
66 Reflector 0.409 0.319 0.375
29 Reflector 0.382
57 Theorist 0.36331 Reflector 0.329
35 Pragmatist 0.282
38 Activist 0.339
67 Reflector 0.300
48 Activist 0.391
10 Activist 0.271
74 Activist 0.288 –.481
34 Activist 0.340 0.340 0.403
7 Reflector 0.330 0.345
75 Theorist 0.255 0.541
70 Pragmatist 0.253 0.441 0.284 0.444
73 Pragmatist 0.305 0.39241 Reflector 0.316 0.318
42 Theorist 0.567
68 Theorist 0.554
65 Pragmatist 0.500 0.257
14 Theorist 0.494
12 Theorist 0.470
35 Pragmatist 0.470
27 Pragmatist 0.421
61 Theorist 0.415
54 Pragmatist 0.405
80 Pragmatist 0.380
49 Pragmatist 0.360
47 Theorist 0.349
5 Pragmatist 0.347 0.291
22 Theorist 0.319
77 Theorist 0.319
8 Theorist 0.343
20 Theorist 0.367 0.526
19 Pragmatist 0.307 0.523
1 Theorist 0.251 0.277 0.424
30 Theorist 0.339
(continued on next page)
154 A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 9/17
with 19 items from the combined Activist–Reflector dimension, item 10 from the Activist scale
was negatively correlated with the factor. Factor 2 consisted of 19 item coefficients greater than
0.25, with all items from the Theorist–Pragmatist dimension. However, three items (1, 19 and 20)
were negatively correlated with the factor. Factor three, consisted of 19 items, with 13 items from
the Activist–Reflector dimension negatively correlated with the factor. Of the residual six items,
five were from the Theorist–Pragmatist scale and were all positively correlated with the factor.
The remaining item from the third identified factor was from the Activist scale and positively
correlated with the factor. The fourth identified factor consisted of 15 items. Eleven items were
from the Activist scale—each negatively correlated with the factor. Of the residual four items,
Table 3 (continued )
Item Scale Factor
I II III IV V
Item Scale I II III IV V
67 Reflector 0.674
62 Reflector 0.650
58 Activist 0.611
43 Activist 0.572
48 Activist 0.561
23 Activist 0.516
10 Activist 0.508
72 Activist 0.461
11 Pragmatist 0.429
8 Theorist 0.374
79 Activist
00.357
0.32932 Activist 0.320
17 Activist 0.483
4 Activist 0.403
71 Activist 0.378
45 Activist 0.368
64 Activist 0.367
40 Activist 0.351
6 Activist 0.345
24 Activist 0.260
69 Pragmatist 0.269 0.585
63 Theorist 0.545
21 Pragmatist 0.541
18 Theorist 0.490
78 Theorist 0.487
50 Pragmatist 0.456
51 Theorist 0.453
59 Pragmatist 0.449
44 Pragmatist 0.350
37 Pragmatist 0.346
38 Activist 0.250
% Variance Explained 7.59% 7.23% 4.99% 3.04% 2.92%
Principal components analysis; oblimin rotation; factor structure matrix coefficients less than 0.25 omitted.
A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163 155
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 10/17
two were from the Reflector scale and again, negatively correlated with the factor. Factor five
consisted of 18 items with factor structure matrix coefficients greater than 0.25. Fifteen items in
this identified factor lie within the Pragmatist–Theorist dimension and each was positively corre-
lated with the factor.Considering the results of the EFA along with the analysis of the LSQ scale correlation matrix,
only limited evidence exists to support the hypothesised two learning dimensions and four learn-
ing styles. The descriptive statistics and results of the confirmatory factor analyses are shown in
Table 4. The fit indices for both the two-factor and four-factor model indicates that the data are a
poor fit to the two hypothesised learning dimensions and four learning styles. Fit indices are
shown for both the two-factor and four-factor models after item analysis. Inspection of the fit
indices after the item attrition exercise indicates both models estimated still failed to fit the data.
The results of the tests of factorial invariance are shown in Table 5. For both analyses, the fit
indices of the total invariance models were worse than the fit indices for the models with invariant
factor parameter estimates that were worse than fit indices for the models with no invarianceconstraints across the two groups. In both cases, models constraining all parameters to be the
same across the two groups of students had poor fit indices.
Table 4
Confirmatory Factor Analyses of the Learning Style Questionnaire (LSQ)
Model Alpha Mean S.D. No. Items
One-factor model 0.783 48.436 8.510 80
w2=11470.967; d.f.=3161; w2/df=3.629; TLI=0.278; RNI=0.288; AGFI=0.539
Two-factor model
Activist–Reflector 0.789 25.204 5.790 40Pragmatist–Theorist 0.689 23.232 5.092 40
w2=6422.316; df=3082; w2/df=2.084; TLI=0.346; RNI=0.362; AGFI=0.633
Two-factor model after item analysis
Activist–Reflector 0.817 18.186 5.016 27
Pragmatist–Theorist 0.769 8.601 3.622 16
w2=2261.967; d.f.=819; w2/df=2.762; TLI=0.461; RNI=0.487; AGFI=0.739
Four-factor model
Activist 0.681 10.55 3.46 20
Pragmatist 0.516 11.79 2.91 20
Reflector 0.731 14.66 3.39 20Theorist 0.577 11.44 3.07 20
w2=6358.546; d.f.=3077; w2/d.f.=2.066; TLI=0.357; RNI=0.374; AGFI=0.642
Four-factor model after item analysis
Activist 0.681 10.55 3.46 20
Pragmatist 0.638 4.76 2.11 9
Reflector 0.750 11.56 3.22 16
Theorist 0.661 4.79 2.24 9
w2=3238.583; d.f.=1373; w2/d.f.=2.359; TLI=0.421; RNI=0.444; AGFI=0.725
TLI, Tucker-Lewis Index; RNI, Relative Noncentrality Index; AGFI, Adjusted-Goodness of Fit Index.
156 A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 11/17
Table 5
Two-group Analyses of Invariance for Business and Health Studies Students
Model w2 d.f. w
2/df AGFI RNI TLI
Two-factor model
2gp (no inv) 10531.673 6164 1.709 0.564 0.245 0.226
2gp (fl inv) 10733.557 6239 1.720 0.563 0.224 0.213
2gp (tot inv) 11026.942 6319 1.745 0.561 0.187 0.187
Four-factor model
2gp (no inv) 10339.496 6154 1.680 0.585 0.277 0.257
2gp (fl inv) 10554.325 6219 1.697 0.578 0.251 0.239
2gp (tot inv) 12056.621 6353 1.898 0.513 0.015 0.020
TLI, Tucker-Lewis Index; RNI, Relative Noncentrality Index; AGFI, Adjusted-Goodness of Fit Index.
Fig. 2. Business studies students’ learning styles and academic performance.
A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163 157
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 12/17
The standardised estimates of the SEM for the sample of business students is shown in Fig. 2.
The squared multiple correlation between academic performance (the observed endogenous
variable) and the four learning styles (the observed exogenous variables) is 0.09, indicating 9% of
academic performance is accounted for by its predictors (learning style dimensions). Examiningthe standardised regression weights, the strongest positive predictor variable is Theorist with a
weight of 0.18. Notably, both Pragmatist and Reflector have negative standardised regression
weights of 0.28 and 0.23, respectively.
Fig. 3 displays the results of the standardised estimates of the SEM for the sample of 145 health
studies students. The squared multiple correlation between the observed exogenous predictor
variables (learning style dimensions) and observed endogenous variable (academic performance)
is 0.04, indicating only 4% of academic performance is accounted for by its predictors (learning
style dimensions). Examining the standardised regression weights, the strongest weight is between
the Pragmatist scale and academic performance (0.18). The standardised regression weights for
the Activist and Theorist scales are negligible at
0.01 and 0.05, respectively.
Fig. 3. Health studies students’ learning styles and academic performance.
158 A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 13/17
Comparing the path diagrams—shown in Figs. 2 and 3—for the business students and the
health studies students respectively, reveals a number of significant differences. Considering the
relationship between the observed exogenous predictor variables (learning style dimensions) and
the observed endogenous variable, the strongest predictor is Pragmatist. However, the standar-dised regression weight for the health studies is 0.18 and for the business students, 0.27. The
strongest positive standardised regression weight between the predictor variable and academic
performance for the business students is Theorist (0.18). Theorist is a poor predictor of academic
performance amongst the health studies students with a standardised regression weight of only
0.05. An analysis of the correlation coefficients between the learning style dimensions shown in
Figs. 2 and 3 reveals a number of significant differences. Notably, the highest correlation between
the learning style dimensions for the health studies students is between Reflector and Pragmatist
(r=0.50); for business studies students this relationship is negligible (r=00.03).
Other notable differences include: Activist–Pragmatist (business r=0.22, health studies
r=
0.21); Reflector–Theorist (business r=0.41, health studies r=0.08); and Activist–Theorist(business r=0.17, health studies r=0.27). In summary, evidence exists to: first, suggest the
relationship between learning style dimensions is different between the sample of business stu-
dents and health studies students; second, indicate that learning style is only a weak predictor of
academic performance; and third, that the learning style predictors of academic performance are
different between the samples of business and health studies students.
4. Conclusion
Results from the multi-analytical research techniques used in the present study with higher
education students from two different faculties and undertaking study in two different deliverymodes indicate that the scores produced by the LSQ have limited reliability and validity for this
population and that the LSQ’s multidimensional structure is not invariant across faculty groups.
Although an item pruning exercise improved the internal consistency reliability evidence, three
(of the four) scales failed to meet a level of minimum internal integrity. The results of the CFAs
on both the two-factor and four-factor models after the item analysis, indicated a poor fit to the
data. In summary, no evidence is found for the four learning styles or two bipolar dimensions
hypothesised by Honey and Mumford (1992). Our study indicates even after modification the
LSQ cannot be safely used with samples of UK undergraduate students.
The use of CFA as a technique, along with supplemental analyses of the invariance of the
parameter estimate across multiple groups, extends the psychometric research surrounding theinstrument. Our study supports the EFAs of De Ciantis and Kirton (1996) sampling UK and Eire
managers.
A further issue concerning the LSQ is that the four learning styles are orthogonal to one
another; that is a person may score high on one and low on the others, or high or low all four. A
problem arises with individuals who have only low or moderate preferences for any particular
learning style, implying a low level of preference for any form learning as conceived by Kolb’s
ELM. Explanations for this could include that the instrument is not sufficient to identify students’
dominant learning style or they do not interact with the learning model of higher education. De
Ciantis and Kirton (1996) contend that the level (abilities) in which individuals engage in each
A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163 159
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 14/17
stage of the learning process (the learning cycle) is unrelated to their learning style and their
effectiveness in completing all or part of the process, each of which should be assessed separately.
Despite the measurement problems surrounding the LSQ (and LSI/LSI-1985), Kolb’s ELM has
been applied in schools, colleges and universities and in vocational education as well as in man-agement development, where it has gained unquestioned status at the core of the profession’s
pedagogical apparatus (Reynolds, 1997). One of the reasons Kolb’s work has caught on so dra-
matically is that the research on which it is based suggests that persons who choose one or
another occupational field tend strongly to share that learning style. This is useful knowledge for
those who are interested in career choice and pedagogy in professional fields. Hopkins (1993)
identifies a profound difference between measured experience and lived experience—a difference
that makes Kolb’s formulation so deeply problematic. Nelsen and Grinder (1985) identify the
fundamental problem with Kolb’s ELM as the failure to untangle experience from learning and
structure from process. Although Kolb has developed measures of the four learning styles,
thereby operationally distinguishing the structures, no method or operation for investigating theprocesses underlying these structures is presented.
In conclusion, it is suggested that the LSQ is based on a model (the ELM) which is not suffi-
ciently sophisticated to describe the learning that takes place in higher education. The LSQ is
defined in terms of a management trainee’s learning rather than that of a student in higher edu-
cation. Caution should be employed if adopting the LSQ to select appropriate instructional
methods or to categorise individual students. The findings indicate the LSQ is not a suitable
alternative to either the LSI or LSI-1985.
Appendix. Revised scale items after item analysis
Activist–Reflector Theorist–Pragmatist
Present study DeCiantis and
Kirton (1996)
Present study De Ciantis and
Kirton (1996)
2 2 5 5
3 8 8
4 11
6 127 7 14
10 21
13 26
15 15 27 27
16 16 30
17 17 35 35
(continued on next page)
160 A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 15/17
Activist–Reflector Theorist–Pragmatist
Present study DeCiantis and
Kirton (1996)
Present study De Ciantis and
Kirton (1996)
23 37
24 42 42
25 25 47
28 28 54 54
29 29 56
31 31 61 61
32 65
34 34 68
36 69
38 70 7039 39 73
40 77
41 41 80
43 43
45 45
46 46
48 48
52
55 55
58 5860
62 62
64
66 66
67 67
71
72 72
74 74
79 79
27 items; =0.82 36 items; =0.88 16 items; =0.77 15 items; =0.69
References
Allinson, C. W., & Hayes, J. (1988). The Learning Styles Questionnaire: An alternative to Kolb’s Inventory? Journal of
Management Studies, 25, 269–281.
Allinson, C. W., & Hayes, J. (1990). Validity of the learning styles questionnaire. Psychological Reports, 67 , 859–866.
Arbuckle, J. L. (1999). Amos users’ guide version 4.0. Chicago, IL: SPSS.
Appendix. (continued )
A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163 161
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 16/17
Byrne, B. M. (1989). A Primer of LISREL: basic applications and programming for confirmatory factor analytic models.
New York: Springer-Verlag.
Cattell, R. B. (1978). The scientific use of factor analysis. New York: Plenum Press.
Cornwell, J. M., & Dunlap, W. P. (1994). On the questionable soundness of factoring ipsative measures data: aresponse to Saville & Wilson (1991). Journal of Occupational and Organizational Psychology, 67 , 89–100.
De Ciantis, S. M., & Kirton, M. J. (1996). A psychometric reexamination of Kolb’s experiential learning cycle con-
struct: a separation of level, style and process. Educational and Psychological Measurement, 56, 809–820.
Dunlap, W. P., & Cornwell, J. M. (1994). Factor analysis of ipsative measures. Multivariate Behavioural Research, 29,
115–126.
Freedman, R. D., & Stumpf, S. A. (1978). What can one learn from the Learning Styles Inventory? Academy of Man-
agement Journal , 21, 275–282.
Freedman, R. D., & Stumpf, S. A. (1980). Learning Styles Inventory: less than meets the eye. Academy of Management
Review, 5, 445–447.
Furnham, A. (1992). Personality and learning style: a study of three instruments. Personality and Individual Differences,
13, 429–438.
Furnham, A. (1996). The FIRO-B, the Learning Style Questionnaire and the Five-Factor Model. Journal of Social
Behaviour and Personality, 11, 285–299.Furnham, A., Jackson, C. J., & Miller, T. (1999). Personality, learning style and work performance. Personality and
Individual Differences, 27 , 1113–1122.
Furnham, A., & Medhurst, S. (1995). Personality correlates of academic seminar behaviour: a study of four instru-
ments. Personality and Individual Differences, 19, 197–220.
Geiger, M. A., Boyle, E. J., & Pinto, J. (1992). A factor analysis of Kolb’s Revised Learning Styles Inventory. Educa-
tional and Psychological Measurement, 52, 753–759.
Geiger, M. A., Boyle, E. J., & Pinto, J. (1993). An examination of ipsative and normative versions of Kolb’s revised
Learning Style Inventory. Educational and Psychological Measurement, 53, 717–726.
Geller, L. (1979). Reliability of the Learning Style Inventory. Psychology Reports, 44, 555–561.
Honey, P., & Mumford, A. (1992). The manual of learning styles. Maidenhead: Peter Honey.
Hopkins, R. (1993). David Kolb’s Experiential Learning Machine. Journal of Phenomenological Psychology, 21, 46–62.
Horn, J. L. (1965). A rationale for the number of factors in factor analysis. Psychometrica, 30, 179–185.Hoyle, R. H., & Panter, A. T. (1995). Writing about structural equation models. In R. H. Hoyle (Ed.), Structural
equation modeling (pp. 76–99). Thousand Oaks, CA: Sage.
Jackson, C. J., & Lawty-Jones, M. (1996). Explaining the overlap between personality and learning styles. Personality
and Individual Differences, 20, 293–300.
Jung, C. G. (1921). Psychological types. NJ: Princeton University Press.
Kaiser, H. F. (1960). The application of electronic computers to factor analysis. Educational and Psychological
Measurement, 20, 141–151.
Kirton, M. (1994). Adaptors and innovators: styles of creativity and problem solving (2nd ed.). London: Routledge.
Kolb, D. A. (1976). Learning style inventory: technical manual . Boston, MA: McBer & Company.
Lewin, K. (1936). The principles of topological psychology. New York: McGraw-Hill.
Loo, R. (1999). Confirmatory factor analyses of Kolb’s Learning Style Inventory (LSI-1985). British Journal of Edu-
cational Psychology, 69, 213–219.
Marsh, H. W. (1994). Confirmatory factor analysis models of factorial invariance: a multifaceted approach. Structural
Equation Modeling, 1, 5–34.
Marsh, H. W., Balla, J. R., & Hau, K. T. (1996). An evaluation of fit indices: a clarification of mathematical and
empirical processes. In G. A. Marcoulides, & R. E. Schumacker (Eds.), Advanced structural equation modeling tech-
niques, 4 (pp. 59–98). Hillsdale, NJ: Lawrence Erlbaum.
Marsh, H. W., Balla, J. R., & Hau, K. T. (1997). ‘ Is more ever too much: the number of indicators per factor in
confirmatory factor analysis’, Paper presented at the Annual Meeting of the American Educational Research
Association, Chicago.
Marsh, H. W., & Hau, K. T. (1999). Confirmatory factor analysis: strategies for small sample sizes. In R. H. Hoyle
(Ed.), Statistical strategies for small sample research (pp. 251–284). Newbury Park, CA : Sage.
162 A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163
7/23/2019 duff-y
http://slidepdf.com/reader/full/duff-y 17/17
McDonald, R. P., & Marsh, H. W. (1990). Choosing a multivariate model: Noncentrality and goodness-of-fit. Psy-
chological Bulletin, 107 , 247–255.
Mulaik, S. A. (1987). A brief history of the philosophical foundations of exploratory factor analysis. Multivariate
Behavioral Research, 22, 267–305.Myers, I. B., & Briggs, K. C. (1962). Myers-Briggs indicator. C.A: Consulting Psychologists Press Inc.
Nelsen, E. A., & Grinder, R. E. (1985). Toward an ex cathedra doctrine of learning. Contemporary Psychology, 30,
622–623.
Newstead, S. E. (1992). A study of two ‘‘quick and easy’’ methods of assessing individual differences in student learn-
ing. British Journal of Educational Psychology, 62, 299–312.
Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory (3rd ed.). New York: McGraw-Hill.
Reynolds, M. (1997). Learning styles: a critique. Management Learning, 28, 115–133.
Ruble, T. L., & Stout, D. E. (1993). Comments on the use of the LSI in research on student performance in accounting
education. Accounting Educators’ Journal , 5, 35–45.
Schwab, D. P. (1980). Construct validity in organizational behavior. In B. M. Straw, & L. L. Cumings (Eds.), Research
in organizational behavior, 2. Greenwich, C.T: JAI Press.
Sims, R. R., Veres, G. G., & Shake, I. G. (1989). An exploratory examination of the convergence between the Learning
Style Questionnaire and the Learning Style Inventory II. Educational and Psychological Measurement, 49, 227–233.Stout, D. (1991a). E, Ruble, T. L. A reexamination of accounting students’ learning styles. Journal of Accounting Edu-
cation, 9, 341–354.
Stout, D. E., & Ruble, T. L. (1991b). The LSI and accounting education research: a cautionary view and suggestions
for future research. Issues in Accounting Education, 6, 41–52.
Tepper, B. J., Tetrault, L. A., Braun, C. K., & Romero, J. E. (1993). Discriminant and convergent validity of the
problem solving style questionnaire. Educational and Psychological Measurement, 53, 437–444.
Thompson, B., & Daniel, L. G. (1996). Factor analytic evidence for the construct validity of scores: a historical over-
view and some guidelines. Educational and Psychological Measurement, 56, 197–208.
Tucker, L. R., & Lewis, C. (1973). A reliability coefficient for maximum likelihood factor analysis. Psychometrika, 38,
1–10.
Willcoxon, L., & Prosser, M. (1996). Kolb’s Learning Style Inventory (1985): review and further study of validity and
reliability. British Journal of Educational Psychology, 66, 247–257.Yahya, I. (1998). Willcoxon and Prosser’s factor analyses on Kolb’s (1985) LSI data: reflections and re-analyses. British
Journal of Educational Psychology, 68, 281–286.
Zwick, W. R., & Velicer, W. F. (1982). Factors influencing five rules for determining the number of components to
retain. Multivariate Behavioral Research, 17 , 253–269.
Zwick, W. R., & Velicer, W. F. (1986). Comparison of five rules for determining the number of components to retain.
Psychological Bulletin, 99, 432–442.
A. Duff, T. Duffy / Personality and Individual Differences 33 (2002) 147–163 163