Full text of DOI:10.1348/000709900158209

British Journal of Educational Psychology (2000), 70, 405±418 Printed in Great Britain
2000 The British Psychological Society
#
Item-level and subscale-level factoring of Biggs’
Learning Process Questionnaire (LPQ) in a
mainland Chinese sample
John Sachs
University of Hong Kong
Lingbiao Gao*
Institute of Curriculum Studies and Teaching Material Development,
South China Normal University
Background. The learning process questionnaire (LPQ) has been the source of
intensive cross-cultural study. However, an item-level factor analysis of all the
LPQ items simultaneously has never been reported. Rather, items within each
subscale have been factor analysed to establish subscale unidimensionality and
justify the use of composite subscale scores.
Aims. It was of major interest to see if the six logically constructed items
groups of the LPQ would be supported by empirical evidence. Additionally, it
was of interest to compare the consistency of the reliability and correlational
structure of the LPQ subscales in our study with those of previous cross-
cultural studies.
Methods. Confirmatory factor analysis was used to fit the six-factor item level
model and to fit five representative subscale level factor models.
Sample. A total of 1070 students between the ages of 15 to 18 years was drawn
from a representative selection of 29 classes from within 15 secondary schools
in Guangzhou, China. Males and females were almost equally represented.
Results. The six-factor item level model of the LPQ seemed to fit reasonably
well, thus supporting the six dimensional structure of the LPQ and justifying
the use of composite subscale scores for each LPQ dimension. However, the
reliability of many of these subscales was low. Furthermore, only two
subscale-level factor models showed marginally acceptable fit. Substantive
considerations supported an oblique three-factor model.
Conclusions. Because the LPQ subscales often show low internal consistency
reliability, experimental and correlational studies that have used these
subscales as dependent measures have been disappointing. It is suggested that
some LPQ items should be revised and other items added to improve the
inventory’s overall psychometric properties.
*Requests for reprints should be addressed to Dr. Lingbiao Gao, Institute of Curriculum Studies and
Teaching Material Development, South China Normal University, China.

406
John Sachs and Lingbiao Gao
The learning process questionnaire or LPQ (Biggs, 1987) is a self-report inventory
designed to measure three approaches to learning ± surface, deep, and achieving ± in
secondary students. The three approaches reflect the extrinsic, intrinsic, and achieving
motivation, respectively, of the learner (Biggs, 1992, 1995). Each approach to learning
is defined by a question related to the student’s motivation for learning and the strategy
that the student employs to achieve it (Biggs, 1992, 1995). Consequently, the LPQ
measures six dimensions of student approaches to learning (Biggs, 1992): surface motive
(SM), surface strategy (SS); deep motive (DM), deep strategy (DS); achieving motive
(AM), achieving strategy (AS).
Since identifying the approaches students use in learning is of considerable practical
educational importance in both improving student learning and improving the
curriculum, the LPQ has been the source of intense cross-cultural research from its
inception (Andrews, Violato, Rabb, & Hollingsworth, 1994; Biggs, 1987, 1991, 1992;
Kember & Leung, 1998; Watkins & Akande, 1994; Watkins, Akande, & Mpofu, 1994;
Watkins, Dong, & Xia, 1995; Watkins & Mpofu, 1997; Wong, Lin & Watkins, 1996).
All these studies reported on the reliability of the six LPQ subscales and on their factor
structure. None of these studies, however, reported an item-level factoring of the LPQ
except for Biggs (1992, 1993). In those cases, the item-level factoring was done
separately for the items comprising each subscale and not simultaneously for all the
LPQ items. It seems reasonable, therefore, to enquire if the logically constructed item
groups of the LPQ (Biggs, 1992) can be confirmed with empirical data before subscale-
level factoring is attempted.
Inventories of student learning such as the LPQ, the slightly longer version, the study
process questionnaire or SPQ (Biggs, 1992) and Entwistle, Hanley, and Hounsell’s
(
1979) approaches to study inventory (ASI; Entwistle & Ramsden, 1983; Ramsden &
Entwistle, 1981), including the revised version (RSAI; Duff, 1997; Entwistle & Tait,
995), are commonly referred to as student approaches to learning (SAL) inventories
because they emphasise the context within which student learning takes place (Biggs,
993; Entwistle & Waterston, 1988). This is in contrast to inventories of student
learning such as Schmeck, Ribich, and Ramanaiah’s inventory of learning processes
1977; Schmeck & Grove, 1979; Schmeck, Geisler-Brenstein, & Cercy, 1991) that place
1
1
(
greater emphasis on the cognitive aspects of learning such as information processing.
Both the SPQ and ASI are appropriate for tertiary level students and measure similar
constructs (Biggs, 1993; Entwistle & Ramsden, 1983; Entwistle & Waterston, 1988;
Harper & Kember, 1989; Kember & Leung, 1998; Watkins, 1996). Watkins (1996) notes
that the adoption of the deep/surface terminology for constructs measured by these
inventories owes much to the early qualitative work of Marton and SaÈljoÈ (1976).
Since the underlying theory of SAL as assessed by the LPQ has been described at
length in Biggs (1987, 1992, 1993, 1995, 1996), Kember and Leung (1998), and Watkins
(1996), it will not be repeated here. The interested reader is referred to the cited
references. The emphasis in this paper will be on the psychometric properties of the
LPQ within a cross-cultural context. In that regard, the purpose of this paper is
threefold. First, the reliability of the six LPQ subscales will be assessed and compared
with those reported in previous studies such as Wong et al. (1996). Second, item-level
factoring will be conducted to determine if the LPQ does in fact measure six dimensions
of student learning as proposed by Biggs (1987, 1992, 1993, 1995). Third, various

Psychometric properties of the LPQ
407
theoretically derived subscale-level factor models will be fit and compared with relevant
ones reported in previous studies such as Andrews et al. (1994), Wong et al. (1996), and
Kember and Leung (1998). Important cross-cultural differences will be discussed.
Method
Sample
A sample of 1070 secondary students was drawn from 29 classes within 15 secondary
schools in Guangzhou, China. The sample was comprised of 557 males and 459
females, with 54 students failing to indicate their gender. Student ages ranged from 15
to 18 years.
Instrument
The LPQ is a self-report inventory consisting of 36 statements about student
approaches to learning. The scale is usually administered to secondary school students.
Students are asked to respond to each statement by indicating how true that statement
is of them using on a 5-point Likert scale: 1-never or only rarely true of me, 2-sometimes
true of me, 3-true of me about half the time, 4-frequently true of me, 5-always or almost
always true of me. These 36 statements are grouped into six subscales with six
statements (items) per subscale. The six subscales are measures of the six dimensions of
student approaches to learning proposed by Biggs (1987, 1992, 1993, 1995): surface
motive (SM), surface strategy (SS); deep motive (DM), deep strategy (DS); and
achieving motive (AM), achieving strategy (AS).
The LPQ data used in this study were part of a larger study conducted by Gao (1999)
on `conceptions of teaching held by school physics teachers in Guandong, China and
their relations to student learning’. Therefore, where necessary, some LPQ statements
(
items) were slightly reworded so that their reference to learning a subject in general was
focused on physics instead. For example, LPQ item 14, `I find that many subjects can
become very interesting once you get into them’, would be reworded to read `I find that
physics can become very interesting once you get into it’. But LPQ item 18, `I try to do
all of my assignments as soon as they are given to me’ would remain unchanged. This
modification to some LPQ statements is only minor and is unlikely to invalidate
comparisons of our results with results from studies that used an unmodified version of
the LPQ.
Once the necessary subject specific changes were made to some of the LPQ
statements, experts in English to Chinese translation were used to translate all 36 LPQ
statements into Chinese. Care was taken to ensure that each translated statement
retained as similar a meaning as possible to the original.
Design
There were three parts to this study. First, internal-consistency reliability estimates
along with other descriptive measures were obtained for each LPQ subscale. Second,
Biggs’ six-factor model was fit to all 36 LPQ items. We decided to use the correlation
matrix for this item-level confirmatory factor analysis because the categorical 5-point
Likert scale used to measure the item-level responses is clearly arbitrary and because the
interest in this analysis centred on the relationships between the 36 LPQ items. The use

408
John Sachs and Lingbiao Gao
of the correlation matrix instead of the covariance matrix for fitting the confirmatory
factor model was also acceptable since no item loaded on more than one factor in this
six-factor model (see Figure 1), and each factor was standardised (Cudeck, 1989).
Figure 1. Simplified schematic of the six-factor item-level model of the LPQ
We also decided to use unweighed least squares (ULS) estimation as opposed to the
more common maximum likelihood (ML) estimation because item-level data of this
type tends to be skewed and heterogeneous and, therefore, is not multivariate normal.
The use of ML estimation with such non-normal categorical data has been shown to
substantially inflate the chi-square statistic (Babakus, Ferguson, & J oÈ reskog, 1987).
Even with ULS estimation, the chi-square will be inflated. Therefore, overall (or
absolute) model fit was evaluated with the root mean square residual (RMR) and the
goodness of fit index (GFI). Values of 0.05 or less for the RMR and values of 0.90 or
more for the GFI indicate acceptable model fit. Although the RMR tends to be
somewhat inflated with non-normal categorical data (Babakus et al., 1987), the GFI is
robust against non-normality and is independent of sample size (JoÈreskog & SoÈrbom,
1981). But neither the RMR nor the GFI is independent of the number of model
parameters estimated, and the RMR is biased in favour of ULS estimation by
definition. To obviate these problems, we also reported the adjusted goodness of fit
index (AGFI) which does take into account the number of estimated model parameters
and Steiger’s (1990) root mean square error approximation (RMSEA), which is based
on the noncentrality parameter (McDonald & Marsh, 1990). The RMSEA adjusts for
both the sample size and the model degrees of freedom. Values in the range of 0.90 to 1

Psychometric properties of the LPQ
409
for the AGFI and values in the range of 0 to 0.08 for the RMSEA indicate acceptable
model fit.
Third, confirmatory factor analysis of the LPQ subscales was used to fit the five
models shown in Figure 2. These five models were chosen because they were the most
interesting substantive models tested in previous research (Andrews et al., 1994;
Kember & Leung, 1998; Wong et al., 1996). The first four models were two-factor
models while the last was a three-factor model.
Two-factor models
Model 1: A Motive factor (M) and a Strategy factor (S) (Wong et al., 1996).
Model 2: A Surface plus Achieving (S+ A) factor and a Deep (D) factor (Wong et al.,
1
996)
Model 3: A Surface (S) factor and an Achieving plus Deep (A+ D) factor (Wong et al.,
996). Kember and Leung (1998) also tested this model. However, they referred to the
surface factor as a `reproducing’ (RP) factor and the achieving plus deep factor as a
meaning’ (ME) factor.
Model 4: This model is a combination of models 2 and 3. It has an (S+ A) factor and an
A+ D) factor, which means that the LPQ subscales AS and AM crossload on these two
1
`
(
factors (Kember & Leung, 1998).
Three-factor model
Model 5: The three factors are Surface (S: subscales SS, SM), Achieving (A: subscales
AS, AM), and Deep (D: subscales DS, DM) (Andrews et al., 1994; Biggs, 1992; Kember
&
Leung, 1998; Wong et al., 1996). These reflect the three approaches to student
learning as measured by the surface, achieving, and deep scales of the LPQ (Biggs, 1992,
p. 45).
Since subscale scores are composite measures, these five factor models were fitted to
the covariance matrix using the generalised least squares (GLS) procedure in LISREL 8
(
distributions were not normal, and GLS, unlike ML, yields an approximate chi square
JoÈreskog & SoÈrbom, 1993). We used the GLS procedure because the subscale score
2
(
w ) statistic even when the data are not multivariate normal (Browne, 1977). In
2
addition to the conventional w statistic, we also assessed model fit with a number of
other fit indices.
To determine overall or absolute model fit, we used Steiger’s (1990) RMSEA, which
was described earlier. For assessing incremental model fit, a comparison of the fitted
model to a base model, we used the Tucker Lewis index (TLI) (Tucker & Lewis, 1973),
where the base model is the null model (i.e., a model of no relationship). The TLI is also
called the nonnormed fit index (NNFI). Conventionally, values of 0.90 to 1 for the TLI
are indicative of a well-fitted model (Marsh, Balla, & McDonald, 1988; McDonald &
Marsh, 1990), with values in the open interval 0.80 to 0.89 indicating marginal fit. To
adjust for the number of parameters estimated by each of our models, we used the
adjusted goodness-of-fit index (AGFI) described earlier. Finally, since Kember and
Leung (1998) reported the comparative fit index (CFI), we included it here too to
facilitate comparisons with that work. The CFI belongs to the same class of incremental
fit indices as the TLI. Values of the CFI lie between 0 and 1, with values close to 1
indicating better fit (JoÈreskog & SoÈrbom, 1993).

410
John Sachs and Lingbiao Gao
Figure 2. Five subscale level LPQ factor models

Psychometric properties of the LPQ
411
Results
Descriptive measures and reliability
The estimates of internal consistency reliability (coefficient alpha) reported in Table 1
tended to be in line with those reported in previous studies (Andrews et al., 1994;
Kember & Leung, 1998; Wong et al., 1996). Nevertheless, there is considerable
variability between the internal consistency reliability estimates of the LPQ subscales
from one study to the next. Some of this variability no doubt reflects cross-cultural
differences. However, direct comparison of reliability estimates from one study to the
next is generally not advisable since the magnitude of the reliability coefficient is
dependent on the sample variance (Stanley, 1971). A better choice of comparison would
be the standard error of measure (SEM), which tends to be invariant of the sample’s
variance (Nunnally, 1978). For example, when this is done for the LPQ subscale means
of the Andrews et al. (1994) data and the Kember and Leung (1998) data, we see that
the SEMs for the subscale means in both these studies are very similar to those reported
in the current study (see Table 1). Therefore, any observed differences in subscale
reliability reflects greater true score variability (i.e., greater group heterogeneity) in the
study that reports the larger subscale reliability.
Table 1. Reliability and descriptive measures for the LPQ subscales
a
SEM
b
c
d
Subscale
M
SD
Skew Reliability Current
Andrews
Kember
Surface Motive (SM)
Deep Motive (DM)
3.20 .69 7.104
3.54 .64 7.136
Achieving Motive (AM) 3.52 .64 7.379
.53
.50
.51
.41
.62
.61
.47
.45
.45
.47
.41
.43
.49
.40
.43
.39
.44
.41
.46
.43
.44
.46
.39
.40
Surface Strategy (SS)
Deep Strategy (DS)
2.49 .61
3.18 .67
.291
.073
Achieving Strategy (AS) 3.39 .67 7.215
a
Standard error of the measure (SEM
mean subscale score.
x
= SD
x
H17RELx
). All SEMs were computed using the
x
b
SEMs for current study.
SEMs computed from Andrews et al. (1994) data.
SEMs computed from Kember and Leung (1998) data.
c
d
The LPQ subscale means and subscale skew are also reported in Table 1. Inspection
of the means shows that respondents tended to use the centre of the item scale when
responding with a slight upward bias between `3-true of me half the time’ and `4-
frequently true of me’. This was true for all the subscales except SS. Most of the LPQ
subscale distributions were negatively skewed except for subscales SS and DS. Also
subscale SS had the lowest reliability (coefficient alpha = 0.41). Internal consistency
reliability estimates as low as 0.32 have been reported for this LPQ subscale (see Wong,
et al., 1996).

412
John Sachs and Lingbiao Gao
Item-level factoring
Biggs’ six-factor item-level model fits the data reasonably well. The RMSEA and the
RMR were 0.048 and 0.053, respectively. Even though the RMR is a little above 0.05,
this is acceptable because of the tendency of the RMR to become slightly inflated with
non-normal data. Both the GFI and AGFI were 0.93 and 0.92, respectively. These
values are well above the 0.90 value indicative of acceptable model fit. The ULS
coefficients (loadings) of the items on their respective factors are shown in Table 2.
Most items had factor loadings of 0.30 or above on their respective factors. The
remaining items had loadings above 0.20 on their factors, while three items (16, 21, and
25) had factor loadings below 0.20 but above 0.10. The only really problematic item was
item 10, which had a zero loading on its factor. Item 10: `I find that the best way to
learn physics is to memorise by heart’, belongs to the SS subscale, and its corrected
item-total correlation with that subscale was only .08. As was noted earlier, the SS
subscale also had the lowest reliability estimate of all six subscales. On these grounds,
we would be justified in deleting this item. This item, however, has not been shown to
be problematic in previous research reported by Biggs (1992, 1993), although in a factor
analysis of this subscale, Biggs (1993) did report obtaining a second eigenvalue of
almost 1 (i.e., 0.99), suggesting that a second factor might be present. But item level
data of this type are inherently noisy (Nunnally, 1978), often resulting in spurious
factors especially when item distributions are heterogeneous (Bernstein & Teng, 1989).
Furthermore, Biggs (1993) did not identify any SS items associated with this putative
second factor. Therefore, the poor psychometric properties of item 10 could reflect
either a characteristic peculiar to our sample or it could reflect some real cross-cultural
difference. The latter is a distinct possibility since work by Marton, Dall’Alba, and Tse
(
1996) suggests that memorising for many Asians is perceived as being similar to
understanding. But because the reliability coefficient increased by only a small amount
0.02) when this item was deleted, and because deleting this item would make it difficult
(
to compare our results with previous work, we decided to retain it when computing the
SS subscale scores.
Subscale-level factoring
While fitting our five subscale-level factor models, it was found that the error variance
for the SS subscale was always negative. Rather than setting its error variance to zero,
we fixed its value to a computed estimate of its error variance by multiplying its
observed score variance by one minus its reliability. Therefore, the reported results for
all models have one degree of freedom less than would be expected.
Inspection of the results in Table 3 shows that none of the models tested fit
particularly well. If a choice had to be made between these models, then models 4 and 5
are to be preferred since they show the best fit. But since both these models fit equally
well, a choice between them has to be made on purely substantive grounds. Therefore,
model 5 might be preferred since it corresponds to the relationship between the three
approaches to learning (surface, deep, and achieving) and the subscale type (motive or
strategy) outlined by Biggs (1992, p. 45).
Although the model fit results are not particularly good, they do agree with results
summarised in Wong et al. (1996) for a Hong Kong (HK1) and a Beijing (BEIJ) sample.
Also, the fit of model 4 was similar to that reported by Kember and Leung (1998) for

Psychometric properties of the LPQ
413
Table 2. Standardised ULS coefficients (loadings) for Biggs’ six-factor LPQ model on
6 LPQ items
3
LPQ FACTORS
SS
Item No.
SM
DM
0.26
AM
0.48
DS
AS
1
2
3
4
5
6
7
8
9
0.20
0.45
0.00
0.11
0.52
0.32
0.35
0.21
0.57
0.60
0.59
0.44
0.39
0.33
0.58
0.45
0.51
0.48
0.41
0.45
0.56
0.43
0.19
0.48
0.42
0.42
0.46
0.32
0.45
0.58
0.25
0.12
0.44
0.59
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
their model 7. Using the Tucker-Lewis index (TLI) and the adjusted goodness-of-fit
index (AGFI), the fit of models 4 and 5 can best be described as marginal. The fit of
models 4 and 5 could have been improved somewhat by allowing various subscale error
terms to correlate. This practice, however, seems unjustified without strong substantive
reasons to support it.

4
14
Table 3 Summary of fit for LPQ subscale-level factor models
Degrees of Freedom and Fit Indices
John Sachs and Lingbiao Gao
2
a
b
c
d
Model
d.f.
w
RMSEA
TLI
AGFI
CFI
1
2
3
4
5
. M ± S
9
9
9
7
7
210.99
202.61
185.64
115.83
115.83
0.14
0.14
0.14
0.12
0.12
0.77
0.78
0.80
0.84
0.84
0.85
0.85
0.86
0.89
0.89
0.86
0.87
0.88
0.92
0.92
. S+ A ± D
. S ± A+ D
. S+ A ± A+ D
. S ± A ± D
a
b
c
Root mean square error approximation.
Tucker Lewis index or nonnormed fit index.
Adjusted goodness-of-fit index.
d
Comparative fit index.
Discussion
An obvious problem with the LPQ is that its subscales are often reported as having low
reliability, usually less than 0.70 and often even less than 0.50. There are of course
exceptions. Wong et al. (1996) reported LPQ subscale reliability estimates for a Hong
Kong international school (HK-I) sample that were mostly greater than 0.80, and
Andrews et al. (1994) reported four reliability estimates of greater than or equal to 0.70
for their Canadian data. The very low reliability of the SS subscale in our study
(coefficient alpha = 0.41) is particularly noteworthy since both the Andrews et al.
(1994) and the Kember and Leung (1998) studies reported reliability estimates of less
than 0.50 for this subscale, too. Referring to Biggs (1993), Kember and Leung attribute
this low reliability to the fact that this subscale contains two components. As discussed
earlier, however, Biggs’ (1993) evidence for a second factor is not compelling.
The poor reliability of the LPQ subscales no doubt explains the disappointing results
associated with their use as dependent measures in a number of studies. For example,
Lai (1991) introduced cooperative learning in one class of secondary 6 (S6) geography,
while another class of S6 geography was taught in the conventional manner. After eight
weeks she found that the cooperative learning group performed better in geography,
but no differences in LPQ subscale scores were found even though differences in
strategies were reported in interviews with students from both groups. Leung (1992)
applied the small group `learning how to learn’ approach with an S4 economics and
public affairs (EPA) class. A significant increase in EPA test performance was observed
while no significant differences in the LPQ subscale scores were observed. Again,
students did report using different learning strategies during the time the new teaching
approach was used. Gao (1999) categorised physics teachers into three groups in terms
of their conceptions of teaching. It was expected that students being taught by teachers
using a `developmental’ conception of teaching over the year would adopt a more deep
learning approach, and those students who were taught by teachers using a `knowledge
delivery’ conception of teaching would adopt a more surface learning approach. The
third group of teachers adopted a `conflict’ conception of teaching, which was not
expected to impact on their students’ learning approach. However, when LPQ subscale

Psychometric properties of the LPQ
415
scores were used as predictors of students’ learning approach, no significant
relationship was found between teachers’ conceptions of teaching and changes in
students’ approaches to learning even though significant relations between teachers’
conceptions of teaching and students’ learning outcomes were identified.
Watkins (1996), in summarising the results from a number of studies on LPQ
correlates with achievement, reports mean correlations (over 16 studies) of only 70.11,
0
.20, and 0.19 between measures of academic achievement and the surface, deep, and
achieving approachs, respectively. Citing Fraser, Walberger, Welch, and Hattie’s (1987)
extensive synthesis of educational research on academic correlates, Watkins (1996) notes
that correlations of 0.20 are worth pursuing. Fraser et al. cite Rosenthal and Rubin’s
(
1982) use of the binomial effect size to show that a correlation as small as 0.20 is
associated with a 20% improvement in the rate of learning. Looked at in this way,
correlations of 0.20 with any of the LPQ subscales are, therefore, of considerable
substantive importance. But many other correlates of academic achievement reported in
Fraser et al. are much higher than 0.20. For example, the average correlation of student
characteristics was found to be 0.40. This correlation translates into a binomial effect size
of 40% , double that associated with a correlation of 0.20. Surely, one could consider a
student’s approach to learning to be a student characteristic; therefore, expectations of
larger correlations with academic achievement seem reasonable. If the reliability of the
LPQ subscales were improved, higher correlations with standardised measures of
academic achievement would be expected, which in turn would translate into greater
power and larger effect sizes in studies using the LPQ subscales as dependent measures.
Improved reliability of the LPQ subscales would also improve item level factoring
results. As it stands now, the unidimensionality of some LPQ subscales ± SM, DM, and
SS ± is questionable. Using a composite score for such subscales makes little sense. For
example, Biggs’ (1996) principal components analysis of the SM subscale would lead
one to construct two subscales: a two-item subscale (items 7 and 19) and a four-item
subscale (items 1, 13, 21, 31). Yet a two- or even a four-item subscale hardly seems
adequate for any serious enquiry into student approaches to learning. A major problem
with item level factoring, as mentioned earlier, is that items are inherently noisy because
responses are categorised and because item response distributions are often very
heterogeneous, even for items measuring a similar concept. Heterogeneity of response
distributions for items measuring a similar concept will attenuate their correlation. This
was certainly true of our observations of the item distributions within each LPQ
subscale and could account for the low measures of internal consistency reliability
(
alpha) reported for many of the subscales.
Subscale item heterogeneity probably also accounts for the fact that not only the
magnitude but also the sign of the covariances (correlations) between the LPQ subscales
can vary greatly from one study to the next. For example, the DM and SM subscales
show a significant negative relationship in Andrews et al. (1994) and a significant
positive relationship in the current study (see Appendix). One could argue that this
reflects some cultural difference. If this were true, then one would expect to see greater
agreement between the covariance matrix of this study and one based on Hong Kong
data such as Kember and Leung’s (1998) study, but the opposite is the case.
From a psychometric viewpoint, the problems outlined above represent some
shortcomings of the LPQ and would suggest the need for improvement. An obvious

416
John Sachs and Lingbiao Gao
first step would involve the rewriting and or removal of items that have been shown to
be problematic in the past. Another possibility would be the addition of one or two
items to each subscale to improve the overall reliability of the subscales. A slightly
larger item pool would have the added advantage of making it possible to drop a
subscale item if it were found to function poorly within a particular cultural context
without adversely affecting the subscale’s validity.
One could argue, however, the shortcomings of the LPQ discussed here, especially
those not reported in other studies, are more indicative of sampling differences than of
any inherent psychometric deficits. For example, the current sample consists of
mainland secondary school science students. Science students are usually more
quantitatively oriented than arts students. Furthermore, the mainland school system
is different from that in Western countries and even from that in Hong Kong. These
sample specific characteristics need to be kept in mind when making any generalisations
about the psychometric properties of the LPQ, especially when these generalisations
would suggest changes to the inventory’s item content.
Although student approaches to learning (SAL) inventories such as the LPQ
emphasise the context within which learning takes place, this does not mean that a
student’ s generally preferred approach to learning is dependent on a particular subject
that students might happen to be studying. If this were not the case, then inventories of
SAL would be of little utility in identifying preferred approaches to learning. So
differences associated with the subject learned are usually not an issue. The second
sampling issue relates to the characteristics of the mainland school system and what
impact this might have had on the results.
The teaching environment in both the mainland and Hong Kong is characterised by a
formal teaching style that is largely curriculum bound and examination oriented,
especially when compared to the West. The situation in the mainland is perhaps even
more examination oriented and curriculum bound than that in Hong Kong. However,
Biggs and Watkins (1996) argue that even with basic differences in Hong Kong and
mainland China, the `Chinese Learner’ is essentially a product of a `Confucian-heritage’
(CH) culture and, therefore, shares important similarities. At least as far as the LPQ is
concerned, this argument is supported by the Wong et al. (1996) study which found
basic similarities in LPQ confirmatory factor analysis results for Hong Kong and
Beijing samples. More importantly, they found that LPQ confirmatory factor analysis
results were similar for Western, Asian, and African samples.
In conclusion, although evidence for the LPQ’s cross-cultural validity is generally
positive, the low reliability associated with of some of its subscales and their
corresponding low correlations with measures of academic achievement needs to be
addressed. This is especially true if the inventory’s continued use is to prove useful in
experimental and correlational studies on student approaches to learning.
References
Andrews, J., Violato, C., Rabb, K., & Hollingsworth, M. (1994). A validity study of Biggs’ three-
factor model of learning approaches: A confirmatory factor analysis employing a Canadian
sample. British Journal of Educational Psychology, 64, 179±185.
Babakus, C.E., Ferguson, J.R., & JoÈreskog, K.G. (1987). The sensitivity of confirmatory

Psychometric properties of the LPQ
417
maximum likelihood factor analysis to violations of measurement scale and distributional
assumptions. Journal of Marketing Research, 24, 222±228.
Bernstein, I.H., & Teng, G. (1989). Factoring items and factoring scales are different: Spurious
evidence for multidimensionality due to item categorization. Psychological Bulletin, 105 (3),
4
67±477.
Biggs, J.B. (1987). Student approaches to learning and studying. Hawthorn, Vic.: Australian
Council for Educational Research.
Biggs, J.B. (1991). Approaches to learning in secondary and tertiary students in Hong Kong:
Some comparative studies. Educational Research Journal, 6, 27±39.
Biggs, J.B. (1992). Why and how do Hong Kong students learn? Using the learning and study process
questionnaires. Education Paper 14, Faculty of Education, University of Hong Kong.
Biggs, J.B. (1993). What do inventories of students’ learning processes really measure? A
theoretical review and clarification. British Journal of Educational Psychology, 63, 3±19.
Biggs, J.B. (1995). Learning in the classroom. In J.B. Biggs & D.A. Watkins (Eds.), Classroom
learning (pp. 147±166). New York: Prentice Hall.
Biggs, J.B. (1996). Western misperceptions of the Confucian-heritage learning culture. In D.A.
Watkins & J.B. Biggs (Eds.), The Chinese learner: Cultural psychology and contextual influences
(pp. 45±68). Hong Kong: CERC and ACER.
Biggs, J.B., & Watkins, D.A. (1996). The Chinese learner in retrospect. In D.A. Watkins & J.B.
Biggs (Eds.), The Chinese learner: Cultural psychology and contextual influences (pp. 269±285).
Hong Kong: CERC and ACER.
Browne, M.W. (1977). Generalized least-squares estimators in the analysis of covariance
structures. In D.J. Aigner & A.S. Goldberger (Eds.), Latent variables in socio-economic models
(pp. 205±266). Amsterdam: North-Holland.
Cudeck, R. (1989). The analysis of correlation matrices using covariance structure models.
Psychological Bulletin, 105, 317±327.
Entwistle, N.J., & Ramsden, P. (1983). Understanding student learning. London: Croom Helm.
Entwistle, N.J., & Tait, H. (1995). The Revised Approaches to Study Inventory. Edinburgh: Centre
for Research on Learning and Instruction, University of Edinburgh.
Entwistle, N.J., & Waterston, S. (1988). Approaches to studying and levels of processing in
university students. British Journal of Educational Psychology, 58, 258±265.
Entwistle, N.J., Hanley, M., & Hounsell, D. (1979). Identifying distinctive approaches to
studying. Higher Education, 8, 365±380.
Duff, A. (1997). A note on the reliability and validity of a 30-item version of Entwistle & Tait’s
Revised Approaches to Study Inventory. British Journal of Educational Psychology, 67, 529±539.
Fraser, B.J., Walberg, H.J., Welch, W.W., & Hattie, J.A. (1987). Synthesis of educational
productivity research. International Journal of Educational Research, 11, 145±252.
Gao, L. (1999). Conceptions of teaching held by school physics teachers in Guandong China and
their relations to student learning. Unpublished doctoral dissertation, University of Hong Kong,
SAR, China.
Harper, G., & Kember, D. (1989). Interpretation of factor analyses from the approaches to study
inventory. British Journal of Psychology, 59, 66±74.
JoÈreskog, K.G., & SoÈrbom, D. (1981). LISREL versions V and VII user’s guide (1st ed.). Chicago,
IL: International Educational Services.
JoÈreskog, K.G., & S oÈ rbom, D. (1993). LISREL 8: Structural equation modeling with SIMPLIS
command language. Chicago, IL: Scientific Software, Inc.
Kember, D., & Leung, D.Y.P. (1998). The dimensionality of approaches to learning: An
investigation with confirmatory factor analysis on the structure of the SPQ and LPQ. British
Journal of Educational Psychology, 68, 395±407.
Lai, L.Y.E. (1991). Effects of cooperative learning on students’ learning outcomes and approaches to
learning in S3 geography. Unpublished master’s thesis, University of Hong Kong.
Leung, W.Y.W. (1992). Effects of student-student interaction on approaches to learning and on
academic performance. Unpublished master’s thesis, University of Hong Kong.
Marsh, H.W., Balla, J.R., & McDonald, R.P. (1988). Goodness-of-fit indexes in confirmatory
factor analysis: The effect of sample size. Psychological Bulletin, 103 (3), 391±410.

418
John Sachs and Lingbiao Gao
Marton, F., & SaÈljoÈ (1976). On quantitative differences in learning ± I: Outcome and process.
British Journal of Educational Psychology, 46, 4±11.
Marton, F., Dall’ Alba, G., & Tse, L.K. (1996). Memorizing and understanding: The keys to the
paradox? In D.A. Watkins & J.B. Biggs (Eds.), The Chinese learner: Cultural psychology and
contextual influences (pp. 69±84). Hong Kong: CERC and ACER.
McDonald, R.P., & Marsh, H.W. (1990). Choosing a multivariate model: Noncentrality and
goodness of fit. Psychological Bulletin, 107 (2), 247±255.
Nunnally, J.C. (1978). Pychometric theory (2nd ed.). New York: McGraw-Hill.
Ramsden, P., & Entwistle, N.J. (1981). Effects of academic departments on students’ approaches
to studying. British Journal of Educational Psychology, 51, 368±383.
Rosenthal, R., & Rubin, D.B. (1982). A simple general purpose display of magnitude of
experimental effect. Journal of Educational Psychology, 74, 166±169.
Schmeck, R.R., & Grove, E. (1979). Academic achievement and individual differences in learning
processes. Applied Psychological Measurement, 3, 43±49.
Schmeck, R.R., Geisler-Brenstein, E., & Cercy, S. (1991). The revised inventory of learning
processes. Educational Psychology, 11, 343±362.
Schmeck, R.R., Ribich, F.D., & Ramanaiah, N. (1977). Development of a self-report inventory
for assessing individual differences in learning processes. Applied Psychological Measurement,
1
, 413±431.
Stanley, J.C. (1971). Reliability. In R.L. Thorndike (Ed.), Educational measurement (2nd ed.) (pp.
56±442). Washington, DC: American Council on Education.
3
Steiger, J.H. (1990). Structural model evaluation and modification: An interval estimation
approach. Multivariate Behavioral Research, 25, 173±180.
Tucker, L.R., & Lewis, C. (1973). The reliability coefficient for maximum likelihood factor
analysis. Psychometrika, 38, 1±10.
Watkins, D.A. (1996). Learning theories and approaches to research: A cross-cultural
perspective. In D.A. Watkins & J.B. Biggs (Eds.), The Chinese learner: Cultural psychology
and contextual influences (pp. 3±24). Hong Kong: CERC and ACER.
Watkins, D.A., & Akande, A. (1994). Approaches to learning of Nigerian secondary school
children: Emic and etic perspectives, International Journal of Psychology, 29 (2), 165±182.
Watkins, D.A., & Mboya, M. (1997). Assessing the learning processes in Black South African
students. Journal of Psychology, 131 (6), 632±640.
Watkins, D.A., Akande, A., & Mpofu, E. (1994). The assessment of learning processes: Some
African data. IFE Psychologia, 2 (2), 1±18.
Watkins, D.A., Dong, Q., & Xia, Y. (1995). Towards the validation of a Chinese version of the
self-description questionnaire. Psychologia, 38, 22±30.
Wong, N.Y., Lin, W.Y., & Watkins, D.A. (1996). Cross-cultural validation of models of
approaches to learning: An application of confirmatory factor analysis. Educational
Psychology, 16, 317±327.
Received 26 April 1999; revised version received 12 January 2000
Appendix
Covariance matrix for LPQ subscale total scores (N = 1070)
SS
SM
DS
DM
AS
AM
SS
13.245
4.252
71.971
72.111
71.352
0.025
SM
DS
DM
AS
17.359
70.298
1.545
3.204
5.277
16.112
7.072
5.473
4.709
14.570
4.557
6.352
16.194
5.416
AM
14.975