Beyond Mars and Venus: An Examination of Gender Differences in Academic Performance of High School Students - Innovative Management in Information and Production - page 35

4 Discussion

This study examines the structural equivalence of a statewide test for males and
females. The sample used for this study was randomly selected from all the
examinees of a statewide test in 2004. Several analyses were performed which
included the selection of best-fitting models as well as the examination of structural
invariance for males and females.
The results of this study show that the two-factor model with GPA as an
exogenous variable provides the best fit. It is because the scale for GPA is quite
different from the scale score used in the statewide tests. Although treating GPA as
a third variable might also be a possible solution, the fit indices suggest that treating
GPA as an exogenous variable is the best way to describe the data.
Based on CFA tests of invariance of the factor structure across groups, the
factor patterns were found to be not invariant across the gender groups. Because
of limitations of the chi-square tests of model fit with large sample sizes, other
descriptive model-fit indices were also utilized. Those descriptive indices suggested
that the covariance for the two groups generally holds, but not for the factor patterns.
Since the lack of invariance in the factor structures across groups is sufficiently
large, it is reasonable to conclude that the factor structure of the math and English
ability is not invariant across the gender groups.
There are several explanations that could explain the difference in factor
structures of math and English ability for males and females. First, male and
female students typically have different experiences that may be relevant to
the development of math and English skills. Male students typically receive more
and higher levels of mathematics instruction than females. And females are more
often encouraged to take courses related to literature or reading ( Femmema and
Sherman 1977 ). In addition, the social and cultural assumption and stereotypes
about the difference in male and female abilities could also be the cause of the
differences in their interests and performance. The other possible reason could
be the incorporation of GPA as an indicator that the scale for GPA being much
smaller than the scale for other subtests could be problematic. In addition, the
unique variances of GPA are quite large for both groups. This implies that the
error variance of GPA could be very large. There are two limitations in this study.
First, this study only considered six variables, and each factor only has two or
three indicators. Given the large sample size, more variables should be incorporated
to lessen the chances of the improper estimation of parameters ( Anderson and
Gerbing 1984 ). It would be desirable to add more academic subject areas and
examine the equivalence of factor structure across genders further. Second, the
methods used to estimate the free and constrained parameters in the model are
only based on the maximum likelihood estimation (MLE) method. Other estimation
methods, such as generalized least-square, and unweighted least square methods
should also be examined as a comparison to test the consistency of the results
estimated by MLE.