L.E. Kost et al. Phys. Rev. ST - Phys. Educ. Res. v.5, p.010101 (2009).
Abstract: Previous research [S. J. Pollock et al., Phys. Rev. ST Phys. Educ. Res. 3, 1 (2007)] showed that despite the use of interactive engagement techniques, the gap in performance between males and females on a conceptual learning survey persisted from pretest to post-test at the University of Colorado at Boulder. Such findings were counter to previously published work [M. Lorenzo et al., Am. J. Phys. 74, 118 (2006)]. This study begins by identifying a variety of other gender differences. There is a small but significant difference in the course grades of males and females. Males and females have significantly different prior understandings of physics and mathematics. Females are less likely to take high school physics than males, although they are equally likely to take high school calculus. Males and females also differ in their incoming attitudes and beliefs about physics. This collection of background factors is analyzed to determine the extent to which each factor correlates with performance on a conceptual post-test and with gender. Binned by quintiles, we observe that males and females with similar pretest scores do not have significantly different post-test scores (p>0.2) . The post-test data are then modeled using two regression models (multiple regression and logistic regression) to estimate the gender gap in post-test scores after controlling for these important prior factors. These prior factors account for about 70% of the observed gender gap. The results indicate that the gender gap exists in interactive physics classes at our institution but is largely associated with differences in previous physics and math knowledge and incoming attitudes and beliefs.
It seems that if both males and females student have the same pre-test scores, they both will also have the same post-test scores, meaning there's no gender difference in the way they learn. It appears that the background of the students have a significant role in how they perform, rather than gender.
Both the multiple-regression and the multiple regression models confirm this interpretation, showing that a majority of the gender gap can be accounted for by factors other than gender explicitly. From the multiple regression analysis we find that only 3 points of the 11 point gender gap cannot be accounted for by background factors. From the logistic regression analysis we find that the odds of a male and a female scoring above 60% on the post-test are not statistically different once background factors are accounted for. Taken together, the results of these models suggest that the persistence of the gender gap is due in large part to differences in males’ and females’ preparation and background coming into the introductory course and not explicitly due to their gender.
Interesting reading if you have the time.