A great deal of research has focused on developing and improving prediction models of academic achievement (Schneider & Preckel, 2017), which are used in various practical applications, depending on the educational context in which they are used. In open access systems, there are no limitations based on previous achievements (e.g., no entry exams) (OECD, 2020). Here, modelling future academic achievement is crucial for the development of study orientation tools (Fonteyne et al., 2017). In contrast, closed access systems have stricter admission procedures, such as high-stakes testing, which is common in Anglo-Saxon education (OECD, 2020). In this context, predictive models of academic achievement are used for selection procedures in higher education (Kuncel et al., 2001).

Various cognitive, affective, and demographic factors are typically included in predictive models of academic achievement (e.g., cognitive ability, academic self-efficacy, test anxiety, and many more) (Schelfhout et al., 2022). Research has reported gender differences in some of these predictors. For example, female students score higher on test anxiety measures than male students do (Núñez-Peña et al., 2016). Furthermore, interaction effects have been found between gender and separate predictors of academic achievement (Steinmayr & Spinath, 2008). For instance, research has shown that academic self-efficacy is a stronger predictor of academic achievement for male students (Huang, 2013; Komarraju & Nadler, 2013). Unfortunately, the majority of research has studied these interactions in isolation, and no studies yet have researched whether gender also affects the predictive value of such factors in the context of prediction models of academic achievement. This is valuable, as prediction models allow us to account for the complex interrelationships among a system of variables (Ruban & McCoach, 2005). As such, the present study first explored how gender interacts with predictors of academic achievement in these prediction models.

As mentioned earlier, predictive models of academic achievement could have far-reaching outcomes as they are used in various high-stakes applications. If gender actually affects the composition of these prediction models or their prediction accuracy, the current models fail to take this into account, which could result in students’ projected academic achievement being over- or underestimated. An example of this is the ‘Female Underprediction Effect’, which entails that women achieve higher college grades than predictions based on their SAT-scores would suggest (Hyde & Kling, 2001). Such effects could be due to real gender differences in study motivation (Freudenthaler et al., 2008), but also to a previously undetected effect of gender on the predictive models of academic achievement. Because of this, the present study explored whether predictive models of academic achievement exhibit gender differences in the accuracy of their predictions, and how prediction accuracy relates to the gender balance in a study program. Furthermore, we examined if different factors are predictive of academic achievement for boys and girls within the same study program. Such differences are currently undetected, thus possibly resulting in prediction accuracy differences. Finally, if it appears that gender does indeed affect the composition of the factors included in such prediction models, it is crucial to assess whether gender-specific prediction models of academic achievement would allow for more accurate predictions. Therefore, we assessed whether gender-specific prediction models of academic achievement make more accurate predictions than models built across genders do.

The present study was performed in the framework of the longitudinal SIMON project of Ghent University (ARWU top 100 of the Shanghai ranking of worldwide universities). This project exists within an open access study environment: all students with a degree from secondary school can start any program in higher education, except medicine, dentistry, and performance arts (OECD, 2020). The SIMON project was developed to help timely degree attainment in Flanders, by providing students with non-binding study (re)orientation advice after they complete an internet-based self-assessment tool. Participation is strongly encouraged among first year university students, thus resulting in high response rates (Fonteyne et al., 2017). The present data set includes SIMON data, collected within the SIMON-project over the time period 2016-2018. As such, we had access to a huge database that contains an extensive amount of information on (predictors of) academic achievement. To develop our prediction models of academic achievement, we considered 3 categories of predictors: cognitive (e.g., score on a mathematics test), affective (e.g., academic motivation), and demographic (e.g., socio-economic status). Grade point average (GPA) was used as a measure of academic achievement. Data from 5,016 first year bachelor students across 16 programs (both STEM and non-STEM) was considered. We developed program-specific prediction models of academic achievement, both across genders and for male and female students separately, using the Akaike’s Information Criterion stepwise selection procedure. This procedure ensures that the best possible prediction model is chosen, by minimizing the model’s prediction errors and consequently, the information loss (Burnham & Anderson, 2002). To assess differences in prediction accuracy between the various prediction models, we used independent samples t-tests to compare the mean prediction errors. Finally, to calculate the correspondence between the composition of the male, female, and general program-specific models of academic achievement, we divided the total amount of shared predictors between the considered models, by the total amount of predictors present across both models.

We found that gender often interacts with cognitive and affective predictors of academic achievement, but interactions with demographic predictors were rather rare. We compared the prediction accuracy of the general model for male data with that for female data. Significant differences were found in 18.75% of the programs, with effect sizes ranging from small to medium. In other words, predictive models of academic achievement do not make equally accurate predictions for male and female students in some study programs. We propose that this accuracy difference could be related to two mechanisms. First, the correlation between these effect sizes (expressing the difference in prediction accuracy for male and female data) and study programs’ gender balances revealed that the more one gender is present in a program, the less accurate the predictions of the model are. Second, we developed gender- and program-specific prediction models of academic achievement and discovered that the correspondence between the predictors present in the composition of the male and female program-specific models of academic achievement is low (M=0.22, SD=0.12). Similar results were found when comparing the composition of general and gender-specific models. The correlation between the correspondence measures and the gender balance measures suggests that a stronger gender imbalance leads to more diverse gender-specific prediction models. However, we also found low correspondence measures in study programs with balanced gender proportions. Indeed, it appears that the correspondence between the composition of general, male, and female models of academic achievement is remarkably low, irrespective of the gender balance of a study program. Finally, we compared the prediction accuracy and explained variances of the gender-specific and general models (on gender-specific data). We found that using gender-specific models improved our predictions, as such models generally lowered prediction errors and increased explained variances.

Burnham, K. P., & Anderson, D. R. (2002). Model Selection and Multimodel Inference: A practical information-theoretic approach (2nd ed.). Springer-Verlag. Fonteyne, L., Duyck, W., & de Fruyt, F. (2017). Program-specific prediction of academic achievement on the basis of cognitive and non-cognitive factors. Learning and Individual Differences, 56, 34–48. https://doi.org/10.1016/j.lindif.2017.05.003 Freudenthaler, H. H., Spinath, B., & Neubauer, A. C. (2008). Predicting school achievement in boys and girls. European Journal of Personality, 22(3), 231–245. https://doi.org/10.1002/per.678 Huang, C. (2013). Gender differences in academic self-efficacy: A meta-analysis. European Journal of Psychology of Education, 28(1), 1–35. https://doi.org/10.1007/s10212-011-0097-y Hyde, J. S., & Kling, K. C. (2001). Women, motivation and achievement. Psychology of Women Quarterly, 25(4), 364–378. https://doi.org/10.1111/1471-6402.00035 Komarraju, M., & Nadler, D. (2013). Self-efficacy and academic achievement: Why do implicit beliefs, goals, and effort regulation matter? Learning and Individual Differences, 25, 67–72. https://doi.org/10.1016/j.lindif.2013.01.005 Kuncel, N. R., Ones, D. S., & Hezlett, S. A. (2001). A comprehensive meta-analysis of the predictive validity of the graduate record examinations: Implications for graduate student selection and performance. Psychological Bulletin, 127(1), 162–181. https://doi.org/10.1037/0033-2909.127.1.162 Núñez-Peña, M. I., Suárez-Pellicioni, M., & Bono, R. (2016). Gender Differences in Test Anxiety and Their Impact on Higher Education Students’ Academic Achievement. Procedia - Social and Behavioral Sciences, 228, 154–160. https://doi.org/10.1016/j.sbspro.2016.07.023 OECD. (2020). Education at a Glance 2020: OECD Indicators. OECD Publishing, Paris. https://doi.org/10.1787/69096873-en Ruban, L. M., & McCoach, D. B. (2005). Gender differences in explaining grades using structural equation modeling. Review of Higher Education, 28(4), 475–502. https://doi.org/10.1353/rhe.2005.0049 Schelfhout, S., Wille, B., Fonteyne, L., Roels, E., Derous, E., de Fruyt, F., & Duyck, W. (2022). How accurately do program-specific basic skills predict study success in open access higher education? International Journal of Educational Research, 111. https://doi.org/10.1016/j.ijer.2021.101907 Schneider, M., & Preckel, F. (2017). Variables associated with achievement in higher education: A systematic review of meta-analyses. Psychological Bulletin, 143(6), 565–600. https://doi.org/10.1037/bul0000098 Steinmayr, R., & Spinath, B. (2008). Sex differences in school achievement: What are the roles of personality and achievement motivation? European Journal of Personality, 22(3), 185–209. https://doi.org/10.1002/per.676