Scarce evidence exists on gender disparities as a result of school time interventions or reforms. This is surprising for different reasons. First, gender equality is a central goal of all countries committed to human rights (United Nations General Assembly, 1948). Secondly, gender equality can contribute to economic growth (Altuzarra et al., 2021; Santos Silva and Klasen, 2021), particularly through increased participation in STEM jobs (Maceira, 2017; Hammond et al., 2020), which critically depend on achievement, self-concept, and course choices of STEM subjects in school (Updegraff et al., 1996; Parker et al., 2012; Watt et al., 2012, 2017; Schoon and Eccles, 2014. Referring to these arguments which underscore the relevance of monitoring effects of educational initiatives and reforms on gender disparities in general, it seems reasonable to believe that the G8-reform might specifically affect gender disparities in STEM. As girls and boys report different levels of self-concept and interest in math-intensive domains of STEM, which are central for subsequent achievement (e.g., Else-Quest et al., 2010; Hübner et al., 2017b,2019; Eccles and Wigfield, 2020; Wu et al., 2021), it is important to investigate if the reform-induced intensifications/compression in lower secondary school might affect gender disparities in STEM-related achievement and motivation.

Several studies found differential effects of instructional time reforms for high- and low-performing students. For instance, Nomi and Allensworth (2009) investigated the effect of the “Double-Dose” algebra reform in Chicago, which required Grade 9 students with test scores below the national median to participate in additional algebra courses. The authors found a stronger positive effect for students close to the median, compared with students who performed much lower. In the same vein, Huebener et al. (2017) found small and sometimes non-significant changes in mathematics and science achievement for lower deciles of the performance distribution in the course of the G8-reform in Germany, whereas effects were larger for higher deciles. To our knowledge, that study is the only one in which the potential effects of the G8-reform on gender disparities were examined in science, reading and mathematics achievement for Grade 9 students. Interestingly, the findings suggested no statistically significant differential effects on girls and boys in Grade 9. The timing of assessment is important to consider when interpreting results of different G8-studies, because G8 students in Grade 9 have had substantially more instructional time compared with G9 students in Grade 9. However, by the end of upper secondary school both cohorts have received a more comparable amount of instructional time.

In another study, Lavy (2015) reported that the treatment effect of increased school time was larger in higher performing countries, using PISA data. The author accounted for systematic differences between different countries by applying a country fixed-effects approach. These results provide tentative evidence of effect heterogeneity as a result of school time reforms, depending on students’ level of achievement.

Many school time studies and reforms focused on changes in STEM achievement of high and low performers, while gender disparities, for instance on motivational outcomes or wellbeing, have been rarely investigated. This constitutes an important limitation of many prior studies because gender disparities in STEM are well documented: The OECD (2019) reported a mathematics advantage for boys in 32 economies/countries (of 78; 14 economies/countries reported advantages for girls) and a science advantage for girls in 34 countries (of 78; 9 economies/countries reported advantages for boys). Notably, the differences were small on average (d = 0.05; ranging from d = 0.22 in Colombia to a non-significant difference of d = 0.01 in the Netherlands), and recent research suggests closings of these gaps, for instance in science achievement (e.g., Meinck and Brese, 2020). There are also meta-analyses that essentially found very small gender differences in math achievement but substantial variability across countries (e.g., Else-Quest et al., 2010). However, robust and systematic gender differences favor boys for math self-concept and interest in adolescence (e.g., Watt, 2004; Else-Quest et al., 2010; Frenzel et al., 2010; Nagy et al., 2010; Hübner et al., 2017b,2019; Widlund et al., 2018; Parker et al., 2020; Mejía-Rodríguez et al., 2021; Wu et al., 2021).

Probably most important in the context of this study, prior research using rich data from the end of German upper secondary school has provided evidence for substantial differences between boys and girls on a broad variety of mathematically intensive STEM outcomes, even after controlling for cognitive abilities. For instance, Hübner et al. (2019) found girls to have statistically significantly lower achievement in mathematics (d ≥ 0.45, p < 0.05) and physics (d ≥ 0.63, p < 0.05), compared to boys, whereas no such gender differences were found in biology. In addition, differences in mathematics in advantage of boys seem to be pronounced in Germany already by Grade 4 in elementary school (d = 0.18; Stanat et al., 2017).

Women and men differ substantially in regard to their mathematical and mathematics-intensive STEM educational pathways and career aspirations (Watt et al., 2012, 2017; Schoon and Eccles, 2014; Lazarides and Lauermann, 2019; Makarova et al., 2019; Lazarides et al., 2020). This process has been referred to as the leaky STEM pipeline (Jacobs and Simpkins, 2005). Prior research has found that central to the choice of advanced course enrollments are students’ subject-specific achievement (Updegraff et al., 1996; Parker et al., 2012) and self-concept and values (Watt et al., 2012), even after controlling for prior achievement levels in the domain (Watt et al., 2017). These motivational variables have been linked not only to school enrollment but further to aspired educational and occupational pathways in mathematics and STEM subfields (Watt et al., 2012, 2017). Choosing advanced courses in high school constitutes a key factor for subsequent enrollment in STEM subjects at university (Ma and Johnson, 2008; Eccles and Wigfield, 2020; Lazarides et al., 2020). Thus, if a reform has differential effects on girls and boys (e.g., increases or decreases to their motivation), it is likely to affect subsequent decisions for or against related courses in high school or later on at university (e.g., Hübner et al., 2017b; Biewen and Schwerter, 2021).

This line of argumentation can be extended and implications can be derived more theoretically: Expectancy-value theory (Eccles, 1983; Eccles and Wigfield, 2002, 2020) outlines that key elements for choices are students’ expectations of success and task values and that both are influenced by prior achievement. Empirical evidence for this assumption can be found, for instance, in literature on the reciprocal effects model between self-concept and achievement (Marsh and Craven, 2006; Seaton et al., 2015). Self-concept is defined as students’ perceptions about their abilities, which develops via engagement with others (Shavelson et al., 1976; Marsh, 1990; Marsh et al., 2016). Task values, the other important set of variables to explain choices, consist of four components: intrinsic, attainment, utility, and cost values. Intrinsic value refers to students’ enjoyment when performing a specific task, attainment value refers to the personal importance a student attaches to a task, and utility value refers to its usefulness; researchers have combined attainment and utility values and referred to “importance value.” Costs, on the other hand, refer to the perceived negative consequences of task engagement, for example, effort or psychological and social costs (Watt et al., 2019).

Regarding subject-specific self-concept and interest, prior research suggests differences between girls and boys, which typically follow stereotypic patterns: Boys tend to report higher self-concept and interest in math-intensive STEM subjects compared to girls, whereas these effects are typically zero or in favor of girls in subjects such as biology (e.g., Denissen et al., 2007; Hübner et al., 2017b,2019; Watt et al., 2017, 2019; Parker et al., 2020; Mejía-Rodríguez et al., 2021). Therefore, if school time reforms force girls to learn similar content in less amount of overall time in subjects they are less interested in and in which they have lower perceptions of their own abilities (e.g., girls in math-intensive STEM subjects), this might even reinforce such less positive perceptions (e.g., Hübner et al., 2017b,2019). In addition, if the reforms differentially affect boys’ and girls’ STEM achievement this might also foster further disparities, for instance regarding students’ self-concept, as these variables are reciprocally related (e.g., Marsh and Craven, 2006; Arens et al., 2017; Wu et al., 2021).

Other variables that are important to consider in the context of an intensified learning environment include students’ perceived stress and health. These variables might be particularly relevant in the context of increasing instructional time because it is intended that students spend more time with learning in school, which might lead to reduced or even too little leisure time to recover (Milde-Busch et al., 2010; Hübner et al., 2017a). As outlined in prior research, mental health is also associated with student achievement (e.g., Tuominen-Soini and Salmela-Aro, 2014; Fiorilli et al., 2017). For instance, Agnafors et al. (2021) found that students with mental health problems in very early years more often performed below grade level later on. Another study by Fiorilli et al. (2017) suggests that students’ burnout is highly relevant for student achievement, both directly and indirectly. The importance of considering wellbeing as a foundation for students’ aspirations was underscored in a study by Widlund et al. (2018) of Finnish students. Depending on the age group, the authors were able to identify either three (Grade 7) or four (Grade 9) latent profiles based on students’ attainment and self-concept in mathematics, their engagement, and three burnout subscales. They found that students with negative academic wellbeing had statistically significantly lower aspirations compared to thriving students. Interestingly, they found that girls were overrepresented in the negative academic wellbeing profile, which is in line with prior findings on gender disparities in school burnout (Salmela-Aro and Tynkkynen, 2012).

Beyond these studies, further research has produced evidence suggesting that girls generally do develop higher stress levels, compared with boys (e.g., Hampel and Petermann, 2006; Moksnes et al., 2010). Studies that have focused on investigating school stressors found schoolwork pressure to partly explain psychological complaints and psychosomatic pain (Hjern et al., 2008), and girls reported higher levels of performance-related stress at school (Moksnes et al., 2010). Finally, prior research provides evidence that increasing learning time might lead to more stress-related health problems (e.g., Marcus et al., 2020). Related to this, Quis (2018) investigated gender-specific differences between G8- and G9-students on school-related stress and health among students at the end of upper secondary school. She found considerable differences in school-related stress and mental health before and after the reform, mainly driven by girls (health) or boys and girls (stress). However, uncertainty exists whether such effects result from increases on a majority of stress facets (e.g., feelings of exhaustion, achievement-related overburdening, or not being able to recover in leisure time), or particularly on specific facets and not others.

We used data from the Additional Study Baden-Württemberg (Blossfeld et al., 2011) from the National Educational Panel Study (NEPS; Scientific Use File 3.2.0). The dataset contains representative data for Baden-Württemberg, assessed from four different cohorts in the final semester of upper secondary school. Two cohorts completed German lower secondary school before the reform and two completed it after the reform. We compared outcomes of these cohorts assessed at the end of upper secondary school (G9: Grade 13 or G8: Grade 12). This design is typically referred to as a cohort control design (Shadish et al., 2002). Overall, students from 44 high track upper secondary schools participated in the study: Cohort 1 (before the reform): n = 1,226 (55% girls); Cohort 2 (before the reform): n = 1,179 (55% girls); Cohort 3 (after the reform): n = 1,205 (56% girls); Cohort 4 (after the reform): n = 1,208 (55% girls). Before the reform, students graduated after 9 years of high track upper secondary school, whereas after the reform students graduated after 8 years. The first cohort of students graduated in 2011, the second (Grade 13) and third (Grade 12) in 2012, and the fourth in 2013. Notably, in Baden-Württemberg, Grade 11 was abolished to implement the G8-reform (Kühn et al., 2013). Data were collected in the final semester of the last year of upper secondary school. Students in Germany are required to spend at least 265 h per week each year in school. This means that G9 students are required to spend on average 265/9 = 29.44 h per week each year in school, whereas G8 students are required to spend 265/8 = 33.13 years per week each year in school, reflecting a difference of 3.69 additional hours that students in G8 are required to spend per week in school. Overall, cumulated mandatory hours were 11 h higher for G8 students from grade 5 to grade 6, and 16 h higher for G8 students from grade 7 to grade 10 in Baden-Württemberg (Homuth, 2017).

In all cohorts, identical instruments were administered to assess subject-specific standardized achievement, self-concept, and interest in the subjects mathematics, biology and physics, as well as to assess school-related stress and health. The questionnaire is available in the NEPS data center¹.

The main analysis proceeded in two steps. First, we estimated multiple-group models for the eight different groups (4 cohorts × gender) in Mplus 8.6 (Muthén and Muthén, 1998–2017). We did this separately for standardized test achievement, self-concept, interest, school-related stress, and health. For achievement, we used multidimensional (multiple-group) item response theory (IRT) models (see Jöreskog and Goldberger, 1975; Hübner et al., 2020). For the remaining constructs traditional structural equation models (SEMs) were applied. Prior work offered clear guidance on how to define measurement models for the achievement measures, self-concept, and interest (e.g., Marsh, 1992; Eccles and Wigfield, 2002; Duchhardt, 2015; Hübner et al., 2016a); this was not the case for the instruments used to assess students’ stress and health which were typically analyzed as a single aggregate score and not with multidimensional models (e.g., Bergmüller, 2003; Hübner et al., 2017a; Quis, 2018). In this study, we utilized a data-driven procedure to explore the underlying factor structure of stress and health items using exploratory structural equation models (ESEMs). As outlined by Marsh et al. (2014), ESEMs combine useful features of exploratory and confirmatory factor analysis (EFA/CFA) such as confirmatory tests of factor structures and associations between different latent factors, and they allow small cross-loadings. For school-related stress and health we performed ESEMs with geomin rotated factor loadings in a multiple group framework. To identify the most adequate solution, we first specified different (single-group) ESEM models with an increasing number of latent factors, before running ESEMs in a multiple group framework with the eight groups (gender × cohort). Models were constrained to test strong factorial/scalar measurement invariance, which is required to meaningfully compare latent means across groups. To judge model fit, we considered the Comparative Fit index (CFI), the Tucker-Lewis index (TLI), the Root Mean Square Error of Approximation (RMSEA), and the Standardized Root Mean Square Residual (SRMR). Based on prior research (MacCallum et al., 1996; Hu and Bentler, 1999; Yu, 2002), we considered the following cutoffs to indicate good model fit: CFI and TLI ≥ 0.95, SRMR and RMSEA ≤ 0.05.

Using these models, we compared the means or—for models with covariates—intercepts of the latent outcomes between the resulting groups using the delta method (Oehlert, 1992) by applying the MODEL CONSTRAINT option in Mplus. Statistically significant differences between the specific group differences constitute interaction effects. We estimated (a) gender differences in G9 cohorts and (b) gender differences in G8 cohorts, and one interaction effect: (c) the difference between a and b (reform × gender). We specified unadjusted models without covariates and adjusted models including covariates (e.g., cognitive abilities, socioeconomic background; see “Instrument” section) to check the robustness of our results. To better interpret our findings, results were transformed into a metric with an overall M = 500 and SD = 100 for achievement and to a metric with an overall M = 50 and SD = 10 for the remaining constructs, using the pooled variance of the latent variables from the unadjusted models. For consistency, we report two-sided p-values throughout, although prior studies suggest a directional hypothesis for stress, health, and math-intensive STEM self-concept in disadvantage of girls. We therefore interpret one-sided p-values to judge statistical significance for those constructs (one-sided p-value = two-sided p-value/2). For all other outcomes, no consistent directional hypothesis could be derived from the literature. All models were specified using full information maximum likelihood estimation (FIML; Enders, 2010), robust standard errors (McNeish et al., 2017), and survey weights.

First, we inspected descriptive statistics. As shown in Table 1, overall, differences between the two cohorts were small. Only with regard to perceptual speed, students in G9 scored slightly higher. Further, students in G8 repeated classes slightly less often than students in G9. This resulted from a generally low repetition rate due to a specific feature of the reform implementation: If students from the last G9 cohort were required to repeat a grade, they had to move from, for instance, the end of grade 10 to the beginning of grade 9, because the respective grade 10 cohort in G8 would have already been ahead of the grade 10 in G9, which the student should repeat (due to the additional hours per week in lower secondary school). These differences were controlled for in the adjusted models as outlined below.

Next, we inspected gender-specific differences. As visible from Table 2, we found substantial differences between girls and boys, both before and after the reform.

As outlined above, we found large disparities between girls and boys at the end of upper secondary school on STEM-related outcomes. In most cases, these disparities followed stereotypical patterns: Overall, girls performed less well on standardized tests in math-intensive STEM subjects. In addition, girls reported lower self-concept and interest than boys in mathematics and physics, whereas there were no significant gender-related disparities in biology. When integrating our findings into the theoretical model (see Figure 1), we can summarize that in most cases we found evidence for the perpetuation model. Disparities before the reform on the respective outcomes were pronounced, and these differences did not change much after the reform. Our findings extend prior findings in three regards: They are based on a later period in the education system (end of upper secondary school, right before the transition to university), a broadened set of outcomes, and a more fine-grained investigation of school-related stress and health.

As we outlined in the theoretical background, several prior studies had suggested treatment effect heterogeneity for high and low achievers (e.g., Nomi and Allensworth, 2009; Lavy, 2015; Huebener et al., 2017), which is why we expected we would find a pattern of results in line with the accumulated advantages model (Figure 1) for STEM outcomes. However, aside from few stress and health facets, we did not find any changes when comparing gender disparities before and after the reform. This might have had different causes—for instance, students in our sample were older at the end of secondary school, compared with students in the reviewed studies. Therefore, our sample might constitute a positive selection of higher performing students as some lower performing students might have dropped out before or in early upper secondary school or might have switched to vocational upper secondary schools, where this reform was not implemented. This might have led to smaller gender differences in upper secondary school than before, in lower secondary school. Further, the major changes of the G8-reform happened in lower secondary school, whereas upper secondary school remained largely unaffected. Therefore, potential interaction effects on STEM outcomes might already have “washed out” by the end of upper secondary school. Most importantly, when comparing differences between G8 and G9 students’ average weekly hours spent in STEM courses, we found negligible differences. This means, that changes in subject-specific instructional time might have been a too small and a central factor for why we did not find any differences on STEM related outcomes. However, this would not explain previously found reform-specific differences between G8 and G9 students for instance in Biology (Hübner et al., 2017a).

In contrast to perpetuating disparities on STEM outcomes after the reform, our study revealed some statistically significant interaction effects on school-related stress and health. Importantly, both girls and boys tended to report more school-related stress and health issues after the reform. However, we did not find interaction effects on all stress and health dimensions, but only on those more related to school, namely the Overload dimension of school-related stress, and the Overburdening and Achievement-related fear dimensions of health. Compared to the perpetuating subject-specific results outlined above, these findings are slightly more in line with the proposed accumulated (dis)advantages model: On average, all students (girls and boys) tended to report higher stress/poorer health after the reform, but particularly those students who were more stressed/had lower health scores before the reform seemed to experience higher school-related stress and poorer health afterward, at least on stress and health facets more closely related to school. These results are in line with prior findings that girls report lower wellbeing scores than boys (e.g., Moksnes et al., 2010; Salmela-Aro and Tynkkynen, 2012; Tuominen-Soini and Salmela-Aro, 2014) and reflect findings from prior studies that students might perceive the remaining leisure time to be too limited to recover from school-related stress (Milde-Busch et al., 2010). The higher average workload per week in lower secondary school as a result of the G8-reform might have been one driver of the unevenly higher stress for girls after the reform. Other potentially relevant stressors than the higher workload could have included longer school days, the abolishment of Grade 11, or completing the same curriculum in a shorter amount of time. However, we cannot trace back which stressors might have ultimately fostered these results, as all of these potential causes are perfectly confounded with the reform (i.e., all changes happened simultaneously), we cannot disentangle their effects.

There are several limitations that are important to consider when interpreting the results of this study. These limitations include potential threats to internal and external validity. Regarding internal validity, it is important to consider that we used data from a cohort control design, whereby two representative cohorts of students from before the reform were compared with two representative cohorts of students after the reform. Although this cohort control design has been discussed as providing a good foundation for the investigation of intervention effects, as it resembles a natural experiment setting (Shadish et al., 2002), it might be possible that the cohorts already differed independent of the reform (e.g., due to historical events). In other words, we did not have a control group who did not receive the treatment at the same time that the students in the treatment group received the treatment (a difference in difference design; e.g., Cunningham, 2021). This of course provides a challenge for all research using reform data because reforms are typically implemented at the same time for all students in a specific state. Therefore, researchers are typically required to consider students from different states or cohorts within the same state (before the reform) as control groups, which in turn introduces different challenges and assumptions, particularly regarding their comparability. To address this potential limitation, we used survey weights to assure representativeness of the different cohorts. Notably, response rates on all assessments were 90% or larger at the student level (e.g., IEA, 2013). In addition, we inspected potential differences between the cohorts and specified adjusted models, in which we controlled for important (presumably relatively time-stable) covariates. All those checks suggested that if selection bias was present in our study, it should have been small at most (e.g., Hübner et al., 2017a).

Furthermore, it is important to underline that our findings are based on self-reports and that we did not have more objective markers to assess stress and health, for instance using data from health insurance agencies, medication records, or cortisol measures. Therefore, it cannot be ruled out that students, at least in part, also reported feeling more stressed because of ongoing discussions with their parents, friends from G9 cohorts, or the media. However, even if part of this effect could have been explained by these aspects, the remaining differences would have still remained of practical significance (e.g., Milde-Busch et al., 2010; Hübner et al., 2017a; Quis, 2018).

Regarding external validity, it is important to keep in mind that we considered representative data of one specific reform in one specific German state (Baden-Württemberg). Therefore, the findings should be generalized cautiously to discussions about effects of changes in instructional time. Most importantly, as shown in prior studies (Else-Quest et al., 2010; OECD, 2019), results on gender-related disparities are very heterogenous in STEM subjects across countries. The authors argue that one of the main drivers of gender differences are differential opportunity structures (e.g., equity in school enrollment). Based on this, it remains to be shown if our findings can be generalized to other countries where gender disparities are less or even more strongly pronounced, compared to Germany. However, doing this would require similar reforms to be implemented in other countries, which we are not aware of, even after consulting a large reform database (OECD, 2015). This also becomes evident when inspecting further related literature. Among others, findings on this topic are based on quite heterogeneous reforms (e.g., Allensworth et al., 2009; Domina et al., 2015; Huebener et al., 2017; Marcus et al., 2020), based on randomized controlled trials (e.g., Meyer and van Klaveren, 2013; Andersen et al., 2016) or cross-sectional secondary data analysis (e.g., Lavy, 2015). Before generalizing results from our study to the general debate about learning time or other environments (e.g., other states or reforms), researchers and practitioners should carefully consider potential similarities and differences.

Finally, the major change implemented by the G8-reform constitutes a school time compression, which was implemented by increasing average time per week spent in lower secondary school (Homuth, 2017). However, beyond these changes, other different elements changed simultaneously with the introduction of the G8-reform, for instance, educational standards were introduced and schools were required to develop a school-specific curriculum (Hübner et al., 2017a). Therefore, although the instructional time change is probably the most dominant feature of the reform, we cannot rule out that other changes might have affected our findings. Results of our study should therefore be interpreted cautiously as reform effects (e.g., a combination of different changes happening at the same time) rather than as pure effects of a change in instructional time.