+++ Abstract submission closed +++ Registration for participation open +++

The titles, authors and abstracts of all paper presentations are listed below in the program overview and can be accessed directly by clicking on the sessions. Presenting persons are marked with an asterisk. The names of the session chairs will be added in due course. Please note that the schedule is still preliminary and may be subject to (minor) changes.

Session 1: NEPS Developments

Thursday | December 11, 2025 | 09:10 am - 10:50 am (CET)

• Changes in the weighting scheme of Starting Cohort 8 in the National Educational Panel Study
  Anne Konrad* (LIfBi Bamberg)
  Ariane Würbach (LIfBi Bamberg)
  ≡ Abstract

  Starting Cohort 8 (SC8) of the National Educational Panel Study (NEPS) follows students from Grade 5 onwards and thus covers the transition from primary to lower secondary education up to upper secondary school or vocational training. Twelve years after the initiation of Starting Cohort 3 (SC3), SC8 was launched. To ensure comparability, the sampling design of SC8 was largely implemented analogously to SC3. The cohort enables the study of competence development, educational processes, educational decisions, and returns to education throughout secondary education. The sample consists of students in regular schools and in special-needs schools with a focus on learning.
  Survey weights are essential to draw valid inferences from the sample to the target population. They correct for unequal selection probabilities and compensate for imbalances between the sample and the population, thereby reducing biases due to nonresponse and coverage errors. In SC8, several changes in the weighting scheme were introduced. First, to avoid extreme weights, response homogeneity groups, that are groups of respondents with similar characteristics, are applied instead of individual response probabilities. Second, calibration is carried out in each wave. In the calibration approach, weights are adjusted such that the weighted totals of auxiliary variables match known population totals from a comprehensive school frame. Third, trimming, which has the potential to introduce ex post bias, is no longer used. Fourth, individual weighting components are provided, allowing users to construct customized longitudinal weights, including those starting from later waves. These changes aim to improve efficiency, enhance precision, and increase usability for the research community. In the first wave, the second and third point are already implemented, while the first and forth will be introduced starting with the second wave.
  To evaluate the properties of the new weighting scheme, we conduct a Monte Carlo simulation based on a synthetic population generated from SC6. Monte Carlo methods enable the assessment of statistical procedures under repeated random sampling. Our study compares alternative approaches and provides empirical evidence on the performance of the weighting scheme in terms of the weight distribution and estimation quality. The paper discusses both the revised weighting scheme of SC8 and the results of the simulation study.



• Assessing trust in science: Development and validation of a short scale for adolescents and adults
  Diana Steger* (LIfBi Bamberg)
  Astrid Schütz (University of Bamberg)
  ≡ Abstract

  Trust in science is a critical foundation for informed decision-making, societal participation, and lifelong learning. In the educational domain, it can affect the motivation to engage with science, subject choice, and openness toward complex topics. Yet, despite its relevance, trust in science and its developmental trajectories, are not yet well-understood—particularly in adolescence and early adulthood. To this end, we developed and validated a short scale suitable for panel studies, enabling systematic investigation of how trust in science evolves and how it relates to educational pathways.
  In Study 1, cross-sectional data from 449 adults (16-75 years, mean=37.66 years, SD=16.97; 52.33% women) were used to derive a 4-item version with good fit in a unidimensional model (CFI=.997, RMSEA=.046, SRMR=.013), excellent reliability (ω=.835), and homogeneous factor loadings (minλ=.730), together with meaningful correlations with personality traits (e.g., r=.19, p<.05 with agreeableness) and socio-political attitudes (e.g, r=-.31, p<.05 with conspiracy mentality). In Study 2, we replicated the model using data from 298 parent-adolescent dyads (children’s age ranged from 13-15 years, mean=13.77 years, SD=0.71; 47.9% female). The scale was invariant across age groups, gender, and educational background, which renders it suitable for assessing trust in science in heterogeneous samples across the lifespan. Importantly, analyses of the dyadic data revealed a substantial association between parents’ and adolescents’ trust in science (r=.66, p<.05), indicating a substantial relationship between parents’ and children's trust in science.
  On this basis, we provide a reliable and efficient tool to assess trust in science across the lifespan. We show that trust in science is an educationally relevant construct with implications for both formal and informal learning environments. Moreover, the data show that adolescents’ trust in science is shaped by family contexts, related to broader personality traits, and socio-political attitudes. Future research could examine how trust in science develops along educational trajectories and how it might be fostered through targeted interventions.



• Not the way we thought it was? Subjective pandemic experiences and panel data in a mixed-methods design
  Piotr Kocyba (Leipzig University)
  Steffen Wamsler (LIfBi Bamberg, University of Bamberg)
  Johannes Kiess (Leipzig University)
  Gundula Zoch (University of Oldenburg, LIfBi Bamberg)
  ≡ Abstract

  The recent Covid-19 pandemic mounted substantial social and economic burdens on citizens. Many people faced substantial difficulties in making ends meet, closed school and childcare facilities caused huge challenges for parents to reconcile the needs of their children with those of work, lockdowns and curfews brought cultural life to a grinding halt, and social distancing measures contributed to a breakdown of social networks and widespread isolation. Numerous studies have since shown that these consequences of such an unprecedented crisis also induced dynamic shifts in a variety of political attitudes, such as trust, interest in politics, susceptibility to conspiracy beliefs, or individual perceptions of efficacy. Yet, qualitative and quantitative approaches in political sociology are frequently at odds as to both causes and magnitudes of such changes. In this paper, we draw on a unique, recent dataset, where a sample of former panellists from the German National Educational Panel Study’s Starting Cohort 5 (NEPS, SC5) was invited to participate in qualitative group discussions taking place in two different cities in early 2025 (N=16). This dataset enables us to explore individual-level mechanisms behind attitudinal shifts (i.e., a loss of political efficacy) during the pandemic period against the backdrop of previously collected longitudinal data in a mixed-methods approach. Thus, we are able to provide qualitative depth to a quantitatively well-researched topic and to engage in bottom-up theory-building and a formation of fine-grained types of trajectories. In doing so, our results contribute to extant research by distinct subjective interpretations of the pandemic that correspond to specific trajectories of political efficacy over time. We demonstrate that loss of political efficacy is not a uniform process, but varies by life situation, institutional experiences, and individual coping strategies. While some respondents describe a gradual erosion of trust and agency rooted in perceived institutional failure, others reinterpret political withdrawal as an act of rational distancing or resistance.
  Methodologically, the study showcases the potential of linking rich qualitative accounts with panel data to trace individual-level change and meaning-making processes. This design allows for both bottom-up typologization and theory refinement in political sociology.
  We argue that combining retrospective narratives with longitudinal data offers a promising pathway for bridging the gap between subjective perceptions and structural developments in times of societal crisis.



• Do extra years of schooling reduce susceptibility to conspiracy myths? Evidence from a German compulsory schooling reform
  Nadja Bömmel (IAB Nuremberg)
  Gundula Zoch (University of Oldenburg, LIfBi Bamberg)
  ≡ Abstract

  This paper examines whether additional years of general education reduce susceptibility to conspiracy myths, drawing on a compulsory schooling reform in West Germany. Conspiracy myths are understood as overarching narratives that attribute political and social events to hidden cabals. Endorsing such narratives can undermine trust in political institutions, deepen social divides, and call the legitimacy of democratic processes, such as elections, into question. Understanding whether education can serve as a powerful protective factor is therefore of high societal and academic importance.
  While education is positively associated with a broad range of political attitudes and behaviors (e.g. Bömmel & Heineck 2023), existing research provides mixed evidence on its causal role in shaping beliefs in conspiracy myths. Theoretically, education may reduce receptivity to such narratives by (1) fostering cognitive and civic skills that lower the costs of verifying information and facilitate the rejection of implausible claims, (2) transmitting democratic norms and a respect for scientific procedures that conflict with the logic of hidden‐plot explanations and (3) providing access to broader social network whose prevailing norms discourage the open endorsement of fringe ideas.
  We leverage the staggered extension of compulsory schooling from eight to nine years across West German states between 1949 and 1969 as a natural experiment. Using data on adult respondents from the National Educational Panel Study (NEPS-SC6, 2021-2023) together with Helbig and Nikolai’s (2015) archival reconstruction of the reform, we compare OLS estimates with results from two-stage least-squares models that exploit the reform as an instrument for years of schooling. Moreover, the rich survey information enables us to compare two different indicators of schooling duration: one based on detailed biographical information and the other derived from certificates, the standard indicator often used in the literature.
  Benchmark OLS regressions reveal a strong statistically significant negative association between education and endorsement of conspiracy myths. Results from instrumental-variable regressions confirm a causal effect, highlighting that one additional year of schooling lowers the probability of endorsing conspiratorial narratives measured as conspiracy index by 2.07 points when self-reported schooling is used and by 1.57 points for generated schooling. Robustness tests — including alternative timing schemes, grade-repeater exclusions, trimmed samples, and placebo analyses — confirm these findings. Overall, the results broaden the catalogue of civic returns to education, suggesting that even modest changes in the educational system can strengthen individual resilience against harmful conspiracy narratives decades later.

back to top

Session 2: Familial and Political Influences

Thursday | December 11, 2025 | 11:00 am - 12:40 am (CET)

• The role of parental skills in explaining the intergenerational relationship between parents' and children's education: Findings from Germany
  Thorsten Schneider* (Leipzig University)
  ≡ Abstract

  Parents’ and children’s educational attainment are strongly correlated. Theories such as Bourdieu's cultural capital framework suggest that higher-educated parents possess more (incorporated cultural) skills. These skills support children’s learning both incidentally, through everyday interactions, and intentionally, through deliberate guidance, contributing to their educational success. Parental skills are also crucial for navigating the education system. Yet, few studies on social stratification explicitly examine the role of parental skill measures.
  Higher parental education — and possibly parental skills — often leads to higher income, which may partly explain the correlation between parents’ and children’s education. According to the family investment model (FIM), larger financial resources are used for educational investments, while the family stress model (FSM) posits that financial shortages increase parental stress and inconsistent parenting, hampering child development.
  Using data from the German National Educational Panel Study (NEPS), specifically the sixth starting cohort covering adults (SC6), we analyze over 2,300 parent-child dyads collected between 2014 and 2020. We estimate logistic regression models to examine the likelihood that children attend the most demanding and prestigious lower secondary school type (‘Gymnasium’). The models account for parental educational attainment, language skills measured across multiple subdomains and time points, and disposable household income. We also control for parents’ test scores in perceptual speed and reasoning, which serve as indicators of ‘mechanics’ and ‘fluid intelligence,’ respectively, and which should be largely unaffected by formal education but could bias estimates if omitted.
  Our findings confirm that children of highly educated parents are more likely to attend the most demanding lower-secondary school type than children of less-educated parents. Including parents’ language skills slightly reduces the effect of parental education, which remains large and statistically significant, suggesting that parental education encompasses more than skills. Beyond parental education, we find strong correlations between parental language skills and children’s educational outcomes. These findings persist even after accounting for household income, which shows negative effects for lowest-income families and positive effects for highest-income families, partially mediating the effect of parental education but not that of parental language skills. Overall, these results highlight the importance of accounting for parental skills in studies of intergenerational educational mobility, while also demonstrating that parental educational attainment itself, above and beyond skills, continues to drive status reproduction.



• Competence in the family language in Russian or Turkish as a resource for reading competence in the language of instruction in German? An examination of transfer effects in multilingual primary school children
  Lisa Tinkl* (IQB Berlin)
  Aileen Edele (Humboldt University Berlin, BIM)
  Birgit Heppt (TU Dortmund University)
  ≡ Abstract

  Reading skills in the language of instruction are crucial for both monolingual and multilingual learners’ educational success (Melby-Lervåg & Lervåg, 2011). Multilingual children who speak a family language (e.g., Russian or Turkish) alongside German often develop reading skills under different conditions than monolingual peers and exhibit lower reading skills (Verhoeven et al., 2019).
  According to the Interdependence/Transfer Hypothesis (Cummins, 1979), (meta-)linguistic and conceptual knowledge can transfer between languages as Common Underlying Proficiency. The Simple View of Reading (Gough & Tunmer, 1986) emphasizes the importance of listening comprehension and decoding for reading comprehension. However, the strength of transfer depends on linguistic proximity. German and Russian share Indo-European roots and grammatical similarities (Campbell & King, 2011), suggesting stronger transfer effects than between Turkish and German.
  Previous studies confirm cross-linguistic transfer and the importance of oral language for reading comprehension (Gottardo et al., 2018; Hoover & Gough, 1990). For secondary school students with automated decoding skills, positive associations exist between listening comprehension in the family language and reading comprehension in the language of instruction (Edele & Stanat, 2016). However, it remains unclear whether this effect holds for primary school children with less developed decoding skills (Ebert & Weinert, 2013).
  This preregistered study examines whether listening comprehension in Russian or Turkish (Grade 2) predicts reading comprehension in German (Grade 4). It investigates whether transfer effects persist after controlling for relevant third variables and whether transfer effects differ between language groups. The sample includes 217 Russian-German bilingual and 152 Turkish-German bilingual children from NEPS Starting Cohort 2 (Blossfeld & Roßbach, 2019; NEPS Network, 2024). Regression analyses and a Chow-test (Chow, 1960) were applied.
  Results show a significant association between Russian listening comprehension and later German reading comprehension (β=.23, p<.05), even beyond early German reading comprehension (β=.36, p<.05). For Turkish-German bilinguals, the effect is not significant (β=.18, p>.05); only early German reading comprehension (β=.15, p<.05) and socioeconomic status (β=.02, p<.05) predict later German reading comprehension. The Chow-test reveals no significant group differences (F(10,348)=1.704, p>.05).
  The findings suggest that listening comprehension in the family language can support reading comprehension in the language of instruction, particularly for structurally similar languages like Russian-German. The absence of a significant transfer effect for Turkish-German may stem from greater linguistic distance or smaller sample size. Overall, transfer effects vary between languages. Given the cultural and academic value of family language skills, their promotion should be politically supported, regardless of transfer potential.



• Parental involvement in the context of migration background
  Marie-Claire Schelmat (Technical University of Munich)
  Anna Hartl (Technical University of Munich)
  Doris Holzberger (Technical University of Munich)
  ≡ Abstract

  Parental involvement encompasses all activities undertaken by parents in the context of school and their children to positively influence the academic achievement.
  Hill & Tyson (2009) divide parental involvement into three dimensions: “school-based involvement,” “home-based involvement,” and “academic socialization.” School-based involvement encompasses parental activities at school. Home-based activities are all interactions at home or visits to external institutions. Academic socialization includes parents' expectations of school and related communication (Hill & Tyson, 2009).
  From a conceptual perspective, Hoover-Dempsey & Sandler (1997) describe the relationship between involvement and academic achievement, outlining mechanisms that help explain this link. Modeling refers to children imitating their parents' (supportive) behavior. Reinforcement involves parents praising positive learning behaviors to encourage their repetition. Instruction entails parents providing relevant information and implementing structured, task-related processes during their supportive activities (Hoover-Dempsey et al., 2001).
  With regard to empirical findings, parental involvement has generally been found to correlate positively with achievement (Täschner et al., 2021). However, this relationship appears to depend on the specific dimension of parental involvement and family background characteristics. Holzberger et al., (2023) found a weaker correlation for students from families with low socioeconomic status. It remains unclear whether similar differences exist for students with a migration background.
  We will therefore ask the following research questions: (1) Does parental involvement differ in the individual dimensions depending on whether or not the parents have a migrant background? (2) Is the relationship between the individual dimensions of parental involvement and achievement moderated by a migrant background?
  To answer the research questions, NEPS data (NEPS-Netzwerk, 2024) will be reanalysed. A regression analysis will be used to calculate the correlation between parental activities from the three dimensions by Hill & Tyson (2009) and achievement. For the school-based dimension the variables contact with school will be used. Frequency help with homework/other school exercises will be used for the school-based dimension and parental monitoring for academic socialization. The migration background serves as a moderator. Data will be analysed after pre-registration is completed.



• Political context and adolescent refugees’ vocational education and training following secondary school
  Benjamin A. Korman* (LIfBi Bamberg)
  ≡ Abstract

  Like most western countries, Germany faces a severe worker shortage due to ongoing demographic change. The large influx in asylum applications from young, predominantly Middle Eastern refugees to Germany between 2015 and 2021 therefore offered the country a unique opportunity with which to address its rising labor market shortages. For these refugees, taking up vocational education and training (VET) has also been considered the most promising means of achieving successful socio-economic integration. However, refugees face unique hurdles, such as discrimination, when transitioning from secondary school into the labor market. For example, experimental work has demonstrated that refugee applicants are perceived as symbolic and realistic threats, inducing intergroup anxiety in host country nationals charged with making recruitment decisions. The discrimination of immigrants and refugees is however unlikely to be evenly distributed given that political rhetoric regarding these individuals can shape how they are perceived and treated by host country nationals. In line with this are recent studies linking higher regional support for the Alternative for Germany (“AfD”) to immigrants’ lower economic integration, increased social undermining by coworkers, and higher likelihood of being the target of a hate crime. Based on this prior work, I propose that the political context influences refugees’ decision to enter the labor market. Specifically, I hypothesize that refugees’ likelihood of starting VET (i.e., entering the labor market) following secondary school, as opposed to starting a prevocational course (i.e. postponing their labor market entrance), decreases as regional support for the AfD increases.
  I find support for my hypothesis in a sample of 394 youth refugees who were part of Cohort 2 (Adolescents) of the study Refugees in the German Educational System (ReGES) and whose survey data were matched to publicly available German federal election results (https://www.bundeswahlleiterin.de/). The findings of the current study are important because 1) many Western countries are becoming increasingly reliant on immigrant workers, 2) public support for far-right political parties has been rising globally for years, and 3) in Germany, the regions most expected to be hit by labor shortages in the future are those in which the AfD is most supported. By demonstrating where adolescent refugees postpone entering the labor market, this study highlights the political context as a potential hinderance to their successful socio-economic integration.

back to top

Session 3: Gender Inequalities and Careers

Thursday | December 11, 2025 | 11:00 am - 12:40 am (CET)

• Unpacking the Gender-Equality Paradox as a micro-mechanism: Evidence from Germany
  Mingming Li* (University of Innsbruck)
  ≡ Abstract

  The educational Gender-Equality Paradox (GEP) refers to the counterintuitive phenomenon that gender gaps in STEM (Science, Technology, Engineering, Mathematics) study choices are larger in more gender-equal (and usually more affluent) countries. Economic Opportunity Theory (EOT) explains this by suggesting that greater affluence reduces the necessity to prioritize lucrative STEM fields for both men and women, and thus provides more room for gender-essentialism self-expression. However, the existing research has predominantly tested this theory at the macro-level, which overlooks the crucial micro-level variations in economic security within societies. In addition, these indicators are mostly composite indicators (e.g., GGGI; HDI) that mix the economic and culture factors, which causes difficulties for both theoretical interpretation and empirical research.
  To address this gap, this study proposes a micro-level framework that analyses the role of household affluence in shaping gendered educational choices. Using rich data from NEPS-SC4, I analyze the selection of STEM versus non-STEM majors among 1,680 students eligible for higher education in Germany. Germany offers a compelling context for this analysis due to its mixed findings in GEP and enduring East-West divisions in culture, economy, and gender norms. The analysis focuses on gender, disposable household income, and their interaction terms as core independent variables. I use Heckman selection models to account for non-random selection into higher education, the Karlson-Holm-Breen (KHB) decomposition approach to disentangle direct and indirect effects, and the double/debiased machine learning (DDML) with multiple learners to ensure robustness and minimize model dependence.
  The findings reveal that the gender gap in STEM major choice widens as household income increases, primarily driven by a decreasing likelihood of women, rather than men, choosing STEM in more affluent families. This pattern remains robust after controlling for academic achievement, family background, personal interests, gender-role attitudes, and occupational values. It holds consistently across both East and West German contexts. KHB decomposition results further show that attitudinal and value-based mediators significantly explain the overall gender gap, but do not account for the income-driven widening. Robustness tests using alternative STEM definitions and sample restrictions confirm the main findings.
  These results provide strong micro-level evidence for EOT and its underlying instrumental motivation. Targeted policy interventions are suggested to promote female STEM participation, particularly among women from more affluent households. The findings challenge explanations based solely on culture or attitudes and underscore the enduring influence of economic security on gendered educational trajectories, even in post-materialist societies.



• STEM career persistence against all odds? The role of personality traits
  Antje Stefani (University of Konstanz)
  Susanne Strauß (University of Konstanz)
  ≡ Abstract

  This study examines the role of personality traits in the persistence of women in STEM careers, particularly within challenging environments marked by high workload, competitive climates, and stereotype threats. Despite these obstacles and the persistent gender gap in STEM, women in Germany do not display higher dropout rates during higher education or early career stages (Stefani et al., 2024). However, research indicates that men and women respond differently to STEM-specific challenges, with social support being more crucial for men’s persistence (Horrocks & Hall, 2024; Stefani, 2023).
  Guided by Lazarus and Folkman’s (1984) theory of stress and coping, we hypothesize that gender-specific personality traits influence STEM persistence. Using the Big Five model, we assess traits like openness, conscientiousness, extraversion, agreeableness, and neuroticism, previously linked to career success and stress management. Data from the German NEPS (Starting Cohort 5) of 2,417 women and 1,368 men reveals that high extraversion scores are associated with higher attrition, likely due to a person-environment mismatch, as less extraverted individuals are more likely to choose STEM. Lower agreeableness and higher conscientiousness further support women’s persistence, particularly at job entry, suggesting resilience in socially challenging environments.



• The gender-specific returns to ICT skills: Do women and men profit equally from skill acquisition over their careers?
  Stephan Bischof (BIBB Bonn)
  Martin Ehlert* (WZB Berlin, Freie Universität Berlin)
  Timo Gnambs (LIfBi Bamberg)
  ≡ Abstract

  In times of rapid digitalization, ICT skills are often cited as important future skills that can help workers advance their careers. Accordingly, many studies found a strong positive association between ICT skills and wages among adults, although causality remains an unresolved issue in this literature (Falck et al., 2021; Woessmann, 2025). However, previous studies assumed that returns to skills are uniform across individuals. Yet, gender-sensitive accounts suggest that women may not profit equally from education and skills compared to men due to structural constraints often referred to as “glass ceilings” or “sticky floors” (Cotter et al., 2001; Purcell et al., 2010). In line with this, studies found that computer use yields lower returns for women than for men (Cheng et al., 2019; Kristal et al., 2024). However, the gendered impact of measured ICT skills on wages has so far not been analyzed.
  In this study, our research question is therefore: Does the effect of ICT skills on wages differ between men and women? We base our analysis on data from the adult cohort of the German National Educational Panel Study (NEPS SC6). Specifically, the NEPS SC6 features measurements of ICT skills in two waves (5 and 14). The measurements are based on objective tests assessing skills and knowledge in using standard ICT software (Gnambs & Senkbeil, 2023). Furthermore, we generate 30 plausible values in a fully imputed data set of individuals employed in wave 5 and 14 (N ≈ 2,500) to analyze true score effects adjusted for measurement error. Our strategy to identify causal effects of ICT skills on wages relies on individual fixed-effects models and further controls of observed time-varying factors from the rich NEPS data.
  Preliminary results indicate that the effect of ICT skills on wages is larger among men. Moreover, while we find a positive effect among men, the estimate for women is even slightly negative. Accordingly, increases in ICT skills over the career among men lead to increases in wages while women’s ICT skill acquisition is not rewarded similarly. Further analyses will assess the validity of the findings given threats to identification, especially with regard to reverse causation, that is transitions to higher paying jobs that lead to the acquisition of ICT skills. Furthermore, we aim to explore individual and structural explanations of the gender difference.



• Gendered returns to further training: Can non-formal training help close the Gender Pay Gap?
  Marco Seegers* (BIBB Bonn)
  Martin Ehlert (WZB Berlin, Freie Universität Berlin)
  ≡ Abstract

  Policy debates have increasingly focused on further training in adulthood as a tool to promote equal career opportunities. Yet it remains empirically unclear whether further training acts as an equaliser, reducing existing wage inequalities, or as an amplifier, reinforcing the Gender Pay Gap.
  This article examines whether wage returns differ between women and men for non-formal job-related further training, i.e. supervised courses, workshops or seminars that do not lead to formal qualifications. Grounded in Relational Inequality Theory, we argue that training only translates into wage gains when individuals have the organisational power and legitimacy to make successful claims—conditions that are often unequally distributed in gendered organisations.
  We use data from the adult cohort of the German National Educational Panel Study (2007-2023) and apply individual fixed-effects panel regression models with a cumulative approach to estimate the long-term effects of participation in further training on wages.
  Our findings show that both men and women benefit from further training. However, men achieve significantly higher wage gains from longer spells of participation, particularly in selective, employer-provided formats. Leadership positions further amplify men’s returns. Although gaps in training participation have narrowed and in part reversed, returns show no equalising potential and instead contribute to the persistence of the Gender Pay Gap. Building on Relational Inequality Theory, our findings contribute to the broader debate on inequality reproduction through gendered claims-making by focusing on further training in organisational and institutional contexts.

back to top

Session 4: Competencies at School Age

Thursday | December 11, 2025 | 02:45 pm - 04:00 pm (CET)

• The longitudinal interplay of math self-concept and competence: An RI-CLPM analysis of intelligence and gender
  Sabine Patzl* (University of Bamberg)
  Janine Täschner (Technical University of Munich)
  Iris K. Gauglitz (University of Bamberg)
  Astrid Schütz (University of Bamberg)
  ≡ Abstract

  The reciprocal relationship between academic self-concept and competence is a key area of educational research, yet prior work typically has two limitations. First, it has often relied on traditional cross-lagged panel models that confound stable between-person differences with true within-person change. Second, the moderating roles of crucial covariates like intelligence and gender remain underexplored. This study addresses these gaps in mathematics, a domain characterized by a "gender paradox," where girls report significantly lower math self-concept (MSC) despite small to nonexistent differences in actual competence. We therefore investigate three core questions: 1) the dynamic within-person interplay between MSC and competence, 2) the role of baseline intelligence in predicting stable versus time-varying components, and 3) whether these reciprocal dynamics differ by gender.
  To address these questions, we analyzed longitudinal data from the NEPS Starting Cohort 3. The final sample comprised 1,892 academic-track students with data from three critical secondary school stages: Grade 5, Grade 9, and Grade 12. We employed a Random-Intercept Cross-Lagged Panel Model (RI-CLPM) to provide a robust test of the reciprocal relationship. This approach isolates genuine within-person changes from stable, trait-like between-person components. It thus allows for a methodologically sound test of the Reciprocal Effects Model (REM) and an examination of how baseline intelligence influences both the stable and the dynamic components of MSC and competence. Gender differences in the dynamic pathways were tested using a multigroup approach.
  Results revealed a significant, reciprocal within-person relationship between MSC and competence. As hypothesized, baseline predictors were crucial for the stable, between-person components: higher reasoning ability and parental SES were significant predictors of students’ trait-like levels of both math competence and MSC. A specific test of baseline IQ on dynamic changes over time, however, found no meaningful predictive effect. Crucially, the reciprocal process itself was equivalent for boys and girls. This finding, combined with a lack of scalar measurement invariance between genders that we found, suggests that the "gender paradox" in mathematics may stem from different, socially influenced starting points rather than from different underlying processes.



• Does schooling increase the maths and language gender gap? Competence development in primary school in Germany
  Wilfred Uunk* (University of Innsbruck)
  Pia Blossfeld (University of Innsbruck)
  ≡ Abstract

  For the US and France, evidence emerged that the primary school stage increases the gender gap in academic competencies. Yet, the studies were restricted to analysing Grades 1 or 2 and focusing on gender gaps in math, neglecting gender gaps later in school and in language competencies. In our study, we aim to advance the literature and address the question to what extent girls and boys differ in math, science, and language competencies before and during primary school in Germany, using data from the Newborns and Kindergarten cohorts of the National Educational Survey (NEPS). Our expectations regarding the development of gender differences in competences differ, depending on the theoretical perspective (schools as equalizers vs Matthew effects).
  Descriptive and fixed effects panel analyses reveal no evidence that during the primary school stage, schooling widens gender gaps in achievement. On the contrary, while the gender gap in science and language (which, for vocabulary, is surprisingly at the advantage for boys) remains stable, the (small) maths gender gap closes in Grade 4. Furthermore, there exists no variation in the effect of gender on competencies across schools.
  Nevertheless, we find that before school entry, at age 4, girls outperform boys in math and science, a gender gap that reverses in Grade 1. This suggests that school-entry rather than the primary school period in Germany contributes to gender-stereotypical gender achievement gaps.



• Are there such things as talent subjects? Genetic and environmental influences on participation and performance in music, art, and sports
  Dave Balzer (Johannes Gutenberg University Mainz)
  Nico Sonntag (Johannes Gutenberg University Mainz)
  ≡ Abstract

  There is a debate among certain German educators and policymakers whether so-called “tal-ent subjects” (e.g., music, art, and physical education) reward natural aptitude more than classroom effort, thereby challenging the meritocratic ideal more than grading practices in core subjects such as German and mathematics. Recent public discussions and media coverage have rekindled calls for solid empirical evidence rather than anecdotal impressions. Our study attempts to fill this gap by offering the first large-scale study of genetic and environmental contributions to both participation in (sometimes elective) as well as grading in such talent subjects.
  In psychology, talent is understood as a relatively stable, and partly innate disposition for high performance in specific domains. Talent represents a potential that, under favorable condi-tions, can translate into exceptional achievement. While genetic factors play an important role, the relationship between genes and performance is not deterministic; environmental factors such as family and school socialization are crucial. Talent can manifest in a wide range of areas. In the German discourse on talent, the focus is most often on the aesthetic-practical and physical-motor domains – namely music, art, and physical education. These subjects, often referred to as “minor” or “non-core” subjects, stand in contrast to the cognitively and theoretically oriented focus of the rest of the curriculum. When looking at findings from empirical educational sociology, one gains the impression that these subjects tend to receive less attention.
  Our contribution examines participation and performance in talent subjects. We first outline the variation in institutional regulations across Germany. Since education policy is set at the federal state level, the Länder have considerable autonomy in defining the curricular status and obligations of these subjects. Using German TwinLife data we then apply the ACE model of the Classical Twin Design to disentangle additive genetic (A), common environmental (C), and unique environmental (E) influences on both participation and academic performance. Monozygotic twins, who share identical genetics, are compared to dizygotic twins with lower genetic similarity, allowing us to estimate the relative importance of genetic versus environmental factors in these outcomes.
  Participation and grading in math and German, and physical education are nearly universal, while participation in music (57%) and visual arts (70%) is more selective. Talent subjects show higher average grades and slightly lower variation, possibly due to positive selection or grading practices. Multilevel models reveal that most variance in both participation and performance stems from between-family rather than within-family differences. ACE decompositions show that shared environmental factors strongly shape participation in talent subjects – especially music and art – highlighting the role of family context. By contrast, genetic influences explain a substantial share of grade variance in music and physical education (around 40%), and to a lesser extent in visual arts (25%). Genetic influences are even stronger in math and German (40–50%), with minimal shared environmental impact. Further analyses will explore gene–environment interactions, particularly whether genetic effects vary by family resources and school contexts. We also propose a method to adjust for selection bias using inverse probability weighting within the ACE framework.

back to top

Session 5: VET - Access, Tasks, Mobility

Thursday | December 11, 2025 | 02:45 pm - 04:00 pm (CET)

• Alternative access to VET and higher education: Pathways, re-entry, and lifelong learning
  Awais Malik (University of Bamberg)
  Miriam Marleen Gebauer (University of Bamberg)
  Silvia Annen (University of Bamberg)
  Melanie Hochmuth (University of Bamberg)
  ≡ Abstract

  The expansion of alternative access routes to vocational education and training (VET) and higher education is a notable feature of educational systems (Powell & Solga, 2011; Wolter & Kerst, 2015). Among these, two particularly relevant pathways are second-chance schooling and professional experience. Second-chance schooling represents a formal institutionalised route for those without standard entry qualifications, while professional experience serves as an informal, practice-based recognition of skills. These routes capture the tension between structured credentialing and experiential learning, making them analytically valuable for understanding lifelong educational participation. From a theoretical perspective, such pathways resonate with Bourdieu’s (1986) concept of cultural and social capital, as they illustrate how individuals without standardized credentials mobilise alternative resources to access education. They also echo the logic of recognition frameworks (Dörffer et al., 2025), in which the legitimacy of prior learning and skills depends on institutional acknowledgment. Furthermore, these routes are embedded in wider debates on the permeability of education systems (e.g., Owens & de St Croix, 2020; Napoletano, 2024) and illustrate the institutional structuring of opportunity for non-linear trajectories, highlighting how educational structures reconcile meritocratic ideals with practices of lifelong learning.
  This study investigates whether individuals who enter VET or higher education through second-chance schooling or professional experience exhibit higher or lower rates of educational re-entry during job transitions compared to those with standard entry qualifications. Situated within the frameworks of lifelong learning and human capital theory (Becker, 1964), the study examines whether non-linear entry routes encourage a tendency of continuous upskilling or, alternatively, whether the weaker recognition of such routes restricts further engagement with educational programs.
  Methodologically, the study applies longitudinal event history methods using NEPS panel data (Blossfeld et al., 2011), operationalised through discrete-time competing-risks models. This enables analysis of the timing and type of educational re-entry across access groups while accounting for job situations and individual characteristics.
  Findings may suggest that recognition frameworks and institutional policies should move beyond treating second-chance schooling and professional experience as fallback options. Instead, they need to ensure that these pathways provide comparable opportunities for continued learning as standard entry routes. This includes, for example, formally acknowledging workplace learning in qualification systems, and offering targeted guidance during job transitions. Such measures can help transform second-chance schooling and professional experience into sustainable drivers of educational participation across the life course, rather than temporary or compensatory solutions.



• Inequality in tasks during VET training in Germany: What differences exist based on the socioeconomic status and school-leaving certificate?
  Anett Friedrich (BIBB Bonn)
  Silvia Annen (University of Bamberg)
  ≡ Abstract

  Much literature deals with individual, institutional or regional characteristics that enlarge or reduce the chances to enter VET (e.g. Beicht & Walden, 2019; Protsch & Solga, 2016). The social background and the school-leaving certificate emerge as important drivers for unequal access opportunities to VET (e.g. Jacob & Solga, 2015; Roth, 2018).
  Much less is known about differences during the VET training phase. The aim of our research is to shed light on inequalities during the training itself by focusing on differences in the socioeconomic status (SES) and concerning school-leaving certificate. Training occupations differ in the skills, knowledge and abilities they convey but also in the cognitive abilities that are needed to learn them (Friedrich et al., 2023). Hence, taking the training occupation into account is decisive for investigating tasks performed during VET.
  We want to answer two research questions: First, if the tasks apprentices perform during their training are connected with their SES and school-leaving certificate. Second, if those connections are mediated by the cognitive requirement level of the training occupation. To answer those questions, we use the starting cohort 4 of the German National Educational Panel Study (Blossfeld & Roßbach, 2019) and estimate multi-level regressions with apprentices clustered in training occupations.
  To address the task apprentices perform during their training, we refer to the classification of Autor et al. (2003). We operationalize non-routine cognitive tasks as complexity and autonomy as well as non-routine manual tasks.
  Theoretically, there should be no significant differences in the tasks of apprentices according their SES and school-leaving certificate once taking the training occupation into account since the training curricula are legally binding and nationwide standardised. Nevertheless, due to mechanisms such as statistical discrimination (Arrow, 1972), differences in the human capital (Becker, 1964) accumulated during school, and self-selection process (Erikson & Jonsson, 1996) differences might exist anyway.
  Our preliminary results show that the SES is not connected with the tasks apprentices perform. This finding indicates that social inequalities due to the SES are not reproduced within the VET system. Further, we find no differences between apprentices with lower and intermediate school-leaving certificates. However, those with Abitur perform less often autonomous and manual tasks. The cognitive requirement level of the training occupation is significantly connected with the tasks but seems not to mediate the connection of having an Abitur and autonomous and manual tasks.



• Spatial mobility and the risk of dropping out of VET: Costs, returns and self-selection
  Linda Hoffmann* (BIBB Bonn)
  Alexandra Wicht (University of Siegen, BIBB Bonn)
  ≡ Abstract

  Spatial mobility is an important means of reducing matching problems in training and labor markets and facilitating youths’ successful transitions into vocational education and training (Matthes & Ulrich, 2018). Research high-lights positive returns to spatial mobility in terms of income, status, or job-match quality (Ganesch et al., 2019; Reichelt & Abraham, 2017; Wicht et al., 2024) while also pointing to mechanisms of social selection (Behringer, 2025; Granato et al., 2015; Winkler, 2018). Drawing on decision-making frameworks (Eccles & Wigfield, 2002; Esser, 1999) spatial mobility can be conceptualized as a strategic cost-benefit decision to increase the probability of a successful transition. Less attention has been given to the costs of mobility, which can increase the risk of dropouts and increase social inequalities.
  Previous research on VET dropouts identifies financial factors as key predictors (Bessey & Backes-Gellner, 2015; Lestari & Setyadharma, 2019). Furthermore, studies reveal an increased risk of dropout among apprenticeship who did not realize their occupational aspirations (Fischer-Browne et al., 2014; Holtmann & Solga, 2024; Wicht et al., 2024).
  Building on this, the present study examines the relevance of mobility-related costs and returns for VET dropouts. Specifically, it investigates whether mobile youths with higher resources (parental income) and higher returns (training wage and realization of occupational aspiration) exhibit a lower risk of dropout.
  Representative longitudinal data from the National Educational Panel Study (NEPS, Starting Cohort 4) (Blossfeld & Roßbach, 2019) on youths’ VET trajectories are used. Logistic regression models are employed to examine the influence of spatial mobility and associated costs and returns on the risk of dropout. Following the concept of functional labor market regions (Kosfeld & Werner, 2012), a distinction is made between spatial mobility within and between regional labor markets (RLMs). Parental financial resources are operationalized using the net equivalized household income. Two indicators are used to measure apprenticeship returns: training income and the realization of occupational aspirations.
  Our findings reveal a significant association between spatial mobility and dropout, particularly for mobility between RLMs. No statistically significant differences are observed dependent on parental resources. The training income emerges as especially relevant for mobility within RLMs: Among youths who are mobile within RLMs, the dropout risk increases with decreas-ing training income. Furthermore, youths who did not realize their occupational aspirations are more likely to drop out of VET when they become mobile between RLMs rather than within RLMs.

back to top

Session 6: Teachers in Focus

Thursday | December 11, 2025 | 04:15 pm - 05:30 pm (CET)

• Student-centered vs. teacher-centered pedagogy: A philosophical divide without outcome differences?
  Brendan A. Schuetze* (University of Utah, University of Potsdam)
  Hyun Ji Lee (Hunter College, The City University of New York)
  Eric M. Anderman (The Ohio State University)
  Katharina Scheiter (University of Potsdam)
  ≡ Abstract

  A long-running and often contentious debate in education centers on the tension between student-centered and teacher-centered instruction: Is it more effective to structure students’ learning through teacher-designed practice activities and direct instruction, or to allow students to generate and discover knowledge in a more independent manner? In the present study, we tested this question by performing a latent class analysis on teachers’ self-reported beliefs about education and then tested whether these clusters, which mapped onto the student-centered versus teacher-centered divide, predicted differences in the achievement and motivational outcomes of students in the respective classes.
  Specifically, with a sample of 305 teachers from the National Educational Panel Study, we used latent class analysis (LCA) with 30 survey items measuring beliefs about teaching, learning, and professional priorities among eighth-grade teachers. From the LCA, we identified two clusters, representing (a) teacher-centered and (b) student-centered approaches to pedagogy, respectively. We then analyzed students’ (N=2,308) outcomes before and after entering these teachers’ eighth grade classrooms. Structural equation modeling (SEM) was used to test whether teachers’ pedagogical philosophies were linked to the changes in students’ math and German motivation (i.e., self-concept, interest), self-regulated learning skills, and math and German achievement from seventh to ninth grade, controlling for students’ previous scores on these outcomes and relevant demographic factors. Results from our SEM models revealed that students’ prior academic ability, self-regulated learning skills, and age predicted the cluster membership of their teacher, but limited evidence of a direct association with student outcomes. Our results indicated that higher-achieving students were less likely to have teachers in the student-centered cluster (as opposed to the teacher-centered cluster), suggesting that teachers’ pedagogical philosophy may be informed by and adapt to the characteristics of incoming students. However, no significant differences in student achievement or motivation dependent on teacher philosophy were found, when controlling for entering characteristics. Furthermore, our post-hoc analysis of a theoretically-relevant mechanism revealed that pedagogical philosophy did not significantly predict student perceptions of teacher autonomy-supportive practices.
  The findings demonstrate that teacher beliefs do not necessarily translate into perceivable differences in instructional practices or significant differences in student outcomes. Rather, clear pedagogical strategies may be required to enact the benefits associated with specific teaching philosophies.



• Can AI help teachers with lesson preparation? Evidence from a Randomized Controlled Trial
  Philipp Appel* (DIE Bonn)
  ≡ Abstract

  Teachers face considerable workload from lesson preparation, which has been identified as a major source of strain (Jauhiainen & Guerra, 2023). Generative artificial intelligence (e.g., ChatGPT and other LLMs) could offer potential support by (a) reducing teachers’ workload and/or (b) enhancing and individualizing instruction (Hays et al., 2024; Kasneci et al., 2023; Grassini, 2023; Lee et al., 2024). At the same time, naïve or uncritical use may entail negative consequences (Busse et al., 2023; Kasneci et al., 2023).
  The present study investigates whether generative AI can effectively support teachers in lesson preparation, and whether participation in AI-Training is beneficial or necessary for such support. Specifically, two research questions are answered: (1) Can elements of lesson preparation be completed more efficiently and/or with higher quality when supported by AI? (2) Does attending a training session on prompting and informed use of AI improve outcomes compared to AI use without training?
  We conducted a preregistered online experiment (RCT) with N=231 teachers, who were drawn from the sample of the TAEPS study (Teachers in Adult Education – A Panel Study). Participants were randomly assigned to one of three experimental groups: (1) Control group without AI access (n=78), (2) Treatment group with direct access to ChatGPT-4o (n=75), and (3) Second Treatment group with access to ChatGPT-4o which had undergone AI-training previously (n=78). All participants completed four tasks related to lesson preparation. Dependent variables included (a) quality of the completed task, assessed through blinded expert ratings based on objective criteria and (b) time spent on preparation (measured in minutes).



• The experience of loneliness among (prospective) teachers
  Till Kaiser* (University of Osnabrück)
  ≡ Abstract

  Although loneliness has received little attention in empirical educational research, recent studies demonstrate its significant relevance. Loneliness correlates at the individual level with psychological and physical well-being and can lead to conditions such as anxiety disorders, depression, and cardiovascular problems (Park et al. 2020). Moreover, its societal significance is evident: loneliness is associated with decreased trust in democratic institutions and reduced participation in political processes, while simultaneously being linked to political radicalization (Langenkamp, 2021, 2025; Langenkamp et al., 2025) – issues that are currently also concerning the school system (Robert Bosch Stiftung, 2025).
  Previous research on loneliness in education has largely focused on students (Schütz & Bilz, 2024), while other actors in school educational settings have been largely neglected. The possibility that (prospective) teachers might also experience loneliness, and that loneliness might be particularly pronounced during transitions from studies, practical training (Referendariat), and entry into the profession, has not yet been investigated in detail. Therefore, the aim of this contribution is to partially address this research gap by exploring the differences in loneliness experiences among teacher education students, trainee teachers, and qualified teachers, and comparing these to other employed individuals.
  To this end, data from the National Educational Panel (NEPS) and the Teacher Education Student Panel are analyzed. Based on the waves from 2016 and 2020, in which loneliness was measured (n=4,769), multilevel logistic regression models (Level 1: observations, Level 2: individuals) reveal that the level of loneliness differs significantly across the various phases of teacher education: Loneliness is most pronounced in the initial phase of teacher education, followed by the practical training phase (Referendariat) and professional life, with the finding that teacher education students, trainee teachers, and qualified teachers are significantly less lonely than other employed individuals in the NEPS. To also analyze intra-individual changes in loneliness following a status change, fixed-effects panel regression models were estimated. The results also point in the same direction as those in the models with random effects, but were not significant, which may be due to the low number of transitions and the relatively large distance between measurements (2016 to 2020). The findings are discussed in the context of professional theoretical considerations regarding well-being, resources, and stressors in the teaching profession (Kaiser & Reintjes, 2025, Rothland & Klusmann, 2023) and current studies on teacher health (Reintjes et al., 2025).

back to top

Session 7: Early Labor Market Trajectories

Thursday | December 11, 2025 | 04:15 pm - 05:30 pm (CET)

• Human capital and motherhood: The extent and implications of child-related gaps in training
  Franziska Riepl* (IAB Nuremberg)
  ≡ Abstract

  This paper studies the dynamics of job-related training in the context of parenthood, and its implications for the diverging labor market trajectories of men and women after childbirth.
  Using linked educational and labor market data from Germany (NEPS-ADIAB-SC6) and a staggered difference-in-differences strategy, I find that the likelihood of women's training participation drops by 17 percentage points (35% of the pre-birth training rate) after childbirth, while men show no significant change. Controlling for reduced employment and hours, especially immediately after childbirth, accounts for 40% of this decline. Both demand and supply factors play a role, as firms provide less financial support for the training of mothers, and mothers participate less even when training is supported. Assuming earnings returns to training of 3-5% each year and monthly motherhood earnings losses of 41%, as estimated within the training dataset, reduced training can explain 5-8% of the motherhood earnings gap over the first eight years. Thus, accounting for changes in training can contribute to understanding long-lasting negative effects of motherhood on earnings.



• Why do young workers start in minimum wage jobs? Education, vocational training, and (mis)match in labor market entry in Germany
  Juliane Pehla* (BAuA Berlin)
  ≡ Abstract

  Transitions from education to the labor market remain a crucial phase in shaping young people’s life courses and long-term employment trajectories. Despite Germany’s well-established vocational training system, which is designed to support smooth school-to-work transitions, a substantial share of young workers still enters minimum wage employment. In 2023, around 17 percent of those earning at or below the statutory minimum wage of €12 per hour were aged 18 to 24, although this group represented only 6 percent of all employees (Earnings Survey 2023, own calculations). This raises important questions about the structural and individual factors contributing to unequal early career outcomes.
  In this paper, I examine the determinants of minimum wage employment among young workers in Germany, focusing on the transition from vocational training to the first job or, for those without training, directly from school to work. Drawing on human capital theory, occupational closure theory, and the model of employment systems, I explore how educational and vocational qualifications, the match or mismatch between qualifications and job requirements, and firm-level and regional labor market structures shape labor market entry.
  For the analysis, I use NEPS-SC4-ADIAB data, a longitudinal dataset that combines survey information from the German National Educational Panel Study (NEPS) with administrative data from the Institute for Employment Research (IAB). This rich data source allows me to trace educational and employment trajectories while incorporating firm-level and regional information.
  Using multivariate regression models, I expect to show that vocational and educational qualifications substantially reduce the likelihood of entering minimum wage employment, although this effect likely depends on the degree of job match and institutional context. I also expect persistent gender differences beyond educational disparities, driven by structural patterns in occupational allocation and job characteristics. By integrating educational, occupational, and institutional dimensions, I aim to contribute to a better understanding of how individual characteristics and structural conditions interact to shape inequality in labor market entry, particularly into minimum wage employment.



• Better off without trying? Ethnic differences in labour market outcomes of different secondary education trajectories in Germany
  Markus Weißmann* (MZES, University of Mannheim)
  ≡ Abstract

  Given the same socioeconomic background and educational achievement, children of immigrants more often choose demanding tracks in secondary education systems. However, especially in highly stratified education systems, these choices come at the expense of not choosing other, possibly better suited education alternatives (Tjaden and Hunkler, 2017). Some researchers have therefore warned against a possible “optimism trap” (Tjaden and Hunkler, 2017) in which children of immigrants end up without vocational qualifications and poor employment outlooks because of these ambitious choices.
  However, research on the impact of different secondary education trajectories (SECETs) on labour market outcomes of young adults with and without migration background is scarce (Birkelund et al., 2021; Weißmann et al. 2023). Therefore, for the German case, I ask what the consequences are of different SECETs for ethnic differences in labour market outcomes in the mid-twenties. SECETs include (1) academic track attendance throughout lower and upper secondary education, (2) changing to vocational training or (3) upper secondary academic tracks after lower secondary non-academic tracks, and (4) lower secondary non-academic trajectories without further qualifying courses. As outcomes, I investigate ethnic differences in labour market participation versus prolonged education or labour market exclusion, and, for the employed, the occupational status (ISEI) and income. I furthermore test for ethnic-specific outcomes of different SECETs and whether any possible differences remain after considering differences in socioeconomic background, human capital accumulation, and aspirations in lower secondary education.
  I rely on data from the CILS4NEPS data project, a novel panel dataset that combines the Children of Immigrants Longitudinal Survey in Four European Countries (CILS4EU) and Starting Cohort 4 of the German National Education Panel Study (NEPS SC4; Dollmann et al., 2024). This unique longitudinal dataset lets us follow a comparatively large number of children with and without immigrant background from the end of lower secondary education at the age of around 15 up until young adulthood at the age of around 27.
  Preliminary results show that differences in SECET do not explain ethnic differences in labour market participation or income in their mid-twenties. However, children of immigrants that have changed to academic tracks instead of vocational training after lower secondary education have jobs with higher ISEI scores and higher earnings compared to young adults without a migration background that took the same educational pathway. While ethnic ISEI premia are due to differences in human capital accumulation and aspirations regarding ISEI, ethnic income premia cannot be explained.

back to top

Session 8: Educational Inequalities and Decisions

Friday | December 12, 2025 | 09:30 am - 10:15 am (CET)

• The primary effects of educational inequalities in German lower secondary education and time-specific mechanisms
  Moritz Fischer* (University of Bamberg)
  ≡ Abstract

  Previous research consistently arrives at the conclusion that differences in a child’s social origin tend to be a key cause of inequality in the development of educationally relevant competencies in Germany. Furthermore, it has been suggested that this is either because of factors in the early childhood or because of the early selection into a highly stratified secondary education system. However, due to the often cross-sectional design of the employed data (e.g., PISA), there is no way of knowing exactly when these estimated inequalities occur during the educational trajecto-ry. Thus, the following research question arises for this thesis: How strong is the influence of a child's social origin on their development of educationally relevant competencies in the time frame of lower secondary education in Germany and what are the respective explanations spe-cifically for this time frame?
  Building mainly on Pierre Bourdieu’s capital and habitus theory, the following time-specific mechanisms that are specifically tailored to the time frame of lower secondary education are worked out: (1) reading frequency, (2) parental help with school related work, (3) motivation to learn, (4) private tutoring, and (5) attended school type.
  In order to test the main time-specific effect and its time-specific mechanisms, German longitu-dinal data from the waves 1 to 6 of Starting Cohort 3 of the National Educational Panel Study (NEPS) is drawn and analysed in multiple linear regression models. The subsequent results first demonstrate a strong total effect of a child’s social origin on their development of educationally relevant competencies up until the end of lower secondary education. Second, it can be deter-mined that the time frame of lower secondary education accounts for around 50 percent of the total effect with regard to reading competence and for around 13-29 percent with regard to mathematical competence. Third, the mediation analyses reveal that none of the factors 1-4 seem to explain any notable proportion of the time-specific effect, while in contrast, the attended school type seems to explain around 49-62 percent of the time-specific effect with regard to reading competence, and around 43 percent with regard to mathematical competence. It can therefore be concluded that a significant proportion of the social inequalities in educationally relevant competencies in Germany seems to occur after the transition to secondary education and the attended school type seems to be the most relevant factor for explaining this.



• Short-term financial stress and student performance
  Lucas Bütje* (University of Konstanz)
  ≡ Abstract

  Students from lower socioeconomic status (SES) families consistently underperform their high-SES peers (e.g. Hanushek, Peterson, Talpey, and Woessmann, 2019; Skopek and Passaretta, 2021). This project investigates the role of short-term financial stress in explaining the socioeconomic achievement gap in Germany.
  For the US, short-term material hardship due to scarcity of Supplemental Nutrition Assistance Program (SNAP) benefits has been shown to negatively affect student outcomes (Bond, Carr, Packham, & Smith, 2021). I explore whether these findings generalize along two dimensions: First, to a country like Germany, where a more comprehensive social safety net may buffer families against such fluctuations. Second, to more general financial stress that goes beyond stress linked to specific in-kind transfers.
  Based on the payment schedules of wages and transfers in Germany, I hypothesize that low-income families are liquidity-constrained at the end of the month, while high-income families can smooth consumption. This could increase the gap in student performance between low-and high-SES families at the end of the month. Drawing on four cohorts of the National Educational Panel Study, I leverage quasi-random variation in the intra-month timing of interviews and cognitive tests within a fixed effects framework. The quasi-randomness is supported by balance tests on a wide range of student and family characteristics.
  My analysis yields two main findings: First, I find evidence consistent with the hypothesis that parents experience liquidity constraints at the end of the month. Contrary to the initial hypothesis, this effect is not restricted to low-SES families, but is also observable in higher-SES families. On average, parents are approximately 0.04 standard deviations less satisfied with their standard of living during the last week of the month. Conversely, I find no statistically significant change in satisfaction for adults without children.
  Second, an investigation of the impact on children’s outcomes reveals no statistically significant average effect. However, the analysis has limited statistical power, resulting in wide confidence intervals that contain the effect sizes reported in the US literature. Therefore, the results should be interpreted as inconclusive rather than precise zeros. Furthermore, there is suggestive evidence of heterogeneity: in line with my initial hypothesis, the effect for children from low-SES backgrounds is statistically significant in some cognitive tests. The magnitude of these effects is comparable to findings in the US literature.
  The project’s next steps to address the statistical imprecision include a formal power analysis and exploring additional datasets to complement my analysis.



• Ambivalent spaces, ambivalent decisions: Spatial contexts, child characteristics, and parental educational strategies at the transition to secondary school
  Eva Fabian* (Martin-Luther University Halle-Wittenberg)
  ≡ Abstract

  Parental decisions at the transition to lower secondary school are not solely driven by children’s performance or family resources. Instead, they are also embedded in specific spatial contexts (Sixt 2013; Helbig 2010). This study builds on Annette Lareau’s concept of parental educational strategies (Lareau 2011) and current research on regional disparities in educational opportunities (Becker et al. 2024; Hartung 2017). It examines how different spatial settings shape educational decisions, with a focus on parents who act against institutional recommendations.
  The analysis uses the second starting cohort of the German National Educational Panel Study (NEPS Network 2024). It is complemented with official spatial classifications from the Federal Institute for Research on Building, Urban Affairs, and Spatial Development (BBSR). By combining settlement structure, rurality, and labour market region type, a differentiated spatial variable was constructed. These variable captures urban, semi-peripheral, and peripheral contexts in relation to their labour market characteristics.
  Results from multinomial regression models indicate that space is neither a linear advantage nor a disadvantage. Instead, it is a context where parental logics of action unfold (Sonnenburg 2019). Children with a recommendation for the Gymnasium who make the transition are more often found in peripheral regions, especially when parents hold higher educational resources. In these areas, fewer children fall into the group without recommendation and without Gymnasium transition. This suggests the Gymnasium is a viable option even for less resourceful families. Conversely, children with a recommendation whose parents opt against the Gymnasium are less common in semi-peripheral regions with weak labour markets. This pattern points to stronger conformity with institutional advice. In semi-peripheral regions with strong labour markets, parents without recommendation who choose the Gymnasium show higher conformity as well. Institutional pressure or alignment with school authority may be particularly pronounced here.
  The next step will investigate whether these patterns can be attributed to the uneven distribution of parental characteristics across spatial types and whether children’s attributes also influence decision-making. In this way, the study links structural spatial contexts with child-specific features, whose meaning may be translated differently into parental strategies depending on the region.
  Overall, the findings demonstrate that space is not merely a background for selection processes but actively moderates educational choices. Semi-peripheral and peripheral contexts emerge as ambivalent spaces, where both ambitious upward mobility and strong conformity with recommendations coexist – underscoring that educational decisions must be understood not only along class lines but also within spatial opportunity structures.

back to top

Session 9: Career Aspirations and Dream Job

Friday | December 12, 2025 | 09:30 am - 10:15 am (CET)

• Career aspirations of students with special educational needs – Effects of institutional conditions during lower secondary education
  Monja Schmitt (LIfBi Bamberg)
  Christoph Homuth (LIfBi Bamberg)
  Sebastian Bauer (LIfBi Bamberg)
  ≡ Abstract

  The proportion of students with special educational needs (SEN) has been increasing for years (bpb, 2024). More than 50% of all students with SEN still attend a special school (Klemm et al., 2023). Attending a special school can be perceived as a risk to further education and training (Blanck, 2020; German Institute for Human Rights, 2023; Jochmaring, 2019; Menze et al., 2021). There is insufficient research on students with SEN who are included in mainstream schools. Existing national studies suggest that inclusive students with SEN tend to be more successful in transitioning to vocational training than students with special education needs. (Nepomyashcha, 2021). One possible explanation for that could be better academic performance (Kocaj et al., 2014). Another explanation could be socially unequal allocation to special schools (Kölm et al., 2017). It remains unclear whether the more favorable transitions are exclusively due to selection effects and/or differences in competence, or whether the school learning context also has additional influences.
  One important mechanism could be career aspirations, which play a decisive role in determining whether and how students enter training and employment after school (Schoon & Lyons-Amos, 2016). However, whether students with SPF who attend inclusive schools develop higher aspirations has been largely uninvestigated to date. Therefore we compare the socio-structural level of idealistic career aspirations (operationalized as ISEI of career aspirations) at the beginning of grade 5 and at the end of lower secondary school in grade 9 for the groups of special education students (as exclusively learning students with SPF) and inclusively learning regular students with SPF at secondary schools, secondary school branches or vocationally oriented school types, and at non-grammar schools in federal states with two-tier secondary school systems. Regular students without SPF serve as our benchmark. Our database is based on the NEPS (NEPS Network, 2024, 2025) Starting Cohorts 4 and 8. We use entropy balancing (Hainmueller, 2012) to account for selection effects between the individual student groups.
  Preliminary analyses show that students with SEN in inclusive settings have higher aspirations than students at special schools, both at the beginning and at the end of lower secondary education. The next step will be to include the students' skills. Our findings to date support the thesis that the differences between students learning in inclusive settings and those at special schools are due not only to selection effects but also to the type of schooling (special school vs. inclusive education).



• Uncertain occupational aspirations in adolescence: Determinants and coping strategies
  Ralf Minor (Friedrich Schiller University Jena)
  Kathrin Leuze (Friedrich Schiller University Jena)
  ≡ Abstract

  The German education and occupational system is characterized by a high degree of occupational specificity and low permeability (Allmendinger, 1989; Hamilton & Hurrelmann, 1994). In this context, employment trajectories of aspirations and choices are consequential. Traditional career choice theories (e.g., Gottfredson (1981)) and empirical studies model such decisions to pursue a particular profession as a process of narrowing down possible options until a final choice is made. This perspective fails to consider that many young people are unable to express any career aspirations by the time they finish school. While this is a normal stage of development, not everyone goes through it successfully and even socio-structural disparities may arise (Gutman & Schoon, 2012). The proportion of young people in Germany who experience this uncertainty (31%) is particularly high compared to the OECD average (15%, based on PISA 2018 data). Common theoretical approaches have neither been adapted to this circumstance, nor are their comprehensive empirical findings on this phenomenon (previous results are relatively old (Fend, 1991) or drawn from selective samples (Dietrich & Kracke, 2009; Dombrowski, 2013; Minor et al., 2023; Richter, 2016)). So, we are conducting a three-year research project with a focus on this group in Germany, starting this year. We will present and discuss an overview of the issues surrounding this phenomenon and outline the approaches we are considering for empirical analysis.
  Our general questions are: Who are these young people and how do they cope with the upcoming transition to vocational training or higher education with regard to their uncertainty?
  Our strategy involves in a first step to typify this group in terms of socio-structural factors, contextual aspects, and temporal patterns of experiencing occupational uncertainty. To this end, we adapt Supers' (1953) developmental approach to this phenomenon. When considering control strategies for young people facing the deadline for transitioning from school to work or from school to higher education in a second step, we draw on the theoretical assumptions of motivation and self-regulation (Heckhausen, 2000). To follow these pathways, our operationalization’s make extensive use of data from the third and fourth cohorts of the National Education Panel (Blossfeld & Roßbach, 2019; NEPS Network, 2024a, 2024b), including the recoding of open-ended responses regarding adolescents' career aspirations.



• Do apprentices change less if they train in their dream job? The case of work values and personality
  Annalisa Schnitzler* (BIBB Bonn)
  ≡ Abstract

  According to the neo-socioanalytic theory (Roberts, 2006), people select themselves into jobs based on their personality and then adapt further to fit into the selected environment due to socialization processes. This leads to rather homogeneous environments in terms of workers’ attitudes, values and personality. Apart from specifics of individual organisations, working environments can be grouped into different types that due to their respective task profiles differ in their members’ characteristics, as described by Holland (1997). Holland argues that when choosing an occupation, people try to enter a working environment type that fits their personal profile of interests, values, etc.. If a person enters an occupation they have deliberately chosen, the gap in the person-environment fit and, following Roberts and Robin (2004), thus the press to adapt should be smaller than for a person having to take up work in an occupation they do not fancy.
  This paper investigates whether adolescents starting a dual apprenticeship in an occupation they declare as their desired profession change less during their apprenticeship than those who do not say so about their training occupation. The focus will be on work values and personality, both of which have been shown to change through entering the work force (Lechner et al., 2017; Roberts & Wood, 2006) and due to specific job characteristics (Johnson, 2001). Looking at an indirect measure of person-environment fit by considering how much the occupation one has entered is considered the desired profession adds to this perspective.
  Analyses are based on NEPS Starting Cohort 4. The sample are those adolescents who successfully completed a dual apprenticeship by wave 9. I calculated changes in work values and the Big Five personality dimensions between grade 10 and grade 9 respectively and four/five years later (wave 9). More than one third does not change at all in each of the six values, for the personality dimensions this holds true for every fifth to fourth participant. In a first step I use t-tests to compare participants who stated that their training occupation was their desired profession to those who negated this with regard to these changes. Preliminary results reveal significantly less change for those in their desired profession only for two values (security and good working hours) and for agreeableness. In a next step, I will also take into account the direction of change and differences between occupations.

back to top

Session 10: Ability Tracking - New Approaches

Friday | December 12, 2025 | 10:35 am - 11:50 am (CET)

• On the wrong track? A comparison of the difference-in-difference method and the value-added approach using the example of the effects of duality in ability tracking with data from the NEPS
  Hartmut Esser* (University of Mannheim)
  ≡ Abstract

  The question of the effects of ability tracking remains unresolved. One reason for this may be, that international comparative studies, such as PISA in particular, have lacked relevant variables to explain the differences between countries and systems: at least the cognitive abilities and cognitive composition of school classes. The difference-in-difference (DiD) method offered a solution: controlling for the “unobserved heterogeneity” of countries or systems (see, for example, Hanushek and Wößmann 2006). The distinctive feature compared to other methods, such as the value-added (VA) approach, possibly combined with multilevel analyses (MLA) for school effects, is that DiD analyses refer theoretically and empirically to countries or systems as aggregates, while VA/MLA approaches to individuals embedded in specific contexts. This raises the question of whether the aggregate relationships found using the DiD method can be transferred to the individual level. Particular efforts have been made to address this issue (Eisenkopf 2007, Bach and Fischer 2021), and the theoretical justifications for differentiation explicitly assume that there may (also) be interaction effects with other influences.
  This raises the question of whether the aggregate relationships found using the DiD method can be transferred to the individual level. This is the problem of a possible “ecological fallacy.” It refers to some general prerequisites for applying the DiD approach: the assumption of parallel trends and stability between levels, i.e., the absence of cross-level interactions (Angrist and Pischke 2009). However, neither of these can be assumed for ability tracking, even in theory: The early separation of educational pathways creates incentives to make a special effort in elementary school (Eisenkopf 2007, Bach and Fischer 2021), and the theoretical justifications for differentiation explicitly assume that there may (also) be cross-level effects with other levels.
  The contribution consists of two parts: a more detailed elaboration of the theoretical and methodological issues on DiD and VA/MLA described above and an empirical application comparing the two approaches based on a reanalysis of a study on the effects of two-tiered system in ability tracking in the German federal states using NEPS data by Matthewes (2022). The result is a significant correction of Matthewes' findings, which can be explained by the incompleteness of his theoretical approach and the inappropriateness of the attempt to identify system effects using the DiD method compared to a VA-MLA approach.



• Simulating the effects of ability tracking on student allocation during the transition to secondary schools
  Marius Kaffai (University of Stuttgart)
  Raphael Heiberger (University of Stuttgart)
  ≡ Abstract

  The effects of ability tracking in German schools remain contested and continue to be the subject of heated scholarly and political debate. On the one hand, the work of Sørensen (1970), Hallinan (1994), Galindo-Rueda and Vignoles (2014), Fu and Mehta (2018), and Esser (2021) has gradually given rise to a coherent theoretical model of ability tracking. On the other hand, a lack of data remains a central problem in analyzing its effects. While initial empirical investigations have been conducted using NEPS data, these are limited not only by small case numbers but also by the fact that data for certain counterfactual scenarios simply cannot be collected.
  To address these limitations, this paper develops an agent-based model that allows for the systematic examination of ability tracking even in scenarios where empirical evidence is lacking. In its current version, the model provides a tool to analyze how ability tracking shapes student allocation at the transition to secondary schools. It further enables the simulation of varying legal frameworks as well as differences in parental aspirations and biases in teachers’ assessments of students’ cognitive abilities. The model is empirically calibrated using NEPS data (SC3).
  An initial application of the model illustrates how the effects of varying degrees of bindingness in school recommendations are shaped by biases in ability perception. Findings suggest that high bindingness can reduce social inequality particularly when socioeconomic status plays a minor role in teachers’ recommendations. Conversely, if parents accurately assess their child’s abilities, excessive bindingness without improvements in the tracking system may actually increase inequality at the transition. Additional scenarios show that legally preventing downward deviations from a recommendation for the Gymnasium — which in reality are used by around 18% of students from lower tracks — not only reduces inequality in the transition process but also substantially improves learning conditions in lower-track schools, while leaving Gymnasium classrooms largely unaffected.
  The paper concludes with an outlook on planned extensions of the model to incorporate network effects within classrooms on student achievement, as well as a discussion of methodological limitations. Beyond these specific findings, the study demonstrates how agent-based simulation can serve as a powerful complement to large-scale assessment data in educational research, providing new insights into counterfactual dynamics that are otherwise inaccessible.



• Content-specific insights into tracking effects: Applying Differential Item Functioning in Change (DIF-C) to NEPS mathematics data in secondary school
  Patricia Martinkova (Czech Academy of Sciences, Charles University Prague)
  Marie-Ann Sengewald (Friedrich Alexander University Erlangen-Nürnberg, LIfBi Bamberg)
  ≡ Abstract

  Research on educational tracking has traditionally focused on differences in total test scores between tracks. However, tracking may influence specific skills in ways that are not visible when analyzing only aggregated scores. In a recent work, Martinková et al. (2020) introduced the method of differential item functioning in change (DIF-C), showing that tracking effects can emerge at the item level even when total gains appear equivalent. Their study of Czech data demonstrated that selective tracks fostered progress in certain higher-order thinking skills despite no overall score differences.
  In this contribution, we extend this approach to German data from the National Educational Panel Study (NEPS) with longitudinal assessments from Grade 5 to 9 (Starting Cohort 3). We apply DIF-C to the assessments of mathematical competence across secondary schools to investigate whether and how tracking in the German school system differentially affects content-related learning gains. Specifically, the NEPS design, with repeated measures of the same students across grades and a coherent assessment framework with the same content area for constructing competence tests, allows us to link content-related responses over time and attribute differences in content-specific growth to instructional environments.
  Theoretically, DIF-C combines psychometric perspectives on differential item functioning with educational theories of instructional sensitivity. By detecting content areas where students from different tracks, despite comparable baseline abilities, diverge in their probability of improvement, the method offers a fine-grained tool for identifying curricular or pedagogical features associated with tracking. We hypothesize that DIF-C will be able to reveal systematic disparities at a finer-grained level, suggesting targeted benefits of tracking that are hidden at the total-score level.
  Beyond substantive findings, we also explore methodological extensions of DIF-C within the NEPS context, such as multiple time points, clustering, and measurement error correction. Our results underscore the importance of analysis beyond aggregated scores for a nuanced understanding of tracking implications.

back to top

Session 11: A Look at Academic Education

Friday | December 12, 2025 | 10:35 am - 11:50 am (CET)

• The influence of meso-level characteristics of universities on trajectories of self-reported student health
  Christian Deindl* (TU Dortmund University)
  ≡ Abstract

  The influence of macro-level contextual factors (e.g., welfare state regimes) on individual health is well documented, while the impact of characteristics of meso-level societal institutions is seldom explored.
  We will examine trajectories of self-rated health of students in relation to contextual characteristics of the university they were enrolled at.
  The National Educational Panel Study (NEPS) provides information about the self-rated health of university students in Germany (from 2005 to 2023), including contextual information about single universities. Using multilevel analysis, we estimate the impact of universities characteristics on students’ health taking individual level determinants of health into account.
  The sample comprised 79,682 observations of 11,130 respondents from 172 universities. Self-rated health of students correlated with socio-economic status of parents, health-behaviour and self-esteem. There was evidence that students’ health was better in universities with a lower teacher-students ratio. Interaction analyses showed that this correlation was more pronounced for students coming from a more privileged background.
  The correlation between institutional characteristics of universities and students’ health indicates that meso-level factors influence health and health inequality. It underlines the importance of health promoting structures in the setting university.



• BAföG’s blind spot: Parental wealth and educational inequality in Germany
  Samuel Jalalian* (Ludwig Maximilian University of Munich)
  ≡ Abstract

  BAföG, the German Federal Student Aid Program, aims to ensure that young people can pursue education aligned with their abilities and interests – even when their parents are unable to finance it. Although parental income is considered in BAföG’s eligibility criteria, parental wealth is not. This omission has attracted little attention in research or public debate, but it may affect educational inequality if parental wealth matters for financing higher education beyond income.
  This paper addresses this gap by examining whether parental wealth affects (a) the perceived ability of parents and students to finance higher education, (b) the application for and receipt of BAföG support, and (c) whether wealth effects differ from income effects across these outcomes. Building on the income and insurance functions of wealth, I argue that wealth matters for education financing beyond income. For example, homeowners can allocate more income to their children’s education than families who rent. Regarding BAföG, subjective expected utility theory points to ambiguous effects: wealthier families may anticipate lower benefits but also lower participation costs of BAföG.
  The analyses draw on data from NEPS Starting Cohort 4. Regression models include parental wealth, income, education, occupation, age, and number of children. Outcomes cover cost perceptions of higher education (e.g., whether parents view higher education as a financial burden), intentions to apply for BAföG, and actual receipt.
  The results point to inverted U-shaped effects of parental wealth on all outcomes, net of income and other controls. Middle-wealth families perceive the costs of higher education as highest and are most likely to apply for and receive BAföG, while both poorer and wealthier families show lower levels. Parental income shows different patterns: it is linearly associated with lower perceived costs of higher education and lower BAföG receipt but follows an inverted U-shape for application intentions. Overall, parental income is the stronger predictor across all outcomes, yet parental wealth has an independent, statistically significant influence.
  The findings highlight a blind spot in BAföG’s design. Parental wealth shapes parents’ (perceived) ability to finance higher education, yet students from the poorest households apply for and receive support less often than those from middle-wealth families. These results suggest that BAföG’s effectiveness in reducing educational inequality could be improved by incorporating parental wealth into eligibility criteria. Future analyses will assess the robustness of these findings with SOEP data and explore underlying mechanisms.



• How do university subjects differ by their cognitive requirement levels? Empirical analyses of the cognitive requirements of university subjects for study aptitude assessment
  Anike Schild* (Federal Employment Agency Nuremberg)
  Erik Sengewald (Federal Employment Agency Nuremberg)
  Nicolas Sander (Federal Employment Agency Nuremberg)
  Dorothea Klinck (Federal Employment Agency Nuremberg)
  ≡ Abstract

  General cognitive ability is the best job-unspecific predictor of occupational success (Sackett et al., 2022). Klemmert et al. (2023) showed that occupational trainings differ in the level of general cognitive ability they require. Considering occupations’ cognitive requirements in relation to individuals’ cognitive aptitudes is therefore beneficial in occupational aptitude assessment. Given that cognitive ability also predicts academic success (e.g. Schneider & Preckel, 2017), our research aim was to examine whether university subjects differ in their cognitive requirement levels and, if so, to group them into levels accordingly. For practical application in study aptitude assessment, we aimed to develop a tool to compare individual aptitudes with the cognitive requirements of chosen fields of study. We develop and use the tool in the Occupational Psychological Service of the German Federal Employment Agency.
  To address the research questions, we empirically analysed the NEPS data from starting cohort 5 (first-year students; NEPS-Netzwerk, 2024; Blossfeld & Roßbach, 2019). We used scores of the test for mathematical skills as a proxy for cognitive ability. An analysis of their relation to academic success revealed that scores in mathematical skills differ significantly between students who completed their studies successfully and those who dropped out (t(3386)=-8.42, p<.001, d=-0.45) with successful graduates scoring higher. Furthermore, a multilevel model confirmed that mathematical skills are a significant predictor of final grades among successful graduates.
  To assess the cognitive requirements of university subjects, we analysed the ability of successful graduates (cf. Klemmert et al., 2023). We grouped participants by their university subjects and calculated statistics describing the distribution of the weighted and z-standardised normed test scores for mathematical skills for each university subject. Based on those results, we clustered university subjects into five cognitive requirement levels. Cluster means range from m=-0.72 for level 1 to m=0.92 for level 5 and differ significantly, confirming that university subjects differ by their cognitive requirements. However, individual ability varies considerably among graduates of subjects with the same level of requirements. For study aptitude assessment, this emphasises that other factors related to academic success must also be considered.
  Additional qualitative expert ratings of subjects’ cognitive requirements correlate highly with the empirical findings. The expert ratings were also used to assign university subjects for which no empirical data was available to the levels.

back to top

Symposium

Friday | December 12, 2025 | 12:30 am - 02:10 pm (CET)

• Response behavior in digital learning and assessments
  Marie-Ann Sengewald (Friedrich Alexander University Erlangen-Nürnberg, LIfBi Bamberg)
  ≡ Abstract

  Modeling response behavior in large-scale assessments and app-based learning provides powerful opportunities to understand the processes underlying test performance and learning. With the growing use of digital assessments and app-based learning programs, response behavior that can be investigated with process data (e.g., repeated mistakes, disengagement, rapid guessing) has gained increasing attention as an important factor influencing the validity of competence measures as well as educational outcomes. This symposium presents four contributions that investigate response behavior and combine advanced psychometric modeling strategies with substantive insights in educational research. The first presentation investigates repeated mistakes in app-based learning, connecting behavioral patterns such as wheel-spinning and gaming to learning gains (Kristensen, Torkildsen, & Andersson). The second contribution examines how response-time indicators of disengagement can enhance the analysis of differential training effects for investigating the heterogeneity in learning outcomes (Sengewald, Torkildsen, Kristensen, & Ulitzsch). The third presentation focuses on large-scale assessments and evaluates various approaches to account for rapid guessing in the estimation of plausible values, highlighting implications for achievement gap analyses (Welling, Zink, & Gnambs). The final contribution introduces a Bayesian multi-state item response model with regime-switching to model test-taking engagement, offering a flexible framework that incorporates respondent, item, and contextual predictors (Aßmann, Gnambs, & Pape). The insights and modeling strategies presented in this symposium offer innovative perspectives on detecting and preventing careless and insufficient effort responding as well as on examining the value of response behavior analysis for advancing app-based learning and cognitive assessments.



• Repeated mistakes in app-based learning
  Jarl Kleppe Kristensen (University of Oslo)
  Janne von Koss Torkildsen (University of Oslo)
  Björn Andersson (University of Oslo)
  ≡ Abstract

  Educational apps can provide engaging learning contexts, which are easy to implement at scale. However, learning gains depend heavily on how learners interact with the tasks and content of the apps. Two of the behavioral patterns that may inhibit learning are gaming the system and wheel-spinning. Gaming behavior refers to rapid guessing or hint abuse with the aim of progressing without engaging with app content. Wheel-spinning refers to getting stuck despite persistent engagement and effort. Both behaviors could result in learners repeating the same mistakes several times. This could either lead to an underexposure to educational content (gaming), or to an overexposure to incorrect input (wheel-spinning). There is, however, a lack of research on why learners repeat mistakes and how it relates to learning.
  To address this gap in the literature, we conducted two studies where we examined repeated mistakes in app-based word learning. Both studies used data from 363 second graders participating in the experiment group of an RCT evaluating the efficacy of Captain Morph, a Norwegian morphology-based word learning app. In Study 1, we examined a) whether the propensity to repeat mistakes was stable across the eight weeks of the intervention, and b) how repeated mistakes related to prior knowledge and learning gains. We used a latent growth curve model (LGM) to examine changes in the propensity to repeat mistakes, and a structural equation model (SEM) to examine the relationship between repeated mistakes, prior knowledge and learning gains. In Study 2, we examined possible associations between task and person characteristics and the number of repeated mistakes, using a generalized linear mixed model (GLMM).
  The LGM in Study 1 indicated that children’s propensity to repeat mistakes is relatively stable over time, and not subject to fatigue or familiarization even after eight weeks of daily app use. Furthermore, the SEM showed that children with lower prior knowledge of multimorphemic words were likely to make more repeated mistakes, and that children who made more repeated mistakes scored lower at posttest than children with equal prior knowledge who made fewer repeated mistakes. In Study 2, the GLMM indicated that task design and content, as well as task position, were related to the number of repeated mistakes children made. Child characteristics associated with repeated mistakes included gender, language background, prior knowledge, word reading efficiency, and non-verbal ability.
  The results will be discussed considering previous research on wheel-spinning and gaming the system.



• Evaluating the differential effectiveness of app-based trainings
  Marie-Ann Sengewald (Friedrich Alexander University Erlangen-Nürnberg, LIfBi Bamberg)
  Janne von Koss Torkildsen (University of Oslo)
  Jarl Kleppe Kristensen (University of Oslo)
  Esther Ulitzsch (University of Oslo)
  ≡ Abstract

  Evaluation studies of educational interventions typically report average effect estimates, which provide insights into the overall effectiveness of a training program. However, focusing solely on average group comparisons can mask substantial heterogeneity in intervention outcomes. Examining differential effects for specific subgroups or even individual learners offers a more nuanced understanding of who benefits most from an intervention and supports more targeted treatment decisions.
  An underexplored information source for differential effect analysis is the rich process data generated by digital learning apps. To investigate the benefit of process data, suitable modeling strategies and are needed. In this talk, we connect psychometric research on modeling response times in cognitive assessments with evaluation research on the heterogeneity of training effects. Specifically, we demonstrate how response-time-based indicators of disengaged responding can improve differential effect analyses.
  We apply this approach to data from Torkildsen et al. (2022), who developed a Norwegian morphology-based word-learning app and evaluated its effectiveness in a randomized controlled trial with 717 second-grade students. We investigate the heterogeneity of training effects in relation to the pre-treatment characteristics of the students and assess the added value of disengagement indicators by examining their impact on explained outcome variance and effect size differences.
  Our findings identify baseline characteristics that predict greater benefits from the training and highlight how different modeling strategies for disengagement indicators influence the conclusions. Beyond the practical insights into the utility of process data, our results illustrate the potential of advanced modeling strategies for improving differential effect analyses in intervention research.



• Comparing different approaches of (not) accounting for rapid guessing in plausible values estimation
  Jana Welling (LIfBi Bamberg)
  Eva Zink (LIfBi Bamberg)
  Timo Gnambs (LIfBi Bamberg)
  ≡ Abstract

  Educational large-scale assessments provide information on achievement gaps, informing politics and shaping educational decisions. However, some of these gaps might partly reflect differences in test-taking motivation rather than in actual achievement. Existing approaches for mitigating the distorting effects of rapid guessing – a common form of low test-taking motivation – focus mainly on manifest competence scores and latent variable models, ignoring the fact that many researchers rely on plausible values as easily applicable competence estimates adjusted for measurement error.
  To close this gap, this study aims to (a) determine the impact of rapid guessing on achievement gaps based on plausible values and (b) propose and evaluate approaches of accounting for rapid guessing in the estimation of plausible values.
  In a simulation study that systematically varied the rapid guessing rates and the size of the achievement gap, four approaches were compared: (1) a baseline model did not account for rapid guessing, (2) a person-level model incorporated rapid guessing as a respondent characteristic in the background model, (3) a response-level model filtered responses with item response times lower than a predetermined threshold, and (4) a combined model combined the person- and response-level approaches. Results show that the response-level and combined model performed best across all conditions, while accounting for rapid guessing solely on the person level did not suffice. An empirical example using NEPS data from the university student cohort (N=478) supported these findings and demonstrated the applicability of all approaches in practice. Recommendations for future research practice are given to improve achievement estimation.



• A Bayesian multi-state item response model with regime-switching for response engagement
  Christian Aßmann (LIfBi Bamberg, University of Bamberg)
  Timo Gnambs (LIfBi Bamberg)
  Markus Pape (Ruhr University Bochum)
  ≡ Abstract

  Disengaged responses in achievement tests can be identified using item response times, where unusually short times indicate rapid guessing rather than solution-oriented test-taking behavior. Educational assessments have often analyzed response times using finite Markov mixture models, also referred to as regime-switching models, modeling distinct types of test-taking behavior as latent state variables. These models were usually estimated via maximum likelihood, and the handling of missing values was not a core issue of the analysis. As the model structure involves latent variables and the data potentially involves missing values, a Bayesian estimation approach appears suitable. Moreover, the state probabilities may depend on observable respondent, item, or contextual characteristics.
  Taking this into account, we propose a fully Bayesian multi-state item response model implemented via MCMC methodology. Our model framework allows for testing how various predictors influence the state probabilities. We discuss a corresponding prior setup, model evaluation through posterior predictive assessments, and a validation using simulated data with convergence diagnostics. In an empirical illustration, we analyze responses on a reading comprehension test from 1,932 German university students in the German National Educational Panel Study.

back to top