Umut Atasever
Research Subjects
Umut Atasever has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Alec I. Kennedy
International Association for the Evaluation of Educational Achievement (IEA), Hamburg, Germany
4 shared publications Nurullah Eryilmaz
4 shared publicationsRolf Strietholt
4 shared publications- Stefan Johansson
Department of Education and Special Education, University of Gothenburg, Gothenburg, Sweden
3 shared publications - Trude Nilsen
University of Oslo
3 shared publications Sabine Meinck
2 shared publications
Latest Publications
- 29 October 2026
- Published by SAGE Publications in European Educational Research Journal
Socioeconomic inequality, early tracking and civic knowledge: A cross-national analysis of adolescents in 43 education systems
This paper examines the potential association between early between-school tracking and socioeconomic inequalities in adolescents’ civic knowledge, a core component of democratic participation and social cohesion. Drawing on pooled data from the International Civic and Citizenship Education Study (ICCS) 2009, 2016 and 2022, we test whether early tracking affects socioeconomic disparities in civic knowledge scores of approximately 260,000 students across 43 education systems. Using three-level multilevel regression models, we find a strong and consistent association between socioeconomic status and civic knowledge. Early tracking is associated with lower average civic knowledge, whereas we find no evidence that it systematically moderates socioeconomic inequalities in civic knowledge. Robustness checks with alternative SES indicators confirm the stability of these findings. Additional analyses for a subsample of early-tracking countries reveal pronounced differences between academic and non-academic tracks, indicating within-system stratification. Overall, the results indicate that early tracking is associated with within-system stratification in civic knowledge and lower average civic knowledge at the system level, but not with a statistically detectable widening of the socioeconomic gradient. This pattern points to the domain-specific nature of tracking-related inequalities: while early tracking has often been associated with stronger socioeconomic disparities in traditional academic domains, we find no evidence that this association extends to civic knowledge.
- 29 October 2026
- Published by SAGE Publications in European Educational Research Journal
Socioeconomic inequality, early tracking and civic knowledge: A cross-national analysis of adolescents in 43 education systems
This paper examines the potential association between early between-school tracking and socioeconomic inequalities in adolescents’ civic knowledge, a core component of democratic participation and social cohesion. Drawing on pooled data from the International Civic and Citizenship Education Study (ICCS) 2009, 2016 and 2022, we test whether early tracking affects socioeconomic disparities in civic knowledge scores of approximately 260,000 students across 43 education systems. Using three-level multilevel regression models, we find a strong and consistent association between socioeconomic status and civic knowledge. Early tracking is associated with lower average civic knowledge, whereas we find no evidence that it systematically moderates socioeconomic inequalities in civic knowledge. Robustness checks with alternative SES indicators confirm the stability of these findings. Additional analyses for a subsample of early-tracking countries reveal pronounced differences between academic and non-academic tracks, indicating within-system stratification. Overall, the results indicate that early tracking is associated with within-system stratification in civic knowledge and lower average civic knowledge at the system level, but not with a statistically detectable widening of the socioeconomic gradient. This pattern points to the domain-specific nature of tracking-related inequalities: while early tracking has often been associated with stronger socioeconomic disparities in traditional academic domains, we find no evidence that this association extends to civic knowledge.
- Open Access
- 28 July 2026
- Published by Center for Open Science in EdArXiv
Striving for More Inclusiveness: Understanding and Addressing Student Exclusion in TIMSS
International large-scale assessments such as TIMSS depend on representative samples and inclusive coverage of the target population to generalize sample results to the underlying population and ensure cross-national comparability over time. In recent cycles, student exclusion rates have increased, raising concerns about representativeness and bias if lower-performing students are disproportionately excluded.Using TIMSS 2023 data, we examine the performance of students classified as “excluded” but who nonetheless participated in the assessment and generated achievement data. We investigate whether they demonstrated meaningful engagement with the assessment, their average performance, and whether their exclusion biased national achievement estimates across 17 education systems at grade four and eight at grade eight.Results show that excluded students generally achieve lower scores and are more likely to be flagged as too low for reliable estimation. However, such cases remain relatively rare, and most excluded students engage meaningfully with the assessment, with many reaching the Low International Benchmark. Because excluded students score systematically below included students, their exclusion introduces a modest upward bias in national achievement, particularly in systems with higher exclusion rates.Overall, this study provides evidence that many excluded students could have been included in achievement estimation without compromising psychometric integrity, highlighting the need to refine exclusion definitions, strengthen accommodation guidance, and enhance the validity and inclusiveness of international assessments.
- 28 July 2026
- Published by Center for Open Science in EdArXiv
Open AccessStriving for More Inclusiveness: Understanding and Addressing Student Exclusion in TIMSS
International large-scale assessments such as TIMSS depend on representative samples and inclusive coverage of the target population to generalize sample results to the underlying population and ensure cross-national comparability over time. In recent cycles, student exclusion rates have increased, raising concerns about representativeness and bias if lower-performing students are disproportionately excluded.Using TIMSS 2023 data, we examine the performance of students classified as “excluded” but who nonetheless participated in the assessment and generated achievement data. We investigate whether they demonstrated meaningful engagement with the assessment, their average performance, and whether their exclusion biased national achievement estimates across 17 education systems at grade four and eight at grade eight.Results show that excluded students generally achieve lower scores and are more likely to be flagged as too low for reliable estimation. However, such cases remain relatively rare, and most excluded students engage meaningfully with the assessment, with many reaching the Low International Benchmark. Because excluded students score systematically below included students, their exclusion introduces a modest upward bias in national achievement, particularly in systems with higher exclusion rates.Overall, this study provides evidence that many excluded students could have been included in achievement estimation without compromising psychometric integrity, highlighting the need to refine exclusion definitions, strengthen accommodation guidance, and enhance the validity and inclusiveness of international assessments.
- Open Access
- 28 July 2026
- Published by Center for Open Science in EdArXiv
An Examination of the Performance of Variance Estimators in International Large-Scale Assessments
Background: The primary objective of this study is to examine the performance of widely used sampling variance estimators in international large-scale assessments (ILSAs): Balanced Repeated Replication (BRR) and Jackknife Repeated Replication (JK2; Half and Full variants). We also investigate the impact of Fay modification factors (10%, 30%, and 50%) and evaluate Bootstrapping as an alternative variance estimator for clustered sampling. Methods: A Monte Carlo simulation mirroring the sampling design of a TIMSS Grade 4 population was conducted. Probability samples were repeatedly drawn under a two-stage stratified cluster sampling design with schools as primary sampling units and classes as secondary sampling units. The simulation varied school sample sizes (30, 50, 100, and 150 schools), considered scenarios with odd numbers of schools within explicit strata, and examined different treatments of school non-response. For each scenario, 1,000 samples were drawn, and the performance of the variance estimators was assessed by comparing the estimated sampling errors with the "true" sampling errors approximated from the sampling distributions of the estimates.Results: The results showed that, for smooth statistics such as means, BRR without Fay adjustment, JK2, and Bootstrapping produced highly accurate variance estimates, with JK2 showing the highest precision across conditions. For non-smooth statistics, including percentiles, Bootstrapping and BRR outperformed JK2 in both accuracy and precision. The JK2 Half and Full variants performed virtually identically, suggesting no precision advantage of the computationally more intensive Full variant. Fay adjustments did not improve JK2 performance, whereas BRR performed better with lower Fay factors (10% and 30%) than with the conventional 50% adjustment. Under school non-response scenarios, common approaches to variance strata construction led to systematic over- or underestimation of sampling variance.Conclusions: Overall, the study provides empirical guidance on the application of replication-based variance estimators in ILSAs under varying statistical conditions and sampling designs.
- 28 July 2026
- Published by Center for Open Science in EdArXiv
Open AccessAn Examination of the Performance of Variance Estimators in International Large-Scale Assessments
Background: The primary objective of this study is to examine the performance of widely used sampling variance estimators in international large-scale assessments (ILSAs): Balanced Repeated Replication (BRR) and Jackknife Repeated Replication (JK2; Half and Full variants). We also investigate the impact of Fay modification factors (10%, 30%, and 50%) and evaluate Bootstrapping as an alternative variance estimator for clustered sampling. Methods: A Monte Carlo simulation mirroring the sampling design of a TIMSS Grade 4 population was conducted. Probability samples were repeatedly drawn under a two-stage stratified cluster sampling design with schools as primary sampling units and classes as secondary sampling units. The simulation varied school sample sizes (30, 50, 100, and 150 schools), considered scenarios with odd numbers of schools within explicit strata, and examined different treatments of school non-response. For each scenario, 1,000 samples were drawn, and the performance of the variance estimators was assessed by comparing the estimated sampling errors with the "true" sampling errors approximated from the sampling distributions of the estimates.Results: The results showed that, for smooth statistics such as means, BRR without Fay adjustment, JK2, and Bootstrapping produced highly accurate variance estimates, with JK2 showing the highest precision across conditions. For non-smooth statistics, including percentiles, Bootstrapping and BRR outperformed JK2 in both accuracy and precision. The JK2 Half and Full variants performed virtually identically, suggesting no precision advantage of the computationally more intensive Full variant. Fay adjustments did not improve JK2 performance, whereas BRR performed better with lower Fay factors (10% and 30%) than with the conventional 50% adjustment. Under school non-response scenarios, common approaches to variance strata construction led to systematic over- or underestimation of sampling variance.Conclusions: Overall, the study provides empirical guidance on the application of replication-based variance estimators in ILSAs under varying statistical conditions and sampling designs.
- 13 July 2026
- Published by Elsevier in Studies in Educational Evaluation
- Vol. 90
School climate and student achievement: Evidence from TIMSS longitudinal
- 13 July 2026
- Published by Elsevier in Studies in Educational Evaluation
- Vol. 90
School climate and student achievement: Evidence from TIMSS longitudinal
- Open Access
- 7 July 2026
- Published by Elsevier in Studies in Educational Evaluation
- Vol. 90
Teacher quality and mathematics achievement gains: Evidence from the TIMSS longitudinal study
- 7 July 2026
- Published by Elsevier in Studies in Educational Evaluation
- Vol. 90
Open AccessTeacher quality and mathematics achievement gains: Evidence from the TIMSS longitudinal study
- Open Access
- 29 June 2026
- Published by Elsevier in Studies in Educational Evaluation
- Vol. 90
Student characteristics and mathematics achievement growth: Evidence from the TIMSS longitudinal study
- 29 June 2026
- Published by Elsevier in Studies in Educational Evaluation
- Vol. 90
Open AccessStudent characteristics and mathematics achievement growth: Evidence from the TIMSS longitudinal study