Tuesday 14th July      Wednesday 15th July      Thursday 16th July      Friday 17th July     

Friday 17th July, 13:00 - 14:30 Room: L-101

Multilevel Models for the Analysis of Comparative (Longitudinal) Survey Data 3
Convenor	Dr Alexander Schmidt-catran (Institute of Sociology and Social Psychology, University of Cologne )
Coordinator 1	Professor Bart Meuleman (Centre for Sociological Research, University of Leuven)

Session Details

Multilevel models have become predominant in analyses of comparative survey datasets, where respondents are clustered in higher-level units like countries or regions. Such models have also long been fitted to longitudinal data, where repeated observations are clustered within units. Additionally, researchers are fitting multilevel models to data that are clustered both ways, such as multiple waves of surveys whose respondents are nested in countries or regions each observed multiple times. These comparative longitudinal survey datasets should be useful resources for studies of social change in the broadest sense, and for drawing inferences previously based on only cross-sectional analyses. This session welcomes papers using multilevel models for the analysis of cross-sectional data, longitudinal data and, in particular, data that is clustered both ways. Papers might address recent methodological advances; present illuminating or innovative applications in some field of the social sciences; and/or discuss limitations and challenges that remain.

Paper Details

1. Multilevel modeling when clusters are heterogeneous. A Monte Carlo comparison of mixed random intercept and slope models, cluster-robust OLS, and two-step approaches
Professor Johannes Giesecke (Humboldt-University Berlin)
Mr Jan Paul Heisig (WZB Berlin Social Science Center )
Dr Merlin Schaeffer (WZB Berlin Social Science Center )

We conduct a series of Monte Carlo simulations to evaluate the performance of three different modeling strategies for multilevel data for the realistic case that effects of controls are allowed to vary across the level two units. We compare random intercept and slope models, pooled OLS with cluster-robust standard errors, and two-step approaches. Our results show that models neglecting cluster heterogeneity can be dramatically less efficient. Only mixed effects models with the correct random effects structure and two-step approaches do not suffer from dramatic efficiency losses. We end with concrete recommendations for applied researchers.

2. Schools vs. classrooms? Sampling design in the OECD’s PISA study
Professor David Reimer (Aarhus University, Department of Education)
Mr Bent Sortkær (Aarhus University, Department of Education)

In contrast to the sampling design of other major achievement studies that sample entire classrooms, the OECD’s PISA study randomly selects from schools. Here we explore to what extent this sampling decision has consequences for research about substantive phenomena that occur at the level of classrooms. Drawing on the so called PISA-Copenhagen study, an extension study of PISA which sampled entire classes, we compare multilevel estimates about the association between disciplinary climate and student test achievement. To this end we aggregate student-self reports about disciplinary climate at the level of schools, and classes, respectively and compare results.

3. How large are school and teacher effects on science achievement? A multilevel modeling perspective
Dr Liang-ting Tsai (Graduate Institute of Educational Information & Measurement, National Taichung University of Education)
Professor Chih-chien Yang (Graduate Institute of Educational Information & Measurement, National Taichung University of Education)

A three-level Hierarchical linear modeling (HLM) was employed to assess the influence of factors at student, classroom, and school level on the science performance of eighth-grade Taiwanese students. The sample in this study comprised 5,042 eighth-grade Taiwanese students nested in 153 classrooms from 150 schools that participated in the Trends in International Mathematics and Science Study (TIMSS) 2011. The results showed that the science performance of eighth-grade Taiwanese students is driven largely by individual factors. Classroom-level factors accounted for a smaller proportion of the total variance in science performance than did school-level factors.

4. Remitting civic participation. A MMMM approach
Dr Bogdan Voicu (Romanian Academy)
Dr Bogdan Voicu (Romanian Academy)

A set of multiple membership multilevel models if employed to test for the changes in the emigration societies due to, on one hand, the returning migrants, and, on the other hand, to mediated contagion with other societies due to migration. Civic participation, said to be embedded in cultures of participation, is used as example for testing. Its measurement is restricted to membership in associations and participation in protest actions. The approach finds its roots in the literature on social remittances, the studies on transnationalism, and the broad social capital debate.

5. Analyzing Redistributional Preferences: An Integrative Multilevel Approach for Cultural and Structural Explanations
Mr Sebastian Hülle (University of Bielefeld, Faculty of Sociology)

In contrast to prior research, this study analyses redistributional preferences by applying not only structural but also cultural approaches and both of them on micro- and macro-level to avoid omitted variable bias.
Applying multi-level modelling, the results indicate a negative of the objective inequality (Gini-coefficient), questioning the median-voter-hypothesis. But using a subjective measure of inequality, there is a positive effect.
While egalitarian justice attitudes increase and anti-egalitarian justice attitudes decrease redistributional preferences. This holds true for individual justice attitudes as well as for dominant justice-ideologies on macro-level.