Multilevel Models for the Analysis of Comparative (Longitudinal) Survey Data 1
|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)|
Social trust, the belief that most other people can be trusted, is remarkably stable over time. Using data from the first 6 rounds of the ESS, this paper examines if the crisis of 2008 has undermined social trust in 15 European countries. Results from Bayesian MCMC multilevel analysis suggest that long-term unemployment rates determine change in trust, whereas changes in GDP or unemployment rates have no impact. The presentation concludes with discussing the robustness of the findings. Special emphasis is paid to the small-N problem in cross-national research.
This paper presents two multilevel techniques to deal with international repeated cross-sectional surveys. The first one accounts for the comparative and dynamic nature of this data. It implies defining multilevel models with distinct levels for individuals and country-years combined (with contextual time-invariant and time-varying covariates). The need and choice of a third level of clustering is discussed. The second strategy implies the nesting of cross-classified random effects models within countries. This approach adds a comparative dimension to the study of APC effects, and also allows for the inclusion of time-varying/-invariant contextual covariates.
This paper shows that the multilevel models used in many published analyses of comparative longitudinal survey data are specified erroneously. They typically omit one or more relevant random effects, leading to downward biases in the standard errors. These biases occur even if the fixed effects are specified correctly; if the fixed effects are incorrect, erroneous specification of the random effects worsens biases in the coefficients. We illustrate these problems using Monte Carlo simulations and two empirical examples, and conclude with straightforward recommendations that will help applied researchers avoid such problems in the future.
As information generated by surveys and polls, sometimes decades old, becomes available, we are now able to perform “historical analyses” of how various socio-political attitudes have changed over time. The presentation shows how multilevel analysis can estimate change over time, taking into account the different goals and wordings of poll questions and the various features of the methods used. Three case studies are presented: a) Change in support for Quebec sovereignty over 40 years, b) Change in voting intentions for Obama in 2012, and c) Change in trust in institutions in Canada over 40 years.