From cross sectional multi level modelling to longitudinal analysis
Project funded by The British Academy
Investigators: David Bartram (PI), Patrick White
Our core idea for this project pertains to research questions asking about the impact of a country-level variable on some sort of individual level outcome. For example: what is the relationship between inequality and status anxiety? For questions with that structure, most researchers use multi-level modelling. A longitudinal analysis seems impossible because there are no individual-level panel data covering a wide range of countries (panel datasets are almost always national entities).
What we typically get, then, is a cross-sectional analysis. Most researchers understand that the results then give us an ‘association’ that can’t be interpreted in causal terms (though sometimes an implication along these lines is evident).
Our core claim is that the obstacle to a longitudinal analysis isn’t lack of data – instead, the obstacle is likely found in implicit assumptions about the control variables ostensibly needed for the models. In most instances, it is almost certainly unnecessary to include individual-level controls – because individual-level factors are very unlikely to influence the country-level variable whose impact we are investigating. These individual-level variables do not fit a clear criterion for selection of controls (in particular, control for antecedents of the focal independent variable).
If we are confident about that view, then a longitudinal approach becomes feasible. For the dependent variable (e.g. status anxiety), we can take data from repeated cross-sectional surveys (e.g. the European Quality of Life Survey) and calculate country-level averages – creating a country-level panel. We can then take time-corresponding measures of inequality (plus any needed country-level controls).
This data structure facilitates a ‘within’ analysis (aka ‘fixed effects’), showing how changes in inequality are (perhaps) followed by changes in status anxiety. This analytical form would be more effective in mitigating against omitted variable bias (for time-constant country-level factors). A causal interpretation is then more reasonable.
The project, carried out in 2022, involved two core components. 1) A research assistant, Sadie Chana, reviewed a large number of ‘in-scope’ studies (i.e., addressing research questions having the indicated structure). The RA paid particular attention to methods sections, to identify the justifications given (or indeed absent) for key analytical decisions. 2) A workshop inviting researchers to Leicester to explore our perspective and approach.
The project has produced the following outputs:
- A report summarising the work of the research assistant
- A draft manuscript identifying ways the perspective might be adopted in quantitative methods courses
- A recording of a talk given by David Bartram to the Centre for Multilevel Modeling at the University of Bristol
These documents will be added here soon.
Additional materials already published include the following journal articles:
- Bartram, D., & Jarochova, E. (2022). A longitudinal investigation of integration/multiculturalism policies and attitudes towards immigrants in European countries. Journal of Ethnic and Migration Studies, 48(1), 153–172. DOI: 10.1080/1369183X.2021.1922273
- Bartram, D. (2022). Does inequality exacerbate status anxiety among higher earners? A longitudinal evaluation. International Journal of Comparative Sociology, 63(4), 184–200. DOI:10.1177/00207152221094815