17 Nov 2025
13:00–14:00
Online
Presenters: Nicolas Sommet (Centre LIVES) & Oliver Lipps (FORS)
Abstract:
Fixed-effects modeling is a powerful tool for estimating within-cluster associations in cross-sectional data and, in particular, within-participant associations in longitudinal data.
In the first part of the session, we introduce a new pedagogical primer for social scientists. The primer uses fictional data, clear graphics, and runnable code in R, Stata, and SPSS. It begins by demonstrating how fixed-effects modeling applies to clustered cross-sectional data, then extends to longitudinal data, with guidance on modeling interactions and discussing limitations related to time-varying confounding and reverse causality. It also introduces more advanced models, including first-difference models, time-distributed fixed-effects models, and within-between models. Finally, the session devotes additional time to simulation-based guidelines for determining the sample size needed to detect within-participant effects of plausible magnitude with adequate power.
In the second part of the session, we address the fact that many social processes are likely to be asymmetric. For example, the effect of unemployment on happiness is probably not the exact opposite than the effect of re-employment. However, this is the standard assumption of fixed effects modelling. First, we show how to model asymmetric effects in first difference models using 2-period data, before extending this to multiperiod data. Finally, we introduce asymmetric fixed effects models and compare them using the example of unemployment and happiness based on data from the Swiss Household Panel.