Social scientists use time series cross-sectional data to explain how macro-social units (typically countries) change over time. But how can political or social change be studied when the outcome of interest is measured at a disaggregated, individual level—like policy preferences, religiosity, political participation, or social capital? Aggregating disaggregated data is an option, but doing so throws away valuable information, and risks committing an ecological or individualistic fallacy. Multilevel (also known as mixed) models, fitted to repeated cross-sectional survey data, have been used to illuminate the individual-level consequences of national-level variables such as GDP/capita and income inequality. But the full potential of such models for explaining cross-national differences in change over time has yet to be realised, particularly because current practice conflates longitudinal and cross-sectional effects. This paper describes straightforward methods for disentangling these two kinds of effects, and for identifying whether change over time is due to time-varying or time-invariant national characteristics (or both). Group mean-centring permits testing of time-varying covariates, while growth curves fitted at the country level allow for change to be a function of time-invariant characteristics. Both techniques can allow for change to be due either to period effects or to cohort replacement, and they are mathematically simple to implement with existing software. They assume that the time series involved are stationary, but this assumption can be tested. The paper concludes with a brief application, using data from the World Values Survey, to the study of why religiosity has declined (or secularism increased) in some countries and not others.
