Not-so-average after all: Individual vs. aggregate effects in substantive research
2025-11-20
In non-linear models, the effect of a given variable cannot be gauged directly from the associated coefficient. Instead, researchers typically compute the average effect in the population to assess the substantive significance of the variable of interest. Based on the average response, analysts often make policy recommendations that are to be implemented at the individual level (i.e. the unit of analysis level). Such extrapolations, however, can lead to gross generalizations or incorrect inferences. The reason for this is that the mean may obscure a large variation in individual effects, in which case the real-world applicability of the average value is limited. Correctly interpreting the average response may prevent unwarranted extrapolations but does not solve the problem of the lack of practical relevance. Particularly when cases carry special meaning (e.g. countries), the political and socioeconomic relevance of research findings should be assessed at the individual level. This article outlines the conditions under which aggregation to mean is problematic, and advocates a case-centered approach to model evaluation. Specifically, we advise researchers to compute and report the quantity of interest for each case in the data. Only by seeing the full spread of cases can the reader assess how well the average summarizes the population. Our approach allows researchers to draw more meaningful inferences, and makes the connection between research and practical applications more realistic.