The treatment of control variables in population science: Deconfounding in Demography 1995-2020

Note: This paper was presented in an oral session on causal inferences at the 2022 Population Association of America in Atlanta, GA. The practice of causal inference experienced significant growth throughout the social sciences in the past twenty-five years. However, it is unclear how these advances have been incorporated into quantitative sociology. To address this, we reviewed publications in Demography in six 5-year waves from 1995 to 2020. We assessed how authors describe the process of deconfounding statistical models. We found that the treatment of confounding factors was highly varied. Inconsistency in the descriptions of variable selection made it difficult to create a reliable quantitative measure. Still, there was minimal evidence that authors coalesced around the improvements in causal inference, at least in how they described control variables. While alignment between demography and improvements in causal inference would help consolidate and improve the science, this alignment is not evident. We explain some benefits of greater consistency and address how peer-review journals could affect the rationale behind variable selection in quantitative sociology.