Abstract
Psychologists are often interested in specific causal quantities (“What is the effect of reducing social-media scrolling by 20 minutes on well-being?”) but tend to focus on vague questions (“Does ‘time spent on social media’ predict well-being?”). There is a mismatch between the causal effects researchers care about and the associations and correlations they compute. In this article, we highlight consistency as a foundational requirement for causal inference that can bridge this mismatch and make statistical estimates causally interpretable. Consistency shows how causal quantities are interpretable only when (a) the causal quantity of interest—the causal estimand—is described precisely enough that it does not bundle different versions of the treatment (e.g., passive scrolling, reading distressing news, or chatting with friends on social media) that have different effects, and (b) the way that the treatment variable changes in the data matches this causal estimand of interest. We clarify these requirements and illustrate how to state and assess them in settings common in psychological research, with a focus on indirect experimental manipulations and multidimensional constructs. Making this step explicit helps researchers articulate their estimands more transparently and is a prerequisite for other concerns such as confounding and model specification.