The Only Thing That Can Stop Bad Causal Inference Is Good Causal Inference

Abstract

In psychology, causal inference—both the transport from lab estimates to the real world and estimation on the basis of observational data—is often pursued in a casual manner. Underlying assumptions remain unarticulated; potential pitfalls are compiled in post-hoc lists of flaws. The field should move on to coherent frameworks of causal inference and generalizability that have been developed elsewhere.

A (failed) attempt to falsify the alliance hypothesis of racial categorization: Racial categorization is not reduced when crossed with a nonalliance category.

“…The present research embodies two recent in the study of cognition: (1) an increased focus on theory-based testing, and not just the reporting of experimental effects (e.g., Gervais, 2021;Smaldino, 2020;Stewart & Plotkin, 2021;van Rooji & Baggio, 2021) and (2) a more explicit focus on causal inference (e.g., Rohrer, Schmukle, & McElreath, 2021). The goal in the present research program is to move beyond characterizing racial categorization as either a generally "fixed" or "flexible" experimental phenomenon (which, depending on one's choice of stimuli, it could be either), and to instead use experimentally-observed changes (or non-changes) in racial categorization as a tool for making inferences about underlying information-processing system(s) that are casually responsible for it.…”

Associations Between Vocal Arousal and Dyadic Coping During Couple Interactions After a Stress Induction

“…The established approach in psychological scientific publications is that one should not make causal statements based on observational data (i.e., correlation is not causation), yet new publications (Rohrer, 2018;Rohrer et al, 2021) suggest that this hinders progress in psychology and it would be much better to make implicit causal assumptions explicit (Grosz et al, 2020;Rohrer, 2018;Rohrer et al, 2021). This suggestion is not just based on research about causal inference over the last decades (Dawid, 1979;Pearl, 1993Pearl, , 2009Peters et al, 2017;Rubin, 1974) but has started to be accepted in psychology (Bullock et al, 2010;Grosz et al, 2020;Rohrer, 2018;Rohrer et al, 2021).…”

“…The established approach in psychological scientific publications is that one should not make causal statements based on observational data (i.e., correlation is not causation), yet new publications (Rohrer, 2018;Rohrer et al, 2021) suggest that this hinders progress in psychology and it would be much better to make implicit causal assumptions explicit (Grosz et al, 2020;Rohrer, 2018;Rohrer et al, 2021). This suggestion is not just based on research about causal inference over the last decades (Dawid, 1979;Pearl, 1993Pearl, , 2009Peters et al, 2017;Rubin, 1974) but has started to be accepted in psychology (Bullock et al, 2010;Grosz et al, 2020;Rohrer, 2018;Rohrer et al, 2021). In contrast to the rather problematic approach in psychology that results from a causal model (i.e., regression models) can only be interpreted as an association, the new approach allows interpreting the results as causal under the assumption that the causal model corresponds to the Data Generating Process (DGP), i.e., that the model reflects the underlying causal process in the real world (Rohrer, 2018), acknowledging that testing this assumption is difficult.…”

A Critical Evaluation of Theories and Their Empirical Testing Methods

“…For example, in the traditional research paradigm, results based on observational data are not interpreted causally (Grosz et al, 2020;Hernán, 2018;Rohrer, 2018). According to meta-researchers, this leads to incoherence: the presence or absence of a causal relationship between variables in a verbal theory may never be explicitly tested within the current research paradigm (Eronen, 2020;Fried, 2020;Grosz et al, 2020;Hernán, 2018;Maier et al, 2021;Pearl, 2009;Rohrer, 2018;Rohrer et al, 2021;Scheel, 2022).…”