Causal Inference
Interpretation of Associational Language in Research Noah Stovitz* Noah Stovitz Ian Shrier Jake Quilty-Dunn Jennifer Hill
Researchers often use ambiguous language, and even clearly written non-causal language may be misinterpreted as causal. The purpose of this study is to evaluate the effect of language on the interpretation of causality in research articles. We surveyed under- or recently graduated university students. We randomized them to see one of 5 different “linking words” between variables that might represents causes and outcomes (ordered by perceived level of causal implication: affects, increases, predicts, increased with, correlated with) within three different contexts which varied by a priori level of perceived causal relationship (exercise and dehydration: likely causal, study abroad and graduation: possibly causal or non-causal, born early in week and intelligence: likely non-causal). In the first 60 respondents with complete data, the proportion of respondents who reported the claim as having a strong causal implication was only 25% for affects (our a priori strongest causal implication), 53% for increases, 35% for predicts, 38% for increased with, and 22% for correlated with. The context affected the interpretation for 45% (27/60) of the participants. Among those participants, 19% (5/27) interpreted the claim as being more causal even though the context shifted from likely causal to likely non-causal. We conclude that interpretations of linking words are complex, with the likelihood of a true causal effect affecting the interpretation.