Abstract Intergroup contact has been promoted as a powerful means to improve intergroup relations, but causal evidence for the effectiveness of contact in affecting outgroup attitudes is scarce. One important and increasingly used approach to estimate causal effects in contact research relies on naturalistic longitudinal survey data. However, causal inference based on longitudinal data depends strongly on the control of (un)observed confounders. The cross-lagged panel model (CLPM), commonly used in contact research, is an analytical model that has been criticised for biasing causal inference by insufficiently accounting for unobserved confounders, and advanced statistical models have been proposed, including the full-forward (FF-CLPM) and random-intercept CLPM (RI-CLPM). To explore the extent of potential bias between the three models introduced by the varying level of control for unobserved confounders, we investigated the robustness of longitudinal positive and negative contact-attitude effects in a systematic re-analysis using 31 datasets from 21 published studies ( N = 33,348). Using internal meta-analysis, we found substantial discrepancies between the models: While the CLPM provided significant longitudinal effects of both non-valenced/positive and negative intergroup contact on outgroup attitudes, analytical models that more strongly controlled for the effects of unobserved confounders found no such longitudinal evidence (FF-CLPM and RI-CLPM), but substantial between-person correlations which cannot be causally interpreted (RI-CLPM). Thus, the potential of intergroup contact to longitudinally affect outgroup attitudes in naturalistic survey data remains questionable. We discuss theoretical and methodological implications of these results and outline potential routes for future research to develop a more robust and informative longitudinal (contact) research.
Friehs et al. (Wed,) studied this question.