In regression (a common statistical practice used in social science research) we often attempt to predict the outcome of a given dependent measure (the DV) based on what we know about other measured variables theoretically related to the DV (the IVs). This common regression method has one problem though: We are predicting values for data that we have already collected. What if we were to engage in actual prediction? That is, what if we attempted to predict the values of a DV that is unknown? How might we do this and what would be the benefit?
This was a fascinating talk presented by Liz Page-Gould of the University of Toronto at the Future of Social Psychology Symposium!
I love the idea of actually predicting the future. Practically, future prediction means that our theoretical models must be able to generalize across samples and studies (theories that do not generalize won’t predict much). Prediction means that theories must also be more sophisticated—instead of focusing on a single predictor of a given DV, models must take care to account for the multiple factors that predict individual outcomes. Focusing on prediction also means that our field must build a cumulative science—updating and integrating existing theoretical models to create predictions that perform better than more incomplete versions. Instead of creating new theories, a predictive science would need to carefully evaluate the predictive validity of new theories before they are integrated into the literature, or deployed as policy. Neat!