Predicting time preference from social media behavior

Abstract

Time preference, or delay discounting, plays an important role in how we make health- and life-related choices. The varying tendency to prefer smaller, immediate rewards to larger, delayed rewards has been found to be strongly associated with addictive, economic, and criminal behavior.We use general digital footprints from social media (i.e., Facebook Likes) to predict individual hyperbolic discount rates, using a sample of 2,378 participants who shared their Likes and completed questionnaires on a monetary discounting task. We employ an automated machine learning approach for the prediction task.We identified a variety of easily interpretable topics that have strong correlations with both high and low time preferences. Using only Likes, we were able to predict individual discount rates with much higher-than-random accuracy (up to r=0.30). We could distinguish a future-oriented person from a more present-biased person with fair accuracy (up to 65%). Using Facebook Likes as predictors was much more accurate than using information on an individual’s substance use behavior, but combining both predictors slightly increased predictive accuracy.Predicting discount rates from social media behavior presents important opportunities for improving individual decision-making, but also comes with potential manipulation and privacy-related pitfalls.