Forecasting long-run causal effects

Randomized controlled trials (RCTs) are the gold standard in causal inference. However, it is difficult for decision-makers to rely on RCTs when evaluating interventions that affect long-run outcomes. We elicit 25,980 forecasts from 505 participants to test the ability of people to predict the long-run outcomes of RCTs whose results were unknown to participants. Experienced forecasters, academic experts, and laypeople predict the outcomes of seven RCTs, each of which measured long-run outcomes at time horizons of at least five years. Experienced forecasters and academics were more accurate than laypeople, and experienced forecasters outperformed academics, at least in part due to their superior calibration. Aggregated forecasts outperform individual forecasts, showing a clear wisdom-of-the-crowd effect, but neither experienced forecasters nor academics consistently outperform simple benchmarks. We see no evidence of improved forecasting accuracy from randomly providing participants with additional information about good forecasting practices, details about the RCT interventions, or specifics about the local study context.