Belief Elicitation with Multiple Point Predictions

We propose a simple, incentive compatible procedure based on binarized linear scoring rules to elicit beliefs about real-valued outcomes - multiple point predictions. Simultaneously eliciting multiple point predictions with linear incentives reveals the subjective probability distribution without pre-defined intervals or probabilistic statements. We show that the approach is theoretically as robust as existing methods, while adapting flexibly to different beliefs. In a laboratory experiment, we compare our procedure to the standard approach of eliciting discrete probabilities on pre-defined intervals. We find that elicitation with multiple point predictions is faster, perceived as less difficult and more consistent with a subsequent decision. We further find that multiple point predictions are more accurate if beliefs vary between participants. Finally, we provide experimental evidence that pre-defined intervals anchor reports.