Combining probability forecasts: 60% and 60% is 60%, but likely and likely is very likely

Abstract

People employ fundamentally different strategies when combining probability forecasts depending on whether information is presented in a numeric or verbal format. While individuals typically average numeric forecasts—maintaining their own judgments near the mean of provided estimates—they tend to “count” verbal probabilities by treating them as cumulative signals. For instance, receiving multiple forecasts that an event is “likely” leads recipients to conclude the event is “very likely,” moving their personal judgments closer to certainty than any single advisor’s estimate. This phenomenon persists across various contexts, including probabilities above and below 50%, real-world events, and both simultaneous and sequential presentations of advice. These distinct combination strategies significantly influence downstream consumer decisions, such as purchase timing and product choices. Experimental evidence fails to consistently support the hypothesis that this behavior is driven by a belief that verbal forecasts provide more independent information or that numeric forecasts are less evaluable. The tendency to treat verbal labels as additive evidence rather than values to be averaged suggests a robust divergence in how uncertainty is processed and aggregated based on communication format. – AI-generated abstract.