Daniil Azarov¹, Adam N. Sanborn², Nick Chater², Robert L. Goldstone¹; ¹Indiana University, ²University of Warwick

According to the "wisdom of the crowd" effect, aggregating multiple estimates leads to a more accurate estimate than most of the individual estimates (Surowiecki, 2004). This effect is observed even when the same person provides multiple estimates (Vul & Pashler, 2008), though it is generally more pronounced when estimates come from different individuals (Dolder & Assem, 2017). Previous research indicates that encouraging individuals to make two highly divergent estimates can lead to greater accuracy when the estimates are averaged compared to providing two similar estimates (Herzog & Hertwig, 2009). Building on this, we hypothesized that the accuracy of averaged estimates would increase as the dependency between sampled estimates decreases, whether due to temporal separation, involvement of different individuals, or estimation of distinct stimuli. In the present study, 56 participants were tasked with estimating the number of objects displayed on a screen (e.g., 90 crackers). For each stimulus set, participants made two estimates either consecutively (e.g., Trials 1 and 2) or with a larger time gap (e.g., Trials 3 and 90). We then aggregated the estimates for each pair of stimuli. Our results show that averaging two estimates from the same individual was beneficial (compared to individual estimates) only when the estimates were separated in time. Additionally, averaging estimates from different individuals produced more accurate results than from the same subject. Finally, accuracy was maximized when the stimuli for which the estimates were made differed (e.g., estimating 90 crackers and 90 butterflies), rather than when the same stimulus was estimated twice, regardless of whether the estimates came from the same or different individuals. These findings underscore the importance of minimizing the sampling dependence between estimates — whether by separating estimates in time, using multiple individuals, or estimating different stimuli — in order to achieve the most accurate estimates.