Central tendency bias in belief elicitation

Abstract

We conduct an experiment in which subjects participate in a first-price auction against an automaton that bids randomly in a given range. The subjects first place a bid in the auction. They are then given an incentivized elicitation of their beliefs of the opponent's bid. Despite having been told that the bid of the opponent was drawn from a uniform distribution, we find that most of the subjects report beliefs that have a peak in the interior of the range. The result is robust across seven different experimental treatments. While not expected at the outset, these single-peaked beliefs have precedence in the experimental psychology judgments literature. Our results suggest that an elicitation of probability beliefs can result in responses that are more concentrated than the objectively known or induced truth. We provide indicative evidence that such individual belief reports can be rationalized by well-defined subjective beliefs that differ from the objective truth. Our findings offer an explanation for the conservatism and overprecision biases in Bayesian updating. Finally, our findings suggest that probabilistic forecasts of uncertain events might have less variance than the actual events.

Online belief elicitation methods

“…This type of central tendency bias has been documented using other belief elicitation methods as well, such as the QSR(Crosetto et al, 2020) …”

The effect of domain and framing on elicited risk aversion

“…identical but are attracted by the focal point when guessing. Notice that the focal point theory is in line with the central tendency bias: when subjects have to estimate or choose among different quantities they tend to exhibit a bias towards the mean of the distribution [46][47][48][49].…”

Objectifying the Measurement of Voter Ideology With Expert Data

“…Our results suggest that this may be (at least partly) unwarranted. This suspicion is fueled by the fact that there has been a comprehensive literature on the so-called 'central tendency bias' (also known as 'regression effect', see Stevens and Greenbaum 1966), a cognitive bias that induces respondents to choose the mean of the distribution they are presented with [see Crosetto et al, 2020]. This bias is particularly well-known for Likert scale items [Douven, 2018], but has long been known to affect many different kinds of responses [see, e.g., Hollingworth, 1910].…”