Anchoring and Subjective Belief Distributions

Low anchor: The subjects are asked if they believe that the a variable is lower or higher than the value of a low anchor
High anchor: The subjects are asked if they believe that the a variable is lower or higher than the value of a high anchor

Mean elicitation The subjects are asked to guess the value of a variable.
Subjective belief distribution (SBD) elicitation: The subjects are asked to input the distribution of a variable using the click-and-drag method (Crosetto and de Haan, 2023)

In a robustness check, we use the number input interface (Crosetto and de Haan, 2023) to elicit distribution.

2023-05-11

2024-07-31

-mean of variable
-spread of variable, measured using its variance and coefficient of variation

-skew of variable
-match of variable to true value

We start by giving all subjects information about central concepts in the survey such as the average and the frequency distribution of a set of values. We also let subjects answer a few control questions that are designed to check that the subjects understand the concepts.

The next step is to inform subjects that we have collected information about a given price distribution. It will be a distribution of historical prices for a one-night hotel room in the city of Rome. This distribution was selected since we want the subject to have some idea about the price distribution but not too much information about it.

Subjects are then randomly allocated to one of the following five treatments, where subjects’ beliefs are elicited with monetary incentives.

1. Control elicitation of Distribution (CD). We elicit the SBDs of the hotel room prices using the procedure suggested by Crosetto and de Haan (2023). There is no anchoring in this treatment.

2. Low anchor elicitation of Mean (LM). The subjects are first asked if they believe that the average price is lower or higher than the value of a low anchor. After that subjects are asked to guess the average price of the hotel room.

3. Low anchor elicitation of Distribution (LD). The subjects are first asked if they believe that the average price is lower or higher than the value of a low anchor. After that, we elicit the SBDs of the hotel room prices by the same technique as in CD.

4. High anchor elicitation of Mean (HM). The subjects are first asked if they believe that the average price is lower or higher than the value of a high anchor. After that subjects are asked to guess the average price of the hotel room.

5. High anchor elicitation of Distribution (HD). The subjects are first asked if they believe that the average price is lower or higher than the value of a high anchor. After that, we elicit the SBDs of the hotel room prices by the same technique as in CD.

We then run a 2nd elicitation so that subjects who received the LM and HM will receive CD and subjects who received CD, LD, and HD will be asked to guess the average hotel prices. (Hence, they receive the same treatment as in LM and HM but without any anchor, which will be denoted as M.) In connection with the 2nd elicitation, we will ask how certain the subjects are about their estimations following the elicitation of cognitive uncertainty (CU) by Enke and Graeber (2022). The 2nd round elicitations are of secondary importance and will only be used in the exploratory analysis.

After the treatment, all subjects answer questions about demographics, cognitive reflection, investment behavior, and financial literacy.

In a robustness check, we use the number input interface (Crosetto and de Haan, 2023) to elicit distribution.

Qualtrics randomizer will be used to allocate subjects into one of the five treatments

Individual

No

750 individuals

750 individuals

150 individuals per treatment

PAP Study 1

PAP Study 2

PAP Study 1 AMENDED

We investigate how the anchoring effect—a well-established cognitive bias—influences subjective belief distributions. While prior research extensively examines the impact of anchoring and other biases on point estimates, their effect on the underlying distribution and its higher moments remains unexplored. Using two pre-registered online experiments (N = 1467) and two elicitation methods, we find that anchoring impacts not just the mean, but also higher moments of belief distributions. Notably, the traditional anchoring effect in means diminishes when eliciting distributions rather than point estimates. We also find that the elicitation method matters: inattentive participants generate spiky distributions when manually entering numbers for many bins, whereas they generate flat distributions using a click-and-drag interface. These findings show that cognitive biases can extend beyond point estimates, and that the elicitation technique may impact results especially among inattentive participants.

Håkan J. Holm, Margaret Samahita, Roel van Veldhuizen, Erik Wengström, Anchoring and subjective belief distributions, Journal of Economic Behavior & Organization, Volume 240, 2025, 107304, ISSN 0167-2681, https://doi.org/10.1016/j.jebo.2025.107304. (https://www.sciencedirect.com/science/article/pii/S0167268125004214)

https://doi.org/10.1016/j.jebo.2025.107304