Everyday econometricians: Selection neglect and overoptimism when learning from others

How to cite

Barron, Kai; Huck, Steffen; Jehiel, Philippe. 2023. "Everyday econometricians: Selection neglect and overoptimism when learning from others." AEA RCT Registry. AEARCTR-0012503. https://doi.org/10.1257/rct.12503-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Kai Barron* , Steffen Huck , Philippe Jehiel

* Primary investigator

Registry ID
AEARCTR-0012503
DOI
https://doi.org/10.1257/rct.12503-1.0
First registered
Last updated
Status
Completed
Trial dates
2016-12-01 – 2023-04-30
Countries
Germany ()
JEL codes
C11, C90, D80, D83
Registry record
http://www.socialscienceregistry.org/trials/12503

Abstract

This is an ex-post registration of Experiment 1, conducted for the paper entitled "Everyday econometricians: Selection neglect and overoptimism when learning from others."

The paper reports the results from two sets of experiments. Both sets of experiments study whether individuals neglect selection effects when learning from data generated by others' decisions.

Experiment 2 was pre-registered in a separate registration file (https://www.socialscienceregistry.org/trials/10536).

Intervention

In Experiment 1: We have 5 treatment conditions.

Scenario 1: Selected: Participants observe selected data when learning from others' decisions - the only observe decisions that are implemented.

Scenario 2: Correlated. Participants again observe selected data when learning from others' decisions. However, the information that learners have is perfectly correlated with past decision-makers.

Scenario 3: Externality. Participants observe selected data when learning from others' decisions, but the degree of selection is exogenously increased by introducing more informed investors.

Scenario 4: Control. Participants observe non-selected data when learning from others' decisions

Scenario 5:Selected NoDGP. Similar to the selection treatment, except participants are provided with less information about the data-generating process.

Experimental design

The experimental design is described in detail in the paper (https://bibliothek.wzb.eu/pdf/2019/ii19-301.pdf and https://shs.hal.science/halshs-04154345/document).

Design details

Randomization method
The randomization is done by a computer programme (oTree).
Randomization unit
We have multiple layers of randomization. Our treatments in Experiment 1 were assigned at the experimental session level. Individuals within the session are then randomly allocated into groups of 2 or 3. Then, in each round, there is randomization in terms of the attributes of the investments that participants choose whether to invest or not invest in. Within each group, these investments are used to compile the databases of the participants to facilitate learning.
Clusters
120
Observations
240
Arms
Each of the five treatments has 48 participants, but due to the experimental design, the number of clusters differs. The cluster is defined by the number of participants who interact within a group. Therefore, in the treatments with groups of 3, we have 16 clusters, in the treatment with groups of 2, we have 24 clusters, while in the treatments where there is no interaction, we have 48 clusters. 16 clusters: Selected, Correlated, Selected NoDGP 24 clusters: Externality 48 clusters: Control
Primary outcomes
Investment propensity, the threshold for investing

Cited by

No citation data available for this registration yet.

Keywords