Learning about One's Self

How to cite

Schwardmann, Peter; Yaouanq, Yves. 2018. "Learning about One's Self." AEA RCT Registry. AEARCTR-0003060. https://doi.org/10.1257/rct.3060-2.1

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

Peter Schwardmann* , Yves Yaouanq

* Primary investigator

Registry ID
AEARCTR-0003060
DOI
https://doi.org/10.1257/rct.3060-2.1
First registered
Last updated
Status
Completed
Trial dates
2018-06-11 – 2018-06-28
Countries
Germany ()
Registry record
http://www.socialscienceregistry.org/trials/3060

Abstract

Previous studies document both present bias and people's naivete about being subject to present bias. However, it is puzzling that naivete that pertains to everyday behavior and a stable trait of the individual can persist. Why do people not learn about their bias? Our experiment seeks to provide a first step to resolving this puzzle empirically. We investigate whether people are able to learn about their present bias if they have the opportunity to engage in an unpleasant task repeatedly. To this end, we develop a framework that allows us to quantify what it means to learn optimally from one's own behavior and to compare individuals' observed learning to this benchmark.

In the experiment, subjects can complete two unpleasant tasks at two separate dates in the future. We elicit subjects' beliefs before and after the first task. During the first elicitation, we elicit a subject's joint prior distribution over her behavior at each of the two dates. After she either completes or fails to complete the first task, we elicit her beliefs about the likelihood of completing the second task. In a treatment, we vary whether the second task is the same task subjects completed at the first date, or a different one.

The resulting dataset allows us to study the extent to which people are able to learn from their past behavior. Moreover, we can investigate three potential drivers of people's failure to learn. First, people may underestimate how informative their behavior tomorrow is about their behavior the day after tomorrow. That is, they may not understand that behavior is driven by deep time preferences and, hence, underappreciate the correlation between their behavior at future dates. Second, people may interpret what task completion today implies for their task completion tomorrow in an upwardly biased way. To investigate this hypothesis, we ask whether a subject's belief after observing her task completion at the first date are compatible with her prior and its implied Bayesian posterior. Third, people may struggle to transport what they learn about their self-control problem in one environment into a (slightly) different environment. By varying the nature of the second task, we ask whether the previous two impediments to learning are more severe if the decision-making environment changes.

Intervention

We investigate whether people are able to learn about their present bias if they have the opportunity to engage in an unpleasant task repeatedly. We develop a framework that allows us to quantify what it means to learn optimally from one's own behavior and to compare individuals' observed learning to this benchmark.

See pre-analysis plan for further details.

Experimental design

Subjects are given the opportunity to complete two unpleasant tasks at two separate dates in the future. We elicit subjects' beliefs before and after the first task. During the first elicitation, we elicit a subject's joint prior distribution over her behavior at each of the two dates. After she either completes or fails to complete the first task, we elicit her beliefs about the likelihood of completing the second task. In a treatment, we vary whether the second task is the same task subjects completed at the first date, or a different one.

Design details

Randomization method
All randomisations are done by a computer, either in real time or before the experiment starts.
Randomization unit
Individual
Clusters
200
Observations
200
Arms
200
Primary outcomes
Beliefs about task completion and implied naivite (Hypotheses 1, 3 and 5 ). Bias in perceived informativeness of behaviour (Hypotheses 2 and 4). See the pre-analysis plan for details on all of our measurements and the construction of key variables.
Secondary outcomes
See pre-analysis plan

Cited by

No citation data available for this registration yet.

Keywords