Elsevier

Economics Letters

Volume 145, August 2016, Pages 45-47
Economics Letters

A random shock is not random assignment

https://doi.org/10.1016/j.econlet.2016.05.022Get rights and content

Abstract

A random shock excludes reverse causality and reduces omitted variable bias. Yet a natural experiment does not identify random exposure to treatment, but the reaction to a random change from baseline to treatment. A lab experiment comparing higher certainty with higher severity of punishment for stealing (holding the expected value of the intervention constant) shows that the difference between the effects of a random shock and random assignment can be pronounced.

Section snippets

The issue

With the credibility revolution, empirical economics has changed fundamentally (Leamer, 1983, Manski, 1995, Angrist and Pischke, 2010). No good publication takes correlation for causation. It seems the enthusiasm for instrumental variables has also somewhat cooled down. Too often their exogeneity is questionable, or they are weak (Bound et al., 1995). A way out are random shocks. They create a “natural experiment” (Meyer, 1995). Researchers have warned that they are not true experiments if the

Design

In the experiment, participants are randomly assigned to groups of two. Each participant is endowed with 20 tokens. Each decides in the active role whether she wants to take any of the passive participant’s tokens. Afterwards, (using the strategy method) in every period the active role is randomly assigned. Participants know they can lose tokens if are found out to have taken tokens from their counterpart. The experiment is framed neutrally and speaks of “taking”, “checking”, tokens that are

Results

For welfare, the sustainable deterrent effect is most relevant. In the experiment, one way to test for it is comparing choices in the first round. In this round I neither find a treatment effect on the frequency of stealing,3 nor on the stolen amount. Alternatively one may compare choices in the entire first phase (of 8 periods). This informs about the deterrent effect on would-be criminals with growing experience. In Table 1,

Conclusions

Mimicking a natural experiment in the lab demonstrates that a random shock is not a panacea. One identifies reactions to a change in circumstances. This change need not be a good proxy for the presence or the absence of the circumstance in question. Essentially the random shock is just an instrument. In the example (with the first change) one would conclude that there is a certainty effect on the frequency of stealing, but a severity effect on the stolen amount. Yet there is no effect at all on

Acknowledgments

Helpful comments by David Jäger, Jonathan Klick, Adrian Hillenbrand and Anna Kochanova on an earlier version, excellent research assistance by Rafael Schwalb, and programming of the experiment in zTree by Anastasia Niechaieva are gratefully acknowledged. The experiment has been funded from the regular budget of the Bonn Max Planck Institute. The Hamburg Econ Lab has kindly run the experiment.

References (18)

  • Olaf Bock et al.

    hroot

    Eur. Econ. Rev.

    (2014)
  • Gary Charness

    Self-serving cheap talk

    Games Econom. Behav.

    (2000)
  • Beatrice Rammstedt et al.

    Measuring personality in one minute or less

    J. Res. Pers.

    (2007)
  • Joshua Angrist et al.

    The credibility revolution in empirical economics

    J. Econ. Perspect.

    (2010)
  • George Antunes et al.

    The impact of certainty and severity of punishment on levels of crime in American states

    J. Crim. Law Criminol.

    (1973)
  • John Bound et al.

    Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variable is weak

    J. Amer. Statist. Assoc.

    (1995)
  • Christoph Engel et al.

    dhreg, xtdhreg, bootdhreg. Programs to implement double hurdle regression

    Stata J.

    (2014)
  • Christoph Engel et al.

    Who is afraid of the stick?

    Rev. Behav. Econ.

    (2015)
  • Urs Fischbacher

    z-Tree

    Exp. Econ.

    (2007)
There are more references available in the full text version of this article.

Cited by (10)

View all citing articles on Scopus
View full text