David Gordon

Introduction

There is a vast body of literature that documents the seemingly irrational phenomena called loss aversion, a subset of prospect theory. Loss aversion is the idea that losses loom larger than otherwise equivalent gains, making it so that an agent would be unwilling to give up something currently in their consumption bundle for something else not currently in their consumption bundle, even though in the absence of both goods and faced with the choice between the two the agent would be indifferent between the two goods. Loss aversion has been shown to be responsible for irrational behavior in a variety of settings, some of them with extremely high stakes such as the PGA tour. Although the effects of loss aversion are well documented when the losses are internal, to my knowledge there are no studies that document if loss aversion is present when one is acting as an agent for another. In the real world we are given many opportunities to act as agents for others, and so if loss averse behavior exists in such situations it might be wise to set up an incentive structure for the agent so that the effects of loss aversion are mitigated. One salient example of being an agent for others are ballot initiatives such as whether or not to permit something like gay marriage because the vast majority of the population is not directly affected by the outcome, but is asked to act as an agent for others in making decisions. The inspiration for the project came from California’s Prop 8 (2008) that sparked controversy over the change in wording. Originally the wording presented the issue without a status quo. The attorney general changed the wording of the ballot measure so that it highlighted the fact that a right was being eliminated, and thereby saying that the status quo was that the right was already granted, which was true, but only in a limited number of municipalities in the state. We wanted to see if such wording changes mattered.

Methods/Results: To test whether loss aversion was present when losses were external to agents we started by using data on Same Sex Marriage ballot initiatives by county. Using Ordinary Least Squares regression we built a predictive model for the share of voters in any county that would vote to support same sex marriage based on demographic variables about the county, and taking into account the region of the state and the share of voters who voted for each of the candidates in the last presidential election. Using this model we were able to account for 69% of the variation in vote shares for each of the counties that had a same sex marriage ballot initiative. Our best estimates suggest that highlighting the loss aversion aspect of this particular issue increased voter support for same sex marriage by 4.7%. This result was significant at the 90% level.

This result although interesting was far from convincing. California was the only state with a biased wording and so putting in a dummy for biased wording is in essence the same thing as including a fixed effect for the state of California, and so this result may be due to unobservable differences that are unique to the state of California and may be unrelated to the wording of the ballot initiative. The project is ongoing, but we plan to find similar data for other issues such as medical and recreational marijuana ballot initiatives and see how loss averse wording affects the outcome of the vote.

To further increase the relevance and robustness of our results we conducted our own experiments using surveys and Amazon’s Mechanical Turk. The surveys asked respondents 5 policy questions about marijuana, tribal gaming, voter identification, same sex marriage, and the ability of the mentally handicapped to vote. The manipulation was that the question that the respondent received was randomized so that the some respondents received a prompt as to the current state of the law

while others received no such prompt. We wanted to see if those who were prompted with the current state of the law as already granting a right were less likely to vote to forbid some sort of behavior or government sanction. We collected a small sample of 300 responses for each issue. So far the results do not seem to be conclusive and we are still working on certain data quality issues as well as issues of how to make the sample more nationally representative since Mechanical Turk samples are characteristically more liberal and younger than the population as whole.

Going forward we plan to increase the sample size and include a measure of the amount of exposure and/or interest that a person has in an issue. We hope that with a larger sample, and a more representative sample we will be able to uncover results that suggest the extent to which loss averse behavior is present even when losses are external.

Discussion

If we can show that agents acting in behalf of others are loss averse, then this will have widespread implications. It will mean that once granted, a right is much harder to revoke than it was to forbid before the right was given. However this finding could have implications in realms other than in ballot initiatives. Indeed, if a portfolio manager is loss averse then they may take unnecessary risks when the market is down in hopes of getting back to the break-even point. Employees of a corporation may be unwilling to exchange efficiency gains in one aspect of the business for smaller decreases in efficiency in another aspect of their work. Indeed if loss aversion exists when one is acting as an agent, then virtually all activities may be evaluated on loss averse and not utilitarian criteria. If this is the case, barring an implausible paradigm shift, it would be necessary to restructure incentives so that the pain of losses is not out of proportion to the actually utility decrease associated with these losses in order to achieve optimal decision making in business, public policy and in a host of other situations