Examining unconditional preference revelation in choice experiments: a voting game approach

Abstract

If respondents are strategic, voting in choice experiments may violate the common modelling assumption that everyone votes for his unconditionally most-preferred alternative. This presents a challenge to accurately estimating welfare measures. We conduct a homegrown-value laboratory experiment designed to mimic a three-alternative choice experiment. Two key pieces of information not previously collected from participants are their (1) unconditionally most-preferred alternative and (2) their perception of the likely vote distribution. We are able to identify whether participants adopt a strategy of voting for their second-most-preferred alternative when their most-preferred alternative is believed likely to garner the fewest votes. We find that many more participants do not vote for their most-preferred alternative than theory predicts. We also test whether prompting participants to think about the likely vote distribution before voting affects their own vote. We find that prompting participants does not affect strategic voting. We find percentages of participants who do not vote for their most-preferred alternative, who cast strategic votes for their second most-preferred alternative, and who cast apparently ‘irrational’ votes are close to estimates from studies in political election settings. The presence of strategic and irrational votes should cause choice experiment researchers to rethink their model assumptions.

Keywords:

• choice experiments
• strategic voting
• lab experiment

Acknowledgements

The authors thank an anonymous reviewer for comments and suggestions which helped us to improve the manuscript.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. As an anonymous reviewer pointed out, the problem of strategic voting in CE is eliminated when there is only a single binary-choice question with only a single policy under consideration (Carson and Groves 2007). However, most choice experiments within the environmental economics literature contain choice sets with three alternatives (e.g. works cited herein). This study therefore investigates strategic voting under current standard practice.

2. Some compare the incentive compatible binomial-choice contingent valuation format with the generally incentive incompatible multinomial-choice format (McNair, Bennett, and Hensher 2011), while others compare hypothetical and binding CE (Lusk and Schroeder 2004; Taylor, Morrison, and Boyle 2010). Some find significant differences in value estimates (e.g. Taylor, Morrison, and Boyle 2010), while some do not (e.g. Lusk and Schroeder 2004).

3. The most closely related homegrown-value lab experiment we have found is Taylor, Morrison, and Boyle (2010), who examine voting outcomes under decision mechanisms with different incentives. However, they do not examine specific voting strategies, in particular, whether participants consider the likely votes of others when casting their own votes. On the contrary, they assume participants vote for their unconditionally most-preferred alternative. Neither do they elicit beliefs about the distribution of preferences or likely distribution of votes.

4. Tyszler and Schram (2011) find that second-favorite strategic voting depends on the closeness in preference of a participant's most and second-most preferred alternatives.

5. One reviewer commented that giving participants an endowment may affect their choices (see Harrison 2007). We did not design our experiment to test for this ‘house money’ effect, however we have no a priori hypotheses about how an endowment would affect strategic behavior in the experiment. Also, unlike as in some other types of public goods experiments, there was no possibility for our participants to keep their entire endowments.

6. Although CE respondents may not have pre-existing familiarity with the alternatives, survey designers generally strive to give respondents as much detail as is feasible to lower the chances of respondents voting under uncertainty.

7. We conducted four sessions of pretests in June 2012 with a wider range of choice sets than were used in the final experiment. Our goal in narrowing down the choice sets to be used in the final experiment was to avoid choice sets in which there would be little opportunity for strategic voting. While we can never be certain of selecting choice sets that maximise the likelihood of observing strategic voting (because this would involve knowing, ex ante, participant preferences, perceptions of the preference distributions, and voting strategies), the voting results of the final experiment (presented in a later section) indicate that we succeeded in avoiding preference distributions in which we would not expect any participants to not choose their unconditionally most-preferred item (e.g. all participants indifferent between two or more items or all participants disfavoring a particular item).

8. Throughout, we use the verb and noun ‘vote’ to refer to one's chosen alternative under the plurality rule specifically, and the verb ‘choose’ and noun ‘choice’ otherwise, including when we refer to their selections under the two rules collectively.

9. It is possible a participant's vote could create or break a tie for the item receiving the most votes. However, when asking about the perceived likely vote distribution, we believed most participants would not have a perception that was this precise. We therefore assume that if they believed a particular item would receive the fewest votes, they believed it would do so regardless of their own vote.

10. A test of parameter equivalence for separate dummy variables for the second and third choice sets was not significant.

11. In the discussion that follows, the word ‘participants’ technically refers to observations of a participant vote. Keep in mind that each of the 99 participants voted on three choice sets, so the 297 ‘participants’ are not necessarily unique individuals.

12. Niemi, Written, and Franklin (1992) estimate the number of voters who self-reported voting for ‘tactical’ reasons, not necessarily second-favorite strategic voting specifically.

13. On the contrary, it is well-known to be a dominated strategy in a plurality voting context (see e.g. Gutowski and Georges 1993, 230). However, Linhart (2009) shows that voting for one's least-preferred alternative can be a rational strategy in other voting contexts (e.g. when there are coalitions).

14. Meffert and Gschwend (2011) find that men are less likely to vote strategically than women.

15. Cheap talk has been used to address other issues, for example, hypothetical bias, in CE (see e.g. Carlsson, Frykblom, and Lagerkvist 2005; List, Sinha, and Taylor 2006).