![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
Abstract
Although citizens are often exposed to conflicting communications from political elites, few studies examine the effects of conflicting information on the quality of citizens’ decisions. Thus, I conduct experiments in which subjects are exposed to conflicting information before making decisions that affect their future welfare. The results suggest that a version of Gresham's Law operates in the context of political communication. When a credible source of information suggests the welfare-improving choice and a less credible source simultaneously suggests a choice that will make subjects worse off, subjects make worse decisions than when only the credible source is available. This occurs because many subjects base their decisions upon the less credible source or forgo participation. This occurs mostly among unsophisticated subjects, who are more easily led astray. These findings reveal important limits to the effectiveness of credible information sources and suggest how political campaigns might strategically use conflicting information to their benefit.
Notes
1. In economics, Gresham's Law refers to the principle that “bad money drives out good money.” In political science, CitationPopkin (1991, p. 79) refers to Gresham's Law of political information—the tendency for small amounts of personal information about candidates to drive more relevant political information out of consideration. Here, Gresham's Law refers to the less credible source of information driving the more credible source out of consideration for some subjects.
2. CitationNicholson (2008) studies the effects of conflicting signals from political parties.
3. For example, CitationCizmar and Layman (2009) suggest that abortion may not be an “easy” issue, contrary to how others typically classify it. CitationMaggiotto and Wittkopf (1981) question traditional classifications of foreign policy as a “hard” issue.
4. In order to use real polls and avoid deception, I use two different sets of math problems. As I discuss in detail below, the relevant comparisons in this study involve subjects making decisions about the exact same problems under different conditions (i.e., one source of information vs. two conflicting sources of information).
5. The endorser makes his or her statement by putting a checkmark beside the answer that he or she wishes to recommend. The experimenter then reads that statement aloud to subjects. This prevents the endorser's tone of voice from confounding the experiment. Similarly, the endorser sits behind a partition so that the endorser's gender, race, and/or age do not affect the extent to which subjects listen to the endorser's statements.
6. This condition is analogous to many real-world political situations, such as when citizens who are concerned about the environment look to the Sierra Club for guidance on how to vote.
7. I do not tell subjects whether the endorser or the poll suggests the correct choice; thus, this aspect of the experiments is not common knowledge to subjects.
8. In equilibrium, the knowledgeable but untrustworthy endorser should mix between making correct and incorrect statements (CitationLupia & McCubbins, 1998). In the experiments, subjects acting as the untrustworthy endorser make predominantly incorrect statements, along with a few correct statements. My conclusions do not change if I limit my analysis to only instances in which the untrustworthy endorser makes incorrect statements.
9. Avoiding deception is particularly important in these experiments because subjects have preexisting knowledge about how to solve math problems. Thus, if I had fabricated poll results for each problem (which would have ensured that each problem had both correct and incorrect poll results and that the same problems could be used in both conflicting information treatment conditions), subjects likely would have realized that some of the poll results were unrealistic. For example, subjects likely would have been skeptical if the poll results for a relatively easy math problem showed that a majority of undergraduates recommended the incorrect answer. Subjects’ realization (or even suspicion) that the poll results were fabricated might have also led to skepticism about other aspects of the experiment and, in turn, affected their behavior. To avoid such confounding factors, I use actual poll results and two different sets of math problems. During informal post-experiment interviews, no subjects expressed skepticism that the poll results were fabricated or not based on responses from undergraduates at their university.
10. To ensure that subjects perceive the poll results as trustworthy, but not necessarily knowledgeable, I ask the following quiz question: “The 66 other UCSD undergraduates earned $0.50 every time that they answered the problem correctly, they lost $0.50 every time they answered a problem incorrectly, and they earned nothing if they chose not to answer the problem. True or false?” Subjects virtually always answer this quiz question correctly. This ensures that subjects understand (a) that the undergraduates who were polled are trustworthy because they had a financial incentive to recommend correct answers and (b) that the undergraduates are from their own university, which allows subjects to make inferences about the undergraduates’ level of knowledge based upon the perceived difficulty or ease of the problems.
11. As a robustness check, I also analyze the data using regressions that control for the difficulty of the math problems and subjects’ levels of sophistication. My conclusions do not change if I use regressions as opposed to difference of means tests.
12. Although the trustworthy endorser always makes truthful statements, it is interesting that subjects do not always base their decisions upon his or her statements. Thus, they do not earn the maximum amount of money (i.e., 50 cents per problem). Post-experiment interviews with subjects reveal that some subjects trust their own ability to solve these problems more than they trust the endorser, which leads them to answer the problems incorrectly.
13. That subjects base their decisions upon the less credible source of information even when the other source is clearly superior suggests that I will observe even worse decisions when there is greater uncertainty about which source is more credible. I will test this proposition in future studies.
14. Easier problems likely bias me against finding that subjects make worse decisions when exposed to conflicting information. Because the problems are easier, it is more likely that subjects will know the correct answers on their own and, thus, less likely that they will be swayed by the presence of a second, conflicting source of information.
15. Again, my results and conclusions do not change if I filter out the few instances in which subjects acting as the untrustworthy endorser make correct statements and analyze only the instances in which the untrustworthy endorser makes incorrect statements. I chose not to filter out the few correct statements because, if anything, this provides a more conservative estimate of the detrimental effects that conflicting information has on subjects’ decisions.
16. There is not a significant difference in the amounts of money subjects earn in the control group and in the untrustworthy endorser condition. This is not surprising, given that subjects who are exposed to an untrustworthy endorser should ignore what he or she says and make their decisions on their own (CitationLupia & McCubbins, 1998).
17. Estimating the loss in income due to non-participation (i.e., leaving the problem blank) is relatively straightforward. I took the difference in the percentages left blank for the “trustworthy endorser” and the “trustworthy endorser + poll” conditions (13%) and multiplied this difference by the expected value of participating in the trustworthy endorser condition (34 cents). This amounts to roughly 4 cents per question (0.13 × 0.34) or 12% of the expected value of participating. I estimated the loss in income due to the greater propensity of selecting the incorrect answer in the “trustworthy endorser + poll” condition as follows. I took the difference in this propensity (i.e., percentage incorrect) for the “trustworthy endorser” and the “trustworthy endorser + poll” conditions (12%) and multiplied this number by the difference in the value of a correct versus incorrect answer ($1.00). This amounts to 12 cents per question or 35% of the expected value of participating. I estimated the losses for the other treatment conditions in the same manner.
18. As CitationDruckman (2004) notes, splitting the sample at the median minimizes measurement error. That is, slight changes in SAT math scores may not capture real differences in sophistication, but a median split should produce groups that are qualitatively different on this dimension. For other uses of a median split when measuring sophistication or knowledge, see CitationRahn, Aldrich, and Borgida (1994) and CitationDruckman and Nelson (2003).
19. It is surprising that sophisticated subjects earn less money than unsophisticated subjects when exposed to only the trustworthy endorser's statements. Post-experiment interviews with these subjects revealed that some sophisticated subjects were overconfident in their own abilities, which led them to ignore the trustworthy endorser's truthful statements and base their decisions upon their own (incorrect) notions about how to solve these problems. Note, however, that the addition of a second, conflicting source of information did not have a significant effect on sophisticated subjects’ decisions, while it significantly reduced the amount of money that unsophisticated subjects earned.