Making Citizens Smart: When do Institutions Improve Unsophisticated Citizens’ Decisions?

Although information shortcuts often help citizens with their decisions, there are circumstances under which they may lead citizens astray. Indeed, it is unlikely that information shortcuts improve the decisions of all citizens in all contexts. Thus, scholars have rightly begun to question the conditions under which information shortcuts help citizens to improve their decisions (see, e.g., Bodenhausen and Lichtenstein 1987; Sniderman et al. 1991; Kuklinski and Quirk 2000; Kuklinski et al. 2001; Lau and Redlawsk 2001; Kam 2005; Merolla et al. 2005; Somin 2006; Friedman 2006; Kinder 2006; Boudreau 2008).

For example, while some political scientists advocate fact-based measures of political sophistication (see, e.g., Luskin 1987; Delli Carpini and Keeter 1996), others measure political sophistication as an index of political knowledge and various forms of political interest (see, e.g., Lau and Redlawsk 1997, 2001). Still other scholars measure political sophistication as the distance between a citizen’s placement of political parties on a 10-point ideological scale and the mean placement of those parties by other citizens (Gordon and Segura 1997).

The Speaker makes his or her statement by putting a checkmark beside the answer that he or she wishes to report. The experimenter then reads that statement aloud to the other subjects. This prevents the Speaker’s tone of voice from confounding the experiment. Similarly, throughout the experiment, the Speaker sits behind a partition so that the Speaker’s gender, race, and/or age do not affect the extent to which subjects trust the Speaker’s statements. Although this communication process is stripped of contextual factors (such as tone of voice, physical appearance, rhetorical skills, etc.) that we know are important in politics, it is necessary for eliminating confounding variables that limit the inferences that can be drawn from these experiments. Indeed, if subjects are able to see, hear, or talk to one another, then it is difficult (if not impossible) to determine whether particular institutions cause subjects to trust the Speaker’s statements or whether some other factor (such as the Speaker’s gender, tone of voice, or rhetorical skill) causes subjects to trust the Speaker. Stated differently, uncontrolled contextual factors reduce the strong internal validity associated with laboratory experiments and make it difficult to isolate the effects of particular treatment conditions. Other scholars use similar communication processes in their experiments, and they also recognize that anonymous, controlled communication environments are essential for isolating the effects of particular treatments and eliminating confounding variables (see, e.g., Wilson and Sell 1997; Lupia and McCubbins 1998; McCubbins and Rodriguez 2006; Ahn et al. 2007).

Speakers (such as political parties, candidates, friends, and family members) frequently make statements to citizens about what choice they should make at election time. When deciding whether to trust a speaker’s statement, citizens can use a speaker’s party identification to determine whether the speaker shares common interests with them (Lupia and McCubbins 1998). Importantly, a speaker’s party identification is a widely used cue—one whose use is not limited to only sophisticated citizens (see, e.g., Brady and Sniderman 1985; Popkin 1991; Sniderman et al. 1991; Lau and Redlawsk 2001; Tomz and Sniderman 2004; Kam 2005). Thus, the common interests condition is met in real world politics whenever citizens use a speaker’s party identification as a basis for trusting his or her statements.

Throughout this paper, I draw upon North’s (1990) definition of institutions, as well as Lupia and McCubbins’s (1998) discussion of how penalties for lying and the threat of verification can be conceived of as institutions. Specifically, North (1990, p. 3) defines institutions as “humanly devised constraints that shape human interaction…In consequence, [institutions] structure incentives.” Lupia and McCubbins (1998) also emphasize that institutions alter the incentives of political, legal, and economic actors. They then describe how penalties for lying and the threat of verification alter the incentives of speakers in electoral, legal, and other political contexts (see Lupia and McCubbins 1998, pp. 205–227).

Speakers in political and legal settings frequently incur penalties if they are caught lying. For example, Lupia and McCubbins (1998) emphasize that jurors know that witnesses for both the prosecution and the defense face penalties for perjury if they lie on the witness stand. Similarly, in political contexts, citizens know that a candidate who is caught lying may suffer a penalty in the form of losing a valuable reputation for honesty (Gerber and Lupia 1995). Importantly, citizens need not be highly politically informed to know that a candidate or witness will suffer a loss of reputation or face monetary sanctions if he or she is caught lying. Candidates and witnesses also know that they can incur a penalty if they are caught lying. Thus, the existence of penalties for lying can be common knowledge in the real world, just as it is in my experiments.

Speakers in political and legal settings are often subject to a threat of verification. For example, attorneys’ cross-examinations of witnesses are a form of verification that can reveal, with some probability, when witnesses have made false statements (Lupia and McCubbins 1998). Similarly, in political settings, other candidates, interest groups, debate moderators, political watchdog groups, and other interested parties can verify a candidate’s statements and reveal when he or she has not been truthful (Lupia and McCubbins 1998). Note also that most citizens know that candidates do not make statements in a vacuum, but rather in an environment in which it is possible for their statements to be verified by others. Most candidates also know all too well that their statements can be verified. Thus, the existence of a threat of verification can be common knowledge in the real world, just as it is in my experiments.

I vary the size of the penalty because penalties for lying in real world political and legal settings also vary in how large they are, relative to what is at stake for a speaker in a particular context. For example, the $15 penalty for lying is a very large penalty, relative to what is at stake for the Speaker in these experiments. The $1 penalty, however, is small relative to the money the Speaker can gain in these experiments. Indeed, in the $1 penalty for lying condition, the Speaker can gain as much as $5.50 if 11 subjects choose the wrong answer to a problem and only lose $1 for lying about the correct answer. Similarly, the penalties for lying that speakers in real world political and legal settings face can be large or small, depending on the nature of the election or trial, the stakes involved, and the value that a speaker places on his or her reputation.

Speakers’ false statements are not likely to be detected or punished with certainty (or with any one probability) in the real world. Rather, the probability of verification is likely to vary depending on the nature of the political or legal context in which a speaker makes a statement. For example, verification may be more likely in high profile court cases with experienced attorneys and in campaigns for national office, as opposed to local office. Thus, I examine the effects of different chances of verification.

The typical number of subjects in each experiment is 12 (11 subjects who answer the math problems and 1 Speaker).

I include a small cost in all verification conditions for two reasons. First, in real world settings, speakers must often pay a cost (be it a loss of reputation or monetary sanctions) if they are verified and found to be misrepresenting the truth. Second, the original experiments that tested the effects of verification also included a small cost (Lupia and McCubbins 1998).

Although there is a significant decrease in the Speaker’s propensity to tell the truth when the 100% chance of verification is reduced, the Speaker still makes truthful statements more often than not even when the chance of verification is as low as 30%. Thus, subjects should never simply assume that the Speaker is lying. Rather, they should base their decisions upon their beliefs about the Speaker, their preexisting knowledge about the math problem, and their beliefs about the difficulty of the problem.

Subjects solving the math problems are significantly more likely to leave the difficult problems blank.

My results are robust to different specifications of sophistication. Specifically, my conclusions do not change if I alter the high and low SAT math scores that I use to classify subjects as sophisticated or unsophisticated.

Interestingly, subjects do not answer every math problem correctly in the 100% chance of verification condition. This seemingly anomalous result reflects the fact that, when using human subjects in experiments, there is always a certain amount of noise in the data and unexpected behavior. I am confident that this result does not stem from confusion about the experiment because, after reading the instructions for a given treatment condition, the experimenter gives subjects a short quiz that asks them about their own incentives, the incentives of the Speaker, and the way that a particular treatment condition works. On the quiz for the 100% chance of verification condition, subjects are asked what percentage of the time the experimenter will verify the Speaker’s statements and announce the correct answer to the math problem (the correct answer to this quiz question is “100%”). To motivate performance on the quiz, the experimenter tells subjects that they earn $0.25 for each quiz question that they answer correctly. After subjects finish taking the quiz, the experimenter corrects subjects’ quizzes in front of them, explains any wrong answers to them, and pays them the money that they earned for their performance on the quiz. Subjects, by and large, answer all of the quiz questions correctly. The experimenter also conducts informal post-experiment conversations with subjects, asking them about their understanding of the Speaker’s incentives, as well as their own incentives and the way that verification worked. In these conversations, no subjects expressed confusion about the treatment condition.

What the informal post-experiment conversations with subjects reveal, however, is that subjects who behave anomalously in the 100% chance of verification condition (i.e. do not trust the Speaker’s statements) are either bored, make a mistake, or want to see what happens if they do not behave as expected. Unfortunately, boredom, making a mistake, and wanting to see how the experimenter reacts to anomalous behavior are factors that play a role in nearly all laboratory experiments. Indeed, I have replicated the 100% chance of verification treatment condition in other, slightly different experiments, and I never observe behavior that is perfectly consistent with my predictions. For example, in other experiments, I show that when a 100% chance of verification is imposed upon two competing Speakers, subjects do not earn the maximum amount of money possible (i.e. they earn, on average, only $0.40, not the maximum $0.50 that I would expect, given that the Speakers’ statements are always verified by the experimenter). Perhaps more astonishingly, Lupia and McCubbins (1998) pay subjects to predict that the outcome of an unseen coin toss is heads (regardless of whether that prediction is correct), and they find that subjects predict heads only 86% of the time (which is significantly less than the expected 100%).