Expressive voting versus information avoidance: experimental evidence in the context of climate change mitigation

Decisions taken via democratic vote are often considered to yield more moral outcomes than individual decisions, in particular when it comes to choices made in a market context. This dichotomy between “electoral choice” and “market choice” (Brennan & Lomasky, 1993) is based on a compelling argument: According to the “low-cost theory of voting” (Tyran et al., 2019), instrumental considerations decrease with pivotality, i.e., the number of voters involved, while the utility from expressing moral principles remains unaffected by the size of the constituency. Thus, with decreasing probability of being pivotal, the ultimate decision is more likely to favor the more moral option (Brennan & Buchanan, 1984; Fiorina, 1976; Tullock, 1971). In market decisions, in contrast, the potential conflicts between instrumental and expressive motives seem to be often resolved in favor of the former, leading to more self-serving, less moral behavior. Recent experimental evidence seems to support this conjecture (Bartling et al., 2014; Falk & Szech, 2013; Falk et al., 2020).

From the outset, this dichotomy has been explained by the notion of decision-makers having to “reduce internal dissonance” (Tullock 1971, p. 387) between their moral preferences and payoff-oriented self-interest. This goes back to the theory of cognitive dissonance, first derived in Festinger (1957), which establishes different individual strategies to resolve such internal conflicts. For example, in situations where moral views and self-interest are in conflict, the individual can resolve this conflict by deciding in favor of either one option or the other. In the context of the low-cost theory of voting, the small probability of being pivotal alters this trade-off in favor of expressing moral principles, leading to more moral decisions.

Another strategy to reduce cognitive dissonance is to not use all available information on the consequences of a decision, even if information is accessible without cost. As was first pointed out by Festinger (1957), people tend to avoid information that might be incongruent with their established attitudes while disproportionately seeking news that is congruent—a tendency that is referred to as selective exposure to information.¹ More recently, economists have identified this behavior as one of the most effective strategies for motivated reasoning to reach favorable conclusions on the effect of own actions on others (Bénabou & Tirole, 2016; Gino et al., 2016; Golman et al., 2017). In the simplest case, individuals can choose to remain entirely ignorant of the nature and scope of external effects that their own decisions have on others and choose according to their narrow self-interests. Thus, by avoiding information, they can circumvent feeling morally obliged to decide in favor of options associated with a larger positive (or a smaller negative) effect on others. They hence exploit moral wiggle room created via their deliberate lack of information (Dana et al., 2007). While there is ample experimental evidence of the exploitation of moral wiggle room in individual decision-making (e.g., Dana et al., 2007; Grossman & van der Weele, 2017; Momsen & Ohndorf, 2020), its importance in the context of incentivized voting decisions has not yet been investigated. In this paper, we present the results of a lab experiment designed to analyze the interactions between expressive voting and information avoidance. Hence, the paper closest to ours is Mechtenberg et al. (2021), which uses a field experiment to investigate moral self-signaling in voting. They find that moral self-image concerns do not lead to biases in information collection, but in information processing and voting behavior. While a dual-self approach has been shown by Grossman and van der Weele, (2017) to be a valid alternative to the more traditional concept of cognitive dissonance when it comes to explaining information avoidance in a formal model, it remains unclear how to apply this model to voting decisions. Hence, in its current form, Mechtenberg et al. (2021) refrains from presenting a formal model that integrates the dual self into the underlying voting game. We believe that the cognitive dissonance approach is better suited to such an integration, as shown within our formal considerations.²

In our experiment, third-party consequences are implemented as contributions to carbon offsets of different sizes, such that decisions in the lab affect climate-change mitigation. The climate change context seems to be particularly prone to the dynamics of information avoidance via self-selection into echo chambers and networks with like-minded individuals.³ In the contemporary academic discourse, such behavior is often associated with biases in information-seeking within social media networks (Bakshy et al., 2015; Halberstam & Knight, 2016). However, the problem of information avoidance is not limited to phenomena of online news consumption. As Boxell et al. (2017) report, polarization of political views might be more pronounced in demographic groups which are least likely to use the internet and social media. In cases where this choice is made willingly and is not compensated by a balanced consumption of traditional media, this would also qualify as a specific form of information avoidance (Klapper, 1960).

In order to examine the behavioral fundamentals of information avoidance, we investigate individual decisions and democratic votes under full information and compare these to a situation where information is initially unobservable but revealable without payoff-relevant cost. We derive predictions from a stylized model of voting decisions where the expressive term is specified as avoided costs from potential cognitive dissonance, which is in turn derived from the formalizations presented in Rabin (1994), Konow (2000), and Spiekermann and Weiss (2016). Formulating expressive voting as a matter of cognitive dissonance not only does justice to the above-cited origins of this theory, but also allows us to compare the effect of differing pivotality under full information to the one under potential information avoidance.⁴ To derive our hypotheses, we integrate this cognitive dissonance model into a simple Bayesian voting game with symmetric strategies and without abstention, which is based on Feddersen and Pesendorfer (1997) and Tyson (2016).

We test our theoretical predictions in an experiment implementing four different decision contexts: individual choice (framed as a consumption decision), majority voting in small and large groups (democracies), and randomized dictatorship. The consumption setting is included to take the initial juxtaposition of market choice and electoral choice made in Brennan and Lomasky (1993) into account.⁵ In all other treatment conditions, decisions affect groups with varying probabilities of a subject’s decision to be pivotal.

Under full information, we observe the lowest share of selfish choices in conflict situations in the individual choice treatment (consumption). In fact, the number of self-serving decisions in this treatment is significantly lower than that with majority voting in small groups and with the randomized dictator condition. Hence, in our experiment, the market context clearly does not lead to a less moral outcome—a finding that apparently contradicts the prediction with respect to markets made in the early literature on expressive voting, in particular in Brennan and Lomasky (1993).⁶ Yet, we surmise that the comparatively low number of “green” choices in the individual choice treatment could be attributed to the effect of rivaling other-regarding preferences in the group settings, as decisions also affect payoffs of other group members.⁷

Interestingly, when comparing our voting treatments under full information, we find a strong effect of pivotality, which is consistent with the low-cost theory of voting. The number of self-serving choices for the full information condition is significantly lower when voting in a larger group than for the treatments with a higher probability of being pivotal. This is in line with identifying a moral bias in larger elections presented in Shayo and Harel (2012), Fischer (1996), Feddersen et al. (2009), and Bischoff and Egbert (2013) and contrasts the studies that do not find an effect of expressive voting (Kamenica & Brad, 2012; Tyran, 2004; Tyran & Sausgruber, 2006).⁸ Framing the decision not as a voting situation, but as a (structurally equivalent) randomized dictatorship does not affect the direction of the effect of pivotality. Furthermore, this effect is stronger for smaller payoff differences, as predicted within our theoretical considerations.⁹

Yet, most importantly, in our treatments with hidden but revealable information, all differences in selfish choices disappear, which suggests that the effect of pivotality is dominated by information avoidance. Indeed, when analyzing the potential occurrence of self-serving information avoidance, we find that this phenomenon arises predominantly in those treatments that yield the largest number of “green” choices under full information, i.e., voting in larger groups and the individual choice condition. In these treatments, exploitation of moral wiggle room is also less pronounced if the difference in payoffs is large, which is in line with our theoretical predictions.¹⁰ Given these results, we conclude that avoiding information is the preferred strategy to deal with cognitive dissonance over all decision contexts investigated here, effectively dominating other strategies, like expressive voting. This result suggests that moral biases might be less likely in elections than previously thought.

The remainder of the paper is organized as follows: The following section describes our experimental design. We derive our behavioral predictions theoretically in Sect. 3. Our results are laid out in Sect. 4. The last section concludes. A translation of the experimental instructions as well as supplementary analyses are relegated to the Appendix.

To investigate the effects of pivotality, information avoidance, and their interaction, we conducted a laboratory experiment where subjects made climate-relevant decisions. The experiment consisted of eight treatments, with each treatment capturing a different decision situation with respect to information and pivotality. In each treatment, subjects made 24 consecutive binary decisions between two allocations that differed in their payoffs and in their contributions to climate change mitigation. The latter was implemented via contributions to reduce global greenhouse gas emissions through the acquisition of high-quality carbon offsets on the voluntary offset market. Hence, decisions made in the experiment had an impact beyond the laboratory.¹¹

Subjects could choose between two options, A and B. The payoffs of the options were drawn from a range of 10 to 90 European Currency Units (ECUs), and within a round, the payoff difference between the two options could be 5, 10, 15, or 20 ECUs. For option A, the contribution to the carbon offset was always equal to 15 ECUs. The contribution associated with option B could be either 0 or 30 ECUs, with each outcome being equally likely. For each round, the associated contribution was independent of the one realized in the previous round, such that subjects should consider each decision separately. The order in which subjects faced the decision situations was randomized, i.e., decision situations were randomly drawn from a pool of possible decision situations without replacement. We implemented two different types of decision situations: In situations with aligned interests, the option with the higher payoff for the subject(s) generated the larger contribution to the carbon offset, whereas in situations with conflicting interests, the option with the lower payoff was associated with a larger contribution. Both types of situations were equally likely to occur, and the order was randomized.

For both phenomena investigated in our experiment, expressive voting and self-serving information avoidance, explanations have been brought forward based on the concept of cognitive dissonance (Festinger, 1957). In our experimental decision situations, such cognitive dissonance can arise if a person experiences a conflict between the motive of maximizing their own monetary payoff and the ideal of providing high levels of environmental benefits. If the effects of their own actions are known, as in our full information treatments, the subject needs to resolve this conflict by either deciding not to maximize their monetary payoff or by not corresponding to their environmental ideals. Both strategies are associated with either monetary or psychological costs, such that the resolution of the cognitive dissonance is always associated with trade-offs. In a group decision with pivotality lower than 1, this trade-off is altered. If the probability of being decisive in a voting context is low enough, the expected impact of the own vote on the monetary outcome is small. In such cases, voting for the “greener” option to correspond to own ideals is associated with lower expected monetary costs. The cognitive dissonance is hence resolved by choosing the green option and the expressive motive of voting dominates considerations on instrumental utility. This reasoning is at the basis of the low-cost theory of voting, predicting that for larger constituencies—and hence lower probabilities of casting the pivotal vote—the number of “moral” votes will be larger than for smaller constituencies or individual decisions (Brennan & Lomasky, 1993; Feddersen et al., 2009; Shayo & Harel, 2012; Tyran, 2004).

If information on the effects of the own decision needs to be revealed by the individual, there exists an additional strategy to reduce cognitive dissonance, as they can choose to only reveal information that is congruent with their ideals. This phenomenon is referred to as selective exposure to information (Festinger, 1957). In its simplest form, a person would simply avoid revealing information on the moral implications of their decision to exploit moral wiggle room (Dana et al., 2007). If the consequences of own decisions for third parties or the environment are unknown, the size of the cognitive dissonance is reduced. In such a case, the individual’s self-image is only challenged by the possibility, not necessarily the fact, that maximizing payoffs corresponds to the less moral option. In our hidden information treatments, it is a priori unclear whether the alternative with the higher monetary payoff is associated with the environmental benefit that is larger (aligned interests) or lower (conflicting interests). Thus, avoiding information and choosing the self-serving option can be a viable alternative to revealing the information and then choosing the greener option even if it is more expensive to do so.

In order to structure our understanding of these different phenomena and to add some rigor to our predictions, it is useful to derive a formal representation of the decisions taken within the experimental treatments presented above. Historically, economic formalization of cognitive dissonance has focused on the strategy of adjusting beliefs and fairness considerations (Konow, 2000; Rabin, 1994; Spiekermann & Weiss, 2016). In the following, we extend this model to also reflect the effects of pivotality and information avoidance in voting decisions. Our main focus is on the context of climate-relevant decisions as implemented in our experimental treatments.

In our analysis of the experimental results, we will first focus on the effect of the different decision contexts on the share of selfish choices under full information (Hypothesis 1). We will then examine behavior under hidden information and explore the exploitation of moral wiggle room within each decision context (Hypothesis 2). Summary statistics for our sample, as well as power calculations, are provided in the Supplementary Material (see Table A.2 and Sect. A.4.2).

We present the results of an incentivized laboratory experiment designed to investigate self-serving information avoidance and moral bias in individual and voting decisions. Subjects choose between two options which affect their own payoff and the amount contributed to carbon offsets such that the impact of the subjects’ decisions reaches beyond the laboratory. Our main interest lies in the resolution of potential cognitive dissonances between the subjects’ narrow self-interest and their self-perception as an environmentally conscious individual.

When information on the contribution is immediately observable, we observe the lowest share of self-serving decisions in the individual choice condition framed as consumption and with voting in large groups (Voting11), while the self-serving option is chosen more frequently if decisions are taken for a group of three, either by a randomized dictator (Dictator) or via voting (Voting3). This contrasts with the initial considerations by Brennan and Lomasky (1993) on expressive voting, who infer from their argument on pivotality that decisions on markets should be less moral than democratic decisions. However, we believe that this contrast needs to be qualified, as the two decision contexts differ with respect to factors besides pivotality. If decisions are taken for the whole group, additional factors are at play that influence the underlying trade-off between self-interest and environmental ideals. It is not unlikely that the difference between our individual choice condition and group decisions is due to the influence of distributive preferences, as a decision that is binding for the whole group also has an impact on the payoff of others. Thus, while the juxtaposition by Brennan and Lomasky (1993) might be normatively appealing, the market does not seem to be the most appropriate baseline to assess potential moral bias within democratic choice.

Comparing our voting treatments yields certainly the most interesting result for the full information case. An increase in the number of voters from 3 to 11 significantly increases the share of “greener” choices. This supports the low-cost theory of voting, as the corresponding decrease in the probability of being pivotal is associated with a decrease in self-serving choices. Hence, the relationship between the share of environmentally friendly decisions under full information and the number of decision-makers is not linear: The share of environmentally friendly decisions is high in individual decision-making contexts, decreases with majority voting in small groups, and increases again in large groups. Notice that there is no reason to assume that the effects causing the decrease in the share of environmentally friendly decisions in small groups disappear for larger groups, such that an isolated effect of the probability of being pivotal might even be larger than reported here.

When information on the externalities is initially hidden, but revealable, differences in self-serving choices between decision contexts disappear entirely, with all decisions being in general more selfish. Furthermore, in the settings where we observed the lowest share of selfish decisions under full information—in the individual choice condition and Voting11—subjects use the possibility of avoiding information as an excuse to behave more selfishly: They exploit moral wiggle room. In both decision contexts, remaining ignorant is used as a strategy to reduce cognitive dissonance, as it allows one to choose the self-serving option without being in open contradiction to environmental ideals. For Voting3 and the Dictator treatment, the observable pattern of self-serving ignorance is less pronounced, as choices are already more selfish under full information. Most interestingly, in the Voting11 treatment, information avoidance entirely substitutes expressive voting as a strategy to manage cognitive dissonance.

This latter result surely needs some discussion, as the idea of electoral choices being more moral is one of the most compelling results of the theory of expressive voting. The possibility of remaining uninformed with respect to the positions of candidates or parties concerning policies that feature a moral dimension is rather the rule than the exception.³¹ Moreover, in recent years, several phenomena have been identified supporting the conjecture that voters tend to actively avoid information. There seem to be, in fact, groups of voters that tend to self-select into echo chambers, isolating themselves from certain information sources.³² To the extent that this self-selection is an active choice, this development could also be interpreted as a more sophisticated form of information avoidance. As we have shown, the possibility of remaining uninformed can lead to a reversal of (revealed) voters’ preferences, as voters with intermediate preference intensity will remain uninformed and choose the option that serves their narrow self-interest. Hence, when it comes to actual elections, such behavior is likely to contribute to a polarization of voters’ preferences.

Note, however, that information avoidance not only provides an explanation for the polarization of preferences, but also could explain the polarization of politicians’ positions and policies themselves. In a Hotelling-Downs framework, policies will converge to the preferences of the median voter if utility functions are concave. In contrast, convex utility functions over an ordered policy range will lead to a polarization in politicians’ positions and their policies (Kamada & Kojima, 2014; Tajika, 2020). It is easy to see that a utility function that is concave over a specific policy range under complete information can become convex if voters remain uninformed with respect to a specific property of these options. In our experiment, for example, the set of policies was characterized by two dimensions, the associated cost and the corresponding effect on the global climate. If, under full information, concavity of utilities is driven by a decreasing marginal utility of environmental utility, self-serving avoidance of this information leads to an evaluation of these options with respect to cost only, for which a convex form is plausible.

Thus, in the context of climate policy, democratic elections will not necessarily lead to an increase in the stringency of mitigation efforts, as the theory of expressive voting would predict (Brennan, 2009). Indeed, this theory does not seem to adequately predict the political development in the political sphere of various countries, where we observe polarization of both voter preferences and proposed policy positions (Karakas & Devashish, 2020). As this discrepancy can be explained via self-serving information avoidance, it seems promising to further investigate these interactions in future research.