Does Altruistic Voting Matter for Election Outcomes?

Transcription

1 Does Altruistic Voting Matter for Election Outcomes? Evidence from Denmark and the US Daniel Mahler February 12, 2016 Abstract Some voters cast their ballots based on selsh concerns whilst others vote for altruistic reasons. This paper analyses how, if at all, altruistic voting aects the outcomes of elections. To this end, a Danish survey with a sample of 2000 representative respondents is conducted. The respondents are asked to identify (1) the party they would vote for, (2) the party they think is best for themselves, and (3) the party they think is best for society as a whole. The political parties are located in various political dimensions such that dierences in where individuals cast their altruistic, selsh, and actual party votes can be compared. The ndings reveal that altruistic voting drive the results to the left and away from extreme candidates. A smaller American survey (n=400) on the 2016 presidential candidates yields similar results. The results imply that - if it is possible to inuence why voters vote - certain candidates may be able to increase their vote share without compromising on political content. I am thankful for helpful comments received at Harvard's Working Group in Political Psychology and Behavior and Harvard's Political Economy Workshop. In addition, I am grateful for comments and advice from Ryan Enos, Chase Harrison, and David Dreyer Lassen. The usual disclaimer applies. This project has received support from the European Research Council. Department of Economics, University of Copenhagen (daniel.mahler@econ.ku.dk) & Institute for Quantitative Social Sciences, Harvard University. 1

2 Once upon a time two boys found a cake. One of them said, Splendid! I will eat the cake. The other one said, No, that is not fair! We found the cake together, and we should share and share alike, half for you and half for me. The rst boy said No, I should have the whole cake! Along came an adult who said, Gentlemen, you shouldn't ght about this: you should compromise. Give him three quarters of the cake. (Smullyan 1980, p. 56) 1 Introduction Some of the most important thinkers in economics and political science over the past century have argued that individuals are able to make judgments based on what they think is best for themselves and what they think is best for society as a whole (Arrow, 1951; Harsanyi, 1955; Sen, 1977; Dworkin, 1978; Elster, 2006). 1 At times individuals will act to their own benet and at other times individuals may act based on altruistic concerns. This dichotomy of behavior has been the subject of intense study in the context of elections. Do individuals vote for selsh or altruistic reasons? Downs' canonical model of economic voting presupposes that when confronted with dierent political platforms, `a rational man always takes the one which yields him the highest utility ceteris paribus; i.e, he acts to his own greatest benet' (Downs, 1957, p. 36). 2 However, decades of literature have shown that voters take account of the how the nation as a whole is fairing and do not vote in line with their material self-interest (Sears et al., 1979; Kinder and Kiewiet, 1981; Chong et al., 2001). In other words, individuals may vote for altruistic reasons. These studies primarily analyze whether and why individually vote sociotropically/altruistically, but refrain from looking at the resulting impact of non-selsh voting. This paper attempts to ll the gap by simply asking: How, if at all, does the dichotomy of ethical voting behavior matter for election outcomes? To this end a Danish survey is designed, wherein a representative sample of more than 2000 Danish voters are asked to identify: the party they would vote for if parliamentary elections were held tomorrow the party they would vote for if they were only to consider what is best for themselves the party they would vote for if they were to consider what is best for society as a whole By comparing the answers to these three questions, the impact of altruistic voting on election outcomes can be analyzed. By altruistic voting is meant individuals who vote for 1 Arrow (1951, p. 18) argues that individuals can order social states according to their tastes, which reect their economic circumstances, and according to their values, to which the individual adds his 'general standards of equity'. Harsanyi (1955, p. 315) distinguishes between individuals' `subjective' preferences, which determine their own well-being, and their `ethical' preferences, which reect impartial and impersonal judgments. Sen (1977, p. 329) argues that besides being motivated to maximize their own welfare, individuals may take actions based on altruistic moral commitments, which create 'a wedge between personal choice and personal welfare'. Dworkin (1978, p.235) distinguishes between 'personal' preferences over an individual's enjoyment of goods and opportunities and 'external' preferences over the assignment of goods and opportunities to others. The latter may be 'altruistic' preferences. Elster (2006, p. 185) contends that individuals can be motivated by interest, which is a 'concern of the absolute welfare of the agent', and reason, which is a 'benevolent and impartial motivation'. 2 To be fair, Downs (1957, p. 27) did acknowledge that [i]n reality, men are not always selsh, even in politics. They frequently do what appears to be individually irrational because they believe it is socially rational i.e., it benets others even though it harms them personally. To simplify the analysis Downs disregarded this. 2

3 a party because they believe this party is best for society as a whole. This is dierent from sociotropic voting, which is simply taking some account [...] of the collective's interest Meehl (1977, p. 14). Sociotropic voting need not reect altruism as individuals may take account of the collective's interest for selsh reasons (Kiewiet and Lewis-Beck, 2011; Kinder and Kiewiet, 1981). In the context of this study, this implies that no claim can be made that individuals that vote for the party they consider is best for society are voting altruistically. They might vote so for selsh reasons. However, if they simultaneously do not vote for the party they think is best for themselves, we know for certain that they do not vote selshly. This divergence allows me to back out what the election outcome would have looked like, if people voted more selshly (or conversely more altruistically). In order to analyze how altruistic voting matters in more detail, the political parties of Denmark will be placed in a political compass such that dierences in where the respondents place their actual, altruistic, and selsh vote can be analyzed. This approach does not allow me to identify who votes altruistically for two reasons. First of all, a large share of the voters (56%) vote for the party they think is best for society and best for themselves. It is unclear whether these vote altruistically, selshly, or something third. Secondly, even if we observe a person that votes for what he thinks is best for society but not best for himself, we cannot conclude that this person voted altruistically, since his motivation is not inferred. He might have voted for the candidate he thought was best-looking and it just so happened to be the candidate he also thinks is best for society as a whole. All that can be detected is the number of people that do not vote altruistically or do not vote selshly. That is, an upper bound of how many selsh and altruistic voters can be calculated. There is, according to the data, at most 82% altruistic voters and 65% selsh voters. The paper nds that if more people had voted selshly, the outcome would have been more more right-winged and more votes would have fallen on more extreme candidates. If more people had voted altruistically, on the other hand, the outcome would have been more left-winged and more centered, but the dierences here are quite small suggesting that most people already vote altruistically. A smaller survey (n=400) on the US primary elections nds very similar results, suggesting that the ndings are externally valid. When facing the three questions individuals may want to appear 'consistent' by aligning their vote choices more than is truthfully the case. If this is the case, any dierences found between the three vote distributions should be considered lower bound. Individuals may also attempt to look non-selsh, by deliberately selecting a party in the selsh choice question that is far away from their actual vote and altruistic vote. This phenomena would be strongest if individuals are asked to place their selsh votes after having placed their altruistic and actual votes. Two points argue against this occurrence. First of all, in 12 cognitive interviews carried out to assess the quality of the questions, this behavior was not detected. Secondly, the question order was randomized, and there were no signicant order eects. Another concern is that the potential divergences in votes may be driven by strategic voting. Bargsted and Kedar (2009) have shown that coalition targeted strategic voting may be present in proportional systems like the Danish one. Four arguments go against this driving the results. First of all, the survey was taken after an election, where no prospects of new coalition building was present. Secondly, as the altruistic and selsh questions do not ask what party do you consider is best for... but who would you vote for if..., strictly speaking the same kind of strategic voting should apply to all three 3

4 questions and hence even out. Thirdly, running the results on the subgroup of respondents that claimed they would never vote strategically yields the same results. Hence strategic voting is not driving the results. Fourthly, the results are similar in the US survey where respondents face very dierent strategic incentives. The outcome where everyone votes selshly or altruistically is naturally a purely hypothetical construct. As such the paper is similar in spirit to the papers trying to determine what election outcomes would have been if everyone voted (Citrin et al., 2003). The ndings here suggest that if certain political parties can capitalize on the duality of voter motivation, they may be able to increase their vote share without compromising on political content. In other words, if party X would receive many more votes if more people voted altruistically, this party should arguably attempt to frame the elections such that people vote for what they think is best for society as a whole. As such, the ndings give rise to a new research question: Can framing impact whether individuals vote for selsh or altruistic reasons? The rest of the paper is organized as follows. Section 2 reviews the existing literature on the topic. Section 3 delineates the hypotheses. Section 4 explains the methodology. Section 5 gives the results. Section 6 discusses the implications of the results and concludes. 2 Related literature An abundance of studies have analyzed whether individuals vote for selsh reasons or altruistic reasons. This section will summarize the main ndings of the dominating approaches to tackle this question. Although they all have issues with identifying altruism and dene altruism in dierent ways, they generally point to voting behavior being predominantly guided by altruistic concerns, and less so by selsh concerns. Sociotropic vs. egocentric voting: Following Kinder and Kiewiet (1979) a huge literature has investigated whether we can best explain individuals' vote choices by their evaluation of their personal income (their egocentric concerns) or by their evaluation of the national economy (their sociotropic concerns). The literature generally nds that individuals are primarily guided by sociotropic concerns and less so by their own pocketbook or other selsh concerns (see e.g. Sears et al., 1979, 1980; Markus, 1988; Sears and Funk, 1990; Lewin, 1991). Several studies have tested if Danish voters are sociotropic or egocentric. A few early studies showed that egocentric voting dominates in Danish elections (Nannestad and Paldam, 1995, 1997). Later studies, however, have largely negated this result. Borre (1997), Lewis-Beck et al. (2013) and Stubager et al. (2014) nd substantial sociotropic voting in the Danish elections from , and no egocentric voting at all. Recently, Alt and Lassen (2014) nd that sociotropic voting has a causal eect on vote intention in Denmark, but that personal economic conditions also matter. Incidence analysis: Another method to determine if individuals' political opinions are selsh or altruistic is to impute what would be in individuals' material self-interest and check if this aligns with their reported attitudes. Or in other words to check if voters vote 4

5 according to their pocketbook. Studies using this approach generally nd that individuals' attitudes do not align much with their material interest. Only when questions are rather specic and precise does self-interest seem to matter (Fisher, 1985; Sudit, 1988; Funk, 2000; Chong et al., 2001). Preferences for redistribution: Yet another method that has been used to study if altruistic concerns matter for voting relies on preferences for redistribution. A number of studies have argued that preferences for redistribution are partly driven by altruism in the form of beliefs and values about distributional justice (Corneo and Grüner, 2002; Dimick et al., 2014; Fong, 2001). Since Rueda and Stegmueller (2014) show that preferences for redistribution matter for voting behavior, indirectly, they infer that voting is driven by altruistic concerns. In more detail, Rueda and Stegmueller (2014) nd that voting for the Democratic Party in the US by high income individuals depends on the state level of inequality. In states with higher inequality, everything else equal, wealthy individuals vote more in favor of the Democratic Party. They interpret this nding as showing that preferences for redistribution - and thereby indirectly altruism - make individuals more likely to vote for a more egalitarian outcome. Eliciting dierent preferences: A nal method used to detect altruistic preferences is to elicit dierent kinds of preferences. This approach is closest to the method of the present paper. Hudson and Jones (1994, 2002) ask their respondents (i) how they would vote with respect to a specic government policy, (ii) how they would vote if the goal was their self-interest, and (iii) how they would vote if considering the `public interest'. They nd that public concerns are a stronger determinant of voting behavior than selsh concerns, suggesting a substantial presence of altruistic voting. Russell et al. (2003) used an experimental set-up to provide empirical support for individuals having a specic kind of altruistic preferences and a set of selsh preferences that dier over the optimal provision of public goods. Summary: To summarize the immense literature on altruistic voting it appears that the results are fairly similar: Individuals vote primarily altruistically and only to a lesser extent selshly. Due to many dierent usages of the word altruism in these studies, a bit of clarication is needed before proceeding. Conceptualizing altruism: A variety of dierent denitions of altruism exists. In this survey altruism will not be measured directly but rather be backed out from the choices of the respondents. As a starting point consider Jencks (1990, p. 53) who denes individuals as altruistic when they feel and act as if the long-term welfare of others is important independent of its eects on their own welfare. Individuals can thus be said to vote altruistically if they vote based on the welfare of 'others' independent of what is best for themselves. Importantly, in the context of this study, a stronger version of altruism is needed, which one could label societal altruism. Individuals may consider the welfare of certain 'others' of society such as their region, ethnicity or another subdivision of society. However, altruistic voting behavior in this study requires individuals to incorporate the entire society as others. Hence, individuals will be said to vote altruistically if they vote based on the welfare of society as a whole independent of what is best for themselves. It 5

6 will not be claimed that any individuals are altruistic, only that they vote altruistically. Whereas altruistic behavior often contains some element of self-sacrice, this dimension is hardly present in altruistic voting. Notice that this denition is much stronger than the denition of sociotropic voting or social preferences. Sociotropic voting only require individuals to take some account of what is good for society. People may do so and still be selsh, if they think that the best way of advancing their own interest is to create a better society. This critique has taken to imply that sociotropic voting need not reect altruism (Kiewiet and Lewis- Beck, 2011; Kinder and Kiewiet, 1981). In a similar vein, my denition of altruism is dierent from social preferences, which loosely means that individuals' put weight on other people's payo when making decisions (Charness and Rabin, 2002). As we do not know if these decisions are made for the sake of their own benet, social preferences need not be altruistic. For a large share of the respondents the altruistic choice, selsh choice, and actual choice will overlap. For these individuals it will not be possible to identify if they voted altruistically or not. This does not matter for the purpose of this study, as altruistic voting behavior of these individuals by construction will not inuence the outcome. These individuals will therefore not drive any of the observed dierences in voting behavior. Finally, it should be noted that it will not be possible to infer the motives for voting altruistically. To use the distinction of Schokkaert (2006), it could be grounded in duties, social pressure, empathy or something else. 3 Theoretical Background The equilibrium eects of altruistic voting behavior is a sparsely studied subject. Here I will distinguish between three ways in which altruistic voting may matter and based on previous literature generate hypotheses to be tested by the data. The rst deals with whether the actual votes are more similar to the altruistic votes or the selsh votes. The second deals with whether in a given political dimension the selsh and altruistic votes are to the left or to the right of the actual votes. The third deals with whether the variance of the dierent vote distributions diers. 1. The rst matter of concern is whether the actual votes are more in line with the altruistic votes or the selsh votes. This is implicitly what most of the literature summarized in the previous section deals with. If many more people vote altruistically than selshly as this literature suggests, we would expect the actual votes to lie closer to the altruistic votes than the selsh votes. The interpretation of this is that if we were to get more people to vote altruistically, the eect would be minor (as most people already vote altruistically). If, on the other hand, we were to get more people to vote selshly, the eect could be quite substantial. Notice, however, that the actual votes do not have to be related to the selsh or actual votes at all. The story in the epigraph about the two boys and the cake reect this possibility. If both boys had been selsh, they would demand the cake for themselves and a compromise would be to split it evenly. If both boys had been altruistic, supposedly they would both want to split the cake evenly and a settlement would be to do so. Once one is selsh and the other altruistic, however, the outcome reects neither the aggregation of selsh preferences nor the aggregation of altruistic preferences. For 6

7 this reason, Elster (1997, p. 14) argues that it can be `a dangerous thing' if some but not all vote according to what they believe constitutes the common good. If some vote selshly and others not, the preferences of the selsh voters will be counted twice, which may generate a socially inferior outcome. The rst aim of the empirical analysis will be to investigate whether such a pattern exists. The literature summarized in the previous section does not suggest so. Rather, it suggests that the actual votes are a combination of the selsh and altruistic votes, albeit much closer to the altruistic votes than the selsh votes. Hypothesis 1: The actual votes will be a combination of the altruistic votes and the selsh votes but much more aligned with the altruistic votes than the selsh votes. 2. Secondly, we want to determine in what political direction the votes would go if more voted altruistically or selshly. Or in other words, would the outcome be more leftwinged or right-winged if more people voted altruistically? In order to derive a hypothesis in this regard, one could set up a simple model. A natural starting point would be modeling one political axis, where voters derive utility from their income alone, and parties or candidates propose dierent income distributions. Individuals could be said to vote selshly if they maximize their own utility, and altruistically if they maximize the utility of the entire society. However, there would be several shortcomings to this approach. First of all, there is more than one dimension in politics. Low income individuals may vote right-wing for selsh reasons because cultural, religious, and social values outweigh economic concerns (Frank, 2007). Secondly, the set-up could change substantially if the model is dynamic rather than static, as individuals may be willing to tolerate inequality if they believe this is necessary to foster growth and and future well-being. Thirdly, there is no reason to believe that the altruistic judgments need to be utilitarian. Individuals may vote on what they think is best for society based on other concepts of justice, fairness or freedom. Individuals may also vote altruistically based on, for example, environmental concerns or other non-person specic outcomes. Finally, individuals may vote altruistically by voting for candidates based on valence judgments, such as their character traits and ability to govern (Stokes, 1963). In sum, it is very hard to make a credible hypothesis about the altruistic outcome through a simple model. Based on the ndings from Rueda and Stegmueller (2014), if altruistic preferences constitute a greater concern for equality, we would expect the altruistic voting to drive the outcome to the left in the economic dimension. Zettler et al. (2011) similarly argue that altruistic voting would drive the outcome to the left due to a concern for social equality, and Hudson and Jones (1994, 2002) nd that preferences for the public interest on average are more supportive of increasing public spending than votes based on self-interest. Based on these ndings we can generate a second hypothesis: Hypothesis 2: The altruistic votes will be more left-winged that the actual votes, and the selsh votes will be more right-winged. 3. Finally, it may be interesting to see if the variance of the vote distributions are impacted by altruistic voting. A lower variance can be interpreted as a greater degree of 7

8 political agreement and a smaller support for extreme candidates. A smaller variance could also indicate that individuals recognize the attitudes of the median voter to be best for society as a whole. Little previous literature is available for guidance in this perspective, and hence it is not possible to create a credible hypothesis. 4 Data & Methodology 4.1 Survey design To study the impact of altruistic and selsh voting on election outcomes, a representative sample of 2000 Danish voters were asked the following three questions: Who would you vote for if the general elections were held tomorrow? Who would you vote for if you only were to consider what is best for yourself? Who would you vote for if you were to consider what is best for society as a whole? The survey was conducted four months following the 2015 election, which saw the center-left government be replaced by a right-wing government. At the time of the survey nine political parties were represented in the Danish parliament, oering combinations of views in terms of both value politics and economic politics. This makes it less likely that voters who have dierent selsh and altruistic judgments, still believe that the same party is most capable of advancing both of these concerns. The survey was conducted by the Danish survey agency, Epinion, which delivers the opinion polls to the Danish Broadcasting Corporation, DR. The respondents are a representative sample of the Danish electorate based on gender, region and age. Survey weights are used throughout the analysis to assure full representativeness over these variables. Besides the three questions highlighted before, the questionnaire contains about 25 questions on voting behavior, political preferences, political knowledge, altruism, and demographics. The questions were to the extent possible adapted from the 2015 Danish Election Survey. 3 The full questionnaire is available in Appendix A.1. The questionnaire was tested using best practices in Cognitive Interviewing as described in Willis (2004). In total 12 cognitive interviews were conducted. The cognitive interviewing resulted in minor rephrasing of questions but revealed no noteworthy problems in terms of understanding or comprehending the three essential questions. Only individuals that do not regularly vote in elections were, unsurprisingly, unable to answer the questions on selsh and altruistic voting. When individuals had to answer what is best for themselves they mostly thought about their own economic or occupational situation. When answering what was best for the world at large they referred to ideology and ethical values when making their choices. None spoke of strategic concerns or misplacing their views so as to appear either selsh or altruistic. Indeed, to most of the interviewees it was not clear that the point of the questionnaire was to analyze altruistic voting behavior. 3 One question on whether voting is a duty or a choice was adapted from the 2010 American National Election Study and a question on distributive preferences was taken from the TV2 candidate test ( 8

9 A pilot of 200 respondents was conducted in October 2015 to ensure that there was enough variation to continue, which there was. Of the 2000 nal respondents, 1600 successfully answered all three central questions. 96% of the respondents who knew who to vote for if elections were held tomorrow were also able to answer either the altruistic or the selsh question, while 85% were able to answer both. This by and large conrms that the respondents were able to comprehend and answer the questions. When individuals have to answer the three central questions it is possible that they try to align their answers so as to appear consistent or that they falsely place their selsh choice so as to refrain from appearing as a selsh voter. This former would tend to diminish the dierences between the various vote distributions and imply that if any dierences are found, these should be considered lower bound. The latter would overestimate the dierence between the selsh votes and the other two votes. Although none of the cognitive interviews pointed to this being a threat, to make this kind of misrepresentation less likely, the respondents were randomly allocated into two groups. One group was asked the best for yourself question before the best for society as a whole question, while the other had the order reversed. It was not possible for the respondents to click back and change their previous answer. The reason for this division is that the respondents' incentives to misplace their true stances becomes more salient when they already have answered one or two of the three questions. Initially, the intention was only to use the answers to the rst question asked. However, a chi-squared test for independence of the party choices from the selsh question depending on which question was asked rst results in p-value of A similar test for the altruistic question results in a p-value of Relatedly, when checking if the share of altruistic and selsh votes for respectively the left and right wing is dependent on the question order, we get a p-value of 0.78 for the selsh question and 0.88 for the altruistic question. Consequently, the ordering does not seem to matter and the whole sample will be used when making inferences. 4.2 Placing the political parties in political axes The answers to the three central questions make it possible to detect dierences in vote distributions by political parties and political blocs. However, in order to gain a deeper understanding of where the votes move, the political parties are located in political axes. This will be done in two ways: 1. One dimension: The respondents are asked to locate each political party on a scale from 0-10, where 0 means extreme left wing and 10 means extreme right wing. Using the average of these locations, the dierent political parties can be placed in a onedimensional space. In order to ensure that the assessments are somewhat reliable, only the respondents who display sucient political knowledge (by correctly reporting the number of members of parliament and the parties partaking in the government) are used. The resulting placements are given in Figure 1. Using this method the four right-wing parties, C, V, O, and I cluster around the same point despite being quite dierent politically. For example, the Danish People's Party (O) is normally considered left of center on economic issues but far right-wing when it comes attitudes towards refugees etc. Liberal Alliance (I), on the other hand, is the most right-wing party economically, but potentially more moderate when it comes to attitudes towards 9

10 Figure 1: Placement of political parties Mean location of political parties 0 Ø F Å A B CVO I 10 Left wing Right wing Notes: The gure shows the average location of the various political parties on a scale from 0 (left wing) to 10 (right wing) using the mean assessment of respondents who displayed a level of political knowledge. refugees. The method fails to capture this due to the reduction of dimensionality. When individuals are forced to locate parties on a uni-dimensional scale they might not share the same interpretation of the scale. The second method deals with this problem. 2. Multiple dimensions: To capture the multiple dimensions present in Danish politics, I utilize candidate tests conducted by the two main Danish television stations, DR and TV2, prior to the 2015 elections. These tests asked all politicians running for parliament to display if they agree with respectively 15 and 42 statements on various political issues such as more jobs should be created in public sector. There were ve possible answer categories ranging from completely disagree to completely agree. Remarkably, respectively 168 and 162 of the 175 elected candidates to parliament answered these questions. Coding their answers makes it is possible to conduct factor analysis to check the number of relevant dimensions in Danish politics. In general there appears to be three dimensions; the economic dimension, the value dimension (containing issues such as immigration, the environment, and crime) and EU politics. As political discourse in Denmark has the that EU politics is not an important determinant of voting behavior, the rst two dimensions will be dealt with in this analysis. Consequently, each politician can be given a factor score in each of the two dimensions, and party averages can be taken such that the various parties can be located in a political compass. The compass using data from TV2 is shown in Figure 2. A detailed derivation of the compass is given in Appendix A.2. This approach easily captures the dierence between the Danish People's Party (O) and Liberal Alliance (I). Nevertheless, it also has some shortcomings. Firstly, the political parties may misreport their true stances in an attempt to be more attractive to the voters. Secondly, in so far as the relative positions of the political parties have changed since the elections, this data is no longer relevant. Thirdly, the questions may have been chosen so as to deliberately highlight political dierences that may not be very salient or important for the voters. The nal compass using data from TV2 deals with this issue by weighting the loadings with the importance of the question as assessed in a separate poll. 10

11 Figure 2: Political Compass (TV2) right wing VALUE SCALE left wing Ø F A Å B O V C I Party A: Social Democrats B: Social Liberal Party C: Conservatives F: Socialists' People's Party I: Liberal Alliance O: Danish People's Party V: Liberals Ø: Red Green Alliance Å: The Alternative left wing ECONOMIC SCALE right wing Notes: The gure displays the nine political parties in the Danish parliament in two political dimensions, the economic scale and the value scale. The data is from answers given to TV2's candidate test and the party positions are derived from an exploratory factor analysis (see Appendix A.2.1 for a detailed derivation). The scales have been standardized such that a value of 1 means one standard deviation from the mean position of the Danish MP's. After the political positions have been calculated using these two methods, the statistical analysis basically amounts to comparing i) the share of votes by party, ii) the mean location of the votes, and iii) the variances of the votes. With regards to comparing the share of votes by party, McNemar's test will be used. When comparing the mean location of the votes paired t-test will be used. For comparisons of variances Levene's test will be used to account for the high degree of non-normality. One may ask why I do not look at changes in the median vote rather than the mean and variance, given that the median voter likely will determine the outcome. The reason is, that the party the median voter favors will at times be unchanged even though there is a great shift of votes, and conversely, may be changed even though there is only a slight change of votes (Höchtl et al., 2012). With only one sample to analyze, this binary measure will not be very indicative. The size of the change in the mean location, on the other hand, will be suggestive of how often we can expect the median vote to change. 11

12 5 Results 5.1 Descriptive statistics Before analyzing how altruistic voting matters it is useful to establish how big an overlap there is between the altruistic votes, selsh votes, and actual votes. Table 1 breaks down the degree of overlap between the three central questions. Table 1: Overlap between altruistic, selsh, and actual votes Actual vote = selsh vote No Yes Sum No 9.3% 8.5% 17.8% Actual vote = altruistic vote Yes 25.8% 56.4% 82.2% Sum 35.1% 64.9% 100% Notes: The table breaks down the degree of overlap between the answers given to the actual voting question, the selsh question, and the altruistic question. n=1600 The table shows that 56.4% of the respondents selected the same party as their actual choice, altruistic choice, and selsh choice. This is hardly surprising as many individuals may convince themselves that what is best for themselves is also best for society at large (Edlin et al., 2007). 35.1% voted for a party they did not believe was best for themselves and 17.8% voted for at party they did not believe was best for society. It is not possible to directly infer if individuals voted selshly or altruistically for two reasons. Firstly, we don't know if the 56.4% voted for altruistic reasons, selsh reason or for a something third. Secondly, suppose we observe a person that votes for party A, selshly prefers party B, and altruistically prefers party A. We still cannot infer that this person voted altruistically, since the person may have voted for party A for a number of other reasons not related to altruism and egoism. However, we can infer that the person did not vote selshly. Hence, we can obtain upper bounds on the share that vote selshly and altruistically. We know for certain that at most 64.9% voted selshly and similarly that at most 82.2% voted altruistically. 9.3% voted neither selshly nor altruistically but potentially a mix of the two. We can try to characterize the voters who are likely to have voted altruistically (here proxied by the 25.8% that vote for what they think is best for society and not for what they think is best for themselves) rather than selshly (the corresponding 8.5%). This is shown in the rst columns of Table 2. Very few variables are able to predict altruistic voting. Only 'being interested in politics' and 'yearly gross personal income' positively predict voting altruistically. The rst may suggest that if you are interested in politics you are more inclined to vote for ideological reasons which need not overlap with what is in your own interest. The second nding, that income predicts altruistic voting, could be because wealthy individuals report that the most right-wing economic party is in their material interest but not in correspondence with their ambitions for what society should look like. Interestingly, two proxy variables for altruistic behavior (whether the person has donated blood or donated to charity) and a variable about preferences for redistribution come out insignicant. The reason for the many insignicant variables may also be that 12

13 Table 2: Who are the altruistic and consistent voters? Altruistic Voters Consistent Voters Coef. Std. err. Coef. Std. err. Female (0.30) 0.11 (0.15) Age 0.00 (0.01) 0.02*** (0.00) Education level (8-point scale) 0.04 (0.09) (0.04) Personal income (8-point scale) 0.23** (0.10) (0.05) Unemployed 0.34 (0.70) 0.24 (0.39) Employee in the public sector -0.58* (0.33) (0.17) Interested in politics (4-point scale) 0.48** (0.20) -0.29*** (0.10) Supporter of a political party (yes/no) (0.29) 0.71*** (0.14) Political standing (from 0=left to 10=right) 0.24 (0.22) -0.40*** (0.10) Political standing squared (0.02) 0.05*** (0.01) In doubt of who to vote for (yes/no) (0.52) (0.35) Political knowledge (yes/no) (0.33) 0.24 (0.17) Donated blood (yes/no) (0.27) 0.22 (0.14) Donated money to charity (yes/no) (0.31) (0.16) Preferences for redistribution (5-point scale) 0.05 (0.15) 0.07 (0.07) Voting is a duty (yes/no) 0.10 (0.28) (0.14) Would vote strategically (yes/no) (0.28) -0.34** (0.15) Observations Notes: * p<0.10, ** p<0.05, *** p<0.01. Based on a binary logit estimation. Altruistic Voters is a binary variable equaling 1 if the respondent votes for a party he thinks is best for society but not best for himself and 0 if he votes for a party he thinks is best for himself but not best for society. Consistent Voters is a binary variable equaling 1 if the respondent chooses the same party for the actual, altruistic, and selsh question and 0 otherwise. The indepenent variable Political knowledge equals 1 if individuals knows the number of seats in parliament and the parties in government, 0 otherwise. strictly speaking we are only detecting non-selsh versus non-altruistic voting, which blurs the picture. We can also see if there are dierences between the ones whose votes overlap (the 56.4% that vote for the same in all three questions) and the ones that don't (the remaining 44.6%). This is shown in the right part of Table 2. Since we here use all respondents that answer all of the demographic and political questions we have more observations. We see that the older you get the more likely you are to report the same answer to all three questions. The same applies to supporters of a political party. This may because people get interested in politics for non-selsh reasons, however, once you belong to a political party, the success of this party becomes your self-interest. Interestingly, individuals with political knowledge are less likely to give overlapping answers. This may be because they are more able to understand the nuances between the parties and hence select dierent ones for the dierent questions. From the variables on where the respondents would place themselves on a left-right scale, we get that individuals that rate themselves in the middle 13

14 are least likely to give overlapping answers. People that place themselves in the extreme ends are more likely to be consistent. Perhaps this is because individuals who are at the extreme ends of the political spectrum can gain the most from politics and hence are most likely to convince themselves that their own gains are also to the benet of society as a whole. Interestingly, we also see that individuals that would consider voting strategically give less overlapping vote choices. One could fear that this will drive the dierences between the altruistic, selsh, and actual vote choices. A robustness check will show that this is not the case. 5.2 Main results Let us next look at the distribution of votes by political bloc as shown in Figure 3. The right wing gets 52.5% of the selsh votes, 47.4% of the actual votes, and 46.1% of the altruistic votes. This suggests that altruistic voting generates a more left-winged outcome in line with hypothesis 2. Figure 3: Distribution of votes by bloc *** ** *** ** Share of votes Left Right selfish actual altruistic Notes: Distribution of selsh votes, actual votes, and altruistic votes by political bloc. Stars indicate signicant dierence from the share of actual votes using McNemar's test. * indicate p<0.10, ** p<0.05, *** p<0.01. n=

15 0.3 *** Figure 4: Distribution of votes by party Share of votes *** ** *** *** *** ** A B C F I O V Ø Å selfish actual altruistic Notes: Distribution of selsh votes, actual votes, and altruistic votes by political party. Stars indicate signicant dierence from the share of actual votes using McNemar's test. * indicate p<0.10, ** p<0.05, *** p<0.01. n=1600. We can break this down by political party as shown in Figure 4. The altruistic votes appear to be much more aligned with the actual votes than the selsh votes. This is in line with hypothesis 1. Only for the Socialdemocrats (A) is there a signicant dierence between the actual votes and the altruistic votes. The Socialdemocrats (A) is the major party in the left-wing bloc, which has governed Denmark in the majority of the past century. We also see that Conservatives (C), Liberal Alliance (I), and the Red-Green Alliance (Ø) receive more selsh votes than actual votes. The rst of two of these have the most right-wing economic policies while the latter has the most left-wing economic policies One dimension In order to break down the results in more detail the political parties are placed on a scale from 0 (left wing) to 10 (right wing) using the mean placement by respondents who displayed a minimal political knowledge. The top part of Figure 5 shows these average placements. The bottom part shows the average location of the actual, altruistic, and selsh votes using the political placements. The gure shows that the selsh votes were signicantly more right-winged than the actual votes. The altruistic votes almost overlap entirely with the actual votes. This suggests that if more people voted altruistically the outcome would hardly change. The gure also plots the election outcomes of the two most recent elections using the party placements weighted with the share of seats each party obtained in the given election. As Denmark shifted from a center-left to a right-wing government from 2011 to 2015, 15

16 Figure 5: Mean location of votes Mean location of political parties 0 Ø Left wing F Å A B CVO I Mean location of votes 10 Right wing p< Altruistic votes Actual votes Selsh votes Election outcomes Notes: The top part of the gure shows the average location of the various political parties on a scale from 0 (left wing) to 10 (right wing) using the mean assessment of respondents who displayed a level of political knowledge. Given this placement of the political parties, the bottom part shows the mean location of the respondents' actual, altruistic, and selsh votes. The altruistic votes and actual votes overlap almost entirely. The p-values are from paired t-tests. The election outcome circles are constructed by using the placements of the political parties weighted with the share of seats each party obtained in the given election. and the dierence between the selsh and actual votes is of greater size, this dierence is of meaningful size. There is a relative big dierence between the 2015 outcome and the 'actual outcome' for a few reasons. First of all the survey was conducted 4-5 months after the election, which allows some voters to have changed their mind. In the intervening months Denmark was met by a refugee crises, a change of leadership in the main opposition party, and a national budget proposal with drastic changes in funding for development aid, environmental issues, education and more. Another reason is that the sample for this survey is only representative by age, gender, and region, which may not be sucient to capture all types of voters. In addition 14% of the voters do not know how to vote for either selshly or altruistically which may give further issues with regards to representativeness. That being said, the within variation that is exploited here should be unaected by this. We can break down the results by dierent types of voters as shown in Figure 6. The top (bottom) part of the gure shows the average location of the altruistic, selsh, and actual votes for the subgroup of respondents who selected a left-wing (right-wing) party for all three choices. The middle part shows the location of the votes for respondents who selected both a right wing party and a left-wing party in one or more of the three questions. A clear pattern emerges. Supporters of either side of the political spectrum believe that what is best for society is towards the center of their own bloc whereas what is best for themselves is further to the extreme in their own bloc. The results from Figure 5 are thus driven by the ones that jump between wings. Figure 6 seems to suggest that the variance of the votes dier. This is tested in Figure 7, which shows the standard deviation of the actual, altruistic, and selsh votes. Clearly, the selsh votes have more variance than the actual votes, and the altruistic votes have less variance. This suggests that the altruistic votes cluster more around centrist candidates, or in other words, that extreme candidates are less chosen. 16

17 Figure 6: Mean location of votes by type Left-wingers 2.9 p=0.24 p< Movers 4.5 p=0.01 p< Right-wingers 8.05 p=0.45p= Altruistic votes Selsh votes Actual votes Notes: Shows the mean location of the actual, altruistic, and selsh votes by voter type. Left-wingers: Individuals whose actual, selsh, and altruistic vote was for a left-wing party. Movers: Individuals whose actual, selsh, and altruistic votes contained both a left-wing party and a right-wing party. Right-wingers: Individuals whose actual, selsh, and altruistic vote was for a right-wing party. The scale is once again from 0 (left wing) to 10 (right wing). The p-values are from paired t-tests. Figure 7: Standard deviation of votes 3.25 p=0.01 p= Altruistic votes Actual votes Selsh votes Election outcomes Notes: The gure shows the standard deviation of the actual, altruistic, and selsh votes using the mean placements of the political parties from Figure 4. The p-values are from a Levene's test. 17

18 5.2.2 Two dimensions A problem with the results presented thus far is that analyzing politics in one dimension may hide relevant information. To deal with this, the mean location of the selsh, actual, and altruistic votes will be compared using the political compass showed in Figure 2. This allows us to break down the party movements by the economic axis and the value axis, as shown in Figure 8. By comparing the distance between the three points on the economic dimension with the distance between the three points on the value dimension, one quickly sees that economic concerns seem to drive most of the dierence. In terms of value politics there is little dierence between which party is chosen for selsh and altruistic reasons. The selsh votes are more right-winged in both dimensions, but particularly in the economic scale. Reversely, the altruistic votes are slightly more left-winged economically, but with no noteworthy dierence in value politics. The reason why there is less of a dierence in value politics may be because values in general incorporate views about society as a whole and to a lesser extent private concerns. It may be harder for individuals to evaluate what values are best for themselves in contrast to evaluating what economic policies are best for themselves. Again, comparing the dierences with the 2011 and 2015 election outcomes, these dierence are of important magnitude. Figure 9 similarly breaks down the results in the two dimensions by variance. Again the distance between the points is larger in the economic scale than the value scale. Selsh votes have a much greater variance in the economic dimension. Hence, when people vote selshly they tend to choose parties who are on the extreme of the economic dimension. Intuitively this makes sense. If you have an above average income, your selsh vote from an economic perspective is the one that proposes the lowest tax rates. If you have a below average income, it could be in your selsh interest economically to vote for the party that favors the most redistribution. The altruistic votes are once again less extreme, this time in particular in the value dimension. This may suggest that when individuals are to think of which values are most best for society as a whole, they try to think of what the average person believes. To sum up, the analysis gives the following four insights: 1) The respondents appear to vote more altruistically than selshly, as the actual votes align closely with the altruistic votes. This means that most people seem to vote altruistically. In addition it implies that the inferior pattern discussed in the epigraph has no empirical leverage. 2) The selsh votes are much more right-winged, and the altruistic votes slightly more left-winged than the actual votes. This is in line with hypothesis 2 and suggests that individuals on average become more egalitarian when thinking about what is best for society as a whole. 3) The selsh votes are placed on more extreme candidates, while there is more agreement within the altruistic votes. Hence, the more altruistically individuals vote the more consensus there will be among the selected candidates and, supposedly, the better equipped they will be to reach compromises. 4) The latter two results apply primarily to the economic dimension of politics. Hence, when individuals think about what is best for themselves or best for society they mostly refer to tax and redistributive policies. 18