Political Behavior

, Volume 35, Issue 1, pp 43–63

The Politics of Interpersonal Trust and Reciprocity: An Experimental Approach

Original Paper

DOI: 10.1007/s11109-011-9181-x

Cite this article as:
Carlin, R.E. & Love, G.J. Polit Behav (2013) 35: 43. doi:10.1007/s11109-011-9181-x


Trust and reciprocity are theoretically essential to strong democracies and efficient markets. Working from the theoretical frameworks of social identity and cognitive heuristics, this study draws on dual-process models of decision making to expect (1) the trustor to infer trustworthiness from partisan stereotypes and thus to discriminate trust in favor of co-partisans and against rival partisans, but (2) the trustee to base reciprocity decisions on real information about the trustor’s deservingness rather than a partisan stereotype. So whereas partisanship is likely to trigger trust biases, the trust decision itself provides enough information to override partisan biases in reciprocity. The analysis derives from a modified trust game experiment. Overall, the results suggest partisanship biases trust decisions among partisans, and the degree of partisan trust bias is consistent with expectations from both social identity theory and cognitive heuristics. When it comes to reciprocity, however, information about the other subject’s level of trust nullifies partisan bias.


Interpersonal trust Partisanship Behavioral economics Political psychology Experiments 

Trust creates opportunities that, when met with reciprocity, result in cooperation. These social preferences and behaviors help resolve myriad collective action problems facing democracies and markets (Almond and Verba 1963; Olson 1965; Arrow 1974; Granovetter 1985; Inglehart 1988; Ostrom 1990; Putnam 1993; Paxton 2002). Scholars disagree on our progress toward understanding cooperative behavior. One review says, “The question of trust is a huge puzzle that is not even near solution” (Nannestad 2008, 432). More optimistically, Science (2005) places the emergence and evolution of cooperative behavior among the twenty-five fundamental questions we should be able to answer by 2030. Speaking to a wide-ranging debate, this study explores the political underpinnings of trust and reciprocity. It asks whether partisan identities, which bolster democratic politics in multiple ways, undercut social norms of trust and reciprocity. If so, strong and polarized partisanship could undermine the efficiency and effectiveness of liberal politics and free markets.

Party identification structures political beliefs and behavior (Campbell et al. 1960; Fiorina 1981; Jacoby 1988; Green et al. 2002; Goren 2002; Bartels 2002), as well as deeper, more enduring political values (Goren et al. 2009). But it also shapes altruism, a social preference (Dawes et al. forthcoming; Loewen 2010; Fowler and Kam 2007; Fowler 2006). Thus, partisanship potentially conditions trust and reciprocity, the key social preferences to well-functioning democracies and markets.

This matter warrants attention because American citizens and elites are becoming more strident partisans (Brewer 2005; Hetherington 2001) while political parties are growing more ideologically polarized (Layman et al. 2010). Scholars debate whether all Americans are more polarized, but few dispute that partisans are (Abramowitz and Saunders 2008; Fiornia et al. 2006; Evans 2003; DiMaggio et al. 1996). Politicians and pundits praise bipartisanship, and citizens dislike partisan bickering and legislative gridlock (Hibbing and Theiss-Morse 1995, 2002). Strengthening partisanship and polarizing parties at once raise the need for bipartisan cooperation and lower its prospects. Yet we do not know whether partisan conflict extends to cooperative social behavior among citizens. If so, partisanship will have penetrated far beyond politics and into the social realm of personal interaction.

This study advances a political theory of trust and reciprocity by testing dual-process strategies based on social identity and cognitive heuristics. It finds partisanship biases interpersonal trust: co-partisans trust each other more than rival partisans. Both Democrats and Republicans behave in this manner. Partisan trust bias rises with strength of partisanship and with political sophistication. Though partisanship highly conditions trust, it does not condition reciprocity. Hence partisan stereotypes hamper cooperation by reducing trust, not by curtailing reciprocity. Simply put, partisanship can be a red herring in trust decisions because trusting actions beget reciprocity regardless of partisanship.

Theoretical Framework

Trust is “a psychological state comprising the intention to accept vulnerability based upon positive expectations of the intentions or behavior of another” (Rousseau et al. 1998, 395). Reciprocity connotes a “pattern of mutually contingent exchange” (Gouldner 1960, 161) compatible with purely rational self-interest, though it need not be tit-for-tat (Axelrod 1984). Rather, this study assumes “relaxed” reciprocity is a continuous decision which can be given in any amount (Kollock 1993). Reciprocity and trustworthiness are thus used interchangeably. These definitions map closely onto the experimental measures of trust and reciprocity employed here. Our expectations vis-à-vis partisanship and trust flow from two venerable and compatible theoretical traditions: social identity and cognitive heuristics.

Social identity theory claims identities shape social perceptions, attitudes, and behaviors, and salient group differences lead people to form psychological attachments to an “in-group” (Tajfel 1978). In-group members (1) magnify differences between themselves and a psychologically relevant “out-group”; (2) exhibit favoritism toward in-group members; and (3) perceive the out-group as undifferentiated, dissimilar, and inferior. Cognitive heuristics are a set of “principles which reduce the complex tasks of assessing probabilities and predicting values to simpler judgmental operations” (Tversky and Kahneman 1974, 1124). Because heuristics help compensate for limited information about or understanding of social and political reality, one need not be a walking encyclopedia to make (boundedly) rational decisions (Kahneman et al. 1982; Lupia 1994; Lupia and McCubbins 1998).

For trust, both theories lead to similar predictions because they translate limited information into stereotypes (Chandra 2003; Burns 2006). Social identity theory predicts individuals will use such information to derive expectations of trustworthiness based on out-group stereotypes. Cognitive heuristics models predict individuals will use stereotypes as informational shortcuts for trustworthiness. Consistent with these expectations, total strangers are often willing to cooperate (Cárdenas and Carpenter 2008; Johnson and Mislin 2009), but trust biases emerge when information about one’s counterpart is shared. Why? In short, we trust some people more than others because we believe or expect they are more likely to reciprocate—to be trustworthy—based on the information available (e.g., Glaeser et al. 2000; Ashraf et al. 2006). As Wilson and Eckel (2006) summarize, “people often trust those they do not know, and they do so on a belief that they can ‘read’ the trustworthiness of others based on something signaled by their counterpart” (189).

Two main types of trustworthiness “signals” may induce trust bias. The first are personal attributes such as race (Burns 2006; Wilson and Eckel 2006), gender (Buchan et al. 2008; Gabarino and Slonim 2009), and age (ibid). Smiles and attractiveness also encourage trust differentiation (Scharlemann et al. 2001; Eckel and Wilson 2003; Wilson and Eckel 2006), and many subjects will even pay for photos of their counterparts before deciding to trust them (Eckel and Petrie 2011). A second potential source of bias is social distance, according to studies of trust across ethno-linguistic, religious (Fershtman et al. 2005; Johansson-Stenman et al. 2009), and class lines (Cardenas et al. 2009). Attributes and social distance can also have interactive effects on trust (Fershtman and Gneezy 2001; Haile et al. 2008). In sum, interpersonal differences “strongly predict a tendency to cheat one another” (Glaeser et al. 2000, 840). Partisanship should condition trust to the extent that it serves as a social identity or a cognitive heuristic. Scholars argue both points.

Greene (1999) considers The American Voter’s definition of party identification—“an affective attachment to an important group object in the environment” (Campbell et al. 1960, 143)—a “precursor of social identity theory years ahead of its time” (394). Because group-differentiation is at the heart of social identity theory, “the group nature of partisanship should naturally create a bipolar partisanship where individuals characterize the political parties into us and them and exaggerate perceived differences to favor their own group” (Greene 2004, 138). Partisan identity metrics imbued with explicit feelings of in-group belonging heighten degrees of inter-party differentiation and strengthen links between partisanship, party perceptions, ideology, and political behavior (Greene 1999, 2004). Experiments in the U.S. (Fowler and Kam 2007) and Canada (Loewen 2010) find greater altruism toward co-partisans than rival partisans. Survey research in Chile finds biases in affect between co-partisans and sympathizers of parties a short distance away in ideological space (González et al. 2008). Partisanship as a social identity may thus shape trust decisions.

Heuristic processing is activated when “substantive information is minimal” and when external cues are prominent and explicit (Mondak 1993, 171). In this sense, decisions to trust partisans resemble political decisions taken under uncertainty and with scant information, e.g., if, what, and for whom to vote (Downs 1957; Popkin 1994; Bartels 1996; Lau and Redlawsk 2001; Lodge and Hamill 1986). Labels also help voters better predict candidates’ positions on the ideological spectrum and issues (Conover and Feldman 1981; Huckfeldt et al. 1999; Koch 2001). Partisanship’s heuristic value is crucial in such complex political decisions, and similar heuristic processing may apply to trust decisions if all one knows is a stranger’s partisanship. Two perspectives describe how partisan stereotypes might act as cognitive heuristics for interpersonal trustworthiness.

Partisanship cues should spur heuristic processing by which individuals extend their views of a political party’s trustworthiness onto people or organizations associated with that party (Mondak 1993; Turner 2007). Co-partisans would thus be considered trustworthy and rival partisans seen as untrustworthy by extension. The heuristic may relate to ideological congruence, reputation, or both (Kuklinski and Hurley 1994; Turner 2007). Party stereotypes are used heuristically because they contain at least some real information on policy, group alliances, traits, and performance. Hence “as ‘best guesses’… about the characteristics of individual party members, especially those individuals wearing the party’s mantle, partisan stereotypes would not be unreliable guides in the absence of other information” (Rahn 1993, 475). Either perspective predicts heuristics to foment partisan trust biases.

Although social identity theory and some cognitive heuristics models may predict group-based reciprocity discrimination, the evidence is thinner. Trust gaps in Belgium between Flemish and Walloon, and between ultraorthodox Jews and secular Israelis, are not matched with corresponding reciprocity gaps (Fershtman et al. 2005). Reciprocity bias occurs in just one of Cárdenas et al.’s (2008) forty-two tests of social distance in Latin America. Eckel and Petrie (2011) show demand for a photo of one’s counterpart is more inelastic for would-be trustors than those faced with the decision to reciprocate. Why do stereotypes give such poor traction on reciprocity?

Actions appear to speak louder than stereotypes. Trustors receive more reciprocity when they have not acted unfavorably to the trustee (Charness and Rabin 2005; Schweitzer et al. 2006), and racial discrimination disappears in the face of task-relevant information (Castillo and Petrie 2010). Dashed expectations of trust, based on trustor attractiveness, lead to a “beauty penalty” in reciprocity (Wilson and Eckel 2006). Such findings run counter to basic social identity theory and cognitive heuristics assumptions, and suggest hard bits of real data trump the less reliable information of stereotypes in strategic cooperative decisions.

For the research questions at hand, stereotypes feed group-based expectations of trustworthiness in the absence of more relevant information. But if real information is present, as it is in reciprocity decisions, stereotypes may play at best a minor role. Hence partisanship should promote trust biases in favor of co-partisans and against rival partisans. Yet the decision to reciprocate is made with more relevant information, and thus partisan stereotypes may not bias reciprocity. These diverging expectations can be characterized by dual-process decision models (cf. Chaiken and Trope 1999), which distinguish theory-driven from data-driven decision-making strategies.

Theory-driven strategies seek to simplify and interpret information by extending stereotypes—knowledge, beliefs, and expectations about a group (Rahn 1993, 474)—to individual members of the group. They focus on information congruent with pre-existing stereotypes and act accordingly, and are optimal for “settings in which the stereotype is available and particular information is lacking, but it may also happen in the absence of an explicit stereotype label if the attributes of the individual target easily ‘signal’ group membership” (Rahn 1993, 477). Data-driven strategies, however, are expected when decision-makers either lack a heuristic, or confront information incongruent with expectations derived from stereotypes. In the latter case, individuals can choose to reject the stereotype as irrelevant upon weighing it against stereotype-incongruent information. For data-driven decisions, then, “inconsistency motivates individuals to pay attention to target attributes, thereby decreasing the influence of the stereotype on judgments” (Rahn 1993, 477).

Rahn’s dual-process model of partisan heuristic processing may indeed be appropriate for political decisions. But its applicability to strategic social decisions, where outcomes are a function of both one’s own and others’ behaviors, is as yet unknown. This study applies such a model to explain trust and reciprocity between partisans. Hence the guiding hypothesis is that partisan stereotypes will influence behavioral elements of cooperation in different ways. When partisanship alone is known, as in the decision to trust, theory-driven information processing should create positive trust bias toward co-partisans and negative trust bias toward rival partisans. Yet when both partisanship and real information are on hand, as in the decision to reciprocate, data-driven strategies could (but may not) dominate. Hence those who risk trusting a rival partisan could see their trust reciprocated. We analyze these expectations empirically with a methodology called the “trust game.”

Measuring Trust & Reciprocity

This study adopts Berg et al.’s (1995) version of the trust game to measure trust and reciprocity. In its basic form, it is played by two randomly assigned anonymous players who do not meet before playing. Player 1 is given a sum of money, say $10, and told she can share some, none, or all of it with Player 2, who was also given $10.1 Player 1 is told that any sum shared will be tripled before giving it to Player 2, and that Player 2 will be given the same options—to return some, none, or all of it to Player 1.Trust is measured as the amount Player 1 sends to Player 2. Reciprocity is measured by the amount Player 2 returns. These measures match the conceptualizations of trust and reciprocity outlined above.

Since Player 2 has no incentive to return any of the money, the unique sub-game Nash equilibrium is for Player 1 to pocket the $10, passing none to Player 2. Most studies, however, find that, on average, both players give above the equilibrium amounts (Cárdenas and Carpenter 2008; Johnson and Mislin 2009). In other words, trusting behavior on the part of Player 1 tends not to be completely misplaced in Player 2. Rather, Player 2 often reciprocates, thereby honoring the trust placed in them by behaving in a trustworthy manner.2

Let us consider internal, construct, and external validity (Shadish et al. 2002)3 in the context of the present trust game study. Internal validity refers to determining causality. As detailed below, in an initial anonymous trust game, Player 1 can choose to give Player 2 something of value with the possibility, but no guarantee, that Player 2 will return something of less, equal, or greater value. The anonymous game establishes baseline levels of trust and reciprocity untainted by partisanship—akin to a pre-test. Subsequent trust games introduce political party identification information in a controlled way and thus create treatment effects—rendering post-test measures of trust and reciprocity. Since partisanship is the only cue Player 1 has about Player 2, one can confidently infer that the relationship between partisanship and trust is causal. In the case of reciprocity, Player 2 has two pieces of information, partisanship and the amount they received from Player 1. Lack of discrimination along partisan lines would suggest greater attention to real information than stereotypes in the reciprocity decision. Not only is the internal validity of the relationships in this study quite high in its own right, it improves mightily on the correlation associations found with cross-sectional survey data that pervade the literature.

Construct validity concerns how well the measures and inferences match working conceptualizations and theory. Here the question is whether the trust game gauges trust and reciprocity, and answering it requires a conceptual discussion and empirical evidence. Conceptually, choices made in the game map directly onto the definition of trust above: Player 1 trustors become vulnerable to and dependent on Player 2 trustees to a degree that corresponds to the amount they send. Since any amount returned qualifies as reciprocity, the game play fits a “relaxed” conception of reciprocity. Empirically, three conclusions about the trust game’s validity are noteworthy. Although Player 1 trusting behavior in the trust game correlates weakly with common trust survey questions, it correlates highly with past reported trusting actions (Glaeser et al. 2000) and scores on a social value orientation scale (Van Den Bos et al. 2009). Functional MRIs of subjects’ brain behavior during the trust game conclude that both trust (Krueger et al. 2007) and reciprocity decisions (Van Den Bos et al. 2009) activate the aMPFC, the area of the brain considered responsible for processing strategic interactions and integrating perspectives of self and other (Harris et al. 2007).

External validity is higher if a causal relationship “exists across a range of experimental and observational datasets” (Morton and Williams 2008). Should the results of this study—conducted via internet with a sample of public university students—be expected to hold more broadly? Student samples, Druckman and Kam (2011) argue, do not undermine external validity because students and non-students differ very little on most politically relevant variables. If strength of partisanship is lower among students, many of whom may still be forging party identities, it should make parties both less salient social groupings and weaker informational cues (Campbell et al. 1960; Druckman and Kam 2011). If anything, a student sample makes for more stringent hypothesis tests in this instance.

An externally valid study should also comport with studies on other samples. Calculations based on Cárdenas and Carpenter’s (2008) extensive review of 28 anonymous trust-game studies on student and non-student samples around the world indicate the proportion sent by Player 1 ranges from .30 to .73 with a mean of .51 and a standard deviation of .10. On average, Player 1s in this study send .44 in the anonymous game, well within a standard deviation of that mean. This result also aligns with trust games on a nationally representative sample in Germany (Fehr et al. 2003). Finally, a major validity concern of internet-based trust games is that subjects tend to trust more when they do not believe they are playing against a real person (Eckel and Wilson 2006). Since the current study does not find higher-than-average levels of trust, online game play does not appear to harm validity.

The trust game has advantages over the standard interpersonal trust measure pioneered by the General Social Survey (GSS): “Generally speaking, would you say that most people can be trusted, or that you can’t be too careful in dealing with people?” Researchers can tailor the trust game to specific groups—in this case co- and rival partisans—instead of the blunt trust object, “most people.”And whereas survey questions tap expressed preferences, the trust game taps revealed preferences, a key distinction since, when it comes to trust, people do not always put their money where their mouth is (Glaeser et al. 2000).4 Lastly, the GSS measure is sensitive to slight changes in question wording and response choices (Renno and Castro 2009) and may lack construct and external validity (Miller and Mitamura 2003). Thus, observed trust game measures of trust and reciprocity are preferable.

Research Design & Hypotheses

To test whether partisanship conditions interpersonal trust and reciprocity, subjects were recruited to play internet-based trust games and to complete a short survey of political attitudes and demographics hosted at a dedicated website. Recruitment was done in person and via email from majors across the social sciences, humanities, and business at the University of Mississippi in April 2009. A total of 138 subjects (69 Player 1s, 73 Player 2s) took part in the study by playing four versions of the trust game. The first game was a classic anonymous trust game (1). But at the start of the next three games the subjects received treatments—they were informed the other player was (2) an Independent, (3) a Democrat, and (4) a Republican. As the protocol (see online supplement) shows, this is the only information subjects were given, and all subjects played the four games in this exact order.5

Conducting trust games via internet presents unique challenges, including the use of incentives and the delivery of pay-outs. Though lab-based trust games can simply use cash, here both players begin each game with an allotment of 10 virtual lottery tickets (Fowler and Kam 2007).6 Each ticket has an equal chance of winning one of three $100 prizes.7 Given the tripling aspect of the game, the maximum number of tickets any subject could obtain is 40 per game, the minimum is 0. Once complete, the lotteries were conducted and the university sent checks to the winners via U.S. mail.

Coordinating the sequential game was another challenge. The trust games reported below were conducted in two waves. In the first, half the sample played as Player 1 trustors. In the second wave, the other half played as Player 2 trustees responding to actual ticket allocations made by first-wave Player 1 trustors. Since the distribution of partisanship between the two samples was not identical8 a problem of sequencing similar to embedding the trust game in a public opinion survey arose. So Player 2s were randomly matched with actual amounts sent by a Player 1 in an ex-post fashion (cf. Fehr et al. 2003). Subjects were told they were interacting with real people and they actually were. Because this study is only interested in the behavior of partisans, the few subjects in the sample who identified as Independents were excluded from the analysis.9

This research design allows tests of three general hypotheses about trust/reciprocity differentiation, and supplementary hypotheses about the effects of social identity and heuristics. Because the dual-processing framework does not make a specific prediction regarding the use of heuristics over information, the two reciprocity hypotheses are competing. The three general expectations, subscripted T for trust and R for reciprocity, are as follows:


Partisans trust co-partisans more than others. Hence Player 1 will send more tickets to a known Player 2 co-partisan than to a known Player 2 rival partisan or to a Player 2 whose partisan identity is unknown (theory-driven, heuristic processing).


Partisanship has no effect on reciprocity. Hence the percentage of available tickets Player 2 returns will not vary according to Player 1’s partisan identify (data-driven, information processing).


Partisans reciprocate more with co-partisans than with others. Hence Player 2 will return a greater percentage of available tickets to a known Player 1 co-partisan than to a known Player 1 rival partisan or to a Player 1 whose partisan identity is unknown (theory-driven, heuristic processing).

To further examine how partisanship shapes trust decisions, a second analysis examines strong partisans. According to self-categorization theory, an offshoot of social identity theory, people who perceive a high degree of similarity between themselves and the prototypical member of a social group are more likely to take on and develop that social identity (Turner et al. 1987; McGarty et al. 1992). If discrimination based on partisanship behaves as a social identity, then strong partisans should exhibit more in-group bias than weak partisans (Fowler and Kam 2007). From the perspective of cognitive heuristics, partisanship should be a more obvious cue to strong partisans than weak partisans. Thus strong partisans should infer an even greater lack of trustworthiness from party signals. Trust expectations consistent with these logics are reported below, superscripted PID-Strength.


Strength of partisanship increases the trust bias towards co-partisans. Hence the difference in tickets Player 1 sends to a known Player 2 co-partisan and to a known Player 2 rival partisan will increase with the strength of Player 1’s party identification.

A related analysis examines the role of political knowledge in shaping partisan trust discrimination. The more knowledgeable a person is regarding politics, the more likely they will use a political heuristic to shape their trust decision. That is, people with more political knowledge are more likely to view a partisanship cue as conveying relevant information about the other person’s trustworthiness. Viewed from social identity theory, more informed citizens should better grasp stereotypical differences between partisans. Thus, we expect greater trust differentiation along party lines by more knowledgeable subjects.


Political knowledge increases the trust bias towards co-partisans. Hence the difference in tickets Player 1 sends to a known Player 2 co-partisan and to a known Player 2 rival partisan will increase with the degree of Player 1’s political knowledge.

In dual-process models, Player 2’s reciprocating behavior may be shaped by Player 1’s partisanship (cue) as well as the degree of trust Player 1 signals (information). Since the reciprocity decision is made in a low-information environment, the amount of trust placed in Player 2 may prove more critical in the overall decision than partisanship. The question, then, boils down to whether partisan stereotypes trump the value of real information regarding the deservingness of reciprocity. If Player 2 returns the same proportion of tickets to Player 1 co-partisans and rival partisans, then party cues are less critical than actual trust signals (consistent with HR1, not HR2). Through this comparison, the analysis will illuminate whether people rely upon heuristics (Player 1’s partisanship) or behavioral information (amount of trust by Player 1) when making anonymous reciprocity decisions.


This section presents broad empirical evidence that partisanship conditions individuals’ decisions to trust, but that behavioral information is more crucial to reciprocity decisions than partisan cues or stereotypes.

Partisanship and Trust

As a baseline for comparison, Fig. 1 displays the average number of tickets sent in each of the four games. The similar trust levels indicate that, in the aggregate, anonymous players, Democrats, Republicans, and Independents are trusted about equally. Significant trust biases only arise when comparing trust between co-partisans, on one hand, and rival partisans and anonymous players on the other hand.
Fig. 1

Average tickets all Player 1s sent to Player 2s in anonymous, independent, republican and democrat trust games

Figure 2 shows the average trust, measured in tickets sent, broken down by partisan identity. Since trust in Independents varies insignificantly by party identification, the bulk of this section focuses on partisans (Democrats and Republicans). The results provide robust evidence that, for partisans, partisanship conditions the degree of interpersonal trust (HT). Democrats give an average of 1.2 tickets less to Republicans than to fellow Democrats. Republicans, in turn, give about 1.3 tickets less to Democrats than to their co-partisans. Abstracting up a level, the average difference between trust in co-partisans and trust in rival partisans is 1.28 tickets (see Table 1)—a substantively high and statistically significant difference according to a parametric matched-pairs t-test and a non-parametric Wilcoxon matched-pairs signed-ranks test (Keele et al. 2008).
Fig. 2

Average tickets democrat and republican Player 1s sent to Player 2s in anonymous, independent, republican and democrat trust games

Table 1

Trust differentiation: Player 1 ticket allocations to Player 2 in the trust game



p value

P value



Wilcoxon signed-ranks test

Co-partisan trust—rival partisan trust*






Co-partisans trust—anonymous trust**






* The difference in tickets sent by Player 1 to Player 2 co-partisans and Player 2 rival partisans

** The difference in tickets sent by Player 1 to Player 2 co-partisans and Player 2s whose partisanship is unspecified

How large are these effects? The effect size of the partisanship information treatment is one of the largest in the literature. The 34% more tickets that partisans send to their co-partisans than to rivals is much greater than what women send to other women versus men (12% increase) (Gabarino and Slonim 2009), the amount people over 50 send to others in their age cohort versus those under 25 (23% increase) (ibid), or the small and inconsistent effects of gender and socio-economic status treatments in six South American cities (Cárdenas et al. 2008). Co-partisans even trust each other more than anonymous players, though the difference (.64 tickets, 15%) is smaller. On the whole, partisans exhibit greater trust in co-partisans than in others. These findings align with prior research suggesting partisanship conditions social preferences (Loewen 2010; Fowler and Kam 2007; Fowler 2006; González et al. 2008).

In light of Cox’s (2004) argument that the trust game taps altruism in addition to trust, and Fowler and Kam’s (2007) evidence of partisan altruism bias in dictator games, one might wonder if the discrimination displayed in Table 1 is indeed trust discrimination. At least two reasons suggest so. First, trust decisions are linked to expected reciprocity in contexts as varied as Germany (Fehr et al. 2003), Russia, South Africa, and the U.S. (Ashraf et al. 2006). That is, the amount of tickets Player 1 sends is a function of the amount of tickets Player 1 expects in return. Cárdenas et al. (2008) report expected reciprocity is a better predictor of trust than social distance or risk preferences in five of the six Latin American capitals they study. And researchers who analyze subjects’ written descriptions of their thought processes in the trust game indicate expected reciprocity is crucial to trust decisions (Chaudhuri and Gangadharan 2003; Lev-On et al. 2010). Even if the amounts sent in the trust game are not entirely based on expected reciprocity, it is unlikely the discrimination results presented above reflect purely altruistic preferences.

Second, the effect sizes reported here are much larger than those recorded in Fowler and Kam’s (2007) dictator game. Indeed, subjects in our anonymous trust game sent significantly more tickets (44%) than in Fowler and Kam’s anonymous dictator game (29.9%). More importantly, partisan discrimination is far greater in the present trust game. In Fowler and Kam’s dictator game, Democrats sent 5.5% points more to fellow Democrats than Republicans, compared to 12% points in our trust game; Republicans have a 3.1%-point altruism gap compared to an 11-point trust gap. As for intrasubject differences, Fowler and Kam find gaps of 5.5% points for Democrats and 2.8 for Republicans. In the trust game these differences are well over twice as large (12 and 13% points, respectively). Thus altruism cannot be the exclusive factor.10

Beyond basic partisan differentiations, a second set of analyses examine whether strength of partisanship and level of political knowledge amplify or muzzle the effect of partisanship on trust (HTPID-Strength and HTKnowledge). A tobit model11 in Table 2 tests both hypotheses. The dependent variable is the difference between the number of tickets subjects sent to co- and rival partisans. A post-test survey measured the independent variables—strength of party identification and political knowledge—and a range of demographic, political, and social controls.12
Table 2

Trust differentiation between co-partisans and rival partisans


Coefficient (standard error)

Political sophistication

2.42* (1.38)


.29* (.78)

Strong partisans

5.38** (1.62)

Moderate partisans

0.34 (.77)

Internal political efficacy

0.6 (.73)


0.18 (.16)


−1.37* (.72)


−.04 (.09)

Religious attendance

.15 (.29)


−1.15 (2.50)




2.75 (.36)

Pseudo R2


Entries are tobit coefficients; 3 observations are right-censored

p ≤ .1 (two-tailed test), ** p ≤ .05

Evidence from this model is consistent with both hypotheses. In line with the strength of partisanship hypothesis (HTPID-Strength), strong partisans more heavily bias trust in favor of co-partisans. Compared to weak identifiers, they send a whopping 5.4 more tickets (out of an initial 10) to co-partisans. Moderate identifiers behave no differently than weak identifiers. And consistent with the notion that those who are more politically informed are more likely to act on party cues (HTKnowledge), for each additional political knowledge question answered correctly, subjects gave .35 tickets more to co-partisans than rival partisans. In other words, going from the lowest level political knowledge to the highest increases trust discrimination by 2.42 tickets. This finding is statistically robust and parallels evidence that political sophisticates use political cues to inform their vote choices (Lau and Redlawsk 2001). Thus the data from Tables 1 and 2 illustrate how partisanship is relevant not only in the political realm of decisions but in the realm of social preferences as well.

Two additional results merit mention. First, Democrats discriminate trust slightly more than Republicans. This seemingly contradicts the Democratic Party’s rhetoric and issue positions, but is consistent with Democrats’ behavior in studies of altruism (Fowler and Kam 2007). Second, men show more partisan bias than women. This is inconsistent with findings that women are more altruistic to co-partisans than rivals (ibid).

Partisanship and Reciprocity

While trust decisions appear highly conditioned on partisanship cues (stereotypes), reciprocity decisions are not. For each of the four games, Fig. 3 reports the average percentage of tickets returned to Player 1 out of the total number of tickets available to Player 2 (i.e., the tripled amount sent by Player 1 plus the 10 tickets given to Player 2 by the researchers). As with trust, Democrats, Republicans, Independents, and anonymous players were, on average, shown roughly the same degrees of reciprocity. The picture barely changes in partisan dyads.
Fig. 3

Average Player 2 percentage allocations in the trust game

Figure 4 displays the average percentage of tickets returned by partisanship in each game. Parametric and non-parametric tests find no significant differences in reciprocity between co- and rival partisans (Table A2). The same is true between co-partisans and anonymous players. Moreover the percentage of available tickets returned is not affected by the number of tickets received from either co- or rival partisans. Thus the information sent in the trusting behavior is not colored by the partisanship of the trustor; trustees treat this information in the same way. The number of tickets returned, however, is closely linked to the number of tickets received (see Table 3). Considered through the dual-processing framework theorized above, Player 2 uses information regarding Player 1’s trust instead of the partisanship cue to determine how much reciprocity is deserved.
Fig. 4

Average Player 2 percentage allocations in the trust game by partisanship

Table 3

Robust regression of percentage and number of tickets returned


% returned to co-partisan

% returned to rival partisan

Difference in % of tickets returned

# of tickets returned to co-partisan

# of tickets returned to rival partisan

Tickets received from co-partisan









Tickets received from rival partisan




















The number of tickets received from co/rival partisans did not affect the percentage returned or the difference in the percentage returned. However, the number of tickets received does affect the number (not percentage) returned

Standard errors are in italics. OLS gives substantively similar results, but robust regression is useful due to the presence of outliers

N = 63 in all models

** p < .05 two-tailed test, *** p < .01

Overall, partisans clearly discriminate trust along party lines. Trust bias among partisans increases with strength of partisan identity and political knowledge, consistent with social identity and cognitive heuristics expectations. Reciprocity does not hinge on partisanship. Players rely upon the cooperation-signaling information imbued in first-mover trust rather than the partisanship cue. One key implication is that rival partisans can arrive at mutually beneficial outcomes under two conditions: (1) when the trustor does not know they are rivals and (2) when the trustor extends trust despite this knowledge, thereby ignoring partisan stereotypes in social interactions. But the dual-process model would suggest this is unlikely in a one-shot game without task-relevant information on hand. By implication, although reciprocity is untainted by partisan bias, distrust of rival partisans reduces the opportunities for Pareto-optimal cooperation across party lines.


The foregoing analysis speaks to the political dimensions of broad multi-disciplinary debates about the origins and nature of cooperation. Theoretical benefits to representative democracy of widespread and strong psychological orientations to political parties must be weighed against at least two previously unknown but potentially unsettling social costs.

First, trust is crucial to social organization because it “enables individuals to take risks in dealing with others, solve collective action problems, or act in ways that seem contrary to standard definitions of self-interest” (Levi 1998, 78). Paired with reciprocity, trust generates the social capital necessary to improve social, economic, and political efficiency (Coleman 1990; Putnam 1993; Fukuyama 1995). Yet partisanship appears to produce “bonding” as opposed to “bridging” social capital (Putnam 2000). Bonding social capital strengthens norms of trust and reciprocity within relatively homogenous social networks and reinforces “out-group antagonism” (23). Bridging social capital spurs social networks that span “diverse social cleavages.” Putnam theorizes bonding networks are crucial to “getting by” whereas bridging networks are “getting ahead” (22). Though this study suggests reciprocity, and thus the potential for cooperation, does not hinge on partisanship, someone must make the first move by extending trust. Therefore strong and polarizing party allegiances, partially a byproduct of the party system and electoral institutions, threaten social cohesion and political cooperation by impeding the formation of trusting and cooperative norms.

Second, each of the democracy-bolstering concepts under study—trust, reciprocity, and partisanship—are in tension. Perhaps the greatest democratic consequence of trust breaking down along party lines is a move away from the Dahlian ideal of the “politics of robust civility” (1997, 372) and closer to the logic of, “what you gain, I lose, and what I lose, you gain” (1971, 153). According to this study, partisans, especially strong ones, are apt to obey this logic. But it is suboptimal. Trusting actions beget reciprocity—even between known partisan rivals—and partisan stereotypes are no substitutes for actual information. Once citizens systematically fail to reciprocate the trust rival partisans place in them American politics will indeed approximate a zero-sum game. Evidently this has not yet happened—even the strongest partisans on average only give about 5 (out of a possible 10) more tickets to co-partisans than rival partisans.

This study raises three sets of questions for future research. One concerns the gamut of social decisions political cues influence. Does partisanship affect other strategic social interactions and thus create an array of collective action problems? Another set of questions concerns who uses political cues and when. If politically knowledgeable people and strong partisans tend not to trust rival partisans, then should we expect electoral “winners” and “losers,” members of opposing political coalitions, or competing interest groups to behave similarly? Are similar effects present in multiparty systems? Finally, if partisanship is a fairly poor clue about trustworthiness, greater cooperation requires overcoming in-group/out-group proclivities and avoiding partisan stereotypes. While difficult to achieve in one-shot interactions, can these norms be strengthened and generalized through repeated interactions? How do the nature of party competition and party system cleavages influence the prospects of cooperation between partisan rivals?

Addressing these questions would surely enrich our understanding of how democratic politics impinge on citizens’ ability to work together for optimal political, social, and economic outcomes. For now, this study concludes that awareness of partisan differences is likely to stymie trust in the absence of other information, and knowledge of a shared partisan identity greatly boosts trust between strangers. These conclusions point to a politics of interpersonal trust and reciprocity which has thus far not received systematic treatment but has potentially far-reaching implications.


That Player 1 knows Player 2 received the same amount is crucial to head off inequality-avoiding behavior.


Individualistic preferences produce the Nash equilibrium of no trust/reciprocity. However, formal and informal institutions in nearly all societies have facilitated the evolution of pro-social preferences (such as trust, altruism, etc.) even in the absence of binding enforcement mechanisms (cf. Henrich et al. 2004; Seabright 2010).


See Morton and Williams (2008) for a thorough discussion.


Indeed, it is uncorrelated with Player 1 trust in the trust game but correlated with Player 2 reciprocity.


A pilot study suggested game order had no observable behavioral effects.


Using lotteries in place of cash may make subjects more risk-neutral (Roth & Rothblum 1982), which should reduce the effects of risk tastes on trust decisions.


The expected value/utility of each lottery ticket is $300/total number of tickets. Since 138 subjects participated, and if the final lottery ticket distribution is uniform, each player's expected utility for participating is $2.17 ($300/138). While these stakes are modest, Camerer and Hogarth (1999) conclude that stake size has no consistent behavioral effects in strategic games. While Johansson-Stenman et al. (2005) provide evidence to the contrary, most Player 1s and Player 2s still send large fractions of their endowment even when the stakes were very high. The fraction subjects in this study sent is near the median amount in the 28 anonymous trust games reviewed by Cárdenas and Carpenter (2008).


For Player 1: 26 Democrats, and 38 Republicans, and 3 Independents; for Player 2: 25 Democrats, 38 Republicans, and 4 independents.


For the purpose of matching Player 1 and Player 2 for the game with Independents, the amounts sent by those who leaned to one party or the other as Independents (weak partisans) were used. For the analyses below, they are treated as weak partisans, not independents. The results are consistent if these subjects are treated as Independents. In the debriefing message, subjects received information that they may have played against a weak partisan or Independent for that particular game (see online supplemental materials).


Our argument that discrimination in the trust game is not driven by altruistic preferences is also supported by a study we conducted in Spring 2011. As part of a larger experiment we asked subjects to play modified dictator games based on Cox’s (2004) dictator game. Unlike the version used by Fowler and Kam, there is no inequality in the endowment given to Players 1 and 2; however, any tickets given by Player 1 were tripled. We found no difference in the number of tickets sent to co- and rival partisans. While tickets were sent, indicating a possible role of altruism in the trust game, such motivations did not affect discriminatory decisions.


Tobit estimation is appropriate since 8% of the dependent variables in the sub-sample of partisans are right-censored, i.e., the subjects gave ten tickets to their co-partisan and zero tickets to the other party; they would have liked to increase the difference in tickets given but were restricted by the pool of tickets the game supplied. Such situations are ideal for tobit models (Gabarino and Slonim 2009; Cox 2004; Burns 2006; Haile et al. 2008).


Survey questions and response sets are taken verbatim from the 2008 ANES and reported in the online supplement. For partisanship strength, the out-group is weak identifiers. Political knowledge is measured on a 7-point scale created by summing correct answers (worth 1 point each) to seven political knowledge questions. This scale is empirically reliable (α = .73). It is re-scaled 0–1 to make its coefficient comparable to the strong and moderate partisan dummies. Among partisans, knowledge had a median of 5.0 and a mean of 4.7.



The authors would like to thank Fernanda Boidi, James Fowler, Cindy Kam, Peter Loewen, Brian Paciotti, Jason Reifler, Sean Richey, Elizabeth Zechmeister, and the three anonymous reviewers for their assistance and insightful comments.

Supplementary material

11109_2011_9181_MOESM1_ESM.pdf (108 kb)
Supplementary material 1 (PDF 107 kb)

Copyright information

© Springer Science+Business Media, LLC 2011

Authors and Affiliations

  1. 1.Department of Political ScienceGeorgia State UniversityAtlantaUSA
  2. 2.Department of Political ScienceUniversity of MississippiUniversityUSA

Personalised recommendations