Assessing the Effects of Partisan Bias at the Group Level of Analysis: A Hidden Profile Experiment

Abstract

Although there is some evidence in the political arena that pooling information can overcome individual biases to improve decision-making accuracy, research from the group communication and psychology arenas suggests otherwise. Specifically, research on the hidden profile, a group-level decision-making problem, suggests that groups are decidedly biased when making decisions. This laboratory experiment tested whether or not partisan biases manifest at the group level of analysis. In the main, it was found that groups composed of either all Republican or all Democratic group members were likely to make a decision that was consonant with their party’s political ideology, which ultimately impacted hidden profile solution rates (i.e., decision accuracy). Moreover, supplemental analyses suggest that Republican and Democratic groups reached their biased decisions through different means. A discussion is provided in which the implications of these results are considered.

Keywords

motivated reasoning partisan bias decision-making hidden profile groups

Despite concerns about biased reasoning (Kunda, 1990; Taber & Lodge, 2006), there is some evidence in the political arena that pooling information appears to overcome individual biases to improve decision-making. For example, the forecasting successes of political markets, in which markets pool private information to provide accurate election forecasts, suggest that, in the aggregate, people’s judgments are accurate (Berg, Forsythe, Nelson, & Rietz, 2008; Wolfers & Zitzewitz, 2004). Similarly, classic work in public opinion suggests that in spite of the idiosyncrasies of individual political attitudes (Converse, 1964), in the aggregate, public opinion is stable and responds to real world events in a rational manner (Page & Shapiro, 1992).

Ultimately the ability to discuss and share information in groups could be important in reducing the impact of individual biases in political decision-making. Deliberation forums, for instance, in which people discuss controversial political issues in a structured setting, have been shown to influence people’s attitudes (Fishkin & Luskin, 2005). Moreover, allowing people to consider multiple dimensions of controversial issues rather than considering only one salient dimension has been shown to benefit decision-making quality (Druckman & Nelson, 2003).

Nevertheless, literature on the hidden profile, a group-level decision-making paradigm (Stasser & Titus, 1985), suggests that, under some general conditions, groups fail to overcome individual biases. As a decision-making problem, the hidden profile task typically unfolds in two phases. First, groups are required to assess the favorability of multiple hypothetical decision alternatives (Wittenbaum, Hollingshead, & Botero, 2004). Second, members are asked to convene and choose the group’s preferred alternative (e.g., best job candidate, Cruz, Boster, & Rodriguez, 1997). Notably, some alternatives are assigned more positive attributes (optimal alternatives) when compared with others (suboptimal alternatives), which implies that, by weight of the evidence, some alternatives are better options than others (Wittenbaum, 2010).

Presumably, when all information about each of the alternatives is available and pooled, choosing the optimal alternative is a simple task. In designating candidate attributes as either shared (known by all members) or unshared (known by a single member), however, the difficulty of solving the hidden profile task increases substantially. Specifically, despite the existence of an optimal solution, information about each of the alternatives is largely unshared and dispersed throughout the group (Stasser & Titus, 2003; Wittenbaum, 2010; Wittenbaum et al., 2004). Hence, solving the hidden profile and thus making an optimal decision require that members exchange unshared information and engage in the “systematic and balanced exploration of relevant issues” (Stasser & Titus, 1985, p. 1467).

Notably, Stasser, Taylor, and Hanna (1989) found that group decision-making accuracy was impaired because group members failed to share unique information with one another. Reviews by Stasser and Titus (2003) and Wittenbaum et al. (2004), in addition to meta-analyses by Lu, Yuan, and Mcleod (2012) and Reimer, Reimer, & Czienskowski (2010), reinforce this conclusion. Despite being presented with a relatively straightforward task, in which members need merely to share private information, many tests of the hidden profile show that groups fail to select the optimal alternative.

Researchers have attempted to uncover the conditions that affect the probability of finding the optimal solution in the hidden profile. Emich (2012), for instance, found that highly task interdependent groups were more likely to uncover the hidden profile and make better decisions when compared with groups low in task interdependence. Lam and Schaubroeck (2000) found that online tools meant to structure decisions and facilitate group communication increased the odds of solving the hidden profile when compared with face-to-face groups. Moreover, past reviews have discussed numerous variables that affect hidden profile solution rates (see Stasser & Titus, 2003; Wittenbaum et al., 2004), and recent meta-analyses by Lu et al. (2012) and Reimer et al. (2010) have also shown that numerous variables moderate a group’s ability to solve the hidden profile (e.g., group size negatively impacts hidden profile solution rates).

One process that has received less attention is the biased interpretation of information resulting from prior beliefs (Greitemeyer & Schulz-Hardt, 2003; Kunda, 1990; Wittenbaum et al., 2004). In their seminal study on attitude polarization, Lord, Ross, and Lepper (1979) found that information deemed inconsistent with one’s prior beliefs was scrutinized heavily and generally discounted as being low in believability. Conversely, preference-consistent information was scrutinized less and seen as more convincing.

Kunda (1990) postulates that such findings are a function of a directional goal bias, such that the interpretation and evaluation of information is affected by the biased processing of information due to one’s prior belief system. If the information being presented is inconsistent with one’s prior beliefs, one is more likely to scrutinize the message, discount it, and counterargue more vigilantly. Conversely, in the event that the presented information is preference consistent, the information is scrutinized less rigorously and seen as more convincing.

Myriad empirical studies have since buttressed the initial findings of Lord et al. (1979) and theoretical musings of Kunda (1990). Ditto and Lopez (1992), for instance, found that preference-inconsistent information was met with greater skepticism. Similarly, Edwards and Smith (1996) found that preference-inconsistent arguments were rated as weaker than preference-consistent arguments, scrutinized for longer periods of time, and counterargued to a greater extent.

Given this general thesis and pattern of results, its application to the hidden profile paradigm is both relevant and necessary in light of the nature of the task (e.g., Scholten, van Knippenberg, Nijstad, & De Dreu, 2007). Because the information distributed in candidate profiles can be judged as either preference consistent or preference inconsistent, findings from the motivated skepticism literature become pertinent to any argument designed to explicate the mechanism(s) by which group decision-making is affected.

One potential bias is the well-documented partisan bias in political decision-making (e.g., Bartels, 2002; Campbell, 1980; Nyhan & Reifler, 2010). The process of motivated skepticism has been cited extensively as the explanatory mechanism responsible for the effects of partisan bias on information processing (e.g., Taber & Lodge, 2006). In a study assessing the effects of motivated reasoning on political decision-making, Redlawsk (2002) found that participants processed incongruent information about a desired candidate for longer amounts of time and put greater effort into searching for information favorable to their desired candidates. Taber and Lodge (2006) similarly found that participants were more likely to list denigrating thoughts about preference-inconsistent information on issues they favored, and that they were more likely to “read an argument of a sympathetic source than expose themselves to an opposing viewpoint” (p. 764). Similar asymmetries in the interpretation of information have also been found when examining the effects of political ideology. For instance, Gaines, Kuklinski, Quirk, Peyton, and Verkuilen (2007) found that different interpretations of the same objective facts were a function of political ideology. Specifically, although both liberals and conservatives agreed upon the tally of deaths in Iraq, they interpreted the magnitude of the statistic differently (liberals perceived the death toll as larger than conservatives did). For additional review of similar results, see Leeper and Slothuss (2014; see also Erisen, Redlawsk, & Erisen, 2018; Kuru, Pasek, & Traugott, 2017; Suhay & Erisen, 2018).

Because the hidden profile paradigm generally stipulates that positive information about the suboptimal alternative is largely shared, whereas positive information about the hidden optimal alternative is largely unshared (cf. Greitemeyer & Schulz-Hardt, 2003), the political partisan bias holds implications for hidden profile solution rates and overall decision-making quality in political groups. Specifically, this dynamic implies that under conditions in which a Conservative (Liberal) candidate is portrayed as the suboptimal alternative (i.e., but is in fact the hidden optimal alternative), Republican (Democratic) group members will view the profile as preference-inconsistent information and discount or misconstrue its valence (i.e., discount shared information, but favor the positive unshared information). Conversely, when a Conservative (Liberal) candidate is portrayed as the optimal alternative (i.e., but is in fact the suboptimal alternative), Republican (Democratic) group members will view the profile as preference consistent and, thus, accept the information and its valence (i.e., discount unshared information, but favor the positive shared information). Because focusing on unshared information is critical to solving the hidden profile problem, decision-making groups should thus be more likely to work at the problem and get the correct answer when the hidden optimal alternative is the alternative that matches the group’s party ideology (see Table 1). Of note, this logic presumes that political affiliation is associated with political ideology. This assumption is warranted, however, as past research has shown that those that identify as Republican (Democratic) generally adhere to Conservative (Liberal) ideologies (e.g., Baldassarri & Gelman, 2008; Levendusky, 2009).

Table 1.

Predicted Pattern of Results.

	Republican group	Democratic group
Conservative Candidate Correct	Most accurate	Least accurate
Liberal Candidate Correct	Least accurate	Most accurate

Note. No differences were expected between Republican and Democratic groups in the full information conditions.

Alternatively, suppose that each group member is exposed to all pertinent information about each alternative, what is termed commonly the full information condition. Experiments find consistently that groups are much more likely to make the correct decision when provided full information (Lu et al., 2012). This finding is expected to generalize to the political arena and attenuate the effects of partisan bias. As Kunda (1990) argues and Ditto and Lopez (1992) demonstrate, although cognitive biases affect cognitive processing, it is unlikely that these biases provide people with the license to adopt desired conclusions when faced with strong contradictory evidence.

Hence, consider a hidden profile experiment in which politically partisan groups are formed and given the task of deciding which of two candidates to endorse for a political office. In addition, for some of these groups the better of the candidates holds a political ideology that is consonant with group members’ political party affiliation, whereas for others the profile of the better candidate has a different political ideology that is incongruent with the group’s political affiliation. Under these conditions, it is expected that groups composed of Republican members (Democrat members) will be more successful at working at the problem and uncovering the hidden profile when the Conservative (Liberal) candidate is the optimal alternative but is portrayed initially as the suboptimal alternative (for reasons explicated previously). Moreover, as implied in the aforementioned argument these effects are expected to be mediated by expressions of biased processing that arise before or during discussion. Common mediators analyzed in this arena typically include the extent to which groups focus on unshared information during discussion, as well as the extent to which members hold heterogeneous candidate preferences before entering discussion (i.e., prediscussion dissent; see Lu et al., 2012; Schulz-Hardt, Brodbeck, Mojzisch, Kerschreiter, & Frey, 2006). That is to say, within the context of this experiment, motivated reasoning in groups was expected to produce disagreement regarding the optimal solution and also facilitate the discussion of unshared information between group members. Moreover, although decision accuracy was expected to be a function of both prediscussion dissent and the sharing of unique information (cf. Lu et al., 2012; Schulz-Hardt et al., 2006), whether Republican or Democratic groups arrived at their group-level decisions through different means was entertained. That is to say, although no differences were expected, the extent to which Republican and Democratic groups differed in regard to their discussion dynamics was investigated.

With respect to the full information condition, no differences in solution rates were expected. Indeed, regardless of group and candidate affiliation, it was expected that groups in the full information condition would outperform those in the hidden profile condition (i.e., a replication of the classic hidden profile finding; Lu et al., 2012).

In sum, this research investigated the extent to which different information environments impacted decision accuracy in homogeneous partisan groups. In addition, the specific mechanisms responsible for eventual group-level decisions were also explored. Ultimately, such an investigation was expected to inform the recent and growing literature that documents the effects of partisan discussion and group structure on attitude strength and preference formation (see Druckman, Levendusky, & McLain, 2018; Klar, 2014; Klar & Shmargad, 2017; Levendusky, Druckman, & McLain, 2016; Robison, Leeper, & Druckman, 2018).

Method

Participants

This study was conducted at a large Midwestern university, and a convenience sample of undergraduate students from a participant pool volunteered in exchange for course credit. Based on the results of an online political party screener item, students were scheduled to come into the laboratory. In total, 30 three-person groups participated in the experiment (n = 90), with 15 groups composed of members identifying as Democratic or Democratic-leaning (10 hidden profile groups, five full information groups) and 15 groups composed of members identifying as Republican or Republican-leaning (10 hidden profile groups, five full information groups).

Design

This experiment employed a 2 (Republican vs. Democratic group) × 2 (hidden profile vs. full information) × 2 (Liberal as optimal solution vs. Conservative as optimal solution) mixed design. The political affiliation and profile inductions were independent-groups factors; the optimal solution factor was a repeated measure.

The profile narratives used in both hidden profile and full information conditions contained two alternative profiles (i.e., Candidates A and B for an electable Federal political office). Each of the hypothetical candidates had 14 pieces of information describing them, some positively and some negatively valenced (see Table 2).

Table 2.

Distribution of Information.

Group member	Valence	Candidate A (Conservative/Liberal)		Candidate B (Conservative/Liberal)
Member 1		Shared	Unshared	Shared	Unshared
	Positive	0	3	5	0
	Negative	5	0	0	3
Member 2		Shared	Unshared	Shared	Unshared
	Positive	0	3	5	0
	Negative	5	0	0	3
Member 3		Shared	Unshared	Shared	Unshared
	Positive	0	3	5	0
	Negative	5	0	0	3
Full Group	Positive	9		5
Full Group	Negative	5		9

Note. At the group level, the correct answer is always Candidate A. As described in the Method section, the political ideology of Candidate A was counterbalanced across group discussions.

In the full information conditions, participants received all information about each candidate. In the hidden profile condition, each participant received eight information items, three of which were unshared and five of which were shared. The information distribution was such that, if all information was pooled, one candidate would emerge as the optimal candidate (i.e., nine positive attributes and five negative attributes) when compared with the other candidate (i.e., five positive attributes and nine negative attributes). Profiles were constructed to bias each group member toward the suboptimal candidate. Two pilot studies were implemented to produce these profiles.

Pilot Studies

In producing both candidate profiles, an initial pilot study (n = 127) evaluated the favorability of 25 pairs of characteristics associated with politicians or political behavior. Each pair was composed of a positive characteristic and its polar opposite (e.g., effective vs. ineffective leadership), meaning participants assessed a total of 50 different characteristics. One-sample t tests against the midpoint of the scale (3) produced a set of characteristics that were perceived as being either significantly positive or significantly negative (p < .05). In all, 14 positive attributes and 14 negative attributes were extrapolated from the list of 50 to create the experimental materials. For example, a negative piece of information was, “while working in these positions Candidate A has been criticized for engaging in unethical behavior at times,” whereas a positive piece of information was, “in past positions Candidate A has spearheaded initiatives that have demonstrated effective managerial skills.”

A second pilot study (n = 131) assessed whether or not the hidden profile and full information materials were biasing subjects in the intended direction. In the main, subjects generally picked Candidate B (suboptimal alternative) when assessing hidden profile materials, and, conversely, generally picked Candidate A (optimal solution) when examining full information profiles, χ²(1, N = 90) = 12.44, p < .001, r = .37. Moreover, this pattern of results emerged in the data regardless of whether Candidates A or B were labeled as either Conservative or Liberal (for both hidden profile and full information profiles). Thus, the creation of both candidate profiles was deemed a success. For verbatim candidate profiles, contact the first author.

Experimental Procedure

When participants arrived at the laboratory, they were greeted by the experimenter who verified that each participant was the correct group member by confirming participant name, sex, and party identification. Of note, the confirmation of party identification, which occurred individually, was expected to prime political party affiliation (cf. Forster & Liberman, 2007). Thus, unless group members disclosed their political affiliations, group members were initially unaware of each other’s political affiliations.

Upon the arrival of all three group members, participants were told that they would be participating in a group experiment. Specifically, participants were told that they would be asked to evaluate the favorability of two candidate profiles, and also that their group would need to decide on which of the two candidates was deemed preferable. Next, materials were distributed, including candidate profiles and prediscussion preference sheets. Of note, the instructions listed in the candidate profiles indicated that some of the candidate information may or may not have been the same. When members had examined the profiles and indicated their prediscussion preference, materials were collected and group discussion commenced.

This procedure was repeated with two different profile sets. Specifically, although groups were assigned to either hidden profile or full information conditions, the within subjects factor (i.e., Liberal as optimal solution vs. Conservative as optimal solution) dictated that each group engaged in two separate discussions. Thus, groups in the hidden profile condition engaged in two hidden profile tasks, whereas groups in the full information condition engaged in two full information tasks. The main difference between both discussions lay in which political candidate was the correct choice. Hence, in one set of hidden profiles the Conservative candidate was the optimal alternative, whereas in a second set of hidden profiles the Liberal candidate was the optimal alternative. Of note, although the same characteristics were used to construct two different sets of profiles, when applicable, the unique information was distributed to make it seem as though members were receiving information about two different political candidates. In addition, the order in which profiles were received was counterbalanced across both hidden profile and full information conditions to control for order effects.

Instrumentation

The focal dependent variable was if the group selected the optimal or the suboptimal alternative (decision accuracy). In addition, several features of the discussion were coded to assess the extent to which discussion content mediated the effect of the experimental inductions on decision accuracy.

Coding

Both discussions were video-recorded so that the impact of mentioning unshared information could be investigated. Two coders, blind to condition and to hypotheses, unitized the discussion data and then coded the content of each unit. To accomplish this task, coders were provided a list of all 28 pieces of information and asked to itemize them in sequential order as they arose during discussion.

Both coders were in general agreement regarding the existence versus absence of an informational unit κ = .75. Disagreements regarding unitization were resolved by the coders via discussion before establishing the final set of units, and the first author acted as a tiebreaker if the disagreement between the coders was unresolvable.

Likewise, coders were in general agreement regarding the specific content of the informational units, κ = .95. Similar to the unitization of the discussion data, disagreements regarding unit content were resolved via discussion before establishing the final set of utterances, which became the measure employed in subsequent analyses.

Unshared information and dissent

The two mediators of interest included the extent to which group members discussed unshared information during discussion, as well as whether group member preferences were homo- or heterogeneous before entering discussion. The extent to which group members discussed unshared information was calculated by dividing the amount of unshared information into total information discussed. Thus, if groups uttered three pieces of information and they were all unshared, the conversation between members was focused entirely on the discussion of unshared information. This variable was calculated both with and without repetitions; informational units were coded as repetitions only if a different piece of information was mentioned in between the unit in question (e.g., Schulz-Hardt et al., 2006). A decision was made to keep repetitions in the analysis because model fit remained unaffected by their inclusion or exclusion.

Prediscussion dissent was defined dichotomously, based on whether group members’ prediscussion preferences were homogeneous (coded as 0) or heterogeneous (i.e., one or more members had different preferences, coded as 1; Lu et al., 2012). Results are described below.

Results

The results of the experiment are presented in Tables 3 and 4. From Table 3, two important patterns of results may be observed. First, a cursory inspection of percentages in each of the conditions shows that the probability of a group selecting the optimal candidate was much higher when given full information (.85) than when administered a hidden profile (.08). Second, group composition and the ideology of the optimal alternative appeared to combine nonadditively to affect responses. Specifically, groups composed of Republicans were more likely to select the best alternative when the candidate was Conservative (.47) than when the candidate was a Liberal (.20). Conversely, groups composed of Democrats were more likely to select the best alternative when the candidate was Liberal (.40) than when the candidate was a Conservative (.27; see Figure 1).

Table 3.

Probability of Choosing the Optimal Candidate by Condition.

	Republican group				Democratic group
	Hidden profile		Full information		Hidden profile		Full information
	Liberal best	Conservative best	Liberal best	Conservative best	Liberal best	Conservative best	Liberal best	Conservative best
Correct	0%	20%	60%	100%	10%	0%	100%	80%
Incorrect	100%	80%	40%	0%	90%	100%	0%	20%

Table 4.

Probit Regression Analysis Results.

	Robust coefficients	SE	Z	p	95% CI
Congruent	4.96	0.40	12.36	< .001	[4.17, 5.74]
Full information	6.12	0.33	19.86	< .001	[5.87, 7.16]
Affiliation	−0.02	0.53	−0.05	.96	[–1.07, 1.02]
Constant	−5.98	0.60	−10.01	< .001	[–7.15, –4.81]

Note. CI = confidence interval.

Figure 1.

Decision-making accuracy percentages.

In providing a more rigorous assessment of these data, a probit regression model was estimated so that the effects of each variable on decision accuracy could be explored (Long, 1997). Specifically, indicators were created for information condition (full information condition = 1, otherwise = 0), and group party identification (Democratic identification =1, Republican group identification = 0). In addition, an indicator variable was created for whether the group’s party affiliation was congruent with the ideology of the optimal alternative, as doing such simplified the analysis considerably (i.e., congruent vs. not congruent). Hence, accuracy of the group judgment was regressed on the indicators for congruency, information condition, and group party identification. Moreover, each outcome was treated as a separate case and standard errors were estimated clustering on groups.

The results of the probit regression analysis confirmed our initial inspection (n = 60; pseudo R² = .62; p < .001). Specifically, the coefficient estimated for the congruency variable was 4.96 (SE = .40; p < .001), thus indicating that groups with party affiliations congruent with the ideology of the optimal alternative were much more likely to select the optimal alternative than those that were not congruent. Moreover, groups assigned to the full information condition were much more likely to arrive at the correct judgment than those who were not (B = 6.62; SE = .33; p < .001), thus replicating the classic hidden profile effect (Lu et al., 2012; Reimer et al., 2010).

Examining the possibility that features of the discussion mediated the impact of the affiliation induction on decision accuracy in the hidden profile condition produced the model presented in Figure 2. In using path analysis to test for mediation (see Hunter, 1985; Hunter & Gerbing, 1982; Hunter & Hamilton, 1987), these data indicated that both the percentage of unshared information discussed and prediscussion dissent mediated this effect, χ²(2) = .13, p = .94, root-mean-square error = .09. Specifically, the model indicates that groups composed of Democrats focused less on unshared information than did Republican groups, but that they were more likely to enter the discussion with dissenting opinions than the Republican groups. Moreover, and quite notably, decision accuracy was higher when group members focused on unshared information during discussion and when there was more prediscussion dissent. Thus, although both Democratic and Republican groups engaged in biased processing when making decisions, the means by which this occurred differed by group party affiliation.

Figure 2.

Causal model, χ²(2, N = 16) = .13, p = .94, root-mean-square error = .09.

Discussion

The results are consistent with expectations. First, the well-established hidden profile effect was found. Groups in the hidden profile condition were much less likely to select the optimal alternative than groups in the full information condition. Second, there was evidence of biased processing at the group level of analysis. Groups composed of Republican members were more likely to select the Conservative candidate than the Liberal candidate, and groups composed of Democratic members were more likely to select the Liberal candidate than the Conservative candidate. But, interestingly this pattern was characteristic not only of those in the hidden profile condition, but also of those in the full information condition (see Table 3). In 10 discussions, Republican groups did not find the hidden profile in which the Liberal candidate was optimal, but found it twice when the Conservative candidate was optimal. Conversely, Democratic groups found the hidden profile in which the Liberal candidate was best only once in 10 discussions, but never found it when the Conservative was the optimal candidate. In addition, in the full information condition Republican groups selected the optimal Liberal candidate only 3 times in five discussions, but always selected the optimal Conservative candidate; Democratic groups selected the optimal Conservative candidate 4 times in five discussions, but always selected the optimal Liberal candidate. Thus, the same pattern holds for both hidden profile and full information discussions, although groups generally were more likely to select the optimal alternative in the full information condition.

These results have implications for the motivated reasoning literature, the political communication literature, and the group decision-making literature. The data provide yet another examination of the motivated reasoning effect, add to what is known about partisan bias, and extend the hidden profile literature by demonstrating how prior beliefs can bias discussions. Perhaps the more important implication, however, is integration of these three scholarly venues that may be considered largely independent of one another. This experiment demonstrates that the manner in which people reason may be germane to understand the ways in which partisans and nonpartisans make political decisions, and it is germane to the manner in which groups, in general, make decisions. Partisan bias provides an especially challenging problem to understand for those who examine motivated reasoning, and it provides an equal challenge for solving group decision-making tasks such as the hidden profile.

The hidden profile problem also challenges motivated reasoning scholars to study larger social units and also provides insights to political scholars seeking to understand decisions made by homogeneous partisan groups; specifically, the current research moves beyond research on motivated reasoning at the individual level to show that political attitudes can also influence group decision-making. Because many political judgments involve group decision-making, including town hall meetings to make local policy decisions (Mansbridge, 1983), face to face deliberation on policy issues (Gastil, 2008), and presidential caucuses (Redlawsk, Tolbert, & Donovan, 2010), among others, social scientists would benefit from expanding the application of the influence of motivated reasoning to explore group behavior in decision-making contexts.

The finding that groups composed of Democrats discussed less unshared information than Republican groups is perhaps explained by different personality characteristics of conservatives and liberals. People higher in conscientiousness tend to identify as conservative (Gerber, Huber, Doherty, Dowling, & Ha, 2010), which may explain why Republican groups took a different approach to the group decision-making process, including following rules, prioritizing tasks, and engaging in other goal-directed behavior associated with the personality trait of conscientiousness. Moreover, there is some evidence that extraversion is associated with conservatism (Gerber et al., 2010), which may explain why Republican groups were more likely to discuss than Democratic groups. Another possibility is that homogeneous groups of Republicans may, in many ways, be more homogeneous than groups of Democratic identifiers. As Grossmann and Hopkins (2016) argue, the Republican Party is committed to the ideological goal of limited government, while the Democratic Party consists of a coalition of single-interest groups (such as specific minority groups). This would mean that there could be greater perceived similarity among homogeneous groups of Republican partisans than among groups of Democratic identifiers, potentially contributing to greater discussion among these groups. However, we note that the difference in group discussion styles was unexpected and that these explanations are ad hoc and necessarily tentative.

There are a number of limitations of this experiment that provide opportunities for future research. One important limitation is the student sample employed. Although this limitation may be minor in some cases, it might be more important in this research arena because student involvement in politics is lower than other age groups (Blais, 2000; Verba, Sclozman, & Brady, 1995; however, see Druckman & Kam, 2011). To provide a richer test of the substantive reasoning leading to the hypotheses, it is important for future research to address this question. Doing so adequately would require studying older participants whose political preference is strong and has been consistently for some time.

Second, there are a number of limitations revolving around various dimensions of size. For example, although it is a common feature of group dynamics experiments, sample size was modest. The ability to conduct larger scale research efforts in future study would be rewarded in that sampling error would be reduced and parameter estimates would become more precise. In addition, groups were composed of only three members. Because groups of people discussing politics often exceed this figure, and because group dynamics change in important ways with increasing group size (e.g., Lu et al., 2012), it would be informative to attempt to replicate this result with groups composed of more members. Finally, only two candidates were considered. Although in the United States only two candidates are usually credible, minor parties’ candidates having a low probability of being selected, the same cannot be said of other countries. Moreover, even in the United States, there may be cases in which there are many more than two viable candidates running for office during primary elections. More generally, group decisions other than political ones involve frequently choosing among three or more options.

Third, this experiment is limited to discussions among partisans (i.e., homogeneous partisan groups). Put differently, there were no mixed groups composed of Republicans and Democrats or Republicans, Democrats, and Independents examined in this study. Recent research on this topic suggests that groups composed of members with heterogeneous party affiliations are decidedly less biased and more critical of the issues being discussed (e.g., Druckman et al., 2018; Klar, 2014; Levendusky et al., 2016). Although these dynamics are expected to attenuate the discussion bias and thus increase decision accuracy, their impact on other group dynamics remains unclear. Specifically, it is unclear whether heterogeneous partisan groups would be able to engage in such dissent without spurring other, more detrimental types of conflict (e.g., relationship conflict; cf. de Wit, Greer, & Jehn, 2012; Manata, 2016; see also Mannix & Neale, 2005). Additional research is required to inspect this possibility, as well as the extent to which heterogeneous partisan group composition impacts decision accuracy. Other related matters that may be of interest include predictions about time taken to make a decision, as well as whether or not a consensus can be reached.

Finally, although the model presented in Figure 1 fit the data well, it is exploratory. Given the present concern about replication in social science (see Pashler & Wagenmakes, 2012), it has become clear that results such as this one need to be replicated before they can be considered highly credible.

Nevertheless, the preceding model and results provide a foundation upon which to further study the effects of motivated reasoning and political bias at the group level. The results of this experiment indicate that group members’ political affiliations can impact the decision-making process as well as the means by which these groups make their eventual judgments. Future scholars are encouraged to continue exploring the additional factors that might further inform these results, as doing so would illume the processes by which political biases manifest and operate at the group level of analysis.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Author Biographies

Brian Manata (PhD, Michigan State University, 2015) is an assistant professor in the Department of Communication at Portland State University. His research focuses on organizational behavior and group and team dynamics. His work has appeared in Communication Yearbook, Group Dynamics, and the Journal of Management in Engineering.

Franklin J. Boster (PhD, Michigan State University, 1978) is a professor in the Department of Communication at Michigan State University. His research program focuses on the study of social influence and group dynamics and their intersection.

Gwen M. Wittenbaum (PhD, 1996) received her doctoral training in social psychology at Miami University and is an associate professor of Communication at Michigan State University. Her research interests span social influence and communication in groups and organizations. She serves on editorial boards for Group Dynamics and Group Processes and Intergroup Relations and was a founding board member of the Interdisciplinary Network for Group Research.

Daniel E. Bergan is an associate professor in the Department of Communication and James Madison College at Michigan State University. He studies public opinion, advocacy campaigns, and voter mobilization.

References

Baldassarri

Gelman

(2008). Partisans without constraint: Political polarization and trends in American public opinion. American Journal of Sociology, 114, 408-446. doi:10.2139/ssrn.1010098

Bartels

L. M.

(2002). Beyond the running tally: Partisan bias in political perceptions. Political Behavior, 24, 117-150. doi:10.1023/A:1021226224601

Berg

Forsythe

Nelson

Rietz

(2008). Results from a dozen years of election futures markets research. Handbook of Experimental Economics Results, 1, 742-751. doi:10.1016/s1574-0722(07)00080-7

Blais

(2000). To vote or not to vote: The merits and limits of rational choice theory. Pittsburgh, PA: University of Pittsburgh Press.

Campbell

(1980). The American voter. Chicago, IL: The University of Chicago Press.

Converse

P. E.

(1964). The nature of belief systems in mass publics. In Apter

(Ed.), Ideology and discontent (pp. 206-261). New York, NY: Free Press.

Cruz

M. G.

Boster

F. J.

Rodriguez

J. I.

(1997). The impact of group size and proportion of shared information on the exchange and integration of information in groups. Communication Research, 24, 291-313. doi:10.1177/009365097024003004

de Wit

F. R. C.

Greer

L. L.

Jehn

K. A

. (2012). The paradox of intragroup conflict: A meta-analysis. Journal of Applied Psychology, 97, 360-390. doi:10.1037/a0024844

Ditto

P. H.

Lopez

D. F.

(1992). Motivated skepticism: Use of differential decision criteria for preferred and nonpreferred conclusions. Journal of Personality and Social Psychology, 63, 568-584. doi:10.1037/0022-3514.63.4.568

10.

Druckman

J. N.

Kam

C. D.

(2011). Students as experimental participants: A defense of the “narrow data base.” In Druckman

J. N.

Green

D. P.

Kuklinski

J. H.

Lupia

(Eds.), Cambridge Handbook of Experimental Political Science (pp. 41-57). New York, NY: Cambridge University Press.

11.

Druckman

J. N.

Levendusky

M. S.

McLain

(2018). No need to watch: How the effects of partisan media can spread via interpersonal discussions. American Journal of Political Science, 62, 99-112. doi:10.1111/ajps.12325

12.

Druckman

J. N.

Nelson

K. R.

(2003). Framing and deliberation: How citizens’ conversations limit elite influence. American Journal of Political Science, 47, 729-745. doi:10.2307/3186130

13.

Edwards

Smith

E. E.

(1996). A Disconfirmation bias in the evaluation of arguments. Journal of Personality and Social Psychology, 71, 5-24. doi:10.1037/0022-3514.71.1.5

14.

Emich

K. J.

(2012). How expectancy motivation influences information exchange in small groups. Small Group Research, 43, 275-294. doi:10.1177/1046496412438269

15.

Erisen

Redlawsk

D. P.

Erisen

(2018). Complex thinking as a result of incongruent information exposure. American Politics Research, 46, 217-245. doi:10.1177/1532673x17725864

16.

Fishkin

J. S.

Luskin

R. C.

(2005). Experimenting with a democratic ideal: Deliberative polling and public opinion. Acta Politica, 40, 284-298. doi:10.1057/palgrave.ap.5500121

17.

Forster

Liberman

(2007). Knowledge activation. In Kruglanski

A. W.

Higgins

E. T.

(Eds.), Social psychology: Handbook of basic principles (2nd ed., pp. 201-231). New York, NY: Guilford Press.

18.

Gaines

B. J.

Kuklinski

J. H.

Quirk

P. J.

Peyton

Verkuilen

(2007). Same facts, different interpretations: Partisan motivation and opinion on Iraq. The Journal of Politics, 69, 957-974. doi:10.1111/j.1468-2508.2007.00601.x

19.

Gastil

(2008). Political communication and deliberation. Thousand Oaks, CA: SAGE.

20.

Gerber

A. S.

Huber

G. A.

Doherty

Dowling

C. M.

S. E.

(2010). Personality and political attitudes: Relationships across issue domains and political contexts. American Political Science Review, 104, 111-133. doi:10.1017/s0003055410000031

21.

Greitemeyer

Schulz-Hardt

(2003). Preference-consistent evaluation of information in the hidden profile paradigm: Beyond group-level explanations for the dominance of shared information in group decisions. Journal of Personality and Social Psychology, 84, 322-339. doi:10.1037/0022-3514.84.2.322

22.

Grossmann

Hopkins

D. A.

(2016). Asymmetric politics: Ideological Republicans and group interest Democrats. New York, NY: Oxford University Press.

23.

Hunter

J. E.

(1985). PATH: A least squares static path analysis program [Computer software]. East Lansing: Michigan State University.

24.

Hunter

J. E.

Gerbing

D. W.

(1982). Unidimensional measurement, second order factor analysis, and causal models. Research in Organizational Behavior, 4, 267-320.

25.

Hunter

J. E.

Hamilton

M. A.

(1987). PATH: A least squares static path analysis program (Rev. version). East Lansing: Michigan State University.

26.

Klar

(2014). Partisanship in a social setting. American Journal of Political Science, 58, 687-704. doi:10.1111/ajps.12087

27.

Klar

Shmargad

(2017). The effect of network structure on preference formation. Journal of Politics, 79, 717-721. doi:10.1086/689972

28.

Kunda

(1990). The case for motivated reasoning. Psychological Bulletin, 108, 480-498. doi:10.1037/0033-2909.108.3.480

29.

Kuru

Pasek

Traugott

M. W.

(2017). Motivated reasoning in the perceived credibility of public opinion polls. Public Opinion Quarterly, 82, 422-446. doi:10.1093/poq/nfx018

30.

Lam

S. S. K.

Schaubroeck

(2000). Improving group decisions by better pooling information: A comparative advantage of group decision support systems. Journal of Applied Psychology, 85, 565-573. doi:10.1037/0021-9010.85.4.565

31.

Leeper

T. J.

Slothuss

(2014). Political parties, motivated reasoning, and public opinion formation. Political Psychology, 35, 129-156. doi:10.1111/pops.12164

32.

Levendusky

M. S.

(2009). The partisan sort: How liberals became Democrats and conservatives became Republicans. Chicago, IL: The University of Chicago Press.

33.

Levendusky

M. S.

Druckman

J. N.

McLain

(2016). How group discussions create strong attitudes and strong partisans. Research & Politics, 3, 1-6. doi:10.1177/2053168016645137

34.

Long

S. J.

(1997). Advanced Quantitative Techniques in the Social Sciences: Vol. 7. Regression models for categorical and limited dependent variables. Thousand Oaks, CA: Sage.

35.

Lord

C. G.

Ross

Lepper

M. R.

(1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37, 2098-2109. doi:10.1037/0022-3514.37.11.2098

36.

Yuan

C. Y.

McLeod

P. L.

(2012). Twenty-five years of hidden profiles in group decision making: A meta-analysis. Personality and Social Psychology Review, 16, 54-75. doi:10.1177/1088868311417243

37.

Manata

(2016). Exploring the association between relationship conflict and group performance. Group Dynamics: Theory, Research, and Practice, 20, 93-104. doi:10.1037/gdn0000047

38.

Mannix

Neale

M. A.

(2005). What differences make a difference? The promise and reality of diverse teams in organizations. Psychological Science in the Public Interest, 6, 31-55. doi:10.1111/j.1529-1006.2005.00022.x

39.

Mansbridge

J. J.

(1983). Beyond adversary democracy. Chicago, IL: The University of Chicago Press.

40.

Nyhan

Reifler

(2010). When corrections fail: The persistence of political misperceptions. Political Behavior, 32, 303-330. doi:10.1007/s11109-010-9112-2

41.

Page

B. I.

Shapiro

R. Y.

(1992). The rational public: Fifty years of trends in Americans’ policy preferences. Chicago, IL: The University of Chicago Press.

42.

Pashler

Wagenmakes

(2012). Editors’ introduction to the special section on replicability in psychological science: A crisis in confidence. Perspectives on Psychological Science, 7, 528-530. doi:10.1177/1745691612465253

43.

Redlawsk

D. P.

(2002). Hot cognitions or cool considerations? Testing the effects of motivated reasoning on political decision making. The Journal of Politics, 64, 1021-1044. doi:10.1111/1468-2508.00161

44.

Redlawsk

D. P.

Tolbert

C. J.

Donovan

(2010). Why Iowa? How caucuses and sequential elections improve the presidential nominating process. Chicago, IL: The University of Chicago Press.

45.

Reimer

Czienskowski

(2010). Decision-making groups attenuate the discussion bias in favor of shared information: A meta-analysis. Communication Monographs, 77, 121-142. doi:10.1080/03637750903514318

46.

Robison

Leeper

T. J.

Druckman

J. N.

(2018). Do disagreeable political discussion networks undermine attitude strength? Political Psychology, 39, 479-494. doi:10.1111/pops.12374

47.

Scholten

van Knippenberg

Nijstad

B. A.

De Dreu

C. K. W.

(2007). Motivated information processing and group decision-making: Effects of process accountability and information processing and decision quality. Journal of Experimental Social Psychology, 43, 539-552. doi:10.1016/j.jesp.2006.05.010

48.

Schulz-Hardt

Brodbeck

F. C.

Mojzisch

Kerschreiter

Frey

(2006). Group decision making in hidden profile situations: Dissent as a facilitator for decision quality. Journal of Personality and Social Psychology, 91, 1080-1093. doi:10.1037/0022-3514.91.6.1080

49.

Stasser

Taylor

L. A.

Hanna

(1989). Information sampling in structured and unstructured discussions of three- and six-person groups. Journal of Personality and Social Psychology, 57, 67-78. doi:10.1037/0022-3514.57.1.67

50.

Stasser

Titus

(1985). Pooling of unshared information in group decision making: Biased information sampling during discussion. Journal of Personality and Social Psychology, 48, 1467-1478. doi:10.1037/0022-3514.48.6.1467

51.

Stasser

Titus

(2003). Hidden profiles: A brief history. Psychological Inquiry, 14, 304-313. doi:10.1080/1047840x.2003.9682897

52.

Suhay

Erisen

(2018). The role of anger in the biased assimilation of political information. Political Psychology. Advance online publication. doi:10.1111/pops.12463

53.

Taber

C. S.

Lodge

(2006). Motivated skepticism in the evaluation of political beliefs. American Journal of Political Science, 50, 755-769. doi:10.1111/j.1540-5907.2006.00214.x

54.

Verba

Sclozman

K. L.

Brady

(1995). Voice and equality: Civic voluntarism in American politics. Cambridge, MA: Harvard University Press.

55.

Wittenbaum

G. W.

(2010). Hidden profile task. In Levine

J. M.

Hogg

M. A.

(Eds.), Encyclopedia of group processes & intergroup relations (pp. 399-401). Thousand Oaks, CA: SAGE.

56.

Wittenbaum

G. W.

Hollingshead

A. B.

Botero

I. C.

(2004). From cooperative to motivated information sharing in groups: Moving beyond the hidden profile paradigm. Communication Monographs, 71, 286-310. doi:10.1080/0363452042000299894

57.

Wolfers

Zitzewitz

(2004). Prediction markets. Journal of Economic Perspectives, 18, 107-126. doi:10.3386/w10504