Productive conflict in group decision making: genuine and contrived dissent as strategies to counteract biased information seeking

https://doi.org/10.1016/S0749-5978(02)00001-8Get rights and content

Abstract

Decision-making groups in organizations are often expected to function as a “think tank” and to perform “reality testing” to detect the best alternative. A biased search for information supporting the group's favored alternative impairs a group's ability to fulfill these requirements. In a two-factorial experiment with 201 employees and managers from various economic and public organizations, genuine and contrived dissent were investigated as counterstrategies to biased information seeking. Genuine dissent was manipulated by forming three-person groups whose members either all favored the same alternative individually (homogeneous groups) or consisted of a minority and a majority faction with regard to their favored alternative (heterogeneous groups). Contrived dissent was varied by the use or nonuse of the “devil's advocacy” technique. The results demonstrate that heterogeneity was more effective in preventing a confirmatory information-seeking bias than devil's advocacy was. Confidence was identified as an important mediator. Implications for the design of interventions aimed at facilitating reality testing in group decision making are discussed.

Introduction

Important decisions with far-reaching consequences, for example, in business, politics, or public administration, are frequently made by groups (Huber, 1980; Robbins, 1992; Vroom & Jago, 1988). Because of this, a large body of research deals with the question of which factors determine the quality of group decision making and how to improve it (e.g., Ganster, Williams, & Poppler, 1991; Guzzo, 1986; Rohrbaugh, 1981; Timmermans & Vlek, 1996). As 't Hart (1998) has pointed out, answers to this question largely depend on the underlying assumptions about the role of decision-making groups in organizations: considering conflict management and social support as the main functions of the decision-making group obviously leads to different criteria for “good” group decision processes than seeing the group as a motor for organizational action.

A very popular and common model of high-quality group decision-making is the, as 't Hart (1998) calls it, “think tank” metaphor. According to this model, effective (group) decision making calls for careful definition of the decision problem, the generation and scrutinization of different solutions, the gathering and discussion of relevant information, and, based on this information, the choice of the alternative with the most positive expected consequences (e.g., Janis, 1989; Janis & Mann, 1977; Stanley, 1981). All in all, the group as a think tank does some sort of “reality testing” to find the best possible solution to a decision problem, and the fact that a group of individuals, in general, possesses more knowledge about reality and has a higher capacity to test this knowledge makes group decision making an attractive choice for organizations (e.g., Huber, 1980; Robbins, 1992).

However, several lines of research point out that groups often fail to meet these requirements. For example, groupthink theory (Janis, 1982) and research on groupthink (Esser, 1998) stress that formal and informal conformity pressures and the desire to preserve harmony within a group can override the motivation to critically appraise the relevant facts, thus (often) leading to poor decisions. Research on biased information pooling (for overviews see Stasser, 1992; Wittenbaum & Stasser, 1996) has consistently shown that groups mainly discuss and make use of information that was available to all group members before the discussion (so-called “shared information”); they thus largely fail to exchange and discuss information that was originally accessible only to one particular individual member (“unshared information”) and which is therefore new to all the others. For certain distributions of the original information among the group members (so-called “hidden profiles”), this reliance on commonly shared information is strongly associated with inaccurate group decisions.

Recently, Schulz-Hardt, Frey, Lüthgens, and Moscovici (2000) demonstrated another process diminishing a group's ability to function as a reality-testing problem-solving unit, namely biased information seeking. In their experiments, they transferred the methodology from dissonance studies on “selective exposure to information” (e.g., Frey, 1986; Jonas, Schulz-Hardt, Frey, & Thelen, 2001) to the field of group decision making: participants were confronted with a decision case from economics or politics (e.g., whether a particular company should invest in a particular developing country) and first made an individual decision for one of the two alternatives. After that, groups were formed that either consisted of members favoring the same alternative or contained a majority and a minority faction. After having discussed the case study and having reached a preliminary group decision (either by consensus or by majority vote), new pieces of information (namely articles written by experts) about the decision case could be requested to prepare the final decision. The selection of this additional information was done on the basis of short summaries of each article from which it was apparent whether the article supported or conflicted with the preliminary group decision.

In all three experiments, a clear preference for information supporting the preliminary group decision was found. This so-called “confirmation bias” (preference for information confirming one's position)1 was most pronounced in groups in which all members had favored the same alternative individually (so-called “homogeneous groups”) and still significant when the minority faction was clearly smaller than the majority faction (one person vs four persons). However, when faction sizes were more similar (three vs two or two vs one person), the preference for supporting information was no longer significant. This effect of preference heterogeneity was demonstrated for student samples and for groups of managers.

Up to this point, these effects need not necessarily have been evidence for a new group process; they could simply have reflected individual confirmation biases in a group setting. The latter has been demonstrated before, for example, in the verdict-driven jury deliberation style as shown by Hastie, Penrod, and Pennington (1983). In verdict-driven juries, the jury members enter the deliberation with specific individual verdict preferences and start with a public ballot; subsequently, members of each faction collect evidence in favor of their position. Although Hastie et al. (1983) provide no systematic data on the issue, one would expect that the larger the majority faction is, the more strongly evidence collection should be biased toward the dominant position—just as in the Schulz-Hardt et al. (2000) studies.

However, the data from the Schulz-Hardt et al. studies go beyond this. In their third and final experiment, group members indicated their individual information requests prior to group discussion. On the basis of these requests, a statisticized group information search was calculated by simply counting how many pieces of information supporting or conflicting with the (subsequent) group choice were chosen by at least one group member. This statisticized group information search reflected how the group would search for information if nothing happened at the group level and, thus, group search was simply conducted by summing up the individual information requests. Because the individual group members were biased toward information supporting their individual preferences, and because in heterogeneous groups the minority bias runs counter to the majority bias, it is not surprising that homogeneous groups had a (slightly) higher confirmation bias in this statisticized group search than heterogeneous groups. However, the results from this third experiment demonstrated that the homogeneous groups' real information search was more strongly biased toward supporting information than their statisticized search; that is, these groups had a stronger confirmation bias than one would have expected on the basis of their members' individual biases. On the contrary, the heterogeneous groups' real information search showed a smaller confirmation bias than their statisticized baseline. Thus, the differences in confirmation bias between groups with homogeneous vs heterogeneous preferences were not simply due to aggregated individual confirmation biases in a group setting, but instead due to group processes that had not been demonstrated previously: preference homogeneity exaggerated the confirmation bias at the group level, whereas preference heterogeneity debiased group information search.

The data also indicated mediating mechanisms for this effect. Initial preference heterogeneity was a necessary condition for disagreement at the time when the group started its information search. This consistent disagreement led to lower commitment to the preliminary group decision and lower confidence about having already found the best alternative (compared to homogeneous groups), and these two processes, in turn, debiased group information seeking.

Of course, biased information seeking need not necessarily be a detrimental feature in group decision making—for example, it may lead to rapid implementation of a decision or may foster success if decision outcomes are contingent on the decision-maker's conviction of doing the right thing. However, if the dominant rationale for using groups as decision-makers is to guarantee a careful and intensive reality-testing process, then biased information seeking prior to reaching a final decision violates these standards: reality testing requires a person to deal with his/her preferences in a self-critical manner, whereas biased information seeking leads to premature self-confirmation.

The question then is how to reduce biased information seeking in group decision making if a balanced, self-critical decision process is desired. As the results of the above-mentioned experiments demonstrate, conflict in the decision-making process is a critical feature with regard to this: biased information seeking was reduced or prevented when at least one dissenting opinion was present at the start of the group information search. Thus, instigating and upholding task-oriented conflict in the decision-making process can be a strategy to counteract biased information seeking.

As the above-mentioned results indicate, one way to facilitate this task-oriented conflict is to select members with heterogeneous decision preferences when forming groups. With this strategy, the means of provoking task-oriented conflict during group discussion is genuine dissent since the group members really do hold different opinions. Several findings in the group decision-making literature support the view that genuine dissent leads to a more open-minded decision-making process in groups. For example, prior studies have shown that groups whose members hold different preferences, judgments, or decisions before entering the group discussion show less overconfidence (Sniezek, 1992), are less prone to underestimating risks (Williams & Taormina, 1993), reach more accurate judgments (Sniezek & Henry, 1989), generate more emergent hypotheses during collective hypothesis testing (Crott, Giesel, & Hoffmann, 1998), exchange information more extensively (Parks & Nelson, 1999), and produce better solutions in problem solving (Wanous & Youtz, 1986) than groups with shared initial viewpoints. Recently, Brodbeck, Kerschreiter, Mojzisch, Frey, and Schulz-Hardt (2002) showed that heterogeneity of preferences facilitates the consideration of unshared information during group discussion and improves a group's ability to solve hidden profiles.

Tjosvold and colleagues have also consistently demonstrated that members of groups with conflicting individual positions are more open-minded during group discussions than members of groups with converging individual positions (e.g., Tjosvold, Johnson, & Lerner, 1981). The former actively ask for diverging arguments and integrate these arguments into their view of the decision problem. High-quality decisions were shown to result from this open-mindedness (Tjosvold, 1982; Tjosvold & Deemer, 1980). Similar processes are highlighted by the literature on minority influence: consistently and flexibly arguing minorities instigate a thorough, intensive elaboration of the problem (Moscovici, 1980) and stimulate divergent thinking (Nemeth, 1986), leading to problem solutions characterized by more creativity and higher quality (Nemeth, 1986; Nemeth & Kwan, 1987) than those of persons who were not exposed to this minority influence.

However, stimulating conflict in the decision-making process by ensuring that heterogeneous preferences are held by the group members (genuine dissent) is not without problems. On the one hand, in many decision-making groups and committees this intervention may simply not be possible because group composition is fixed. If, for example, a decision has to be made about whether a company should expand into a new market, then the board of directors is responsible for discussing and deciding this issue. Learning that all directors favor an expansion prior to discussing this issue at a board meeting would, of course, not justify replacing members of the board. It is also hardly practicable to enlarge the board to include dissenters in the decision-making unit.

On the other hand, even if it were possible to ensure genuine dissent, the above-mentioned positive effects of this preference heterogeneity are often accompanied by negative side effects. For example, groups with strongly diverging attitudes among their members have been shown to lack cohesion (Jackson, 1992; Terborg, Castore, & DeNinno, 1976). The lack of social integration in such groups can lead to an increase in turnover among group members (O'Reilly, Caldwell, & Barnett, 1989). If we take into account that maintaining or enhancing the members' capability to work together in the future is an important dimension of group performance (Hackman, 1987), then these effects have to be taken seriously. In addition, heterogeneous groups are slower to implement decisions than homogeneous groups (Hambrick, Cho, & Chen, 1996; White, Dittrich, & Lang, 1980). Especially in competitive environments this delay can be disadvantageous (Judge & Miller, 1991) and this disadvantage may thus outweigh the informational gains in the decision process of heterogeneous groups. All in all, genuine dissent has its costs, and an organization facing the problem of optimizing decision processes in its decision-making units may judge this price to be too high.

Due to these possible problems, another approach to instigating conflict in the group decision process has gained importance, namely contrived dissent. Various procedures have been developed to stimulate debate by assigning controversial roles to the group members, regardless of their real opinion (for overviews see Katzenstein, 1996; Mason & Mitroff, 1981). Among these, the so-called “devil's advocacy” procedure (Herbert & Estes, 1977) has received particular attention in theory as well as in empirical research (e.g., Schwenk, 1984, Schwenk, 1990; Schwenk & Cosier, 1993). In devil's advocacy, a group member or a subgroup is assigned the role of the devil's advocate whose task is to criticize proposals made by the group. When consensus on a particular solution to the decision problem has emerged, the devil's advocate generates counterarguments to this solution and tries to identify all weaknesses inherent in it. The group then has to react to this criticism and check whether the arguments put forward by the devil's advocate can be invalidated. Following this check, the proposed solution is either chosen or rejected.

A number of empirical studies have demonstrated that devil's advocacy facilitates the quality of group decisions. Student samples (Schweiger, Sandberg, & Ragan, 1986; Schwenk & Cosier, 1993; Schwenk & Valacich, 1994; Valacich & Schwenk, 1995) as well as managers (Schweiger, Sandberg, & Rechner, 1989) discussed decision cases in small groups either using a consensus approach or by acting out a devil's advocacy debate.2 These studies consistently demonstrated that groups using devil's advocacy made decisions of higher quality than the consensus groups did. In some of the studies, these gains in decision quality were accompanied by negative side effects; for example, compared with members of consensus groups, members of groups using devil's advocacy showed lower satisfaction with the group process and outcome (Schweiger et al., 1986; Schwenk & Cosier, 1993), less acceptance of the decision (Schweiger et al., 1986), and less desire to work with each other in the future (Schweiger et al., 1986; Schwenk & Cosier, 1993). However, other studies revealed no such effects (Schweiger et al., 1989; Valacich & Schwenk, 1995) or even reported higher satisfaction among people using a devil's advocacy method (Schwenk, 1984). In addition, the data provided by Schweiger et al. (1989) point to the possibility that increasing familiarity with dialectical decision techniques like devil's advocacy may counteract negative side effects on group members' motivation or satisfaction. In sum, contrived dissent seems to have lower costs with regard to group climate than genuine dissent and, thus, organizations may be less reluctant to use the former than the latter as a tool to debias group information seeking.

However, the research on devil's advocacy does not allow any conclusion to be made about whether devil's advocacy has a debiasing effect on group information search. In his book on groupthink, Janis (1982) (cf. also Janis, 1989) assumes that devil's advocacy prevents a group from falling prey to premature consensus and thus allows the group to perform an unbiased decision process. But, to our knowledge, no empirical tests of this assumption have taken place so far. Chen, Lawson, Gordon, and McIntosh (1996) found no differences between a devil's advocacy condition and a control condition with regard to groupthink tendencies. However, the manipulations used by these authors do not conform to any version of devil's advocacy documented in the literature, so we will not consider these results here.

In addition, hardly any studies have compared the effects of genuine and contrived dissent on group decision making. In a recent experiment, Nemeth, Connell, Rogers, and Brown (2001) exposed participants to a dissenting opinion that was labeled as the opinion either of an “authentic” minority or of a devil's advocate; in a control condition no dissent was introduced. Whereas the former depolarized individual thoughts (i.e., no dominance of supporting over conflicting thoughts occurred in this condition), the latter even slightly increased cognitive bolstering compared to the control condition. In addition, only authentic minority dissent led to a significant opinion change in the direction of the dissenter; the effect for the devil's advocate was in the right direction, but not significant.

Whereas these results may suggest a superiority of genuine over contrived dissent with regard to debiasing, it is difficult to generalize them to group decision making because in the Nemeth et al. (2001) experiment no group interaction or group discussion took place (and, thus, no group-level data were provided), nor did the devil's advocate have a chance to fulfill his or her predominant function of systematically reacting to and criticizing the other group members' proposals. Thus, it remains to be seen whether contrived dissent resulting from devil's advocacy counteracts biases like group-level confirmatory information seeking in the same way as genuine dissent resulting from heterogeneity of individual preferences does.

To compare the effectiveness of genuine and contrived dissent in debiasing group information search, we decided to fully match both factors orthogonally in an experiment. As in the Schulz-Hardt et al. (2000) experiments, genuine dissent was operationalized by composing groups of members with consensual individual prediscussion preferences (homogeneous groups) or with a majority and a minority faction (heterogeneous groups). Contrived dissent was induced by the use of the devil's advocacy procedure. With regard to the practical relevance of the experiment, it was deemed necessary to induce devil's advocacy not only in homogeneous but also in heterogeneous groups, because the decision to use devil's advocacy in a decision-making group is often made before anything is known about the homogeneity or heterogeneity of members' preferences regarding the decisions to be made.

In our experiment, we measured to what extent the information search in each of the four conditions is biased in the direction of the alternative preferred within the group, that is, to what extent the groups request more pieces of information supporting their preferred alternative than pieces of information conflicting with their preference. However, as outlined above, simply comparing this confirmation bias across conditions may be misleading, especially in the case of genuine dissent: if, for example, each group member favors information confirming his or her individually preferred alternative, then combining three individuals with the same preferred alternative (homogeneous group) automatically leads to a stronger confirmation bias (in favor of the group decision) compared to a group containing a minority member who favors the opposite information. Thus, to provide a fair test, it is necessary to measure individual information requests prior to the group discussion, to combine these requests to form a statisticized group information search, and to use this as a baseline against which the real groups' information search is compared. The change in the confirmation bias from the statisticized to the real group information search, which we will label net group bias, then reflects group-level processes.

Three different effects are of interest to us: a general debiasing or bias-preventing effect of genuine dissent would mean that, irrespective of whether devil's advocacy is used, heterogeneous groups show a smaller net group confirmation bias (or even a net group disconfirmation bias) than homogeneous groups. Further on we will refer to this as the GD effect. In turn, a general debiasing or bias-preventing effect of contrived dissent would mean that, irrespective of group composition, groups with devil's advocacy exhibit a smaller net group confirmation bias than groups without devil's advocacy. This effect will be referred to as the CD effect. Finally, the effectiveness of genuine and contrived dissent may depend on each other. For example, devil's advocacy may reduce the net group confirmation bias in homogeneous groups, whereas in heterogeneous groups (due to the genuine dissent already present) no such effect occurs. Interactive effects like this will be labeled the CD×GD effect.

In addition to detecting and comparing the effects of genuine and contrived dissent on group information seeking, we also wanted to identify mediators for those effects that occur. Two variables seem to be of particular relevance in this context: controversy of group discussion and confidence about the correctness of the preliminary group decision. On the one hand, both forms of conflict should lead to a more controversial group discussion. This enforced exchange of arguments for and against both alternatives could extend to information seeking and thus induce requests for information both supporting and conflicting with the preliminary group decision. On the other hand, genuine and contrived dissent could reduce confidence about the correctness of the preliminary group decision. As Schulz-Hardt et al. (2000, Experiment 3) have shown, heterogeneous groups are less confident about the correctness of their preliminary decision than homogeneous groups, and this partially mediates the lower confirmation bias in the former groups. However, it is not known whether the same happens when conflict is based on devil's advocacy instead of genuine dissent.

To have an indicator for possible costs of genuine and/or contrived dissent, we also measured to what extent genuine and contrived dissent reduced group members' satisfaction with the group process and outcome. Negative side effects like these are important for decisions about whether one or both forms of dissent can be recommended as a counterstrategy to biased information seeking.

Section snippets

Overview

After having read an economic case study, the participants made an individual decision and were given the opportunity to select additional pieces of information about the decision case. These additional pieces of information either supported or conflicted with their prior choice. Afterward, groups of three persons were formed which either contained members who preferred the same alternative (homogeneous groups) or consisted of a two-person majority and a one-person minority (heterogeneous

Manipulation checks

Of the 201 individuals, 88 decided for country A and 113 decided for country B. No significant differences between these two groups were observed in the individual information search. In addition, neither the age nor the gender of the participants influenced individual information seeking. As mentioned above, three groups had to be eliminated from the group-level data analyses because they failed to fill out parts of the questionnaires correctly. Thus, 64 groups remained, 16 in each

Discussion

Our experiment was designed to test whether and (if so) to what extent two forms of conflict, namely genuine dissent and contrived dissent, counteract biased information seeking and, thus, facilitate reality testing in group decision making. Genuine dissent was stimulated or prevented by manipulating group composition, that is, forming groups of persons who either favored the same alternative individually (homogeneous groups) or constituted a majority and a minority faction with respect to

Acknowledgements

We gratefully acknowledge the help of Felix Brodbeck, Adrienne Huggard, and two anonymous reviewers who made helpful suggestions on an earlier version of the manuscript.

References (66)

  • D. Timmermans et al.

    Effects on decision quality of supporting multiattribute evaluation in groups

    Organizational Behavior and Human Decision Processes

    (1996)
  • J.S. Valacich et al.

    Devil's advocacy and dialectical inquiry effects on face- to-face and computer-mediated group decision making

    Organizational Behavior and Human Decision Processes

    (1995)
  • R.A. Baron et al.

    Behavior in organizations: understanding and managing the human side of work

    (1989)
  • R.M. Baron et al.

    The moderator–mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations

    Journal of Personality and Social Psychology

    (1986)
  • F.C. Brodbeck et al.

    The dissemination of critical unshared information in decision-making groups: The effect of prediscussion dissent

    European Journal of Social Psychology

    (2002)
  • S. Chaiken et al.

    Heuristic and systematic information processing within and beyond the persuasion context

  • Z. Chen et al.

    Groupthink: deciding with the leader and the devil

    Psychological Record

    (1996)
  • R.A. Cosier et al.

    Agreement and thinking alike: ingredients for poor decisions

    Academy of Management Executive

    (1990)
  • H.W. Crott et al.

    The process of inductive inference in groups: use of positive and negative hypothesis and target testing in sequential rule-discovery tasks

    Journal of Personality and Social Psychology

    (1998)
  • A.R. Dennis

    Information exchange and use in small group decision making

    Small Group Research

    (1996)
  • K. Edwards et al.

    A disconfirmation bias in the evaluation of arguments

    Journal of Personality and Social Psychology

    (1996)
  • R.B. Gallupe et al.

    Unblocking brainstorms

    Journal of Applied Psychology

    (1991)
  • D.C. Ganster et al.

    Does training in problem solving improve the quality of group decisions?

    Journal of Applied Psychology

    (1991)
  • J.W. Graham

    Principled organizational dissent: a theoretical essay

  • R.A. Guzzo

    Group decision making and group effectiveness in organizations

  • J. Hackman

    The design of work teams

  • D. Hambrick et al.

    The influence of top management team heterogeneity on firms' competitive moves

    Administrative Science Quarterly

    (1996)
  • P. 't Hart

    Preventing groupthink revisited: evaluating and reforming groups in government

    Organizational Behavior and Human Decision Processes

    (1998)
  • R. Hastie et al.

    Inside the jury

    (1983)
  • T.T. Herbert et al.

    Improving executive decisions by formalizing dissent: the corporate devil's advocate

    Academy of Management Review

    (1977)
  • G.P. Huber

    Managerial decision making

    (1980)
  • S.E. Jackson

    Team composition in organizational settings: issues in managing an increasingly diverse work force

  • Cited by (0)

    The research reported in this article was made possible by grants from the German Research Foundation (Deutsche Forschungsgemeinschaft; [DFG]), Project No. Fr. 472/12-1. The data reported have been part of the first author's doctoral dissertation and the second author's diploma thesis.

    View full text