Economics

Depending on how one defines the question, there is a universe of empirical findings on the behaviour of corporate actors in economics. In a way, much of econometrics is concerned with corporate actors in that it looks at firms. In most industries, firms are not manager-owned. However, this work is not behavioural in the sense of this paper. It is interested in establishing how firms react to changes in restrictions, not to find systematic deviations from the predictions of the rational choice model.

But there are exceptions. Most pertinent are the scattered experiments that have created ‘firms’ in the lab. In the first such experiment, groups of three students were given the task to be firms in an oligopolistic market with a homogeneous good. Subjects were not allowed to talk across firm boundaries. There was little tacit collusion. Over time, results came closer to the Nash equilibrium. Initially, ‘firms’ were mainly following routines. Although the setting was relatively simple, it took considerable time until firms figured out that they could calculate best responses. Once they had found this technology, firms tended to stick to it. There were, however, motives beyond profit maximization. For instance, ‘rivalistic’ action played a role, as did attempts at testing the grounds, or at minimizing average cost (Selten and Sauermann, Reference Selten and Sauermann1959).

A later experiment had a similar setting, but allowed for process innovation. Again, collusion played no role. A single ‘firm’ made such a move. But the other firms did not trust its cooperative intentions. Outcomes progressively converged to the Nash equilibrium. But the underlying process was not an attempt at calculating the equilibrium. Rather firms extrapolated the trend, and they assumed that their competitors would behave in pretty much the same way as they themselves (Selten, Reference Selten and Sauermann1967b).

In a third experiment, firms were again composed of three students. The product was homogeneous. Firms could change prices and invest in capacity. Information about the behaviour of competitors was costly. Again, there was very little collusion. However, results deviated from the Nash equilibrium, somewhat to the advantage of firms. The dominant motive was not maximizing profit; this motive only accounted for about 10% of the decisions. It was by far dominated by the intention to sell more, and to stabilize market shares. This motive accounted for more than two thirds of the decisions (Selten, Reference Selten and Sauermann1967a).

These experiments establish cognitive and motivational deviations from the standard economic model. It took time before firms found profit-maximizing strategies. They relied on routines and were content with extrapolating trends. The intention to maximize sales, and rivalistic behaviour, deviated from the motive of profit maximization. All of this has also been reported for individuals.

Psychology

Psychology defines itself as the science of individual behaviour. This explains why there is at best some work related to psychology on the behaviour of corporate actors. Closest are attempts at finding the cause for Bowman's paradox: in most industries, returns are lower, the higher the exposure of a firm to risk (Bowman, Reference Bowman1980). Using the industry's median return on equity as the reference point, an econometric study shows that the correlation between the exposure to risk and returns is negative only if the firm is below the reference point for the industry. The paper interprets this as evidence for loss aversion in corporate actors (Fiegenbaum and Thomas, Reference Fiegenbaum and Thomas1988). A later econometric study adds two more factors to the picture: firms do also take higher risks if their industry performs poorly, compared to other industries. And firms become more risk prone if there is little room for slack (Bromiley, Reference Bromiley1991). This psychological research thus shows that a bias known from individuals is also present in firms.

Organisation science

The rich literature on the behaviour of organizations in organization science, and in management science, on which this survey can only touch (Cyert and March, Reference Cyert and March1992; Scott, Reference Scott2003; Berger and Bernhard-Mehlich, Reference Berger and Bernhard-Mehlich2006) provides additional support. Organizations are likely to focus on what has worked in the past (‘exploitation’), and to neglect the need to prepare for the future (‘exploration’) (March, Reference March1991). For instance, data from railroad accidents show that railroad companies basically only learn from their own experience as long as their performance is close to aspiration levels. They become sensitive to others’ experiences only if they are far enough away from the aspiration level (Baum and Dahlin, Reference Baum and Dahlin2007).

Organizations also exhibit a functional equivalent to loss aversion (Kahneman and Tversky, Reference Kahneman and Tversky1979). This could be shown with data from the US radio broadcasting industry. Firms regularly receive data on their audience, in a format that facilitates comparisons over time, and comparisons with competitors. If one uses these data to construct an aspiration level, firms are much more likely to engage in risky change if they are at or below the aspiration level (Greve, Reference Greve1998). These studies too show that corporate actors are liable to the same biases as are individuals.

Law

The empirical law movement is still nascent. Potentially, legal scholars might, however, be in a good position to generate empirical knowledge on the behaviour of corporate actors. Not only is corporate law a rich field. More importantly even, regulation frequently targets firms. Since it is much easier to govern a small number of professionals, regulation often targets firms as their proximate goal, even if the social problem originates in individual behaviour. A good illustration is car exhausts. Instead of trying to change driving habits, environmental law obliges car manufacturers to endow automobiles with catalytic converters.

In this spirit, it has been shown that corporations are often not deterred by legal sanctions in the way expected utility models predict (Braithwaite and Makkai, Reference Braithwaite and Makkai1991; Axelrad and Kagan, Reference Axelrad and Kagan2000). A study has used a questionnaire to ask farmers, homebuilders, and the owners of marine facilities why they comply with regulation specific to their business, if they do. Marine facilities typically are corporations, while farmers and homebuilders tend to be manager owned entities. While more than 80% of farmers and homebuilders say they comply since this is their civic duty, only 60% of boatyard owners say the same (May, Reference May2005: 330). Moreover, the paper shows that this sense of civic duty is driven by different forces. In farmers, reputation plays practically no role. In homebuilders, reputation is mildly important. In boatyards it is crucial (May, Reference May2005: 334). We thus have found a first systematic difference, both in terms of behaviour and of driving forces. Yet, strictly speaking, it is not one between individual and corporate actors, but rather one between different degrees of corporatization; farmers and homebuilders too run a commercial enterprise.

Manager-owned firms

A study tested Dutch hog farmers, who were all running their own businesses. In computer-guided interviews, they were asked to indicate how they would decide on lotteries capturing the key task of their professional lives: at which price to trade their hogs. Their trading behaviour was precisely predicted by prospect theory. They were risk averse in the domain of gains, and risk prone in the domain of losses. The reference point narrowly coincided with production cost (Pennings and Smidts, Reference Pennings and Smidts2003). In this respect, the corporate setting did thus not change behaviour.

Managerial behaviour

If individual managers have power to act for the corporation, the rich body of knowledge on the behaviour of experimental (student) subjects is a good first approximation (Kiesler and Sproull, Reference Kiesler and Sproull1982; Schwenk, Reference Schwenk1984; Horide, Reference Horide2003; Bazerman, Reference Bazerman2006). Yet managers act on behalf of a legal person. In so doing, they engage foreign money. Their decisions change their employees’ lives. The decisions are taken as signals on the labour market for managers. Boards, and ultimately shareholders, can later hold them liable. Does any of this systematically affect deviations from the standard rational choice model?

The effect that is best studied in managers is overoptimism (Moore, Reference Moore1977; Langevoort, Reference Langevoort1997b). Many managers believe that their companies, and they themselves, are better than average (Larwood and Whittaker, Reference Larwood and Whittaker1977). A third of them say that their probability of failure is just zero; more than 80% say that this probability is between 0 and 30% (Cooper et al., Reference Cooper, Woo and Dunkelberg1988). Managers systematically underestimate the strength of their competitors (Zajac and Bazerman, Reference Zajac and Bazerman1991). Overoptimism is stronger if there are no clear benchmarks (Alicke et al., Reference Alicke, Klotz, Breitenbecher, Yurak and Vredenburg1995); this effect is much stronger in managers than in ordinary subjects (Palich and Bagby, Reference Palich and Bagby1995). Overoptimism leads to too much investment from internal funds, and to too little outside financing (Malmendier and Tate, Reference Malmendier and Tate2005). It also makes bidders overly aggressive in corporate takeovers (Roll, Reference Roll1986; Hayward and Hambrick, Reference Hayward and Hambrick1997). Interestingly, managers seem to distinguish between corporate and personal risk. On average, they are not willing to invest more of their own money in their firm when they are directing the firm into major acquisitions (Boehmer and Netter, Reference Boehmer and Netter1997).

Managers are prone to self-serving reinterpretations of reality (Cox and Munsinger, Reference Cox and Munsinger1985: 88; Langevoort, Reference Langevoort1997a: 638). They take credit for good outcomes and lay blame on the environment for bad outcomes (Clapham and Schwenk, Reference Clapham and Schwenk1991). They tend to pay selective attention to regulatory concerns, like safety (Mendeloff and Gray, Reference Mendeloff and Gray2005), and to perceive reality in a filtered way (Starbuck and Milliken, Reference Starbuck, Milliken and Hambrick1988b). They even misperceive objective facts that are key to assessing their position in the market, such as last year's sales, or the percentage change in their industry's sales in the previous year (Mezias and Starbuck, Reference Mezias and Starbuck2003). In all of these respects, manager behaviour does not seem to systematically deviate from the behaviour of independent individuals. The managerial setting just gives the respective bias its specific form.

This is different in the following findings. Managers act more strategically than non-managers in a game that models the interaction with regulators, the so-called ratchet effect game. In this game, players are assigned the roles of firms and of central planning agencies. Firms have high or low productivity, which is their private information. Firms and planners interact sequentially. Planners set targets. If firms produce above target, they receive a bonus. Initially, most firms meet the target if they can do so. Yet implicitly, that way they reveal their type and give up their information rent. In the next round, planners appropriate the rent by raising targets. After a number of repetitions, high productivity firms understand the incentive structure. They stop increasing output. Ultimately, typically planners are forced to treat high and low productivity firms alike, i.e. the pooling equilibrium is reached. In principle, student subjects and professional managers act the same way. However if the planning frame is added, professional managers as planners set much higher targets, and as firms they converge quicker to the pooling equilibrium (Cooper et al., Reference Cooper, Kagel, Lo and Gu1999).

Also, the endowment effect can be made to disappear when subjects do not have to decide on their own property, but are assigned the role of agents. Specifically subjects were first endowed with a mug, as in the original experiment of Kahneman et al. (Reference Kahneman, Knetsch and Thaler1990). They then could venture to manage a firm. If they did, they were told that their salary was contingent on the firm's profit, and that the firm needed the mug as a production input (Arlen et al., Reference Arlen, Spitzer and Talley2002). These manipulations made the endowment effect disappear.

Behaviour of representatives

There is a solid body of evidence on behaviour within organizations (for an overview see Greenberg, Reference Greenberg2003). But most of this research is concerned with internal viability. It only indirectly touches upon the question how organizations behave in their dealings with the outer world. Yet a number of studies directly address the behaviour of representatives.

One study asked high-level managers and partners of accounting firms to decide, in a deliberately complex case, whether they would issue an unqualified opinion, or whether they would express concern regarding the viability of the firm as a going concern. The experiment manipulated the order in which, at the end of the experiment, five additional pieces of evidence were presented. Half of the participants got this evidence in positive/negative order, the remaining half in opposite order. Participants were indeed significantly influenced by the last pieces of information they received. There thus was a ‘recency’ effect, much as it is known from experiments with (inexperienced) student subjects (Arnold et al., Reference Arnold, Collier, Leech and Sutton2000).

Another study has found that fund managers in Japanese institutional investors exhibit short-termism, herding, and self-marketing to improve the appearance of portfolio performance (Suto and Toshino, Reference Suto and Toshino2004). Marketing personnel exhibit overconfidence (Mahajan, Reference Mahajan1992). Again, the corporate setting does thus not induce systematic differences in behaviour.

Cognition

A whole industry has tested individual subjects against the rational choice model, taken as a norm for good behaviour. Long lists of biases have been compiled (see only Conlisk, Reference Conlisk1996; Kahneman and Tversky, Reference Kahneman and Tversky2000). Are groups prone to the same cognitive effects? There is at least some comparative evidence (for an overview see Kerr et al., Reference Kerr, MacCoun and Kramer1996).

The following study is characteristic of this literature. In probability judgements, students have been demonstrated to be overly influenced by representativeness, and therefore to violate Bayes’ rule by neglecting base rates (Kahneman and Tversky, Reference Kahneman and Tversky1972). In this study, groups of five were asked to estimate the probability that a candidate for admission to a club was an engineer. Participants were either told that 21 of 30 applicants were engineers, and nine of them were physicians, or the other way round. They were given three vignettes, one with character traits expected in an engineer, one with traits expected in a physician, and a neutral one. Groups were compared to isolated individuals deciding silently; individuals instructed to think aloud (to control for verbalization effects), and coacting individuals (to control for social facilitation). Groups neglected base rates in about the same magnitude as individuals when deciding on one of the representative vignettes. They however were significantly closer to the base rate when deciding on the neutral vignette (Argote et al., Reference Argote, Rukmini and Melone1990; also see Argote and Seabright, Reference Argote and Seabright1986).

Groups are sensitive to framing. In some studies, no characteristic differences between groups and individuals could be found (Tindale, Reference Tindale and Castellan1993). But the loss aversion of one member of the three-person group could be overridden by a non-loss majority (Kameda and Davis, Reference Kameda and Davis1990), and the bias could be attenuated by shifting the choice to the group (Neale et al., Reference Neale, Bazerman, Northcraft and Alperson1986; Sutter, Reference Sutter2007). However, group discussion amplified loss aversion, i.e. the dependence of risk preference on the reference point (McGuire et al., Reference McGuire, Kiesler and Siegel1987), and led to escalating commitments (Whyte, Reference Whyte1993). Yet this effect can be mediated by social and cultural norms, and also by the salience of the new information in group discussions (Paese et al., Reference Paese, Bieser and Tubbs1993).

Individuals and groups violate expected utility theory in pretty much the same way when comparing two lotteries that both have the same expected value: a riskier one with a higher gain in the event of success, and a less risky one that, however, also yields a smaller amount of money if the subject wins. If the probability of winning on the safer bet is high, individuals and groups prefer this one. They prefer the riskier one otherwise. Group interaction does not make individuals more consistent with expected utility theory (Bone et al., Reference Bone, Hey and Suckling1999).

Like individuals, groups tend to reverse preferences in risky tasks. They tend to choose to play more conservative (low variance) gambles, but to set higher selling prices for riskier (high variance) gambles, despite the bets having equal expected values. Their choices depend on how the task is presented to them. This effect is more pronounced in groups (Mowen and Gentry, Reference Mowen and Gentry1980). However, in a later study, group discussion weakened the effect. The moderation was due to the fact that some choices were easier to justify (Irwin and Davis, Reference Irwin and Davis1995).

Groups are not spared from the hindsight bias. They do slightly better in that it is more likely that at least one of the group members recalls the initial assessment correctly and influences the group. The advantage goes away, however, if one controls for decision time (Stahlberg et al., Reference Stahlberg, Eller, Maass and Frey1995). Groups also exhibit overoptimism (Jourden and Heath, Reference Jourden and Heath1996). Being a member of a group buffers individuals from later deception (Heath and Jourden, Reference Heath and Jourden1997). Also ‘groupthink’ (Janis, Reference Janis1972) matters: pessimism is interpreted as disloyalty. Those who show pessimism risk being shunned (Kahneman and Lovallo, Reference Kahneman and Lovallo1993).

In Western societies, individuals have a strong tendency to attribute behaviour to personality, rather than to situational factors (Nisbett et al., Reference Nisbett, Peng, Choi and Norenzayan2001). This tendency is attenuated if judgement is shifted to groups (Wright and Wells, Reference Wright and Wells1985; Wittenbaum and Stasser, Reference Wittenbaum and Stasser1995).

Groups are not immune, but less liable than individuals to overconfidence (Sniezek and Henry, Reference Sniezek and Henry1989; Plous, Reference Plous1995). In defining an aspiration level, groups do not significantly differ from individuals. In one study, the goal was defined with respect to the perceived probability of success, and with respect to its attractiveness. A more difficult goal was more attractive than an easier one. The group raised the aspiration level if it had been successful, and it lowered it after failure. Goals tended to be slightly above the previous level of performance (Zander and Medow, Reference Zander and Medow1963).

In terms of cognition, we thus get a mixed result. In principle, cognitive biases known from individuals also work on groups. Yet it is possible that they are moderated by specific rules for taking group decisions, by group discussion, and by norms prevalent in a group.

Risk preferences

Initially, comparisons between individuals and groups seemed to show that groups are consistently more risk prone (for a review see Vinokur, Reference Vinokur1971b; Zaleska, Reference Zaleska1976). However, at closer sight, instead of a ‘risky shift’, experimenters sometimes found a ‘cautious shift’ (for a review see Cartwright, Reference Cartwright1971; Fraser et al., Reference Fraser, Gouge and Billig1971).

One colourful study uses data from a favourite pastime of Alaskans. For a small fee, people can bet on the exact date and the hour when the ice in the Tanana River will break in spring. The individual that comes closest to the truth wins the jackpot. A considerable number of bettors are betting pools, which can be reliably identified by their names. Comparing their guesses with those handed in by individuals, it turns out that groups are more sensitive to the distribution of historical break-up dates, and that the distribution for groups has less mass in the tails and exhibits lower variability. This implies that groups are neither more risk prone nor more risk averse, but more risk neutral than individuals (Adams and Ferreira, Reference Adams and Ferreira2009).

Recently, experimental economists have also become interested. They have found that individuals holding constant absolute risk aversion become risk neutral maximizers of expected utility if deciding collectively (Bone, Reference Bone1998). Another study does not find a shift towards expected utility. But when deciding collectively, subjects make more profit since they come closer to the norms of portfolio selection theory (Rockenbach et al., Reference Rockenbach, Sadrieh and Mathauschek2007). However, investment clubs have been shown to perform less well than individual investors (Barber and Odean, Reference Barber and Odean2000).

The underlying mechanism is complicated (Clark, Reference Clark1971). It is not due to leadership of particularly risk prone members (Hoyt and Stoner, Reference Hoyt and Stoner1968). The aggregation of information within the group matters (Vinokur, Reference Vinokur1971a), as does group homogeneity (Crott et al., Reference Crott, Zuber and Schermer1986), the impact on members’ affects (Vinokur, Reference Vinokur1971b), group norms (Brown, Reference Brown1965), framing (Paese et al., Reference Paese, Bieser and Tubbs1993), reference points (McGuire et al., Reference McGuire, Kiesler and Siegel1987), the tendency to perceive a ‘false consensus’ (Ross et al., Reference Ross, Greene and House1977), and the way interaction within the group is organized: Is there open discussion (Zaleska, Reference Zaleska1974)? Is there an (informal) leader (Brown, Reference Brown1965)? Is individual responsibility for group outcomes clearly established (Wallach et al., Reference Wallach, Kogan and Bem1964)?

Most importantly, in the generation of the group's risk attitude, polarization takes place (Doise, Reference Doise1969; Moscovici and Zavalloni, Reference Moscovici and Zavalloni1969; Myers and Lamm, Reference Myers and Lamm1976; Isenberg, Reference Isenberg1986). The group is more radical than the aggregation of individual attitudes supports. Recently, economists (Cason and Mui, Reference Cason and Mui1997; Luhan et al., Reference Luhan, Kocher and Sutter2007) and lawyers have also become interested in the phenomenon (Sunstein, Reference Sunstein2002). For the explanation of the phenomenon, two hypotheses compete. In order to preserve self-esteem and influence within the group, group members compare their behaviour to the behaviour of other group members (Brown, Reference Brown1965). Alternatively, group discussion asymmetrically exposes members to persuasive arguments in favour of either caution or risk taking (Burnstein and Vinokur, Reference Burnstein and Vinokur1973).

Cooperation

The most prominent contribution of experimental economics to group research has been testing groups on standard games. In the dictator game, in one study groups were more selfish than individuals (Luhan et al., Reference Luhan, Kocher and Sutter2007). However, in another study, the most other-regarding member of the group won the day (Cason and Mui, Reference Cason and Mui1997). In the ultimatum game, groups offered less than individuals (Robert and Carnevale, Reference Robert and Carnevale1997), but they were also willing to accept less than individuals (Bornstein and Yaniv, Reference Bornstein and Yaniv1998). This held irrespective of the internal rules for making decisions (Messick et al., Reference Messick, Moore and Bazerman1997). Also in signalling games, groups played more strategically than individuals. They were more successful, and they learned more effectively (Cooper and Kagel, Reference Cooper and Kagel2005); likewise, in a gift exchange game. The rationalizing effect of group decision was robust to changes in the internal procedure for making decisions (Kocher and Sutter, Reference Kocher and Sutter2007). In a trust game, groups were less trusting than individuals, but just as trustworthy (Kugler et al., Reference Kuglar, Bornstein, Kocher and Sutter2007; also see on the relationship between interpersonal and interorganisational trust Zaheer et al., Reference Zaheer, McEvily and Perrone1998). However, an earlier study on the trust game had found little difference between individuals and groups (Cox, Reference Cox, Zwick and Rapoport2002; see also Insko and Schopler, Reference Insko, Schopler and Sedikides1998). In a power to take game, groups took and destroyed about the same as individuals (Bosman et al., Reference Bosman, Hennig-Schmidt and Winden2006).

When playing a one shot prisoner's dilemma, groups were less cooperative than individuals (for a summary of the evidence see Schopler and Insko, Reference Schopler and Insko1992; Lodewijkx et al., Reference Lodewijkx, Wildschut, Syroit, Visser and Rabbie1999; Schopler and Insko, Reference Schopler, Insko and Foddy1999; Bornstein, Reference Bornstein2003). If given a choice, groups were more likely to use threats, rather than promises, to bring cooperation of their competitors about (Lindskold et al., Reference Lindskold, McElwain and Wayner1977). The effect even remained stable if a dilemma game alternated with a bargaining game (Morgan and Tindale, Reference Morgan and Tindale2002), where groups were more effective than individuals (Polzer, Reference Polzer1996). Likewise, in a common pool dilemma, although groups were less myopic than individuals, they were more competitive and therefore less efficient (Gillet et al., Reference Gillet, Schram and Sonnemans2007).

To study the underlying forces, an experiment using the minimum effort game is revealing. In this game, each member of a larger group has to indicate an integer between 1 and 7. The group payoff is distributed according to these numbers. Yet the group payoff is the higher, the lower the minimum number posted by a group member. The game has seven equilibria in pure strategies, where all members choose the same integer. If the game is repeated, mean and minimum choices quickly converge to a fairly low value, leaving group members with a low payoff (Van Huyck et al., Reference Van Huyck, Battalio and Beil1990). If, however, two groups compete such that the one with the minimum integer gets all, mean integers are almost stable, and minimum integers go up over time. The competition motive helps groups coordinate on a much more favourable equilibrium (Bornstein et al., Reference Bornstein, Gneezy and Nagel2002).

Intergroup competition also improved intragroup cooperation (Bornstein et al., Reference Bornstein, Erev and Rosen1990; Bornstein and Erev, Reference Bornstein and Erev1994). The effect was particularly strong if groups were allowed to communicate internally before acting (Bornstein and Rapoport, Reference Bornstein and Rapoport1988). If group size increased, however, group members were less prepared to overcome the internal dilemma since they perceived their personal contribution as less relevant (Kerr Reference Kerr1989; also see Levine and Moreland Reference Levine and Moreland1990). If two groups of unequal size competed, the effect was stronger on the larger group (Rapoport and Bornstein, Reference Rapoport and Bornstein1989). These findings have triggered a line of research on the relationship between firm size and performance (Gooding and Wagner III, Reference Gooding and Wagner1985; also see Gladwell, Reference Gladwell2000).

Group cooperation remained stable in repeated interaction if communication was permitted (Bornstein et al., Reference Bornstein, Erev and Goren1994); cooperation eroded otherwise (Bornstein et al., Reference Bornstein, Winter and Goren1996). In repeated interaction, in one study groups cooperated more effectively than individuals (Pylyshyn et al., Reference Pylyshyn, Agnew and Illingworth1966). Individual group members contributed significantly more if they could have defected individually, but if this would have reduced group success in an intergroup prisoner's dilemma game (Erev et al., Reference Erev, Bornstein and Galili1993; Bornstein and Ben-Yossef, Reference Bornstein and Ben-Yossef1994). The dilemma was overcome by reciprocity (Goren, Reference Goren2001).

However, in a centipede game, groups exited significantly earlier than individuals (Bornstein et al., Reference Bornstein, Kugler and Ziegelmeyer2004). And groups were less effective in bringing tacit collusion about in a repeated Bertrand game (Bornstein et al., Reference Bornstein, Budescu, Kugler and Selten2008). They converged faster to the competitive solution than individuals (Bornstein and Gneezy, Reference Bornstein and Gneezy2002). However, in a repeated Cournot game, i.e. when teams compete in quantity, there is no difference between individuals and groups (Morrison and Kamarei, Reference Morrison and Kamarei1990; Raab and Schipper, Reference Raab and Schipper2004).

In a chicken game, groups were more aggressive than individuals (Bornstein et al., Reference Bornstein, Budescu and Zamir1997). However, in a game of assurance, they reached the effective result. Groups were thus able to overcome the urge to compete for fear, but not for greed (Bornstein and Gilula, Reference Bornstein and Gilula2003). In a beauty contest game, teams were more effective than individuals since they learned faster (Kocher and Sutter, Reference Kocher and Sutter2005). Specifically, teams of four outperformed not only individuals, but also teams of two (Sutter, Reference Sutter2005). In common value auctions, groups suffered from the winner's curse at least as much as individuals. They did even worse if they had access to a lot of information about the auctioned item (Cox and Hayne, Reference Cox and Hayne2006). Likewise, in an ascending sealed-bid English auction, groups stayed longer and paid significantly higher prices than individuals (Sutter et al., Reference Sutter, Kocher and Strauss2005).

The greater competitiveness of groups rests on a combination of fear and greed (Insko et al., Reference Insko, Schopler, Hoyle, Dardis and Graetz1990; Schopler et al., Reference Schopler, Insko, Graetz, Drigotas, Smith and Dahl1993; Wildschut et al., Reference Wildschut, Pinter, Vevea, Insko and Schopler2003). Fear results from the fact that there is greater distrust in intergroup relations, compared to interpersonal relations (Insko et al., Reference Insko, Kirchner, Pinter and Efaw2005). Greed results from the fact that there is more social support for aggressiveness in the dealings of the group with its competitors (Wildschut et al., Reference Wildschut, Insko and Gaertner2002), and that individual contributions to group decisions are not identifiable by members of the outside group, which makes it impossible to hold individual group members accountable for harm inflicted on outsiders (Schopler et al., Reference Schopler, Insko, Drigotas, Wieselquist, Pemberton and Cox1995).

Group decision making as a political process

All of the previous explained group behaviour from an individualistic perspective. One gains additional insights into the driving forces for the differences between individual and group behaviour if, following the sociological tradition, one interprets group decision making as a political process (March, Reference March1962; Shapira, Reference Shapira1993: 92; Miller, Reference Miller1998). In this process, communication matters (Simon, Reference Simon1997: 208–249). It increases risk taking (Wallach and Kogan, Reference Wallach and Kogan1965), competitiveness (Wildschut et al., Reference Wildschut, Pinter, Vevea, Insko and Schopler2003), and makes groups more aggressive in a prisoner's dilemma (Goren and Bornstein, Reference Goren and Bornstein2000). While the mere reading of alternative arguments has hardly any influence on individual decision making, group discussion has (Bishop and Myers, Reference Bishop and Myers1974). Persuasive argumentation increases polarization (Burnstein and Vinokur, Reference Burnstein and Vinokur1977; Fitzpatrick, Reference Fitzpatrick1989). In buying stocks, compared to individuals, investment clubs are much more influenced by what sounds like a good reason, although such stocks do not, on average, perform better (Barber et al., Reference Barber, Heath and Odean2003). The reported performance of other group members becomes a benchmark and makes group members more risk prone (Payne et al., Reference Payne, Laughhunn and Crum1981). Interestingly, however, if an outsider listens to a group discussion, this produces pretty much the same risky shift as in group members (Lamm, Reference Lamm1967).

In the sociological literature on organizations, the claim features prominently that reality is constructed (Daft and Weick, Reference Daft and Weick1984; Meindl et al., Reference Meindl, Stubbart and Porac1994; Lant and Shapira, Reference Lant, Shapira, Lant and Shapira2001; March, Reference March, Lant and Shapira2001), and that organizations are institutions for making sense out of an overly complex, uncertain and ambiguous environment (Weick, Reference Weick1995; Meindl and Stubbart, Reference Meindl and Stubbart1996). This leads to progressive uniformity (Sutcliffe and Huber, Reference Sutcliffe and Huber1998), ‘organizational languages’ (Wernerfelt, Reference Wernerfelt2004), causal maps that change over time (Barr et al., Reference Barr, Stimpert and Huff1992; Barr, Reference Barr1998).

If, in this discourse, a socially relevant risk is coded as an opportunity for learning, organizational decision-makers are much less cautious than if it is constructed as a threat. Past bad experiences have a stronger effect on future organizational decisions if, in organizational memory, risky hazards are stored as graphic narratives (Tamuz, Reference Tamuz, Lant and Shapira2001). The disaster of the space shuttle Challenger provides an illustration of the effects. NASA had successfully managed so many space missions, it had proven so many internal and external sceptics to be wrong, that the faulty belief could emerge that some of the safety precautions were no longer necessary (Starbuck and Milliken, Reference Starbuck and Milliken1988a).

Horizontal structure

Corporate actors have explicit rules for accepting new members. They can exploit these rules to select members with desired personality traits. For instance, the life insurer Met Life made a point out of hiring people for their sales force that had an optimistic bias (Seligman, Reference Seligman1990).

Corporate actors also have explicit rules for aggregating their members’ views and wishes into collective choice. Different rules produce different group choices (for a review see Kerr et al., Reference Kerr, MacCoun and Kramer1996). Take the following experiment as an illustration of this literature. Participants were randomly assigned to groups of six. The decision task was determining the colour of a jar. Individual payoffs depended on whether the group got the colour right. If the jar was red, it held seven red and three blue balls. If it was blue, it held seven blue and three red balls. Each participant privately drew one ball. Decisions were taken either by majority or by unanimity. In the unanimity case, the group decision was red only if all participants voted red. In the majority condition, the very large majority voted in favour of their private signal, whereas in the unanimity condition, quite a few participants who had a blue signal nonetheless voted red (Guarnaschelli et al., Reference Guarnaschelli, McKelvey and Palfrey2000).

If members exchange information on evidence, a majority rule improves accuracy more than if members exchange hypotheses (Laughlin, Reference Laughlin1988). If group members are presented with easily interpretable, symmetrically informative signals, the majority rule works as a tool for debiasing. The majority rule loses its advantage for accuracy, though, if members receive asymmetrically informative signals (Bottom et al., Reference Bottom, Ladha and Miller2002).

Straw polls may bias group decisions. This for instance happened in a jury like situation when, early on, the provisional overall assessment was asked for. When a considerable number of members had consistently voted for one verdict, later voters were likely to swing. The effect was stronger if the initial row of voters had pleaded non-guilty. This reflected the fact that convicting an innocent is seen as the much graver mistake (Davis et al., Reference Davis, Stasson, Ono and Zimmerman1988).

In cognitive tasks, groups are more effective if, internally, they are not allowed to interact freely. One superior technique is ‘nominal group technique’. In the first phase, members independently and silently generate their ideas in writing. Subsequently, each member presents exactly one idea to the group, one speaking after the other. Eventually, members independently rank the ideas. Another superior technique is the Delphi method. All interaction is in writing. In the first phase, members independently respond to a questionnaire. From these responses, a feedback report is generated. Members independently react to this report in writing (van den Ven and Delbecq, Reference Van Den Ven and Delbecq1974; Surowiecki, Reference Surowiecki2004).

In any group of more than two, there is room for the formation of subgroups and coalitions (for a review see Goodman et al., Reference Goodman, Ravlin and Schminke1987). In a cognitive task, the predetermined majority and minority influenced each other by the exchange of evidence; there was no influence from the exchange of hypotheses; the majority had more influence on the minority than the other way round (Laughlin, Reference Laughlin1988). In another cognitive task, being exposed to alternative minority views improved accuracy (Nemeth and Kwan, Reference Nemeth and Kwan1987).

Vertical structure

Corporate actors with a purely horizontal structure are relatively rare. Small partnerships are the best illustration. Usually, some of the structure is vertical, and there is some form of hierarchy. Employees depend on employers. Managers depend on boards and shareholders. Due to a contract, agents have to act on behalf of principals.

The psychology of principal–agent relations is still relatively tentative. It is being speculated that the main concern of agents is not being held liable (Shapira, Reference Shapira1993); this might be backed by the general research on accountability (for overviews see Tetlock et al., Reference Tetlock, Skitka and Boettger1989; Lerner and Tetlock, Reference Lerner, Tetlock and Ramachandran1994; Lerner and Tetlock, Reference Lerner and Tetlock1999). The effect might be aggravated by the fact that most organizational reward systems punish failure much more than they reward success (Shapira, Reference Shapira1995). This might increase loss aversion (Kahneman and Lovallo, Reference Kahneman and Lovallo1993), and might induce uncertainty avoidance (Cyert and March, Reference Cyert and March1992: 167), narrow framing, such that managers can present success stories at every next reporting date (Kahneman and Lovallo, Reference Kahneman and Lovallo1993), and ultimately even outright cheating (Harris and Bromiley, Reference Harris and Bromiley2007).

The effects of team structure on performance depend on further variables. One study tested medical students on team diagnosis. Three students were presented with information on one and the same patient. Each of them held some pieces of information in private. For one condition, one of them was ex ante determined to be the leader. The presence of a leader improved accuracy for two reasons: leaders took care that the group did not overlook shared information and, once it had been presented for the first time, made sure that unshared information was sufficiently taken into account (Larson et al., Reference Larson, Christensen, Franz and Abbott1998). In another study, hierarchical teams of one leader and three subordinates were better at generating new knowledge if they knew each other well, if team membership was stable, and if the team had time to gain experience (Hollenbeck et al., Reference Hollenbeck, Ilgen, Sego, Hedlund, Major and Phillips1995).

There is also some data on the psychology of principals. When a principal has to generate a single estimate from conflicting evidence agents have reported, principals have a cautious bias. From the same list of numbers they tend to privilege the higher one if the task is predicting cost, and they tend to privilege the lower one if the task is predicting sales (Cyert and March, Reference Cyert and March1992: 80–85). If the principal is aware of a tendency, in different agents, to bias reports, she quickly learns to compensate for the bias, so that her ultimate assessment is unbiased (Cyert and March, Reference Cyert and March1992: 85–92). However, superiors overrate their ability to detect cheating agents (DePaulo et al., Reference DePaulo, DePaulo, Tang, Swaim, Giacalone and Rosenfeld1989; Ekman and O'Sullivan, Reference Ekman and O'Sullivan1991).

Identity

If the narrow incentive view of principal–agent models were a true description of corporate reality, most corporations would be doomed to failure. There is a high premium for firms that are able to induce non-self-interested internal cooperation (Miller, Reference Miller1998). The most important tool to bring this about is the formation of corporate identity (Akerlof and Kranton, Reference Akerlof and Kranton2005). A good illustration is Henry Ford's five dollar day. At a time when this was outrageously more than any worker earned in the car industry, Henry Ford created a long-term commitment to a system of political authority. For Ford workers, the labour market no longer mattered (Miller, Reference Miller1998: 11).

One way of experimentally testing the effect is making different aspects of identity salient for different treatment groups. The task was making money while preserving the sustainability of a common pool resource. Common pool resources have a non-linear harvest function. The non-linearity is due to the fact that the resource is replenishable, but only at an exogenously given rate. Consequently, in optimality total catch is defined by maximum sustainable yield. In the experiment, participants could go on harvesting only as long as the stock was not empty. Two groups had access to the resource: students and elderly citizens. If the fact was made salient that both were members of the same community, participants were much more cautious than if the experimenter indicated that he wanted to study ‘how the behaviour of young people compares to the behaviour of elderly persons’ (Kramer and Brewer, Reference Kramer and Brewer1984).

In line with this, members of college fraternities have been more willing to cooperate with members of their fraternity in a prisoner's dilemma (Kollock, Reference Kollock and Danielson1998). However, making group identity salient was counterproductive if the good was framed as a public good, and if the group was large. Since group members felt that their own contribution was negligible anyhow, they became greedy (Brewer and Kramer, Reference Brewer and Kramer1986). Also, the effect is weaker the more attractive the outside options (Berninghaus et al., Reference Berninghaus, González, Güth, Ramser, Franz and Stadler2004). The willingness to engage in costly altruistic punishment is much higher if the victim is a member of the group that defines the identity of the punisher. The effect has been shown in tribal Papua New Guinea (Bernhard et al., Reference Bernhard, Fehr and Fischbacher2006) and in the Swiss army alike (Goette et al., Reference Goette, Huffman and Meier2006). The causal arrow may even be reversed. Not only does a sense of identity induce members to cooperate more. If subjects are continuously exposed to a social dilemma, this also creates an identity (van Dijk et al., Reference van Dijk, Sonnemans and van Winden2002; Sonnemans et al., Reference Sonnemans, van Dijk and van Winden2006).

Organizations do not only have formal structure. Over time, their members also assume roles, which add informal structure to the organization. Roles do not fully determine behaviour, but they influence how organization members see the task at hand (Simon, Reference Simon1997: 24 f.). They shape normative expectations (Gross, Reference Gross1958), but may also engender role conflicts (Kahn, 1964; for alternative role concepts see Biddle and Thomas, 1979; Biddle, 1986). For instance, when presented with the rich narrative of an artificial business case, executives had a strong tendency to see the central problem in their own area of expertise (Simon, Reference Simon1997: 299 f.). With fixed roles, internal coordination is smoothed by tacit routines (Rick et al., Reference Rick, Weber and Camerer2006). In coordination-intense groups, leaders emerge, even without explicit institutional intervention, and even if group members are not allowed to talk (Guastello and Bond, Reference Guastello and Bond2007). If leaders are highly typical for the in-group, they are strongly endorsed irrespective of their distributional fairness. Leaders who are less prototypical must compensate by stronger distributional fairness for being equally endorsed (Platow and van Knippenberg, Reference Platow and van Knippenberg2001).