Social psychology perspective on collective intelligence
From Handbook of Collective Intelligence
We begin with some definitions:
- Social psychology is the study of human cognition and behavior in the context of social groups.
- Group. Two or more people who interact, depend somewhat on each other, and have common goals, "those social aggregates that involve mutual awareness and potential mutual interaction," McGrath (1984).
- Team. A group with more persistent membership
- Collective. Two or more people, but with little direct contact with each other.
The differences are not material enough for us to make a distinction, so we use "group." See Kerr and Tindale (2004).
Format of this wiki
In the following, we document the following types of information for behaviors (e.g., group goal setting), phenomena (e.g., team performance), or concepts (e.g., process gain), pathologies (e.g., social loafing), biases (e.g., loss aversion).
For each, we list one or more of the following, depending on the richness of available research, listing both theories and empirical evidence where available:
- Typology (what is it?)
- Origins, mechanisms, mediators (how does it work?)
- Issues, pathologies, biases (what are the problems with its workings?)
- Determinants, moderators (what affects it, directly or indirectly?)
In the psychology literature, group work and team performance are closely related topics. Both are very general topics, from which other topics in this wiki follow. Here, we focus on the nature of group work.
Steiner (1972) proposes a typology of group tasks:
- Divisibility: divisible (doing a math proof) versus unitary (e.g., reading a page)
- Optimizing: maximizing/optimizing (optimizing is a term used for qualitative measures, such as [writing a good essay]) or minimizing
- additive: group result is the sum of those of its members—e.g.: group pulling a weight using a rope
- compensatory: group result is the average of those of its members—e.g., estimating a pig’s weight based on farmers’ guesses
- conjunctive: group result requires every member to contribute—e.g.: a relay race
- disjunctive: group result is a chosen outcome from some member
- discretionary: group result depends on how individuals’ results are weighted.
Hackman (19680  offers a different typology for what he calls intellective tasks:
Davis et al. (1976)  also offer a typology:
- two-person two-choice
- coalition formation
Shaw (1971)  divides tasks by their dimensions:
- solution multiplicity
- intrinsic interest
- population familiarity
- cooperative requirements
- intellective vs manipulative requirements.
McGrath (1984)  divides group tasks by the processes involved:
The generate-negotiate processes are sometimes laid along a collaborate-coordinate-conflict resolution dimension, and the choose-execute processes along a cognitive-behavioral dimension. These dimensions form what McGrath calls a circumplex.
Tasks can be classified by the primary process involved:
- generate: creativity, planning
- choose: intellective (have demonstrably correct answer) and judgment (does not)
- negotiate: cognitive conflict, mixed motive
- execute: psychomotor, contests
Setting of group goals
A stream of research looks at what happens when a group's goal is set endogenously by the group, rather than exogenously given to the group.
Since Lawrence and Smith (1953), the weight of the evidence is that a group that collectively sets its own goal tends to perform better. In a recent interesting study, Ludwig and Geller (1997) first observe the level of safety precautions used by pizza deliverers (use safety belt and signal indicators, stop at traffic junctions) at 2 stores. Then 1 store is given an exogenous goal of a higher level, but deliverers at the other store are allowed to participatively set their own goal. The latter does significantly better.
The consensus is that the link from participative goal setting to group performance is mixed. For example, Pritchard et al. (1988), in a longitudinal study over 9 months, report that participative goal setting improves performance by an astonishing 75%. But Mann et al. (1998) report that, in an experiment in which 138 branches of a company is randomly assigned to different ways of goal-setting, from exogenously set benchmarking and endogenous participative goal-setting to a control condition, it is benchmarking that results in higher sales performance. They attribute this to benchmarking's being a more informative: the group knows that the target is achievable and can learn from best practices.
Why and how do group goals impact influence?
- Increase effort and persistence, focus attention. See Bryan and Locke (1967)
- Collective efficacy, which is the group's shared perception that of its ability to perform. Prussia and Kinicki (1996) report data consistent with social cognitive theory, in which feedback and goal setting mediates through collective efficacy to determine performance.
- Sense of justice. In a study involving 235 students, Roberson et al. (1999) find that the link from participative goal setting to performance is mediated by a sense of justice.
Erez and Arad (1986) test three possible explanations for why participation in goal setting might facilitate performance, in an experiment involving 96 white-collar employees. These possibilities are:
- Social factor of group discussion.
- Motivational factor of involvement.
- Cognitive factor of information exchange.
They find that the first two, but not the last, are responsible for enhancing output, learning, and member satisfaction.
- Incentives. Guthrie and Hollensbe (2004) report that the higher power the incentives, the more higher goals groups will set for themselves.
- Social dilemmas.Seijts and Latham (2000) set up 274 subjects in 3- and 7-person groups to invest in personal or group accounts with "social dilemmas." For example, subjects have to decide between low but personal returns versus high but shared, group returns. They report that a social dilemma appears to be a boundary condition for the normally positive effect of group goal setting on group performance."
- Task complexity. Wood et al. (1987), after reviewing 125 papers on the subject, conclude that goal effects are the strongest for simpler tasks such as brainstorming and tasks measured on reaction time, but less on more complex tasks such as business game simulations and scientific production. This is because, in a simple task setting, goals help focus attention and resources, but in a complex task setting, goals can lead to too much focus on one aspect of the complex task at the expense of others.
- Task interdepedence. Mitchell and Silver (1990) report that task interdependence moderates the link from goal setting to performance. Like goal setting itself, task interdependence helps focuses attention and resources. They investigate four conditions: (1) group goal, (2) individual goals, (3) both, and (4) none. They find that all but (2) give the same performance, suggesting that group goals are not particularly additive in its contribution to performance. Conversely, Matsui et al. (1987) find that group goals do work well when tasks are independent.
Group decision-making, aggregation of member inputs
We focus here on how groups' decisions arise out of members' inputs. A special case is when a judge (which could also be an automatic algorithm) alone makes the decision, based on advice from advisors. Important reviews are in http://www.citeulike.org/user/rlai/article/1399470 Cooke (1991)] and Wallsten et al. (1999).
Quality of aggregation
Wallsten et al. (1999) propose that an aggregation scheme is of high quality if it produces outputs that are:
- yield maximum expected value; see also Clemen and Murphy (1990). This means outputs are either accurate (if the event space is well-defined in the form of equivalence classes) or maximally diagnostic (or highly resolved, in the language of Brier score partitions, if the space is not well-defined); see Yates (1988).
Some examples are described below.
This aims to provide a formal theory how individual preferences are aggregated to form a collective preference. It is originally motivated by the Condorcet voting paradox: if voter 1 prefers A to B to C, 2 prefers B to C to A, and 3 prefers C to A to B, any of A, B, or C are equally good if decision is by majority vote.
In psychology, the models evolve to what is today called social decision schemes.
This is the process by which a group develops generalizations from observed data. This work is pioneered by Laughlin and Shippy (1983). Laughlin (1999) summarizes the research stream in the form of 12 postulates, more or less reproduced below:
- Cooperative decision-making groups may resolve disagreement among their members in formulating a collective group response in five ways:
- Random selection among proposed alternatives.
- Voting among proposed alternatives.
- Turntaking among proposed alternatives.
- Demonstration of preferability of a proposed alternative.
- Generation of a new emergent alternative.
- The five ways of resolving disagreement may be formalized by social combination models:
- Random selection: equiprobability model.
- Voting: majority and plurality models.
- Turntaking: proportionality model.
- Demonstration: truth-wins and truth-supported wins models.
- Generation of a new emergent alternative: specified probability of an alternative not proposed by any member.
- Cooperative group tasks may be ordered on a continuum anchored by intellective and judgmental tasks. Intellective tasks are problems or decisions for which there is a demonstrably correct response (e.g., algebra problems). Judgmental tasks are evaluative, behavioral, or aesthetic judgments for which there is no demonstrably correct response (e.g., jury decisions).
- A demonstrably correct response requires four conditions:
- Group consensus on a conceptual system.
- Sufficient information.
- Incorrect members are able to recognize the correct response if it is proposed.
- Correct members have sufficient ability, motivation, and time to demonstrate the correct response to the incorrect members.
- The number of group members that is necessary and sufficient for a collective decision is inversely proportional to the demonstrability of the proposed group response.
- Inductive tasks are both intellective and judgmental: nonplausible hypotheses may be demonstrated to be nonplausible (intellective), but correct hypotheses may not be demonstrated to be uniquely correct relative to other plausible hypotheses that also fit the evidence (judgmental).
- If at least two group members propose correct and/or plausible hypotheses, the group selects among those hypotheses only (demonstration); otherwise, the group selects among all proposed hypotheses.
- If a majority of members propose the same hypothesis, the group follows a majority social combination process (voting); otherwise, the group follows a proportionality process (turntaking) and proposes an emergent hypothesis with probability 1/(H+1), where H is the number of proposed hypotheses (group members).
- Given sufficient information and time, collective induction is comparable to the induction of the best of an equivalent number of independent individuals.
- Collective induction is improved more by multiple evidence than by multiple hypotheses.
- There is more group influence on individuals than individual influence on groups in simultaneous collective and individual induction.
- Positive hypothesis tests are generally more effective than negative hypothesis tests in collective induction.
Social decision scheme
Some of the earliest work in this area include those by Lorge and Solomon (1955) and Smoke and Zajonc (1962). More recently, Davis’ (1973) proposes a social decision scheme. Suppose a group of r members are considering n alternatives. There are n+r-1Cr; scenarios (call this number m) of how the r members vote for the n alternatives. Put another way, for each scenario i=1…m, there is a probability distribution: each alternative j=1…n has a probability, which we can call d(ij). Davis calls the mxn matrix of d(ij)’s the social decision scheme, or D.
p(i), i=1…m, is the probability that members arrange themselves in scenario i. Note that p(i) = rC r1,..,rn x p1r1x..xpnrn.
So if we denote the group’s probability of choosing alternative j=1…n as P(j), then P[1xn]=p[1xm].D[mxn]. Davis derives different outcomes: (1) truth wins, a particular true alternative is chosen if at least one member votes for it, (2) proportionality (an alternative’s probability of being chosen is proportional to the votes it gets), (3) equiprobability (alternatives with non-zero votes are equally probable), (4) majority (alternative with majority votes definitely selected). Davis, et al. (1970) find that among these schemes, equiprobability is the only one that fits the empirical evidence in a trial involving 38 4-undergraduates groups.
A fascinating prediction of the above formulation is that size parameters r and n are well-defined theoretical inputs into group performance. Davis’ (1973) finds that r=4 is the optimal as n increases.
Social judgment schemes
While it is more common for researchers to investigate the aggregation of information, Davis (1996) proposes a method for aggregating preferences. Davis' proposal is to weight an individual's preference by an decreasing concave function of the distance between that preference and others' preferences. There is some empirical evidence that people actually aggregate in this way, see Davis (1996).
Hinsz (1999) suggests that both the Davis and Crott et al. schemes could be enhanced by refining the weights to take account of members' influence, using measures such as central tendency and faction-attraction.
Sequential aggregation is subject to peer pressure (see the section on Conformity). Davis et al. (1988) study 6-person mock juries in which each jury is pre-assembled with 3 persons who originally lean towards guilty and 3 towards innocent. They find that when 3 persons of the same leanings first voice their opinions, the fourth tends to waver. They find a leniency bias, so that the juries' swings are stronger toward innocent than guilty. The swing is smaller the later the straw poll is taken in the discussion, while the bias is bigger the later the straw poll is taken. Indeed, the group decision is often robust to whether polling is sequential or simultaneous (see Davis et al. (1989)).
The setting here is of one judge who is advised by multiple advisors or jurors. Some of the findings include (see Bonaccio and Dalai (2006) for a review and complete citations):
- Even expert judges are reluctant to completely dismiss advice from novices, Harvey and Fischer (1997)
- Judges give more weight to their own private information (egocentric discounting), to information that is consistent with their own beliefs, and to information from experienced advisors see Yaniv and Kleinberger (2000).
- Judges learn from experience. Experienced judges take twice as much advice from the right sources, Harvey and Fischer (1997)
- Surprisingly, judges are most influenced by confident (even if wrong) advisors; see Sniezek and Van Swol (2001).
Ariely et al. (2000) argue that instead of a human aggregator using sophisticated weights on group members' inputs, it is often more accurate to take the simple average of the inputs. Indeed, "substantial improvement" can be obtained with just 2 to 6 members, and the conclusion requires only light assumptions, such as pairwise conditional independence.
Kameda (1996) find that the Japanese technique of nemawashi, in which a leader gathers support from individual members off-line, actually do worse than dictatorial decision-making.
Hastie and Kameda (2005) evaluate 9 aggregation rules against the criteria of accuracy, maximization of members' utility, and resources to employ the aggregation rule. The 9 are:
- Taking the median
- Using Davis' (1996) social judgment scheme of weightings
- Using Borda ranks, in which members first rank the alternatives, and the alternative with the best total rank (obtained by summing individual ranks) is chosen
- Using the Condorcet majority rule, in which all alternatives are compared pairwise, and the alternative that wins in all comparisons is chosen (there may be no unique winner)
- Taking the majority (also called the plurality rule)
- Best member rule, in which group members are first compared (say based on past track record in accurate prediction), and the best member picks the choice
- Random member rule
- Group satisficing rule, in which a choice is first randomly picked, and a second choice is also randomly picked to compare with the first, and the winner is compared with a third randomly picked choice, and so on, until the winner's value exceeds some aspiration threshold.
Using simulation results, the authors find that the majority rule is the best.
Johnson et al. (2001) report empirical evidence that the average input converges to the truth even if individual inputs are not independent.
There is a large literature on combining absolute (instead of probabilistic) measures. Gordon (1921) reports an experiment in which she asks subjects to rank order lifted weights, and conclude that the group's order is almost always more accurate than individuals' guesses. Lorge et al. (1958) and Zajonc (1962) reviews the literature on the subject, including not just rank ordering but direct guesses of absolute measures.
Wallsten et al. (1999) review the literature on different algorithms and statistics for aggregating members' inputs which are probabilistic. There is overwhelming empirical evidence in a variety of settings:
- Football game scores, Winkler (1971)
- Precipitation, compared with National Weather Service forecasters and actual, Murphy and Winkler (1987)
- Differentiating psychotic from neurotic patients on the basis of their MMPI profiles, compared with judgment of 29 clinical psychiatrists, Goldberg (1970)
- Detecting abnormality from 45 excretory urograms, Curtis et al. (1987)
- Probability of survival in an intensive care unit, Winkler and Poses (1993)
- Economic growth rates, Graham (1996)
There are broadly three types of methods of aggregating expert judgments:
- Axiomatic; see Morris (1974). In this approach, we first specify conditions for our aggregation rule, and then specify a rule that satisfies those conditions. Some example conditions include:
- Unanimity principle: if all experts agree, then the aggregate output is that agreed judgment.
- Independence: if all experts agree that their inputs are independent, then the aggregate output should leverage that agreement.
Cooke differentiates methods based on classical versus Bayesian methods. Geneth and Zidek differentiate between Bayesian and methods.
Wallsten et al. (1999) list some criticisms of the axiomatic approach. One (see Clemen and Murphy (1990)) is that it leaves the aggregator little discretion in say, using her own information in the aggregation rule. Another criticism (see Geneth and Zidek (1986)) is that the axiomatic approach makes unnecessary and unreasonable predictions, such as an expert becoming dictatorial.
- Classical. In this approach, the aggregator uses scoring rules to weight members' inputs. The usual weights account for members' accuracy and the diagnostic value of their inputs (see Cooke's (1991) characterization of these as low entropy).
- Bayesian. The idea here is straightforward: new information such as members' inputs are incorporated into the aggregate by Bayesian updating. Some examples include:
- "Signal detection'. Sorkin et al. (2001) use as weights members' influence over each other. The method is effective for combining inputs that are binary, with yes or no outputs.
- Maximum entropy; see Myung et al. (1996). This uses Shannon's information measure to weight members' inputs and to reflect their interdependence. It aims to capture the maximum amount of information and remain maximally non-committal in areas where information is not available.
Wallsten et al. (1999) argue that the appropriate criterion for aggregating group members' inputs depends on the type of uncertainty: aleatory (associated with external chance factors) or epistemic (associated with internal lack of knowledge); see Heath and Tversky (1991).
Aggregation through brainstorming
See the section on Brainstorming.
Common knowledge effect and aggregation of information
See the section on Common knowledge effect.
Brainstorming (Osborn ,1953) requires that members:
- be expressive
- postpone evaluation
- piggyback on each others’ ideas
- get as many ideas as possible.
Taylor, et al. (1958)  provide the first empirical test of brainstorming. Unfortunately, they find that nominal groups (i.e., individuals working on the same problems, but without communications among the individuals) do better.
At Purdue, Weisskopfjoelson and Eliseo (1961)  find that subjects who are told that ideas will be evaluated give higher quality ideas even if fewer ones, contrary Osborn's original formulation of brainstorming.
Diehl and Strobe (1987) , in their review of this research, suggests that creativity might be enhanced when the groups are told that they are measured on creativity.
Issues and pathologies
Why might brainstroming not work?
- Social loafing. Mullen et al. (1991) study 20 groups and report that individuals collectively produce more ideas, the number of ideas decreases with group size, and the presence of experts hurt group performance. Most ideas are generated when responses are written down, before they are shared. They attribute this result to social loafing.
- Production blocking. Diehl and Stroebe (1987) also find process loss, and that the loss is explained more by brainstorming than whether how performance is rewarded (group credit versus individual credit). They attribute the mediocrity to production blocking.
- Evaluation apprehension. Group members might be weary of being evaluated, Cottrell (1972).
- Derailment. Nijstad et al. (1999) attributes it to interruptions in members’ trains of thought (see work on social cognition, e.g., Levine et al. (1993), Hinsz et al. (1997), Larson and Christensen (1993)).
- Social inhibition, in which members are reluctant to offer too many ideas for fear of being criticized (see Diehl and Stroebe (1991) ). In the psychological safety literature, Edmondson (1998)) attributes it to an unwillingness to contribute ideas because of a lack of psychological safety.
- Social matching. Paulus and Larey (1996), Roy et al. (1996), and Van Leeuwen and and van Knippenberg (2002) propose that members could also match their performance to what is expected of them, or dumb down to the weakest member’s performance because of social comparison.
- Illusion of productivity. Pauhus et al. (1993) report evidence that group members might also think they achieve a lot, because each member takes too much credit for the group’s contribution.
- Topic fixation. The ideas tend to cluster together, indicating some form of groupthink. Some researchers such that brainwriting, in which group members first write out their ideas alone, counter this problem. See Larey and Paulus (1999) .
Researchers have documented moderating factors or modifications of group processes, some of which suggest techniques to counter the issues with brainstorming:
- Use of a facilitator to minimize production blocking and evaluation apprehension (Offner et al., 1996).
- Alternative framing to prevent production blocking. One technique is to decompose problems into subpart, Coskun et al. (2000).
- Electronic brainstorming, to prevent production blocking (Gallupe et al., 1991) or evaluation apprehension (Cooper et al., 1998). Perhaps the most interesting is that electronic brainstorming seems to be more productive with larger groups (Gallupe et al., 1992). This is not because larger groups might generate wilder ideas (Connolly, 1993) but that such groups
- Alternative group processes:
- Interaction process analysis (IPA). This concerns problem-solving among unacquainted members.
- Nominal group techniques
- Expectation-states Theory. Torrance (1954) suggest that higher status members exert greater influence. Ofshe and Lee (1981) report that more assertive members also exert greater influence.
Psychological biases and heuristics
One of the interesting issues is whether and how collective intelligence might accentuate or mitigate psychological biases, and the use of heuristics, in problem solving.
Typology and empirical evidence (individual level)
Examples of biases in individual cognition are below. We note that there are criticisms for some of these, that some of these might disappear with learning (see Tversky and Kahneman (1982) for a review).
This is the tendency to judge the source of others’ errors as dispositional and our own errors as situational. A form of this is the illusion of control. For example, people play craps in casinos tend to throw harder when they want to get larger numbers.
This is the tendency to attribute errors to others and successes to ourselves.
Just world bias
This is the tendency to blame victims for what happens to them, to resolve the cognitive dissonance of believing in a just world and seeing victimization.
This is the tendency to process only the information that confirms our prior beliefs. A form of this is the congruence bias, which is the tendency to test only to confirm a null hypothesis, and not to disconfirm alternative hypotheses. Another version of this is the attention bias. When we see A and B, we say A comes with B (or worse, A causes B), but seldom asks whether how frequent are the other three combinations: not A and B, A and not B, and not A and not B. Yet another version of this is the observer-expectancy or experimenter-expectancy bias: an experimenter unconsciously skews an experiment or his observations to confirm his beliefs. A colorful example of this is Clever Hans, the German horse in the early 1900’s which is purported to answer mathematical questions. Rigorous tests shows that there is no fraud (the horse could answer even if the questions are not posed by the trainer), but it could answer only when the trainer knows the answers. The trainer inadvertently prompts the horse to react. The related subject-expectancy bias is when the subject of the experiment biases his answer to say what he thinks the experimenter wants to look for.
Hindsight and choice-supportive bias
This is the tendency to believe our predictions or choices have worked out, after seeing the outcome.
Trait ascription bias
This is the tendency to see ourselves as more varied in behavior and moods, while others are more predictable (see review in Pronin and Ross (2006)).
This is the tendency to evaluate with an over-emphasis on on conspicuous differences (e.g., both Californians and Mid-westerns say Californians are relatively happier than Mid-westerners, because of the sunny weather in California vs. rainy in London, even though Mid-westerners are as happy on an absolute scale as Californians). It is related to the availability heuristic, in which people over-attribute outcomes to prominent causes and the primacy or recency effects (variations of the serial position effect), in which people over-attribute to initial or recent causes. A version of the primary effect is the reminiscence bump: Janssen, et al. (2005) reviews this literature, in which autobiographies tend to overemphasis early periods of our lives.
This is the tendency to overstate the probability of specific events over general events. It is due to our using the representativeness heuristic, in which people under-estimate the base rate of events happening.
This is the tendency that our willingness to accept compensation for something we have exceeds the willingness to pay for something we do not have. Critics of earlier experiments on this effect argue that the two goods (one we have, one we are asked to pay for) are different in some ways, such as how scarce they are (e.g., Shogren, et al. (1994)). It is also called the reflection effect in prospect theory, caused by the loss aversion bias. A version of this is the sunk cost bias. Horse race gamblers who have already made their bet on a horse gives considerably higher chance to the horse’s winning, versus those who are about but have not bet on the same horse, Knox and Inskter (1968).
This is tendency to reverse our preferences: when comparing payoffs in the near term, we prefer smaller payoffs now, but when comparing payoffs in the future, we prefer delayed larger payoffs. People who exhibit a higher hyperbolic discounting are also those who have less self-control.
This is belief that something is going to happen because it has not happened for some time, even though the events are independent. The inverse gambler’s fallacy is that, on observing that something has happened, we believe that it has not happened for some time. A related bias is the cluster illusion or the hot hand fallacy, in which we see patterns from random series of events (or more colorfully, the Texan sharpshooter fallacy, in which a Texan gunman shoots several shots at a barn door and then draws target rings around those shots). Another version of this is the telescopic effect, in which recent events are thought to have occurred further back and distant events more recent.
Impact or durability bias
This is the overestimate the length or intensity of an event’s impact on our feelings.
This is the attribution of outcomes to our doing, when they are simply regressions to the mean.
This is the tendency to seek information even when it will not affect our decisions.
An example is overly optimistic estimates of project completion (although that could also be attributed to other causes ,such as impression management). Svenson (1981) reports that 80% of the drivers he studies claims that they belong to the top 30% in terms of skill and safety. Interestingly, overconfidence may explain the existence of winner-take-all tournaments, in which the real probability of promotion and reward is limited to only very few. We work hard and join such tournaments because we overestimate our chances of winning. Another version of this is the valence effect, in which we overstate the chance of good things happening and understate the bad. Yet another version of this is rosy retrospection, in which we rate past events more optimistically than we had originally rated them. Also, [ignorance more frequently begets confidence than does knowledge,] so that overconfidence seems to be even more severe among the less competent.
This is the urge to do the opposite of what one is told to do, or what is hardest to get.
Our decisions are influenced by what recent events (this is another version of the availability heuristic at work). Ariely, et al. (2004) asks students to make bids in an auction. But the students are first asked to write down the last 2 digits of their social security numbers. Those students with bigger digits make bigger bids. A related issue is framing. The general result is that less risky choices are chosen when options are presented in positive light, and more risky ones chosen otherwise.
This is the tendency to judge the probability of the parts (say the total probability of death from 3 prominent types of cancer) to be more that of the whole (the probability of dying from any type of cancer).
Also called the Barnum or Forer effects, this is the tendency for us to believe specific things written for us do turn out true.
When we judge a person to be good on one characteristic, this good judgment carriers over to other characteristics that we do not know much about. The opposite—a devil’s effect—has also been observed.
We think we know others more than others know us. Even then, we overestimate other’s knowledge of us. Finally, we think we know ourselves more than others know themselves.
This is our tendency to code our media (e.g., sheet music), believing that it is universal (e.g., Western sheet music cannot capture all music).
(source: Robert Sutton, Stanford Entrepreurial Thought Leaders Podcast). This is the tendency that we get only similar people into our groups, reducing performance because of less diversity. Sutton observes that Amazon gives an award to employees for doing things without first checking with their boss, to counter this bias.
Typology and empirical evidence (group level)
Many of the above seem to also appear at the group level. For example, Argote, et al. (1990) report evidence that groups suffer from the representativeness fallacy. Hinsz and Indahl (1995) report that groups are also susceptible to anchoring, such as juries who anchor their damage awards to the amount requested by plaintiffs. Sunstein et al. (2002) even argue that group juries exhibit more anchoring than individual jurors.
Kerr, et al. (1996) use Davis’ (1973) theory to suggest when groups or individuals might be more biased. Hinsz et al. (2007) suggest that aggregating individual biases to the group level may attenuate or accentuate the biases.
In addition, there are biases that are unique to groups (see review in Hinsz, et al. (2007)):
Idividuals favor their own group members. The bias appears to be reduced with the size of the group. See Sherif's (1936).
Out-group homogeneity bias
Individuals see members of other groups as more homogeneous than those in their own groups. The bias seems to be independent of the size and number of groups, and is not due to the relatively less interaction between the individuals and out-group members.
Groupthink, bandwagon effect, herd behavior
This is the tendency to do what others do.
Facilitation and loafing
A version of this is the contrast effect. For example, a person placed next to a less appealing one is viewed as more appealing than normal, but when placed next to a more appealing one, is viewed as less appealing than normal. See the section on Social facilitation and loafing
This is the tendency for groups to adopt more extreme or riskier positions after discussion. Indeed, the positions are often more extreme than each individual would wants (see also xx). One probably cause is the desire to conform. Another comes from [persuasive argument theory,] in which members suggest more and more reasons to distinguish the options, so that the final option chosen is backed by a lot more reasons than if there were no discussion.
Biased use of information and the common knowledge effect
See the section on Common knowledge effect.
Zajonc, et al. (1968) and others documented that groups tend toward more riskier decisions. In a trial in which groups and coacting individuals choose between getting 0.75 cents if a left bulb lights up and 3 cents if the right lights up, and the subjects know that the left lights up with a probability of 80% and right, 20%, more. As Davis (1973) explains, the standard explanations for this include: (1) riskier members are more persuasive, (2) increased familiarization through group discussions lead members to riskier choices, (3) diffused responsibility also leads members to riskier choices, (4) there is [cultural value inclining subjects toward risk,] (5) unlike the above, the observed group outcome might be due to the social decision scheme rather than shifts in individual preferences (Vinokur (1969)). All but the last two have been doubted, especially because a converse cautious shift is also in the literature (Dion, et al., (1997)). Cartwright (1971) document that 75% of observed group decisions might be due to the social decision scheme, rather than individual preference shifts.
Distortions in multi-level group decisions
Davis (1973) suggests that when there are multi-level group decisions, such as in a democratic political process, the "people’s preference may be very distorted if we use a fair majority social decision scheme. In practice, such distortions might be corrected with minority reports and interest groups. But we are aware of no data that permit a test of the distortions-by-levels argument."
Common knowledge effect
Stasser and Titus' (1985) classic paper point out that groups are inefficient users of information, often under-using private information (the hidden profile) among members. Gigone and Hastie (1993) report evidence that groups focus on shared information, and ignore private information available to members. Winquist and Larson (1998) report that, in a task involving 3-person groups deciding which of 2 professors is to be nominated for a teaching award, unshared information does impact group decision after sufficient discussion.
Researchers have proposed several possible explanations of the common knowledge effect.
Biased sampling model
Stasser and Titus' (1985) propose that members "often fail to effectively pool their information because discussion tends to be dominated by (a) information that members hold in common before discussion and (b) information that supports members' existent preferences…Even though groups could have produced unbiased composites of the candidates through discussion, they decided in favor of the candidate initially preferred by a plurality rather than the most favorable candidate. Group members' pre- and post-discussion recall of candidate attributes indicated that discussion tended to perpetuate, not to correct, members' distorted pictures of the candidates."
Furthermore, information usage is a disjunctive task: the group can use a piece of information as long as at least one member mentions it in the group discussion. Therefore, the probability of shared information being mentioned is higher than that of unshared information.
Costly information exchange
Karau and Kelly (1992) propose that information exchange is costly, both cognitively, socially, and in terms of time taken. Therefore, shared information is that which dominates in group processes.
Wittenbaum et al. (1999) contend that members prefer to "discuss and repeat information known by all members (shared) more than they do information known by one member (unshared)." Doing so enhances one's own credibility and reduces dissonance.
A twist on this is that members who might have valuable unshared information tend to accept a powerful other member, even if acceptance does not fully use the unshared information; see Kameda et al. (1997).
Premature preference negotiation and rigidity in updating
Group members often start with their idiosyncratic preference for how to solve the group's problems. But despite the purported use of the group discussion forum as a means to negotiation and resolve different preferences, the reality is that members tend to stick to their original preferences; see Phillips and Edwards (1966) and Gigone & Hastie (1993). See also the work on preference-consistent evaluation by Greitemeyer and Schulz-Hardt (2003).
Tindale et al. (1996) and his colleagues, in a series of papers, argue that if group members share mental models, then it contributes to the denigration of unshared information and an over-emphasis on shared information.
Type of information: serial position and the picture-superiority
Stewart (2004) report that serial position effects (i.e., primacy and recency effects) and the picture-superiority effect (i.e., the tendency to recall pictures better than words) can impact the recall of unshared information by group members for use in the group discussion. They find that primacy and picture-superiority effects affect recall, but the recency effect does not.
Expertise and role assignment
In an experiment involving 3-person groups, Stasser et al. (2000) assign some groups on two dimensions: having some members as experts (who know the characteristics of hypothetical candidates for student government) and knowing which members are indeed experts. They find that the bias from common knowledge effect is reduced by knowing which members are experts, but not by just having experts in the group.
In a related experiment, Larson et al. (1998) find that team leaders help reduce the bias from the common knowledge effect by repeating and eliciting unshared information. However, because they also do the same for shared information, the accuracy of group work (the medical diagnosis in the experimental setting) is not improved.
Larson et al. (1994) report experimental evidence that discussion time, importance of the task, and training all reduce the bias arising from the common knowledge effect.
Greitemeyer (2006) find that designation of a member as an advocate of information sharing and task experience also reduce unshared information bias.
Hollingshead (1996) finds that asking members to rank all alternatives instead of getting them to decide on the one best alternative also reduces bias. Along similar lines, Schulz-Hardt et al. (2006) find that splitting the group process into a search and dissent pre-discussion followed by a ranking discussion also alleviates the bias.
Importantly, for our agenda in collective intelligence, Hollingshead (1996) also finds that computer-mediated groups do not have this reduction in bias, compared with face-to-face groups, probably because only "procedural aspects of group discussion may help overcome" the bias.
Stasser and Stewart (1992) find that groups whose members know that they have among themselves sufficient information to solve a problem (the problem given is to identify the guilty in a mystery story), they disgorge unshared information into a pool to increase the solution rate to 67%, compared with only 35% for groups whose members are told they do not have sufficient information among themselves.
Conformity is the process and result of our modifying our beliefs and behavior to match others’ expectations.
Groupthink is "a mode of thinking that people engage in when they are deeply involved in a cohesive in-group, when the members' strivings for unanimity override their motivation to realistically appraise alternative courses of action." (see Janis (1972)).
Peer pressure is conformity with similar status others. Unlike groupthink, peer pressure is generally used to refer to situations in which peers' beliefs are revealed sequentially, and the peer is from others rather than the self. The classic demonstration is by Sherif's (1936) and Asch (1948) and theory of cognitive dissonance by Festinger (1954).
Moscovici, et al. (1969) investigate whether 2 confederates could influence 4 subjects into saying that a blue slide is green. 8.4% of the groups decide that the slide is green, and 32% of all subjects report seeing green at some point. The influence is especially strong when the minority confederates are consistent and stable (and therefore needs time to develop). Consistency influences because it signifies certainty and competence.
One implication of minority influence is that leaders are often chosen from a consistent minority than from a majority (see Nemeth and Wachtler (1973)).
- the minority is just one person
- it has a rigid style. Minorities perceived as dogmatic have less influence.
- it moves away from the majority position
- it cannot get even one majority member to agree.
Typology and empirical evidence
Harvard psychologist, Herbert Kelman (1958) identified three major types of social influence.
- Internalization is conforming both publicly and privately. This is also called informational compliance. For example:
- Sherif's (1936) autokinetic experiment asks subjects to estimate how much a light in a dark room has moved. Subjects tend to bunch up their estimates into a narrow range.
- Baron, Vandello, & Brunsman (1996) asks subjects to identify a previously-shown suspect among a line-up of other suspects. Subjects who are told that their input is important and will be used in legal work conform more than those who are told that theirs is only a trial.
- Compliance is to conform only publicly, but keep to one’s own beliefs privately. This is sometimes called normative compliance. For example:
- Asch (1948) asks a subject to identify which of a collection of lines is the same length as a benchmark line. In one version, the subject is asked for his answer after a number of confederates, posing as other subjects, deliberately give wrong answers. Subjects are found to overwhelmingly conform with the wrong answers even though they later say that they privately do not agree.
- Baron, Vandello, & Brunsman (1996), modify their earlier "eye-witness experiment" so that subjects now have a larger time during the first time they see the suspect. In the identification parade, the results are now reversed: the subjects who are told their’s is only a trial conform more.
- Identification is conformity with someone or a group who is liked and respected, see Singer et al. (1965).
Origins and mechanisms
- High group cohesion and homeogeneity
- Charismatic and directive leadership
- Long periods of isolation from credible outside input.
Janis also proposes that groupthink might be detected when we observe:
- invulnerability or unquestioned self-belief among group members, especially with self-appointed "mind guards" (also known as "party whips")
- that warnings are discounted, outsiders are stereotyped, dissenters are ostracized, deviating ideas are shot down
- or in general, deviations are punished and conformance is rewarded.
Different mechanisms for conformity versus minority influence
However, Moscovici and Personnaz (1980) argue that there are differences. For example, the minority pressured by the majority tends to exhibit their compliance publicly even if they keep dissenting beliefs private, while the majority influenced by the minority tends to accept privately, and possibly without public acknowledgment. Maass and Clark (1984) suggest that social impact theory and social influence models predict public compliance, and are better explanations for conformity, while attribution theory predicts internalized, private changes in beliefs, more consistent with minority influence.
Another difference is that minority influence tends to stimulate cognitive activity. One reasonable for this is that minority influence is less credible, and so must be observed only with greater cognitive activity, see Moscovici and Personnaz (1980). Another explanation is that minority opinions are taken more seriously because, according to Kelley's augmentation principle (see section on Attribution theory), minorities have to face off majority pressure. A third explanation is that minorities face greater pressure, and according to Yerkes and Dodson (1908), that generates more demanding validation process, associated with cognition. A fourth explanation is that accepting a minority position runs the risk of being classified as a minority, so that additional risk motivates greater cognition.
Social impact theory
Latane’s (1981)suggests that conformance is higher:
- the more important the group is to the individual
- the closer time and space proximity of the group to the decision
- the larger number of people in the group (although this last reduces conformance when the number gets sufficiently large).
Janis and others suggest other factors that moderate groupthink:
- some members are assigned the role of "critical evaluator" or devil’s advocates; these may be different members each time; likewise, independent outsiders should be invited and group members should be allowed one-on-one as well as sub-group face-time
- the higher status a member has, the later she should express her opinion
- set up independent groups to compete coming up with different solutions
- Each member should discuss the group's ideas with trusted people outside of the group.
Esser (1998), in reviewing the empirical literature, contend that many theoretical predictions are not empirically tested in a rigorous way.
Self-attention theory and the other-total ratio
Scheier and Carver (1979) propose that self-attention (such as that induced using mirrors or a watching audience) can counteract social conformity. Mullen 91983 report that conformity is an increasing concave function of the size of the opposing faction over the total group size (other-total ratio), and a decreasing convex function of the own-total ratio.
Social influence model
Tinford and Penrod (1984) fits a Gompertz growth function to empirical jury decisions data and suggests a function like:
Influence = constant x Exp(-4 x Exp(-1.75 sources/targets))
where sources is the number of people the other faction seeking to persuade our faction, and targets is the number of people in our faction.
This results in an S-shaped relationship between influence and the sources/targets ratio, which is similar to the relationship with the other-total ratio, except that it predicts a sharper gradient.
Nemeth and Wachtler (1977) report that unlike majority factions, the smaller the minority faction, the more influence it has. This is because others judge that a small faction that maintains its position is probably right. However, a countervailing force is that the smaller the minority faction, the lower is its perceived competence. Indeed, the latter is stronger than the former, so that smaller minorities tend to exert less influence.
Besides minority faction size, another determinant of minority influence is how far is its position from the majority. There is a large literature with mixed results (see Maass and Clark (1984) for a review). One consensus seems to be that a small discrepancy is (unsurprisingly) irrelevant, but minority influence exerts itself when there is large discrepancy only if the minority and majority positions are on the same side of a neutral point (see Nemeth and Endicott (1976)).
The normative context is also important. Moscovici, et al. (1969) find that when the environment favors originality and creativity rather than objectivity, minority influence is increased. Similarly, Mugny et al. (2002) report that the majority is more likely to be influenced when their expectations are met, consistent with the correspondence hypothesis.
Double minorities (different from the majority in both belief and membership in a social group) are found to have less influence than single minorities (different only because of beliefs), see Maass and Clark (1984). This is attributed to double minorities being perceived as being self-interested.
- agreed upon by many (consensus)
- maintained by people over time (consistency)
- is distinct from other positions (distinctiveness).
Kelley also proposes that these factors are subject to certain principles:
- Covariation. They interact. For example, a position held by the consensus that is inconsistent is less persuasive. A position held by a consistent minority (distinctiveness) but a different position held by the majority that is also consistent is less persuasive.
- Augmentation. The factors are perceived as more important if there are inhibitors against them. In the minority influence setting, the inhibition is the obvious peer pressure against a distinctive consensus among minority members.
- Discounting. The factors lose their strength when other positions with stronger factors are present.
Social facilitation and loafing
Social facilitation is when we perform better in the presence of others. Social loafing is when we perform worse in the presence of others.
Note that "worse" or "better" isn't necessarily "less active" or "more active".
The consensus seems to be that performance on simple tasks tend to improve when people are under attention, but that on complex tasks tend to deteriorate. For example, Triplett (1898) observes that cyclists do better when there are other cyclists.
Zajonc (1965) demonstrates that this "mere presence effect" or "audience effect" is also observed in rats and cockroaches, suggesting that the effect does not require higher cognition. This is consistent with the Yerkes and Dodson (1908), who suggest that for simple tasks, performance is linear with arousal, but for complex ones, it has an inverted U-shape. In other words, the downward-sloping part of the curve appears only in complex tasks, and is presumably caused by stress on cognition processes, such as "tunnel vision."
The literature offers several theories to explain both social facilitation and loafing.
- Drive theory. Zajonc (1965) argues from an evolutionary perspective, that the presence of others leads to [arousal.] Social facilitation occurs on simple tasks whose dominant response is to improve (in Michaels et al. (1982), better pool players’ dominant response is to improve, so when being watched, they do better) and social impairment occurs on complex tasks whose dominant responses is to deteriorate (in Michaels et al. (1982), poorer pool players’ dominant response is to deteriorate, so when being watched, they do worse).
- Evaluation Apprehension. Cottrell (1972) suggest it is not arousal, but the worry of being judged, that is the origin. A capable person being watched wants to show-off her capabilities, while a less capable person fumbles if being watched. Bartis et al. (1988) report that subjects who are asked to list possible uses for a knife list more users when they are individually evaluated then when their performance is pooled. However, they also report that this is reversed when subjects are asked to list uses that are "as creative as possible."
- Distraction conflict. Sanders et al. (1978) suggest that the audience distracts the subjects and that may increase performance (for simple, but not complext) tasks. They propose several mechanisms through which this might work:
- Dissonance induction. Baron et al. (1973) argue that have an audience distracts subjects, and this is a source of cognitive dissonance
- Overcompensation. Allport (1924) suggests that, when distracted, people might be biased to work harder to overcompensate for the distraction, leading to observed increased in performance
- Uncertainty. Averill (1973) propose that the distraction is a source of uncertainty, so subjects rationally compensates for it by increasing performance.
(Latane et al. (1979)) suggests that people do worse when in a group, because:
- Free riding. People have less incentive when the reward is pooled
- Hard to evaluate contribution. It seems to be mitigated when the task or the group is important.
- Low identification. People do not identify with the group or its task
- Regression to the mean. People identify with the group and regress to the mean or lower standard of performance
The following moderating variables attenuate social loafing. Some of these could be practical ways to counter loafing.
- enhancing individual accountability with well-defined tasks
- increasing intrinsic motivation with a better match between talent and task requirement
- building ownership by giving individuals some control over what they do and when they do it.
Some of the factors above interact. For example, Williams and Karau (1993) report that in cases where members feel that the group task is important to them and that cohesiveness is low, such members actually compensate for the low performance of other members.
Group performance, process loss and gain
Steiner (1972) proposes that group performance can be measured relative to some benchmark, so that is may be classified as process loss or process gain (sometimes also called the "assembly bonus effect").
Lorge and Solomon (1955) propose an informational model that predicts process loss or gain: if individual I has a probability Pr(I) of solving a problem, then a group’s probability of solving the problem is 1 – (1 – P(I)) for all the I’s.
Steiner (1972) is more directly useful in thinking about collective intelligence. He proposes that group performance depends on:
- task demands: these are the planned procedures for completing the task
- process: these are the actual procedures used.
Steiner believes that resources determine potential productivity, while a mismatch of processes and task demands ([faulty processes]) leads to below-potential actual productivity. The latter mismatch could arise from:
- coordination loss
- motivation loss.
This raises a number of issues comparing group with individual performance:
- Given a task, do groups (what size and configuration?) do better than coacting individuals, than single individuals?
- Also, given a task, what kinds of individuals do better in a group (and again, what size and configuration?) versus alone?
- What are efficient matches in the task-group or task-individual mapping?
The above results in a typology of group tasks (see the section on Group work). For Steiner, the impact of group size on performance depends on the type of task. For example, on the dimension of combinability:
- additive: Ringlemann (1913) reports how the tonnage a group of people can pull using a rope. He reports that, unsurprisingly, a group can pull a greater weight than any individual can, but interestingly, each individual pull less hard, so there is decreasing returns to group size. He attributes the latter to coordination loss (people pulling at different times) and motivational loss (social loafing).
- compensatory: Performance increases with group size (if guesses are independent?). Tziner and Eden (1985) report that, among 208 three-man military crews, ability and motivation have compensatory (the authors say "additive" but they mean "average," rather than "total") effects.
- conjunctive: Performance is that of the weakest member, so group size again reduces group performance. Conversely, if less skilled members increase their effort, group performance increases (the Kohler effect, see Kohler (1926)).
- disjunctive: group size increases group performance. Taylor and Faust (1952) report that 4-person groups do better than 2-person pairs, which in turn do better than individuals, in answer "20 questions" puzzles. This is when measuring absolute output (time needed to get right answers). But when measuring man-minutes, when the reverse is true (put another way, it is cheaper to get the job done, if the time needed to get it done is not a constraint).
The above illustrates the distinction between comparing outputs of group (e.g., weight pulled) versus individual(s) (i.e., either the case of picking just one individual, or a combination of coacting individuals), which is the common approach, versus looking at:
- task productivity: output divided by some measure of cost, such as number of members, or effort of members
- individual productivity: from the point of view of one individual, whether her productivity is better in a group or alone
- opportunity costs: whether an individual is better off doing something else, as a measure of cost.
The baseline performance could be measured differently. In the case where it is measured as the performance level of the group’s most capable member, the "ubiquitous finding across many decades of research" ( Kerr and Tindale (2004)) is that there is process loss (e.g., Hill (1982), Laughlin and McGlynn (1986)).
Most using, again, the productivity of the most competent member as the baseline:
- Laughlin, et al. (2002) suggests that the gain arises from "the highly intellective nature of Letters-to-Numbers problems" (mapping A-Z to 0-9).
- Michaelson, et al. (1989), report an average process gain in 97% of the 222 project teams studied. The average process gain appears to be about 40%. They attribute this result to a more realistic setting outside laboratories: subjects were students who work 32 hours together in organizational behavior classes. They also find that group size seems independent of process gain.
- Tindale and Sheffey (2002) tests whether group members should share more information to counter the common knowledge effect, or less, to avoid cognitive load. The tests involve 120 coacting undergraduates and 395 interacting ones in 5-person groups, who are tasked to recall consonant-vowel-consonant trigrams. They find that, for recall, coacting is slightly better than interacting, and partial redundancy (each of 47 trigrams presented to only 2 members) is better than total. The former is probably because of production blocking (one member’s contribution is hampered by another’s; in this case, there was one pencil to write down the trigrams, so all but only one member can write at one time) and social loafing. When reducing "intrusion errors" (trigrams written there are not the in the presented list), interacting is better than coopting and total redundancy is better.
- Hall and Williams (1970) report that 30 groups trained to be effective in areas such as democratic leadership, flexible communications, cooperative problem solving, openness, and responsibility sharing with a protection of individual rights, do outperform 30 untrained group. The task was for individuals and groups to predict the order in which 11 jurors in the film "12 Angry Men" would switch votes from "guilty" to "not guilty."
- Sniezek and Henry (1989) report that [group judgment process is not well described as an averaging] of individual judgements, but are more accurate. Indeed, "30% of the group judgments were more accurate than the group's most accurate individual judgment," which tends to occur when there is high degree of disagreement and group judgments lie outside the range of initial individual judgments. Group judgments are also more confident.
- Indispensability of weak members. In conjunctive tasks, Hertel (2000) report that groups who are resigned to having weak members are likelier to exhibit the Kohler effect.
Some factors that facilitate group performance are known in the literature, although there is doubt on the endogeneity of some of these. For example, Mullen and Copper (1994) suggest that experiments have not fully accounted for how performance feeds back to cohesiveness.
- Giving member task-relevant information, or a frame or reference Laughlin, et al. (1999).
- Developing cohesiveness (see the meta-analyses in Evans and Dion (1991) and Mullen and Cooper (1994)). Lea at al. (2001) suggest that electronic groups might develop stronger identities and norms, contradicting the prediction by the Social Identity Model of Deindividuation Effects (SIDE) that visual anonymity reduces group attraction. Others have found this relationship from cohesivenss to performance, but moderated by other variables:
- Group norms. Langfred (1998) reports a similar result, but uses data from a 1995 survey of Danish military. He also finds that the link from cohesiveness to performance is moderated by the direction of norms (i.e., task-oriented or not).
- Commitment to the group’s goal, Podsakoff et al. (1997)
- Technology mediation. Straus and McGrath (1994) report that in a study of 72 3-person groups, the face-to-face groups do significantly better than the computer-mediated ones, for the more complex tasks.
- Developing a transactive memory of who knows what. Liang et al. (1995) report that subjects in a task of assembling transistor radios, groups that undergo training together to develop transactive memory do better. Moreland and Myaskovsky (2000) suggest that this can be achieved through group training that focuses on members’ competencies, rather than group identification or motivation. See also early work by http://www.citeulike.org/user/rlai/article/1397545 Wegner et al. (1985)] and Wegner et al. (1991).
- Ensuring team leadership. Hackman (2004) and Hackman and Wageman (2004) argue that leaders can spell the difference between success and failure because they determine what kind of team is created and how the team is structured and coached. They also suggest that leaders often make suboptimal decisions.
- Deploying suitable members for group tasks. Some studies identify general traits, such as loquacity, use of reason to influence, dominance, see Littlepage and Mueller (1997), Littlepage et al. (1995).
- Conversely, configuring groups to better recognize and deploy expertise, such as giving explicit instructions to share information and exploit expertise, getting feedback, or investing in working together (e.g., Henry (1995)).
- Brainstorming. However, the consensus today is that nominal groups outperform brainstorming groups, Paulus and Dzindolet (1993).
- Stress. The literature on individual psychology, the consensus is that stress has an inverted-U shape relationship with performance—e.g., Jamal (1984). The relationship also slides rightward as individuals adapt to stress. Further, the increase in performance is biased toward quantity and against quality. The results seem to be also supported among groups, see Karau and Kelly (1992). For example, Kruglanski and Webster (1991] report that within a group, deviating opinions are rejected more with time pressure while conforming ones are accepted more. Stress works through several mechanisms in its link to performance:
- Focusing resources. Karau and Kelly (1992) suggest that stress, in the form of time pressure, forces groups to focus their resources.
- Need for closure. Kruglanski et al. (2006) summarizes their long line of research on how stress increases the need for closure, and therefore greater conformity in the form of a "group mind."
Using diaries from 222 employees in seven companies, Amabile, et al. (2005)  provide empirical evidence that positive affect, including less stress, enhances creativity.
Predictive measures of group performance
We first review the material in individual performance, then on the transferability of research on individual to collective performance, and finally, on collective performance. In the literature, predictors are usually associated with measures such as IQ. We examine this, as well as other measures, such as those related to personality, not just intelligence.
Predictors of individual performance
Intelligence as measured in IQ tests
The best known predictors of individual intelligence is the IQ test, of which there are many variations, such as:
- Stanford-Binet test. This tests 5 factors: Fluid Reasoning, Knowledge, Quantitative Reasoning, Visual-Spatial Processing, and Working Memory; see Terman and Childs (1912).
- Raven's Progressive Matrices. This tests eductive (think clearly, use knowledge) and reproductive (memorize, recall) abilities; see Raven (1938).
- Wechsler test. This tests 7 areas of verbal ability (Information , Comprehension, Arithmetic, Similarities, Vocabulary, Digit span, Letter-Number Sequencing) and 7 areas of performance (Picture Completion, Digit Symbol - Coding, Block Design, Matrix Reasoning, Picture Arrangement, Symbol Search, Object Assembly); see an early proposal in Wechsler (1926).
The underlying analyses in many IQ test is factor analysis. Spearman (1920) proposes that intelligence's main factor, which he calls g, is that which can be measured. Others later partition g into sub-factors, such as gF (fluid intelligence) and gC (crystallized intelligence).
Ree and Earle (1991) report that g is the single best predictor of job performance. Geary (2004) summarizes the empirical literature, which suggests that high g also correlates with a variety of outcomes, such as low rates of school drop-out, divorce, single-parenthood, incarceration, and being on welfare.
Much more controversial is not how good IQ scores can predict future outcomes, but how much of IQ as measured is nature or nurture, and how malleable and how quickly changeable. Some (not all) of the major papers in the debate include Gottfredson (2003), Jensen (1998), and a [of intelligence researchers]. A more measured tally of the controversy is in Neisser et al. (1996). The Flynn Effect (see Flynn, 1984), documenting that overall IQ scores have been rising, is often marshaled to show that IQ is malleable; see also the  entry on the subject.
Thorndike (1920) writes about social intelligence, now more popularly referred to as emotional intelligence. Mayer et al. (1991) develop the modern EQ (emotional quotient) tests for emotional intelligence (labeled the Mayer-Salovey-Caruso Emotional Intelligence Test, or MSCEIT). Brackett et al.(2006) summarize recent research as well as report that MSCEIT corresponds well to others' perception of the subject, but not the subject's self-perception.
In the popular literature, Goleman (2006) writes that "whenever people come together to collaborate in a meeting or as a team, there is a very real sense in which they have a group IQ, the sum total of the talents and skills of those involved. But the single most important element in group intelligence is not IQ in the academic sense, it is EQ…Harmonious groups are helped by having a particularly talented member while those with friction, fear, anger or emotional static under perform." According Goleman suggests that EQ consists of measures of:
- Confidence. A child’s sense of being more likely to succeed than not, and that adults will be helpful.
- Curiosity. Finding out about things is pleasurable and positive.
- Intentionality. The wish and capacity to have an impact. A sense of competence and being effective.
- Self control. The ability to modulate one’s own behaviour appropriately.
- Relatedness. Engaging with, understanding and being understood by others.
- Capacity to communicate. Exchange ideas, feelings and concepts with others. A sense of trust and pleasure in engaging with others.
- Co-operativeness. Balancing one’s own needs with those of others in groups.
Other aspects of performance
Cattell (1940) constructs a test he believes is free of the biases due to education and culture.
Besides general intelligence, there are also measures of specialized capabilities.
Antonietti and Giorgetti (1998) review the literature on tests of an individual's visualizing and verbalizing skills.
Kozhevnikov et al. (2005) review the literature on tests of an individual's spatial and object recognition capabilities.
Individual versus group
Cattell (1948) argues that a fair amount of research into individual psychology might apply to group psychology. The latter might be analyzed at three levels: (1) some statistics describing individual members of the group, (2) the structural relationships among members and between groups (for example, group membership might overlap), and (3) the syntality or behavior of the group. He also provides 7 theorems that determine the syntality of groups:
- dynamic origins of groups
- vectorial measurement of synergy
- subsidiation in the syntal lattice
- subsidiation in the personal lattice
- hierarchies of loyalty from the law of effect
- synergic constancy in a system of overlapping groups
- isomorphism of syntality change and personality change.
Wegner (1987) suggests that the converse, that the complementarity and specialization shown in group work might have facilitated the development of individual minds that have complementary and specialized functions.
Predictors of collective performance
In the psychology literature, Weschsler (1971)  suggests different types of measures, which have since been further developed by others.
Mean, variance, minimum, maximum measures
Previous attempts tend to predict group performance based on some statistic involving members' performances. For example, Tziner and Eden (1985) report that group performance is an average of individual performance. See [[#Group performance, process loss and gain | section on group performance] for details.
Barrick at al (1998) study 51 work teams of 653 employees. They report that average, variance, minimum, maximum of members' ability and personality scores are good predictors of groups' cohesion and performance. This is confirmed by Neuman et al. (1999), who study 328 retail assistants in 82 teams. They report that there is generally a positive relationship between mean and variance in their Big 5 personality traits, such as openness conscientiousness and agreeableness, to group performance.
Woolley et al. (2006) propose that complementarity measures are more important than mean and variance measures. For example, a person good at recognizing shapes (using the ventral visual system) is better paired with another good at recognizing spatial relations (using the functionally separate dorsal visual system) in a task that involves using both capabilities.
Therefore, they argue that the proper matching of members to roles is important (see also Faraj and Sproull (2000)).
Hackman (2002) points out that just having the right team members, whether "right" is measured in average or complementarity terms, is insufficient. The way in which the members work together is also important. Hackman (1968) point out that the type of tasks, describable by their content and process emphasis, describe as much as 50% of group performance. Finally, Hackman et al. (1976) argue that the interaction of task and team are important.
Woolley et al. (2006) also propose this idea in the setting of visual cognition. According to them, a key determinant of joint effects is the degree to which members collaborate, because once members are assigned roles, the tendency is that they each focus only on their own information (see section on transactive memory in this wiki). Importantly, collaboration matters only for groups whose member-role assignment is incongruent. Collaboration has little effect when the assignment is congruent, and indeed, has negative effect when the assignment is homogenous (members have the same skills, so collaborating only wastes time).
Deindividuation and depersonalization
Diener, et al. (1976) observe that children steal more candies in the Halloween trick-and-treats when anonymous than when known to the group.
Wikipedia has a good description of the issues surrounding the questions about the nature of interpersonal beliefs:
- Accuracy: is A’s belief about B accurate?
- Agreement: does A’s belief about B matches B’s belief about himself?
- Similarity: does A’s belief about C matches B’s belief about C?
- Reciprocity: does A’s belief about B matches B’s belief about A?
- Projection: does A’s belief about B matches A’s belief about herself?
- Meta accuracy: does A know how others see herself?
- Assumed projection: does A believe there is reciprocity?
Cronbach (1955) develop measures (Cronbach’s alpha) for similarity. Kenney (1994) suggests that similarity is neither necessary nor sufficient for accuracy. Vazire and Gosling (2004) achieve consensus among 11 people who rate personality perceptions of people with FaceBook webpages. http://www.youjustgetme.com/ is an on-going website in which users guess the personality of people based on the latter’s description on themselves.
groupthink or herd behavior., Group polarization, Pluralistic ignorance, The Wisdom of Crowds, Crowd psychology Abilene paradox is a paradox in which a group of people collectively decide on a course of action that is counter to the preferences of any of the individuals in the group.
Mischel, et al. (1989), in his Marshmallow experiment, give 4-year-olds the choice of eating a marshmallow now or getting 2 if they wait. They find that 14 years later, the eaters do worse than the waiters on a number of dimensions, including emotional stability and SAT scores.
Unresolved or understudied issues
Getting the right people onto a team
J.C.R. Licklider: "Take any problem worthy of the name, and you find only a few people who can contribute effectively to its solution. Those people must be brought into close intellectual partnership so that their ideas can come into contact with one another," quoted by Weiss (2005)[ http://www.citeulike.org/user/rlai/article/346599]. Licklider also suggests that such people—"knowledge workers"—are too independent by nature, so even when they have been identified, it would be a challenge to get them to work together.