Report Consensus Decision Making by Fish

Australia School of Biological SciencesUniversity of SydneySydney, NSW 2006AustraliaSummaryDecisions reached through consensus are often more accu-rate, because they efficiently utilize the diverse informationpossessed by group members [1–3]. A trust in consensusdecision making underlies many of our democratic politicaland judicial institutions [4], as well as the design of webtools such as Google, Wikipedia, and prediction markets[5, 6]. In theory,consensus for the option favored by the ma-jorityofgroupmemberswillleadto improveddecision-mak-ing accuracy as group size increases [2, 4]. Although group-living animals are known to utilize social information [7–10],little is known about whether or not decision accuracy in-creases with group size. In order to reach consensus, groupmembersmustbeabletointegratethedisparateinformationthey possess. Positive feedback, resulting from copyingothers, can spread information quickly through the group,but it can also result in all individuals making the same,possiblyincorrect,choice[8,11,12].Ontheotherhand,ifin-dividuals never copy each other, their decision making re-mains independent and they fail to benefit from informationexchange [4]. Here, we show how small groups of stickle-backs (Gasterosteus aculeatus) reach consensus whenchoosing which of two replica fish to follow. As group sizeincreases,thefishmakemoreaccuratedecisions,becomingbetter at discriminating subtle phenotypic differences of thereplicas. A simple quorum rule proves sufficient to explainour observations, suggesting that animals can make accu-ratedecisions without theneedforcomplicatedcomparisonoftheinformationtheypossess.Furthermore,althoughsub-mission to peers can lead to occasional cascades of incor-rect decisions, these can be explained as a byproduct ofwhat is usually accurate consensus decision making.Results and DiscussionTo investigate consensus decision making, we presentedgroups of three-spined sticklebacks (Gasterosteus aculeatus;a common freshwater fish) with images of conspecifics withdiffering phenotypic traits. An individual’s appearance canconvey information to, and thus have important fitness conse-quences for, an observer [7]. For example, abdomen profilecan imply foraging success, whereas color may relate tofishes’ health and small black spots could indicate infestationby a parasite. In our experiment, we chose four phenotypesthatwerelikelytoconveyinformation:size,corpulence,shade,and spottiness. Eleven different images were created, depict-ingsmall,medium,orlarge;fat,medium,orthin;light,medium,ordark;andspottedorplain(seeExperimentalProcedures).Ineach trial, we presented the focal fish with a choice betweentwo replicas with different traits.Figure1givesthedistributionofthenumberoffishfollowingthemore attractive replica fish for experimental trials with one,two, four, and eight fish and the large versus medium and darkversus light treatments. For single individuals, one of thereplicas was always more attractive (in the sense that it wasfollowed more often) than the other. This bias was preservedas group size increased, with the majority of fish followingthe image that was more attractive in the single fish trials. Asgroup size increases, the distribution of the number of fish fol-lowing the popular leader takes a J shape (test of fit of a bino-mial distribution is significant to p < 0.001 for all treatments forgroup sizes of four and eight). In the majority of trials, either allor all but one of the fish followed the more attractive leader,whereas in a substantial minority of trials, all or all but one ofthe fish followed the least attractive leader. Similar resultswere seen across all experimental treatments (see Figure S2,available online): fish in the same trial tend to follow thesame leader, and the proportion of fish following the moreattractive leader increases with group size. These J-shapeddistributions are characteristic of positive feedback andcopying [11, 13, 14].In order to make more concrete statements about the typeof decision making occurring in these groups, we give threealternative hypotheses about the probability that a fish ina group of size n with replica treatment jfollow the more willattractive replica in that treatment, p