UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

When two heads are better than one : the independent versus interactive benefits of collaborative cognition Brennan, Allison Anne 2014

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2014_september_brennan_allison.pdf [ 1.19MB ]
Metadata
JSON: 24-1.0166036.json
JSON-LD: 24-1.0166036-ld.json
RDF/XML (Pretty): 24-1.0166036-rdf.xml
RDF/JSON: 24-1.0166036-rdf.json
Turtle: 24-1.0166036-turtle.txt
N-Triples: 24-1.0166036-rdf-ntriples.txt
Original Record: 24-1.0166036-source.json
Full Text
24-1.0166036-fulltext.txt
Citation
24-1.0166036.ris

Full Text

 WHEN TWO HEADS ARE BETTER THAN ONE: THE INDEPENDENT VERSUS INTERACTIVE BENEFITS OF COLLABORATIVE COGNITION by ALLISON ANNE BRENNAN B.A., Highest Distinction, The University of Virginia, 2008 M.A., The University of British Columbia, 2010 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACILTY OF GRADUATE AND POSTDOCTORAL STUDIES (Psychology) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) August 2014 © Allison Anne Brennan, 2014   ii Abstract  Previous research has shown that two heads working together can outperform one working alone, but whether such benefits result from social interaction or the statistical facilitation of independent performance is not clear. Here I apply Miller’s (1982; Ulrich, Miller & Schröter, 2007) race model inequality (RMI) model to distinguish between these two possibilities. This model was developed to test whether response times to two signals compared to one were especially fast because the observer could detect a signal in either of two ways (i.e., separate activation models) or because both signals contributed to a common pool of activation (i.e., coactivation models). I explored the independent versus interactive benefits of social collaboration in four experiments.  In a first experiment I replicated Miller’s classic finding that coactivation underlies the faster responses to two targets than one during simple visual search by a single individual. However I found that two-person team performance was no faster than the performance of two independent individuals. Reasoning that the division of the cognitive load between collaborators was important to achieving collaborative performance gains, I employed a more complex enumeration visual search task in three subsequent experiments. With this task I found that performance by two-person   iii teams exceeded the fastest possible performance of two independent individuals. This violated Miller’s RMI and indicated that interpersonal interaction produced the collaborative cognition performance gains. I then linked the magnitude of these collaborative gains to features of the interpersonal interaction between team members, including verbal communication, affiliation, and non-verbal communication such as posture, gesture, and body movement.  Together these experiments serve as an important proof of concept that Miller’s RMI can be applied to differentiate between the independent and interactive benefits of collaborative cognition. In addition they demonstrate that the interactive benefits of collaborative cognition are influenced by features of the social interaction between collaborators.     iv Preface  This thesis describes four studies that took place at the University of British Columbia during 2010-14. The author of this thesis was the primary contributor to the identification and design of the research prgramme and analyzed the data in roughly equal collaboration with Jim Enns. Paul Kealong processed the physiological data in Chapter 5. The author did not collect the data presented here, but supervised the efforts of Chris Yeh, Ema Salja, Veronica Galvin, Sarah MacDonald, and Emily Ryan as they did so. Craig Zhou transcribed the verbal communication. Peter Lenkic provided programming and analysis support. All of the writing in this thesis is the author’s own. A modified version of Chapter 4 authored by A.A. Brennan & J.T. Enns is currently under review.  Chapter 4 was presented at the 35th Annual Conference of the Cognitive Science Society and the abstract titled ‘‘Collaborative coactivation in visual search’’ was published in the conference proceedings. Chapter 2 was presented at the 11th Vision Sciences Society Annual Meeting and the abstract titled ‘‘Collaborative coactivation in search’’ was published in the Journal of Vision. This research was approved by the University of British Columbia Behavioural Research Ethics Board (Real-World Visual Search H09-01732).   v Table of Contents  Abstract .................................................................................................................... ii Preface ..................................................................................................................... iv Table of Contents .................................................................................................. v List of Tables ......................................................................................................... vii List of Figures ........................................................................................................ ix Acknowledgments .............................................................................................. xi Dedication ............................................................................................................ xiii 1 General introduction .................................................................................... 1 Cognitive science goes social .............................................................................. 1 Two heads are better than one ........................................................................... 7 Independent versus interactive effects ...........................................................16 Overview of experiments ....................................................................................24 2 Independence versus interaction  during simple visual search .. 27 Introduction ...........................................................................................................27 Method ....................................................................................................................33 Results .....................................................................................................................35 Discussion ..............................................................................................................44 3 Independence versus interaction  during enumeration visual search by  unfamiliar pairs .............................................................................. 48 Introduction ...........................................................................................................48   vi Method ....................................................................................................................53 Results .....................................................................................................................59 Discussion ..............................................................................................................71 4 Independence versus interaction  during enumeration visual search by  affiliated pairs ................................................................................. 74 Introduction ...........................................................................................................74 Method ....................................................................................................................78 Results .....................................................................................................................82 Discussion ..............................................................................................................93 5 Interpersonal interaction qualities  that facilitate collaborative benefits  in enumeration visual search ............................ 98 Introduction ...........................................................................................................98 Method ................................................................................................................. 104 Results .................................................................................................................. 111 Discussion ........................................................................................................... 129 6 General discussion ................................................................................... 133 Limitations and future directions .................................................................. 137 Implications ........................................................................................................ 139 References .......................................................................................................... 144 Appendices ........................................................................................................ 155 Appendix A. ......................................................................................................... 155 Appendix B. ......................................................................................................... 158 Appendix C. ......................................................................................................... 160    vii List of Tables  Table 1.  Mean RT and accuracy with SE in parentheses for individuals detecting one versus two targets .......................................... 36 Table 2.  Mean RT and accuracy with SE in parentheses for teams detecting one versus two targets................................................................. 40 Table 3. Mean RT and accuracy with SE in parentheses for shared space and shared target teams, by the numbers of targets ................ 43  Table 4.  Mean correct RT with SE in parentheses for unaffiliated pairs in Chapter 3 by the factors: test order, number of targets in the display, and social condition .................................................................. 60 Table 5.  Mean accuracy with SE in parentheses for unaffiliated pairs in Chapter 3 by the factors: test, number of targets in the display, and social condition .......................................................................... 61 Table 6.  Mean correct RT with SE in parentheses for affiliated pairs in Chapter 4 by the factors: test order, number of targets in the display, and social condition .................................................................. 83 Table 7.  Mean accuracy with SE in parentheses for affiliated pairs in Chapter 4 by the factors: test order, number of targets in the display, and social condition .......................................................................... 84 Table 8.  Mean correct RT with SE in parentheses for pairs in Chapter 5 by the factors: test order, number of targets in the display, and social condition ........................................................................ 114   viii Table 9.  Mean accuracy with SE in parentheses for pairs in Chapter 5 by the factors: test order, number of targets in the display, and social condition ........................................................................ 115    ix List of Figures  Figure 1. A. Miller’s model of the independence versus interaction of two target signals prior to the response by an individual observer. B.  Miller’s RMI adapted to test the independence versus interaction between two people prior to a joint response ........................................................................................................ 6 Figure 2. The same distribution of RTs depicted with a PDF and CDF ......................................................................................................................... 19 Figure 3. A diagram of the comparisons made in Chapter 2 .............. 29 Figure 4. CDFs of correct RT during search by individuals for one versus two targets ............................................................................................. 37 Figure 5. A typical enumeration visual search display and the four targets that could appear in each display ........................................ 55 Figure 6. Mean cumulative density functions for correct RT by unaffiliated pairs in Chapter 3 ....................................................................... 66 Figure 7. A scatter plot of the negative association between the frequency at which unfamiliar pairs verbally communicated and the interactive benefit gained by collaborating with another person .................................................................................................................... 68 Figure 8.  A diagram of the associations in Chapter 3 ........................... 70   x Figure 9. Mean cumulative density functions for correct RT by affiliated pairs in Chapter 4............................................................................. 87 Figure 10. Communication similarity and Social affiliation scatter plots ........................................................................................................................ 89 Figure 11. A diagram of the associations in Chapter 4 .......................... 92 Figure 12. Mean cumulative density functions for correct RT by pairs in Chapter 5 ............................................................................................. 120 Figure 13. The effects of social group and partition condition on social coactivation ........................................................................................... 123 Figure 14. The effects of social group and partition condition on Verbal Communication Frequency ............................................................ 125    xi Acknowledgments  A heartfelt thanks to all of the bright and kind individuals who have been sunny spots in days since this project began in 2010.  First and foremost I thank my supervisor Jim Enns who has been a role model and mentor, in research and in life. Jim remained supportive (and maintained a healthy sense of humor) as I decided to change research directions or leave research altogether each time that I read a new book. I will not forget ‘‘all things in moderation, especially excellence,’’ and that ‘‘this is not my life’s work, but my dissertation.’’  That being said, I extend many thanks to my dissertation examination committee members Elizabeth Dunn and Rebecca Todd, external examiner Rick Dale, and comprehensive exam readers Peter Graf, Todd Handy, and Jess Tracy. I also acknowledge the UBC Vision Lab and collaborators, especially Craig Chapman, Steve DiPaola, and Liane Gabora, whose comments and suggestions have greatly improved this work.  With gratitude and respect, I thank the research assistants who have contributed to this project: Chris Yeh, Ema Salja, Veronica Galvin, Sarah MacDonald, Craig Zhou, Emily Ryan, and Paul Kealong. I look   xii forward to hearing about the amazing things that these talented individuals will continue to accomplish. Warm thanks and hugs to my colleagues and friends: Sophie Lanthier, Ana Pesquita, Marcus Watson, Peter Lenkic, Ali Greuel, and Eleni Nasiopoulos. This has been so much more fun with these folks than it would have been otherwise.  Thanks to my family, Anne, Kevin, and Lesley, for never bothering me with questions about my research. Considering the men I acknowledged in this section of my previous two degrees are no longer around, here I dubiously and lovingly thank Greg Boothroyd. I appreciate Ken Beatty for his support along the way with the proposal (i.e., the dissertation proposal).  Funding for this research was provided by the Natural Sciences and Engineering Council of Canada in the form of a Doctoral Fellowship to myself and a Discovery Grant to James T. Enns.    xiii Dedication        To learning.                   1 1 General introduction All of life is social. -Steven A. Frank (2007) Human beings are able to pool their cognitive resources in ways that other animal species are not. -Michael Tomasello (1999) Cognitive science goes social In its nascency, research in cognitive science strived to elucidate the processing architecture of the black box in the mind of the individual between stimulus and response left empty by behaviorism. This was the era in which Sternberg (1966) inferred an internal serial-comparison process from the finding that the time to report whether a test symbol was contained in a memorized sequence of symbols increased with the length of the sequence and Chomsky (1968) argued that a theory like generative grammar, which not only attributed internal representations but characterized their underlying order, was needed to explain language. This was the zeitgeist when Miller (1982) developed the race model inequality (RMI) to test whether response times to two signals compared to one were especially fast because the observer could detect a signal in either of two ways (i.e., separate activation model) or because both signals contributed to a common pool of activation (i.e., coactivation model).    2 In more recent years, cognitive scientists have become interested in how individuals process socially relevant information. Enabled by advances in computing technology, studies of attentional orienting have moved from arrows as directional cues, to eye gaze, pointing human hands, and oriented body postures (Nummenmaa & Calder, 2008) and studies of face perception have moved from schematic line drawings to photos and video clips (Palermo & Rhodes, 2007). Psycholinguists began to explore the natural use of language, for example, revealing links between individuals’ word use, social dominance, and personality (Pennebaker, Mehl, & Niederhoffer, 2003). In this era, Miller’s (1982) RMI was applied to test for independent versus interactive effects in the multisensory integration of emotion expressions: females processed audio-visual emotion expressions more efficiently than men, and this difference was not the merely result of auditory or visual processing time differences (Collignon, Girard, Gosselin, Saint-Amour, Lepore, & Lassond, 2010). Most recently, the field of human cognition has undergone a shift towards adopting more of a distributed approach (e.g., Hutchins, 1995), where cognition is considered in the context of an individual’s physical and social environment. This emerging trend is evidenced by the theme of the Cognitive Science Society’s annual conference in July 2013 of "Cooperative Minds: Social Interaction and Group Dynamics" and the accompanying ‘‘Joint Action Meeting,’’ which brought together researchers sharing an interest in individuals’ ability to think and act together. Here cognitive scientists reported   3 on their studies of the inherently social nature of perception and action (e.g., Knoblich & Sebanz, 2006) and advocated that a dynamical systems approach be utilized to quantify how communication between individuals unfolds over time (Dale, Fusaroli, Duran, & Richardson, 2013). This burgeoning research on collaborative cognition has demonstrated that two or more people working together can outperform an individual working alone in several cognitive domains. For example, two people who communicated their confidence in what they saw to one another outperformed one person in a perceptual detection task with noisy visual signals (Bahrami, Olsen, Latham, Roepstorff, Rees, & Frith, 2010). However, this purported benefit of ‘‘collaboration’’ was also obtained from non-communicating teams by selecting the response of the more confident independent individual (Koriat, 2012). This latter result shows that two people can outperform an individual in the absence of any actual social interaction between the collaborators. These opposing accounts pose a dilemma for research on collaborative cognition. Either the benefits of cognitive teamwork result from the communicative interaction among team members (i.e., there is a mechanism of social exchange underlying the benefit) or they are nothing more than the statistical facilitation expected from aggregating independent responses. For example, one person tossing a coin has a 50% chance of it landing on heads, whereas two people each tossing a coin have a 75% chance of it landing on heads when their coin flips are considered together.    4 In Chapter 1, I first review previous research on collaborative cognition in several domains including the estimation of unknown values, creative problem solving, perceptual detection, enumeration, and visual search. I discuss the characteristics of the tasks and the nature of the interaction between individuals that resulted in either team performance gains or losses, relative to individual performance. For each of the experiments reviewed in this first section, I consider the individual performance benchmark against which collaborative performance was compared.  Next I describe how Miller’s (1982; Ulrich, Miller, & Schröter, 2007) RMI can be applied to this new research on collaborative cognition to provide a more rigorous benchmark against which to judge joint performance. As shown in Figure 1A, Miller’s RMI originally tested whether responses to two signals were especially fast because the observer could detect a signal in either of two ways (i.e., separate activation model) or because both signals contributed to a common pool of activation (i.e., coactivation model). I describe the method I employed to adapt Miller’s RMI to test whether collaborative performance gains resulted from interpersonal interaction between team members (i.e., social coactivation model), or whether the gains merely reflected the statistical facilitation of performance by two people working independently (i.e., separate social activation model; see Figure 1B).  Throughout this thesis I use the term social coactivation to refer to the interactive benefit of collaboration that results from the interpersonal interaction between team members prior to a combined joint response.  Social coactivation denotes a   5 collaborative performance gain beyond what can be accounted for with a separate social activation model of independent performance by two individuals, with responses by the faster independent individual on each trial.      6 Figure 1. A. Miller’s model of the independence versus interaction of two target signals prior to the response by an individual observer. In the separate activation model, the faster of two signals triggers the response on each trial (i.e., there is no interaction between target signals). In the coactivation model, the two signals are combined prior to the response on each trial. B.  Miller’s RMI adapted to test the independence versus interaction between two people prior to a joint response. In the separate social activation model, the faster of two people responds on each trial (i.e., there is no interaction between team members and they do not combine information). In the social coactivation model, the two people combine information prior to a joint response.             7 Last, I provide a brief overview of the experiments in Chapters 2 to 5 in which I used Miller’s RMI to explore the task characteristics and social factors that influenced the independent versus interactive benefits of collaborative cognition. Two heads are better than one As early as 1907, Sir Francis Galton observed that the average estimate of the weight of an ox at a Plymouth county fair by a crowd of non-expert individuals was more accurate than even the most accurate individual guess (Galton, 1907). Since then, this ‘‘wisdom of crowds’’ in making judgments about unknown values and events has been extended to hundreds of other estimation tasks including the success of Hollywood blockbusters (Wolfers & Zitzewitz, 2004) and financial forecasts (Clemen, 1989). In a recent book on the topic, Surowiecki (2004) suggests that in order for crowds to be wise, it is important that people’s opinions are not influenced by the opinions of others, and that for the greatest gains, each person should ideally be a specialist who can access private knowledge concerning the problem. In contrast to this, when people have knowledge about the estimates of others, the diversity of their responses is narrowed and the wisdom of the crowd is undermined (Lorenz, Rauhut, Schweitzer, & Helbing, 2011). Research has shown that the wisdom of crowd effect is based on a simple statistical principle. Averaging multiple estimates boosts accuracy because it reduces the systematic and random error of each individual estimate. The benefit of multiple estimates is that   8 they tend to bracket (i.e., fall on opposite sides of) the true value (Soll, 1999). A low correlation among the errors of individual estimates virtually guarantees that the average estimate is more accurate than a randomly selected individual estimate (Larrick & Soll, 2006) and aggregating a few people’s estimates is usually sufficient to boost accuracy, especially if the people have only modestly correlated errors (Hogarth, 1978). In fact, the ‘‘wisdom of the crowd’’ effect is so deeply rooted in the statistical averaging of errors that estimation accuracy can be improved by averaging in the absence of any actual crowd. Averaging multiple estimates from the same individual, where each estimate is based on somewhat different knowledge, has been shown to be more accurate than any single best estimate (Herzog & Hertwig, 2009). Increasing the independence of each estimate improves individual accuracy in much the same way as it improves the accuracy of estimates from multiple different people: within-person estimates provided with greater temporal separation tend to be less similar, and therefore the average of these estimates is more accurate (Vul & Pashler, 2008).  The benefits of collaboration have also long been asserted for creative problem solving, such that groups have been reported to generate a greater number of solutions than the same number of individuals working independently (the nominal group baseline; Osborn, 1948). To produce this outcome, brainstorming groups are instructed to generate as many ideas as possible, to avoid criticism of any ideas, and to strive to combine and improve on others’ ideas.   9 However, decades of research has demonstrated that brainstorming groups think of fewer and less effective ideas than the same number of people working alone, who later pool their ideas (for reviews see Kerr & Tindale, 2004; Mullen, Johnson, & Salas, 1991). Research has suggested that this is the result of several factors. More than one group member is unable talk at the same time when groups work together; this limits the ability of group members to generate ideas at the same time and independently of the speaker (Diehl & Stroebe, 1987). Once a group member has established a productive line of thought, group discussion has been shown to disrupt individual problem solving (Nijstad, 2000). Group creative problem solving performance is also impaired relative to the nominal group baseline because group members can experience evaluation apprehension, and choose not to share their ideas as a result (Camacho & Paulus, 1995)). However, anonymous electronic brainstorming groups, where individuals type their solutions and can see other group members’ typed solutions only when they choose to do so, can be used to alleviate this apprehension (Cooper, Gallupe, Pollard, & Cadsby, 1998). The format of these electronic groups also allows individuals to generate ideas without interruption from other group members, while allowing access to the ideas of others when desired. Under these conditions, brainstorming group performance has been shown to match (Gallupe, Cooper, Grise, & Bastianutti, 1994) and even to exceed (Dennis & Valacich, 1994) the nominal group baseline.   10 In a recent series of high-impact publications, Bahrami and colleagues have explored the benefits of collaborative cognition in perceptual detection and decision-making tasks (Bahrami, Olsen, Latham, Roepstorff, Rees, & Frith, 2010). In these experiments, pairs of participants viewed brief visual displays containing six vertically oriented Gabor patches, equidistantly positioned around an imaginary circle. Each trial consisted of two observation intervals; either the first or second interval contained a target Gabor patch that had higher contrast than all of the others. The perceptual sensitivity of one or both participants was reduced in randomly selected trials by adding noise to the stimuli. Initially, each individual chose the interval that they thought contained the target without consulting with the other. Next, individual decisions were shared between the individuals by way of verbal communication. If participants disagreed, they discussed the matter until they reached a joint decision and entered a joint response.  Bahrami et al. (2010) found that pairs outperformed the more sensitive individual when both individuals received identical amounts of added visual noise. However when individuals received unequal amounts of noise, the performance of pairs did not exceed the more sensitive individual. These data fit a weighted confidence sharing (WCS) model, which proposes that participants communicated their confidence in what they saw to one another, and the resulting joint decision aligned with the individual whose shared confidence was higher. This model assumes that participants have access to their own perceptual noise or variance on the task,   11 that they believe their decisions are unbiased, and that they can accurately communicate their confidence in their decision. Further support of this model was provided in additional experiments. Verbal communication, but not objective accuracy feedback, was necessary for collaborative performance to surpass the more sensitive individual (Bahrami, Olsen, Bang, Roepstorff, Rees, & Frith, 2012a). When participants were allowed to freely discuss their perceptual decisions, they accrued a larger collective benefit than when they were instructed to use a numerical scale to communicate their confidence (Bahrami, B., Olsen, K., Bang, D., Roepstorff, A., Rees, G., & Frith, 2012b). In a detailed analysis of verbal communication, Fusaroli, Bahrami, Olsen, Roepstorff, Rees, Frith, & Tylen (2012) found that pairs showed a high propensity to adopt each other’s language style. Moreover, they observed a strong correlation between a pair’s use of the same task-relevant vocabulary and the magnitude of collaborative performance gains. In their communication analysis they assessed both local alignment (the transition probability that a given lexical expression of a participant would be a repetition of an expression used by the other participant in the previous interaction trial) and global convergence (the degree to which participants in a dyad converged on a limited functional set of shared expressions rather than drifting between multiple sets of expressions). They also considered both discriminate (only dealing with the expressions of confidence directly relevant to the experimental task) and indiscriminate (all lexical items in a dyad’s transcript) linguistic   12 alignment, finding that discriminate alignment was associated with collaborative performance gains. Conversely, indiscriminate alignment was associated with poorer collaborative compared to individual performance. These associations were found at both the level of local alignment and global convergence While Bahrami and colleagues suggest that it is necessary that individuals communicate their confidence through social for joint performance to exceed the performance of the more sensitive individual, this benefit has also been replicated in the absence of social interaction. This was achieved when two individuals both reported whether they saw a target in the first or second observation interval and their level of confidence in this response on a computer. The program in which the experiment was run automatically selected the response of the more confident individual as the joint response, and as such, individuals did not interact during the experiment (Koriat, 2012). While this paradigm used a computer to aggregate individual responses to produce a collaborative benefit in the absence of interpersonal interaction, it does not deny that when individuals communicate their confidence in an interpersonal interaction it can contribute to their joint performance exceeding the performance of the more sensitive individual.  The WCS model, representing what Bahrami and colleagues consider the upper boundary for collaborative perceptual detection and decision-making performance, has also been tested using   13 collaborative visual enumeration (Bahrami, Didino, Frith, Butterworth, & Rees, 2013). The experimental procedure was similar to that of Bahrami et al. 2010, with the exception that individuals enumerated (e.g., Trick & Pylyshyn, 1994) the number of items in a display. Individuals each saw identical stimuli consisting of between 5 and 16 blue and yellow dots that varied in number by one of four possible ratios: 2:1, 4:3, 6:5, 8:7. Each individual first responded independently whether more blue or yellow dots were present. If there was a disagreement, participants negotiated a joint decision and then entered a single joint response. Joint decisions were found to exceed the decision of the better performing individual, demonstrating that enumeration information can be effectively communicated between individuals. However, collaborative enumeration performance did not reach the level predicted by the WCS model. This indicates that the assumptions of the WCS model, which was developed for collaborative perceptual detection and decision-making, do not apply to collaborative visual enumeration. This therefore limits the applicability of the WCS model of collaborative performance gains to collaborative tasks other than perceptual detection under noisy conditions.  Collaborative gains have also been found when two people work together during visual search. In a series of experiments, researchers at Stony Brook University compared the search performance of individuals to search by pairs across several communication and visual display conditions (Brennan, Chen, Dickinson, Neider, & Zelinsky, 2008; Chen, 2007; Neider, Chen, Dickinson, Brennan, &   14 Zelinsky, 2010). Pairs could communicate in one of three ways: verbally with a microphone and speakers, with a gaze cursor that showed where their partner’s eyes were fixated on the search display in real-time, or with both of these methods. There were also three conditions that did not allow for communication between individuals. In the no communication condition, two-person teams completed the same search task as communicating pairs (i.e., each trial ended when the first of two participants responded). In the one-person condition, double the number of individual participants as pairs each completed the experiments fully independently of other participants (i.e., each individual entered a response on each trial). Nominal pairs were created from the one-person search condition by combining post hoc the responses of two random individuals. The faster of these two individual responses on each trial was used to calculate nominal pair performance. This series of experiments revealed that collaborative visual search by communicating pairs was faster than search by one person. Of the conditions under which pairs communicated, search with shared gaze alone was faster than both search with shared gaze and voice, and search with only verbal communication. Two-person teams that communicated with shared voice also outperformed non-communicating pairs and nominal pairs, although error rates were significantly higher for two-person non-communicating pairs. Even though no interaction was possible between non-communicating and nominal pair members, their performance surpassed search in the one-person condition (Brennan et al., 2008).   15 This line of research has also shown that collaborators employ different strategies to divide the cognitive load depending on the demands and constraints of the visual search task. Pairs who searched for a single visual target that differed from distractors by a one feature (i.e., an ‘‘O’’ amongst ‘‘Qs’’) divided the display spatially, each person searching roughly half the display (Brennan et al., 2008). When searching for multiple different targets (e.g., banana and hammer amongst fire hydrant and barbeque), pairs adopted the strategy of dividing the task by target identity, each searching for half of the possible targets (Chen, 2007). Pairs who were unable to communicate with one another (i.e., seated in the same room, but without either shared gaze or voice) were unable to divide the task.  In related research, visual search by collaborative and nominal pairs was compared using signal detection analysis (Malcolmson, Reynolds, & Smilek, 2007). Participants searched for a visual target among distractors (i.e., circle with one gap amongst circles with two gaps) in briefly presented displays. This made it difficult to find the target and made it possible to measure correct target detection (hits) and false target detection (false alarms) and to calculate estimates of target sensitivity and response bias. Participants pressed a key if the target was present and withheld a response when the target was absent. During collaborative search, participants shared a single target present response button and freely communicated while performing the visual search task together. In the nominal condition, each participant independently performed the search task, which consisted of the same sequence of   16 displays so that it was possible to combine the responses from the two individuals. In both collaborative and nominal conditions, responses were designated as ‘‘present’’ when one or both participants indicated the target was present and designated as ‘‘absent’’ when both participants indicated the target was absent. Signal detection analyses revealed that collaborative pairs were more sensitive to the presence of the target and had a more conservative response bias than the nominal pairs. Although collaborative pairs were less likely than nominal pairs to correctly detect a target (i.e., a hit), they were less likely to make false alarms. This result contrasts with the findings of Brennan et al. (2008), reviewed earlier, who found that collaborative pairs made more false alarms than nominal pairs  Independent versus interactive effects  While previous research demonstrated that two people working together can outperform an individual person, and sometimes they can outperform even the aggregated performance of two non-interacting individuals, whether such collaborative benefits result from the interpersonal interaction between team members or merely reflect the statistical facilitation of performance by two independent individuals (as shown in Figure 1) has not yet been directly tested. Steiner (1972) was privy to this distinction between independent and interactive benefits when he formalized the relationship between actual and potential group performance in order to measure whether group work resulted in overall performance gains or losses. He suggested that potential group   17 performance needed to account for pooling effects, such that groups with more members had a better chance at coming to the correct solution, in addition to their additive effects (e.g., in rope-pulling; the group output is the simple aggregation of individual outputs). As Steiner posed the problem, it is important to be able to determine whether and when group performance goes beyond these additive effects, so that the results can legitimately be attributed to an effect of group interaction.  In the series of experiments presented in this thesis, I adapt Miller’s (1982; Ulrich, Miller & Schröter, 2007) race model inequality (RMI) as a tool for parsing the independent versus interactive benefits of group collaboration in cognitive task performance. Miller’s RMI was originally developed to test between two different models of cognitive processing in an individual person, in an effort to account for the ‘redundant signal effect’ (RSE; Kinchla, 1974). The RSE is a common finding in divided attention, where an individual’s detection responses to redundant signals are faster than to single signals. In this section, I will briefly review the background to Miller’s development of the RMI model, before describing how I have adapted it for use in a social collaborative setting. The first model proposed to account for the RSE in an individual posited that it was based on a simple statistical principle. This race model (Raab, 1962) of the RSE proposed that each signal was detected separately. Response times for single signals trials were the result of the latency of a single detection process, whereas   18 redundant stimuli trial responses were triggered as soon as the first stimulus was detected. Therefore redundant signal response times were the result of the faster of two signal detection processes. Because the average time of the winner of the race between two detection processes was faster than the average detection time of a single detection process, the race model accounts for the RSE. The race model can be tested by examining the distributions of response times. According to race models, observed response time (RT) distributions should satisfy, for every value of t, the race model inequality (RMI; Miller, 1982; Ulrich, Miller & Schröter, 2007): Fz(t) ≤ Fx(t) + Fy(t), t > 0, where Fx and Fy are the cumulative density functions (CDFs) of correct response time (RT) in the two single-signal conditions, and Fz is the CDF of RT in the redundant signals condition. As shown in Figure 2, CDFs are comparable to standard probability density functions (PDFs; e.g., the bell curve) that depict the probability of a variable at a given value, except that that they are cumulative so that the probability at each point in the distribution is the probability of a variable at a value that is equal to or less than a given value. Note that this statistical procedure treats trials as independent events, thus ignoring any temporal dependencies that might exist between trials.  19 Figure 2. The same distribution of RTs depicted with a PDF and CDF.             20 Fz(t) may approach Fx(t) + Fy(t) and still support independent activation from each signal, but if Fz(t) exceeds Fx(t) + Fy(t) the RMI is violated in support of a model in which the activations from both signals are combined, prior to the response. For example, a violation of the RMI has been reported for multisensory processing, such that especially fast responses to multimodal stimuli (i.e., redundant signals presented to two sensory modalities) seem to arise from the multisensory integration of signals prior to a decision being reached, and not the statistical facilitation of independent activation from each signal (Miller, 1991). To determine whether collaborative cognition involves interpersonal interaction between team members, or whether it reflects the statistical facilitation of two people working independently, I substituted the comparison of one vs. two target-signals in previous studies with a comparison of performance by one vs. two people. In Chapters 2-5, I tested for violations of Miller’s RMI in the distributions of correct responses, using the algorithm and MATLAB routines provided in Ulrich, Miller, and Schröter (2007). To adapt the code to compare one vs. two people, I substituted the CDFs of correct RTs in the two single signal conditions, with the CDFs of RTs by the faster and slower of the two individuals in each team, respectively. I also substituted the CDF of RT in the redundant signals condition with the CDF of RTs by each team. To generate the model of statistical facilitation, the CDFs of RTs by the two independent individuals in each team were combined into a fourth CDF, which was truncated at the number of RTs in the two-person   21 team CDF, in order to provide a fair comparison with the total number of trials in the CDF for each team performance. This effectively eliminates the slower tail of the RT distribution generated by combining the responses from two independent individuals. This statistical facilitation benchmark therefore represents the fastest performance that can be derived from the two individuals working independently.  Typically, the RMI is rejected if the statistical facilitation benchmark is significantly violated (i.e., two-person team CDF is faster than the statistical facilitation CDF) at any CDF percentile (Ulrich et al., 2007). If performance by two-person teams does not exceed the statistical facilitation benchmark, the benefit of collaborative cognition merely reflects the statistical facilitation of independent individual performance. Alternatively, if team performance exceeds the statistical facilitation benchmark, the RMI is violated in support of social interaction among team members. Performance by two-people collaborating will therefore be greater than the sum of performance by two independent individuals. Previous research has grappled with the independent versus interactive benefits of collaborative cognition and resolved this in two primary ways (for a review see Hill, 1982). Some researchers opted to compare collaborative performance by two-person teams to the performance of the better of two individuals (e.g., Bahrami et al., 2010). Other researchers opted to compare two-person collaborative performance to the performance of nominal pairs,   22 which were created by selecting the faster of two individual responses on each trial (e.g., Brennan et al., 2008).  It is a useful exercise to consider how each of these two methods compare to Miller’s approach based on the RMI, which compares collaborative performance to the fastest possible performance by two independent individuals. Consider the following scenario. Two individuals and a two-person team performed four target detection trials (RTs are in msec):  Trial 1 Trial 2 Trial 3 Trial 4 Person A =  200 300 400 200  Person B =  300 200 400 500  Person A + B = X X X X  First considering the average performance of the faster of the two individuals as the correct baseline (i.e., Person A in this example), the baseline that a two-person team must beat to show collaboration is [(200 + 300 + 400 + 200)/4] = 275 msec. Collaborative RTs faster than 275 msec would indicate a collaborative performance benefit. Alternately, considering the performance of a nominal pair as the performance benchmark, the individual baseline is [200 (faster of Person A and B on trial 1) + 200 (faster of Person A and B on trial 2) + 400 + 200]/4 = 250 msec. In this case, collaborative RTs faster than 250 msec indicate a collaborative performance benefit. Now consider the benchmark given by the application of Miller’s RMI to the same situation. Here the fastest RT of either of the two   23 individuals on each trial is used, which is [(200 + 200 + 200 + 300)/4] = 225 msec. This estimate was derived using the following method: 1) Combine all RTs from Person A and Person B:  200 300 400 200 300 200 400 500 2) Order these RTs from fastest to slowest:  200 200 200 300 300 400 400 500 3) Select the first n RT’s, where n = the number of trials  performed by the two-person team, and truncate  thereafter:  200 200 200 300 Collaborative RTs faster than 225 msec would therefore be taken as evidence of a collaborative performance benefit, relative to the statistical facilitation benchmark of Miller’s RMI. The first point illustrated by these three examples, is that the performance benchmark for determining whether a team’s performance is ‘‘collaborative,’’ can vary a great deal, depending on the method used. It is therefore critical that an appropriate measure is selected. In this thesis, I advocate the use of Miller’s RMI. It is a more appropriate approach because the alternatives (i.e., the better of two individuals, nominal pairs), can lead to the conclusion that groups are collaborating when in fact their performances still fall within the bounds of a statistical facilitation model (i.e., the race model). In other words, these methods underestimate the independent benefits and therefore overestimate the interactive benefits of collaborative cognition. While there may indeed have been interactive benefits in previous studies of collaborative   24 cognition, I argue that that this claim has not yet been tested against an appropriate measure of individual performance. Overview of experiments In this thesis, I have newly applied Miller’s classic RMI model, originally used for the study of divided attention in a single individual, to the context of collaborative cognition. This tests whether two people working together in a team are truly interacting in their joint performance or whether they are simply working as the statistical sum of two independent people. I begin in Chapter 2 with a replication of Miller’s divided attention paradigm, measuring the performance of individual participants as they detect one versus two target signals. I then apply these same procedures to the analysis of two people working together to perform the same task. The results show that two people working together in a simple search task of this kind do not exceed the benchmark of statistical facilitation, as given by the RMI. In other words, although I find evidence for statistical facilitation in team target detection performance, I do not find evidence for interactive collaboration. In Chapter 3 I adopt an enumeration visual search task that is more cognitively complex than the simple visual search task used in Chapter 2. This task is chosen because it increases the opportunities for two people to communicate and collaborate for optimal team performance, although it does not necessitate interaction between team members. This task provides evidence for collaborative performance that exceeds the statistical facilitation model of the   25 strict RMI test. I therefore subsequently explore the association between verbal communication and the magnitude of the interactive benefit of collaboration in this study. Since Chapter 3 tests collaborative cognition in unaffiliated pairs of participants, in Chapter 4 I give the same enumeration visual search task to pairs of friends, (i.e., affiliated participants). These results show much stronger evidence for collaborative cognition, and the analyses of individual differences in the strength of the affiliation and in the style of communication provide hints as to what underlies these benefits.  In Chapter 5 I compare the combined effects of affiliation and communication by varying both factors orthogonally using an experiment with random assignment. That is, individual participants are randomly assigned to complete the enumeration search task with a friend or an unknown other, and both groups participate with either full opportunities for communication or with a partition that eliminates visible body language between team members. The results show that the strongest collaboration occurs when friends can communicate freely with their partners, with strangers and non-visible friends collaborating at similar lower levels. In addition, I measured physiological variables during these performances (heart rate, galvanic skin response) in an effort to determine how arousal contributes to collaborative performance gains.  Finally, in Chapter 6 I review the findings of Chapter 2 to 5, discussing their implications for the field of collaborative cognition.   26 Among the highlights of these conclusions are that Miller’s RMI, developed to test the independence versus interaction between two signals in the mind of an individual person, can be harnessed to test the independence versus interaction of performance by two people. Utilizing this model provides a method for calculating a more appropriate baseline measure of individual performance against which to compare collaborative performance than has been used in previous research on collaborative cognition.  This novel application of Miller’s RMI demonstrates that two people speed search because collaboration facilitates the division of the cognitive task load between collaborators, allowing two people to process together in parallel what individuals would process serially. Linking the interactive performance gains detected with Miller’s RMI to features of the social interaction, emphasizes the importance of studying human cognition in a social context.     27 2 Independence versus interaction  during simple visual search    Introduction Before I set out to apply Miller’s (1982, 2007) race model inequality (RMI) to test whether collaborative cognitive performance gains result from interpersonal interaction or the statistical facilitation of independent individual responses, I thought it wise to ground this investigation in a replication of Miller’s divided attention paradigm. As discussed in Chapter 1, it is a classic finding in divided attention that when participants are required to respond as quickly as possible to the onset of a target stimulus, they are faster to respond when two targets are presented than when only one is presented (e.g., Hershenson, 1962). This faster response time to redundant stimuli is termed the redundant signal effect (RSE). While Raab (1962) demonstrated how the RSE could result from the statistical facilitation of separate detection processes, Miller (1982) showed that responses to redundant stimuli are often faster than can be accounted for by this separate activation race model. These violations of the RMI show that the especially fast responses to redundant stimuli are the result of coactivation, i.e., the activation   28 from individual stimuli is combined by the system, prior to making a decision to respond.  In Chapter 2 I employed a version of the simple visual search task used by Miller (1982, Experiments 4 and 5). The visual displays had two target positions, one to the left and one to the right of centre. The target could occur in either position (single target), both positions (redundant targets), or neither position (target absent). On each trial, participants responded as quickly and accurately as possible whether either (i.e., single or redundant targets) or neither (i.e., target absent) target was present. The same visual display conditions were used in each of the three comparisons made in Chapter 2. Figure 3 provides a diagram of these three comparisons.    29 Figure 3. A diagram of the three comparisons made in Chapter 2. A) Replicating Miller (1982), the speed at which an individual person searched for one versus two targets was compared. B) Extending Miller (1982), the speed at which two-person teams who shared space or targets searched for one versus two targets was compared. C) Newly applying Miller (1982), the speed at which one versus two people searched for 0, 1, and 2 targets was compared.     30 As shown in Figure 3A, I first compared individual participant’s RTs to one vs. two targets to replicate the classic divided attention finding of a RSE. I predict that RTs to two targets would not only be faster than RTs to one target, but that they would be faster than can be accounted for by the race model, which is premised on separate activation of a response decision for each of two targets. Such a finding would replicate Miller’s (1982) classic finding of target coactivation (i.e., individual RTs to two targets are especially fast because activation from targets is combined prior to a decision) and prefaces the subsequent comparisons I make in Chapter 2 based on team performances.  As illustrated in Figure 3B, I next extended Miller’s RMI to investigate whether the RSE (i.e., faster RTs to two targets than one) occurs when targets are processed by two people working together instead of an individual person. This investigation was prompted by the ‘‘astonishing’’ (Pollmann & Zaidel, 1999, pg. 246) and ‘‘paradoxical’’ (Corballis, 1998, pg. 1795) finding that the RSE was larger in individuals without a functioning corpus callosum compared to individuals in whom this structure functioned normally (Miller, 2004; Pollmann & Zaidel, 1999; Reuter-Lorenz, Nozawa, Gazzaniga, & Hughes, 1995). This exaggerated RSE occurred only when redundant visual stimuli were presented bilaterally (i.e., to different visual hemifields and thus different hemispheres), and not when redundant stimuli were presented to the same visual hemifield (i.e., same hemisphere). This finding of unusually large RSEs and RMI violations for split-brain individuals is surprising in light of the   31 processing architecture of coactivation models, which propose that redundancy speeds RT because activation from the two redundant stimuli is combined. Therefore such effects should be reduced or eliminated when information combination is prevented i.e., when split-brain individuals process bilateral redundant stimuli. I further extend this finding of larger RSEs in individuals with two partially or completely disconnected hemispheres, asking whether RSEs and RMI violations are also larger when two individuals process redundant targets. Models developed to account for the enhanced redundancy gain in individuals without a functioning corpus callosum share the common feature that both hemispheres contribute activation or inhibition to information processing despite being functionally severed. In the simple search task I employed in Chapter 2, I can think of no way that target inhibition or activation could be shared between the two collaborating individuals. As shown in Figure 3, two-person teams shared the detection task with respect to space (each inspecting only one location) or target (each preparing for one target). Because of this division, each individual accumulated activation from only a single target during redundant target detection. While these division-of-labor strategies do not allow target activations to be combined, it should make detection of both single and redundant targets more efficient. Research has shown that object detection is limited to one target at a time (Houtkamp & Roelfsema, 2009) and therefore dividing the task between individuals should permit teams to process together what they   32 would process serially as individuals. With consideration of these features of the simple detection task, I posit that two-person teams will be facilitated for both single and redundant target trials. Because this outcome produces no net change in the magnitude of the redundancy gain, I predict that two-person teams will respond no faster to two targets than one. While I do not predict a RSE during simple visual search by two-person teams, I propose that the overall performance of two-person teams may exceed the performance of two individuals working independently. Previous research has shown response facilitation by two people acting together, because each represents the other person’s task-related actions similar to how they represent their own (Sebanz, Knoblich, & Prinz, 2003; 2005). I applied Miller’s RMI to determine whether such collaborative cognition performance gains are produced by interpersonal interaction between team members (i.e., a sort of ‘social coactivation’) or whether they simply reflect the statistical facilitation of the performance of two people working independently. This is a critical distinction to make when investigating the performance gains of collaborative cognition because it differentiates between the statistical improvement that is possible from two independent heads compared to only one versus the interactive performance gains that emerge through interpersonal collaboration.  As illustrated in Figure 3C, I tested this by substituting the comparison of one vs. two target-signals in previous studies with a   33 comparison of performance by one vs. two people. If performance by a two-person team does not exceed the model of statistical facilitation derived from treating each individual’s performance as an independent signal, the benefit of collaborative cognition merely reflects statistical facilitation. Alternatively, if team performance exceeds the statistical facilitation benchmark, this shows that the performance of teams is greater than the sum of individual team member performance, and implies that the benefits of collaborative cognition involve social interactions among individuals.  Method Participants. Forty-two University of British Columbia students (24 female; age mean = 19.95 years) participated in exchange for course credit. Participants were randomly assigned to participate in the experiment with another person, which created 21 participant pairs who were unknown to one another at the start of the experiment. RSEs resulting from coactivation have been reported with similar numbers of participants as tested in this experiment. Data from one participant pair was excluded due to equipment failure; analyses were conducted on the remaining 40 individuals (20 pairs). All participants provided written informed consent and were debriefed in accordance with APA guidelines. I report all measures I collected in this experiment. Stimuli & Apparatus. As shown in Figure 3, displays depicted photographs of two common objects at the left and right of centre against a white background. 0, 1, or 2 targets (toy bear, coffee cup)   34 and 2, 1, or 0 distractors (e.g., calculator, hammer) appeared in each display, respectively. The same target or distractor never appeared twice in a display. Each target appeared equally often in the left and right position; distractor identity was randomized. Photographs of targets and distractors were selected from a set of non-licensed, publically accessible stock images. Photographs were cropped to be uniform in size, each occupying approximately 250 by 250 pixels (8 degrees of visual angle in diameter). The location at which photographs appeared was randomly jittered on each trial from 0-60 pixels in either the x or y dimension. This made the total distance between images range from 210 pixels (7 degrees of visual angle) to 450 pixels (15 degrees of visual angle).  The experiment was conducted on a 24-in iMac computer (1920 pixels X 1200 pixels screen resolution) using the Python-based OpenSesame experiment building software (Mathôt, Schreij, & Theeuwes, 2012). Two keyboards, one for each participant, were connected to one computer during collaborative visual search; only the first key press on each trial was recorded. Procedure. Participants indicated as rapidly and accurately as possible whether either or neither target appeared on the screen using keys marked with ‘present’ or ‘absent.’ On each trial, participants first viewed a central fixation cross for 650 msec, followed by a two-object display. The experiment automatically progressed to the next trial after 3000 msec if participants failed to enter a response within this time. Participants completed 300 trials:   35 100 each with two targets, one target and one distractor, and two distractors. Trials were grouped into 5 blocks of 60 trials (15 blocks of 60 trials in total) and trial order was randomized and counterbalanced within blocks. Accuracy was displayed on the screen after each block.  Participants completed these 300 trials in each of the three social conditions (900 trials total): (1) as two independent individuals, (2) as members of a two-person team dividing the task by spatial location (one person detecting targets on the left, the other person detecting targets on the right), and (3) as members of a two-person team dividing the task by target identity (one person detecting the toy bear, the other person detecting the coffee cup; see Figure 3 for a diagram of these three social conditions). A research assistant explained the instructions of these three conditions to participants and the test order of the social conditions was counterbalanced across participants. Individuals completed the task on separate computers in the same room. Two-person teams completed the task seated side-by-side at the same computer with two keyboards. Teams were informed that only the first response on each trial was recorded and they should communicate to respond as rapidly and accurately as possible.  Results A) One vs. two targets by individuals. Coactivation, not statistical facilitation, underlies individuals’ faster responses to two targets than one. In accordance with the classic finding of divided attention (e.g.,   36 Hershenson, 1962) and shown in Table 1, I found that individuals detected two targets faster and more accurately than one target.  Table 1.  Mean RT (msec) and accuracy (%) with SE in parentheses for individuals detecting one versus two targets.  RT (msec)  Accuracy (%) One (bear) 561.49 (12.24) 95.14 (0.84) One (coffee) 575.44 (11.15) 95.05 (0.99) Two 496.01 (12.41) 98.17 (0.80) Replicating Miller (1982, 2007), I found that responses to two targets were faster than the separate activation race model could account for (see Figure 4). This demonstrates that coactivation between target signals, not the statistical facilitation of independent signal activation, underlies the faster processing of two targets compared to one by an individual. This finding is supported by the following analyses.    37 Figure 4. CDFs of correct RT during search by individuals for one versus two targets. Faster responses are to the left, slower responses to the right. Replicating Miller (1982), two targets together (red solid line) were detected faster than individual targets (blue and green solid lines) and the statistical facilitation model (dashed black line).    38 I used a repeated measures analysis of variance (ANOVA) to test for differences in correct RT between target conditions (bear, coffee, bear and coffee). This analysis in combination with Tukey’s HSD follow-up tests revealed that two targets were detected faster than one target [F(2, 80) = 60.28, p < .001], such that when the bear and coffee appeared together, they were detected faster than when either the bear [p = .01] or coffee [p < .01] appeared alone. A second ANOVA testing for differences in accuracy showed that two targets were detected with higher accuracy than one target [F(2, 80) = 15.27, p < .001]. Again, this was true when either the bear [p = .043] or coffee [p = .035] appeared alone compared to when they appeared together. Given the finding of faster search for two targets than one, I next investigated whether the separate activation race model was sufficient to account for this performance improvement. I tested for violations of Miller’s RMI using the MATLAB routines provided in Ulrich et al., 2007. This script generated the model of statistical facilitation by aggregating the cumulative density functions (CDFs) of correct response times (RTs) in each of the single-signal conditions (blue and green solid lines). This pool of single-signal correct RTs was truncated at the equivalent number of correct RTs in the two-signal CDF (red solid line), thus eliminating the slower tail of the entire pooled RT distribution. This statistical facilitation benchmark (dashed black line) represented the fastest two-signal RTs possible with separate activation from each signal. Miller’s RMI is violated if the two-signal CDF (red solid line) is significantly faster   39 than the single-signal statistical facilitation benchmark (dashed black line) at any CDF percentile. This indicates that coactivation, not the statistical facilitation of independent signal activation underlies the RSE. Alternatively if the two-signal CDF (red solid line) is no faster than this benchmark, then the separate activation race model provides a sufficient account. Typically, the RMI is rejected if there is a significant violation at any percentile (Ulrich et al., 2007). Using Bonferroni-corrected paired sample t-tests at the 10 CDF percentiles, I found that Miller’s RMI was violated significantly at the .05 percentile [t(20) = 1.96, p = .03]. This violation of the RMI indicates that the faster RTs to two targets than one result from coactivation, not the statistical facilitation of separate target activation. Miller’s RMI was not significantly violated at the .15 to .95 percentiles [all p’s > .08].  B) One versus two targets by teams. Teams respond no faster to two targets than one. I used mixed-design ANOVA with target condition (bear, coffee, bear and coffee shared space, bear and coffee shared target) as a repeated measures factor, and test order (individual before collaborative, collaborative before individual) as a between groups factor. Tukey’s HSD follow-up tests were conducted on significant main effects with more than two levels and interactions. A first ANOVA testing for differences in correct RT revealed a main effect of target condition [F(3, 54) = 20.20, p < .001], such that the bear and coffee together were found by teams sharing targets faster than the coffee alone [p = .05]. However, teams sharing targets were   40 no faster at detecting the bear and coffee together than the bear alone, and teams sharing space were no faster at detecting the bear and coffee together than either the bear or coffee alone [all p’s = NS]. A second ANOVA testing for differences in accuracy showed that two targets were detected with higher accuracy than one target [F(3, 54) = 8.95, p < .001], such that the bear and coffee together were found by teams sharing space more accurately than the bear [p = .02] or coffee [p = .01] alone. Together these findings demonstrate that two-person teams who shared space or targets failed to detect two targets together faster than either target alone. Because there was no RSE during simple visual search by teams, there was no reason to use Miller’s RMI to explore whether the RSE resulted from coactivation or independent target activation. See Table 2 for the mean RT and accuracy of shared space and shared target teams detecting one versus two targets.  Table 2.  Mean RT (msec) and accuracy (%) with SE in parentheses for teams detecting one versus two targets.  RT (msec)  Accuracy (%) One (bear) 517.33 (23.83) 90.34 (2.68) One (coffee) 540.73 (26.79) 89.84 (3.83) Two (shared space) 468.45 (17.04) 97.15 (0.92) Two (shared target) 459.38 (16.79) 99.19 (0.31)    41 C) One versus two people. Simple search by two-person teams was no faster than search by two independent individuals. I used a mixed-design ANOVA with social condition (slower individual, faster individual, shared space team, shared target team) and target number (0, 1, 2) as repeated measures factors, and test order (individual before collaborative, collaborative before individual) as a between groups factor. Tukey’s HSD follow-up tests were conducted on significant main effects with more than two levels and interactions. A first ANOVA testing for differences in correct RT revealed main effects of social condition [F(3, 54) = 8.20, p < .001] and target number [F(2, 36) = 74.45, p < .001]. Follow-up tests determined that detection by the slower independent individual was slower than detection by the faster independent individual [p < .001], and by two person collaborative teams who shared targets [p = .01], and space [p < .001]. Also that 2 targets were detected faster than 1 target [p = < .001] or 0 targets [p = < .001], and 1 target faster than 0 targets [p = .001]. There was no significant interaction between social condition and target number, or main effect or interaction involving test order [ps > .40].  A second ANOVA testing for differences in accuracy showed a main effect of target number [F(2, 36) = 3.59, p = .04], such that 2 targets were detected more accurately than 1 target [p < .001] and 0 targets [p = .02]. This analysis also revealed a significant interaction between target number and social condition [F(6, 108) = 3.36, p < .01], such that 1 target detection accuracy was lower by shared space teams than either the slower and faster individual or shared target teams,   42 but there were no differences in 0 or 2 target detection accuracy between the levels of social condition.  No other main effects or interactions were significant  [ps > .20]. Taken together, these findings show that detection performance by two-person collaborative teams did not exceed the performance of the same two independent individuals. Had team performance exceeded independent individual performance, I would have utilized Miller’s RMI to test whether the performance gain resulted from interaction or the statistical facilitation of independent individual responses. However given that team performance did not exceed individual performance, there was no reason to conduct the RMI analyses. See Table 3 for mean RT and accuracy values.   43 Table 3. Mean RT (msec) and accuracy (%) with SE in parentheses for shared space and shared target teams, by the numbers of targets.   RT (msec)  Accuracy (%) 0 target Slower individual 665.12 (23.24) 90.32 (5.08) Faster individual 572.03 (12.37) 89.77 (5.04) Collaborative (shared space) 618.35 (19.16) 89.14 (5.29) Collaborative (shared target)  592.68 (13.19) 91.36 (5.11) 1 target Slower individual 631.12 (16.61) 90.86 (4.55) Faster individual 504.01 (10.98) 89.50 (4.50) Collaborative (shared space) 542.75 (24.48) 81.90 (5.48) Collaborative (shared target)  513.26 (15.26) 88.27 (4.90) 2 targets Slower individual 544.66 (20.16) 94.32 (4.70) Faster individual 458.48 (12.56) 94.18 (4.67) Collaborative (shared space) 468.45 (17.04) 92.62 (4.72) Collaborative (shared target)  456.76 (16.77) 94.64 (4.62)   44 Discussion  In Chapter 2 I first replicated the classic finding of an RSE, such that individuals responded faster to two targets than one target during a simple visual detection task. Using Miller’s (1982, 2007) RMI I showed that this reported RSE was the result of coactivation (i.e., interaction) between the target signals, not the statistical facilitation of independent target activation. In a second comparison I extended Miller’s RMI to investigate whether two-person teams respond faster to two targets than one target during the same simple visual search task. As predicted, teams sharing the task with respect to space (each inspecting only one location) or target (each prepared for only one target) responded no faster to two targets than one. This suggests that there must be redundancy in the mind of an individual for a RSE to occur. Because teams shared the detection task, each individual accumulated activation from only a single target during redundant target detection.  In a third comparison in Chapter 2, I also applied Miller’s RMI to determine whether collaborative cognition performance gains result from interpersonal interaction between team members or whether they simply reflect the statistical facilitation of the performance of two people working independently. While I predicted that the overall performance of two-person teams would exceed the performance of two independent individuals, I found that team performance was no faster than the performance of the faster of two independent individuals. Had team performance exceeded independent individual performance, I would have utilized Miller’s   45 RMI to test whether performance gains resulted from independent or interactive effects of collaboration. However, I had no reason to conduct the RMI analyses because team performance did not exceed individual performance. I propose that performance by two-person teams failed to exceed independent individual performance because individuals could complete the simple visual detection task efficiently on their own, with no need for additional information from their partner. As a result, there was very little opportunity for collaboration to improve performance. I consider this result analogous to role of set size in visual search performance. During highly efficient ‘‘pop-out’’ search, reducing set size does not speed performance, while reducing the set size during less efficient serial search (i.e., search requiring attention) speeds performance (see Wolfe, 1998 for a review). Previous research demonstrating collaborative benefits during visual search have shown that these benefits arise because teams divide the visual displays, for example, each team member searching for the target in one half of the display (Brennan, Chen, Dickinson, Neider, & Zelinsky, 2008; Chen, 2007; Neider, Chen, Dickinson, Brennan, & Zelinsky, 2010). This sort of division by teams effectively reduces the set size by half, and therefore speeds performance by teams relative to individuals. Because the displays in Chapter 2 contained only two items and were therefore processed very efficiently already, division of the task between team members did not speed collaborative performance.    46 It should be noted that although individual performance was efficient, it was not so efficient that the addition of a second redundant target failed to speed responses and produce a RSE. On the contrary, in Chapter 2, I replicated the classic finding of an RSE and used Miller’s RMI to show that it resulted from interaction between the target signals, not the statistical facilitation of independent target activation. I therefore propose that redundant targets and collaborative detection increase search efficiency through different mechanisms. As such, an increase in detection efficiency from one factor does not necessitate that there will be a proportionate increase in efficiency as the result of the other factor.  Redundant targets speed detection because both targets contribute activation to a common pool of evidence used in coming to a decision (i.e., coactivation), or because of the statistical facilitation of independent detection responses to two targets. On the other hand, two people speed detection because they divide the cognitive task load between them, together processing in parallel what would be processed serially as individuals. Thus, while a second target contributed additional activation to the common pool and speeded individual responses, the task constraints of simple target detection did not permit the division of the task load necessary for a second person to speed performance. Considering this limitation in the applicability of Miller’s (1982) original simple search task to the realm of collaborative cognition, I introduce an enumeration visual search task in Chapter 3.  This task is sufficiently cognitively complex that it allows for the possibility of collaborative benefits without   47 artificially necessitating social interaction or preventing social interaction from occurring. I will continue the investigation of the independent and interactive effects of collaborative cognition with this task in Chapter 3.    48 3 Independence versus interaction  during enumeration visual search by  unfamiliar pairs    Introduction Using a simple visual search task in Chapter 2, I found that team performance was no faster than the performance of the faster of two independent individuals. I posit that performance by two-person teams failed to exceed independent individual performance because individuals were able to complete the simple detection task efficiently on their own. At the same time, I propose that two people are more efficient in some tasks, such as the more complex visual search task of Brennan et al. (2008), because in such tasks collaboration allows the division of the cognitive task load between collaborators, allowing two-people to process in parallel what individuals would effectively have to process serially. From this perspective, the simple search task of Chapter 2 did not permit such a division of the task load that would permit a second person to speed performance.    49 In Chapter 3 I chose to study a more complex enumeration visual search task in an effort to distinguish the effects of independent statistical facilitation from the potential interactive benefits of collaborative cognition. This task required the enumeration of a set of four potential target objects against a busy and changing background of distractor objects. This task required (1) search skill to identify each member of the target set when it was encountered, and (2) enumeration skill to track how many targets had been encountered in each display (e.g., Trick & Pylyshyn, 1994). I selected this task for several reasons. First, visual search is one of the most extensively studied paradigms in cognitive psychology (Wolfe, 1998) so the results can be linked to a rich background of data and theory. Second, successful visual enumeration is limited primarily by speed rather than accuracy so that individuals and pairs can perform the task with equivalently with high accuracy. The question is whether teamwork will speed performance. Third, this task generates many correct RTs for each condition, which is required for applying Miller’s RMI model to any set of data.  Fourth and most important, considering the findings of Chapter 2, enumeration visual search is sufficiently cognitively complex that it allows for the possibility of collaborative benefits (i.e., if communication between individuals adds information and speeds performance) without artificially necessitating social interaction (i.e., if social interaction were essential for successful performance) or preventing social interaction (e.g., if the task could be completed so quickly that social interaction was superfluous). Visual search lends   50 itself well to collaboration because individual attention has a limited capacity for representing visual space and search targets in working memory (Houtkamp & Roelfsema, 2009). Social collaboration has the potential to offset the limited capacity of individual attention, allowing collaborators to divide the cognitive task load. This could allow a process that must be executed serially when done by an individual to be executed more efficiently by a pair of individuals working in parallel.  Research has shown that collaborators employ different strategies to divide the cognitive load depending on the demands and constraints of the task. Pairs who searched for a visual target that differed from distractors by a single feature divided the display spatially, each person searching roughly half the items (Brennan et al., 2008). When searching for multiple targets, pairs adopted the strategy of dividing the task by target identity, each searching for half of the possible targets (Chen, 2007). Collaborators who were unable to communicate with one another did not divide the task (Brennan et al., 2008). If I find that interpersonal interaction underlies the collaborative benefit of enumeration visual search, an important next step is to consider the mechanisms that facilitate successful joint attention on this cognitively demanding task. I posit that verbal communication supports the exchange of information between collaborators. Participants in the collaborative portion of this experiment sat side by side while searching for multiple visual targets on a computer   51 screen. Because the task necessitated that they looked at the computer screen and not each other, I hypothesize that participants used verbal communication to exchange information during the collaborative enumeration visual search task.  Previous research suggests that the content and quantity of verbal communication contributes to the performance benefit of collaboration. For example, verbal communication, but not objective accuracy feedback, was the necessary condition for collaborative performance to exceed the performance of the better pair member working alone in a low-level perceptual detection and decision-making task (Bahrami, Olsen, Bang, Roepstorff, Rees, & Frith, 2012). In a detailed analysis of verbal communication, Fusaroli, Bahrami, Olsen, Roepstorff, Rees, Frith, & Tylen (2012) found that pairs showed a high propensity to adopt each other’s language and that the alignment of particular task-relevant vocabularies communicating the degree of confidence in what they saw strongly correlated with the magnitude of collaborative performance gains. In their communication analysis they assessed both local alignment (the transition probability that a given lexical expression of a participant would be a repetition of an expression used by the other participant in the previous interaction trial) and global convergence (the degree to which participants in a dyad converged on a limited functional set of shared expressions rather than drifting between multiple sets of expressions). They also considered both discriminate (only dealing with the expressions of confidence directly relevant to the experimental task) and indiscriminate (all lexical items in a dyad’s   52 transcript) linguistic alignment, finding that discriminate alignment was associated with collaborative performance gains, whereas indiscriminate alignment was associated with reduced collaborative relative to individual performance. These associations between linguistic alignment and the magnitude of collaborative performance gains were found at both the level of local alignment and global convergence. Although often beneficial, research has also shown that verbal communication can, in certain circumstances, be detrimental to collaborative performance. For example, in the study by Wu & Keysar (2007), performance improved for pairs who communicated using mutually shared information, whereas pairs who communicated using information that was known by only one individual showed reduced collaborative relative to individual performance. In a similar vein, collaborative visual search was faster when completed by pairs who could see where their partner was looking (by way of a gaze cursor that displayed their partner’s eye movements) than by pairs with the ability to verbally communicate with their partner in addition to viewing their partner’s eye movements (Brennan et al., 2008; Chen, 2007; Neider et al., 2010).  In Chapter 3, I implemented two verbal communication measures: 1) total number of distinct utterances made by each team, and 2) similarity in the number of distinct utterances made by one team member relative to the other team member. I explored the frequency of verbal communication because the majority of   53 utterances made by teams performing enumeration visual search communicated the identity or location of targets. Thus the content of verbal communication was similar from one trial to the next, as well as between different collaborative teams. I hypothesize that similarity of number of utterances between team members will be associated with collaborative performance benefits because it relates to division of cognitive task load. Because verbal communication passed information between team members about the identity or location of targets that had been located by one member, members who spoke at a similar frequency presumably located similar numbers of targets. A more equal division of the cognitive task load should result in larger collaborative performance gains compared to teams where one member’s performance exceeded the other’s. This notion is consistent with Bahrami et al.’s (2010) report that collaborators with more equal visual sensitivities demonstrated larger collaborative performance gains than collaborators with disparate visual sensitivities. Method  Participants. Forty-eight University of British Columbia students (37 female; age mean = 23.39) were recruited with advertisements asking them to participate in research with a friend (24 pairs of individuals). Participants were randomly assigned to participate with another person in this group (excluding their own friend) to ensure that pairs were unknown to each other prior to participation. The research assistant confirmed this at the time of the experiment; no   54 pairs reported knowing one another prior to participation. Participants were paid $10 per hour. They provided written informed consent and were debriefed in accordance with APA guidelines. Because one pair did not complete the collaborative portion of the experiment analyses were conducted with 46 individuals (23 pairs of individuals). RSEs resulting from coactivation have been reported with similar numbers of participants as tested in this experiment. I report all measures I collected in this experiment. Stimuli & Apparatus. As shown in Figure 5, the experimental displays depicted shelves containing 82 distractor objects commonly found in a home or office and 0, 1, or 2 of 4 possible targets. Distractors appeared in four different arrangements. The same target never appeared twice in a photograph. Each target appeared equally often in each display quadrant in one target displays and in one of the quadrants not already occupied by the first target in two target displays. This generated 356 search displays in total (4 0-target + 64 1-target, + 288 2-target displays). Displays subtended 40° x 32° visual angle on a 24-inch iMac computer. The experiment was controlled by Matlab 2010a software and Psychtoolbox3. An iSight webcam recorded participants’ head and torso during the collaborative task.   55 Figure 5. A typical enumeration visual search display (top) and the four targets that could appear in each display (bottom).  Participants searched displays and indicated whether 0-, 1- or 2-targets were present. This display contains 2 targets: coffee and penguin.  This task was used in Chapters 3 through 5.  Procedure. Participants indicated as rapidly and accurately as possible the number of targets present. Before testing, they were familiarized with the displays and the four target objects, which remained continuously visible in four pictures placed underneath   56 the computer screen. The reason initially given to participants for video-recording the session was for the purpose of ‘‘knowing where your eyes are looking when you search.’’ Upon debriefing, participants were informed that the real reason was ‘‘to be able to measure the verbal and nonverbal behavior associated with your performance as a team.’’ Participants were then given the option of declining to have their video data be used in the study, but all agreed. Half of the participants first performed the task individually before being tested as a team; the other half performed as a team before being tested individually. In each case, participants searched a subset of all of possible displays, and indicated the number of targets present by pressing keys labeled 0, 1, or 2. Participants completed 90 trials as individuals (30 each with 0, 1, and 2 targets) and 90 trials as a team. I used weighted random sampling of the 356 total search displays to ensure that 0, 1, and 2 target trials appeared with equal frequency. Participants received feedback on their percentage of correct responses every 15 trials. When searching as a team, participants were instructed to use whatever strategy they thought was best to work together. Because there was only one keyboard for response entry, each participant took a turn entering the response, with the keyboard being exchanged after 45 trials.    57 Teamwork Quality Measure. Following completion of the enumeration visual search task, participants completed a questionnaire designed for this experiment to measure how well they worked with their partner as a team. It consisted of 14 questions such as ‘‘My partner and I communicated effectively to complete the search task’’ and ‘‘I enjoyed working with my partner.’’ Responses were made using a 5-point Likert scale (strongly agree to strongly disagree). See Appendix A for the full Teamwork Quality questionnaire. A teamwork quality score for each pair was computed by summing individual participant scores. Communication Frequency Measure. A research assistant blind to the purpose of the study made a written transcript of all verbal communication during the team task. See Appendix B for a representative sample of verbal communication during ten trials of collaborative enumeration visual search. I then calculated the total number of distinct utterances made by each team member. Summing the individual utterance counts of each team member, I created a Communication Frequency measure for each pair. The smaller this number, the less pairs spoke during the collaborative visual enumeration task.  Communication Similarity Measure. A measure of Communication Similarity was created by subtracting the number of utterances made by the less talkative member of each team from the number of utterances made by the more talkative member, and dividing this number by the Communication Frequency (i.e., the total number of   58 team utterances). The smaller this score, the more similar the relative contribution of each team member was to communication during the visual enumeration task. Analysis of independent versus interactive performance benefits. I tested for violations of Miller’s RMI in the distributions of the correct responses, using the algorithm and MATLAB routines provided in Ulrich et al., 2007. CDFs of correct RTs in the two single-signal conditions from Ulrich et al. (2007) were replaced with the CDFs of RTs by the faster and slower of the two individuals in each team, respectively. The CDF of RT in the redundant signals condition of Ulrich et al. (2007) was also substituted with the CDF of RTs by each team. Each of these three CDFs therefore contained a total of ninety correct RTs, less the small number errors of errors that were committed. To generate the model of statistical facilitation, the CDFs of RTs by the two independent individuals in each team were combined into a fourth CDF, which was truncated at the number of RTs in the two-person team CDF to eliminate the slower tail of the RT distribution. This statistical facilitation benchmark therefore represented the fastest performance that could be derived from the two individuals working independently. Miller’s RMI will be violated, and will therefore compel the interpretation of social interaction, if two-person teams are significantly faster than the statistical facilitation benchmark at any CDF percentile. Alternatively if two-person teams are no faster than this benchmark, then the improved performance of two searchers compared to one can be interpreted   59 as statistical facilitation. Typically, the RMI is rejected if there is a significant violation at any percentile (Ulrich et al., 2007). Results Collaborative performance exceeded the performance of either individual considered alone. Consistent with previous research on collaborative visual search, two people working together on the new visual enumeration search task were faster and more accurate than either individual working independently. Mean correct RT and accuracy are shown in Tables 4 and 5, respectively.    60 Table 4.  Mean correct RT (sec) with SE in parentheses for unaffiliated pairs in Chapter 3 by the factors: test order (individual before collaborative, collaborative before individual), number of targets in the display (0, 1, 2), and social condition (slower individual, faster individual, collaborative).   individual  collaborative Test order  collaborative  individual 0 target Slower individual 20.06 (2.25) 13.55 (0.63) Faster individual 14.57 (1.46) 11.18 (0.79) Collaborative  8.35 (0.54) 10.78 (0.93) 1 target Slower individual 19.92 (3.35) 13.13 (0.95) Faster individual 13.64 (1.33) 10.30 (0.70) Collaborative  8.44 (0.59) 10.89 (0.86) 2 targets Slower individual 12.76 (2.02) 7.72 (0.98) Faster individual 8.33 (1.39) 5.47 (0.60) Collaborative  5.15 (0.66) 7.36 (1.06)   61 Table 5.  Mean accuracy (%) with SE in parentheses for unaffiliated pairs in Chapter 3 by the factors: test order (individual before collaborative, collaborative before individual), number of targets in the display (0, 1, 2), and social condition (slower individual, faster individual, collaborative).   individual  collaborative Test order  collaborative  individual 0 target Slower individual 97.58 (1.11) 98.33 (1.39) Faster individual 96.06 (1.79) 99.17 (0.60) Collaborative  100.00 (0.00) 99.09 (0.62) 1 target Slower individual 89.39 (2.50) 88.89 (2.15) Faster individual 86.97 (2.82) 91.11 (2.07) Collaborative  93.33 (1.49) 90.83 (2.10) 2 targets Slower individual 82.12 (2.92) 79.44 (3.45) Faster individual 77.27 (5.52) 79.44 (3.12) Collaborative  88.48 (2.86) 85.00 (2.83)   62 Correct responses were made in an average of 11.14 sec (SD = 5.02). I used a mixed-design ANOVA to test for differences in correct RT, with social condition (slower individual, faster individual, collaborative) and target number (0, 1, 2) as repeated measures factors, and test order (individual before collaborative, collaborative before individual) as a between groups factor. Tukey’s HSD follow-up tests were conducted on main effects with more than two levels and interactions. This analysis revealed that team enumeration by unfamiliar pairs was faster than enumeration by either the slower or faster individual alone [F(2, 42) = 16.26, p < .01]. There was no difference in RT between the slower and faster individual member of each team [p = .88]. It also showed a main effect of target number [F(2, 42) = 78.09, p < .01], such that enumerating 2 targets was faster than enumerating 1 target [p < .01] and 0 targets [p < .01], while there was no difference in RT between enumerating 1 and 0 targets [p = .87]. There was also a main effect of test order: enumeration was faster overall when collaborative enumeration preceded individual enumeration compared to when individual enumeration preceded collaborative enumeration [F(1, 21) = 6.37, p < .01]. This analysis revealed significant 2-way interactions. Team performance maximally exceeded individual performance during enumeration of 0 and 1 targets [F(4, 84) = 9.50, p < .01]. Team performance also exceeded individual performance during enumeration of 2 targets, but to a lesser extent. Team performance exceeded performance by both the slower and faster individual when individuals were tested   63 before teams [F(2, 42) = 14.08, p < .01], while there was no performance difference between individuals and teams when teams were tested before individuals. Response accuracy was high at 90.14% (SD = 10.84). A mixed-design ANOVA with the same factors as correct RT revealed that enumeration accuracy declined as target number increased [F(2, 42) = 59.89, p < .01]. Zero target accuracy (98.41%) was higher than both 1 target (90.10%) and 2 target (81.93%) accuracy and 1 target accuracy was also higher than 2 target accuracy [p’s < .01]. Additionally, there was a main effect of social condition [F(2, 42) = 4.32, p < .01], such that team enumeration was more accurate than enumeration by the faster individual [p < .05], but there was no accuracy difference between the slower and faster individual [p = .88] or the slower individual and the team [p = .14]. There were no significant interactions between the factors of this analysis [all p’s > .19]. Together these findings revealed a speed-accuracy tradeoff such that participants made relatively rapid responses to 2 targets compared to 0 and 1 targets, but at the expense of accuracy. Because only correct RTs are considered in Miller’s RMI, I conducted the following analyses on RTs in the 0- and 1-target conditions where accuracy was highest. Test order effect. The fore-described analysis of correct RT demonstrates that the magnitude of collaborative performance   64 gains relative to independent individual performance was influenced by the test order (individual before collaborative, collaborative before individual) on.  Team performance was faster than individual performance when individuals were tested before teams, while there was no performance difference between individuals and teams when teams were tested before individuals.  Given that cognitive task performance generally improves with practice, especially when the task is new, it is not surprising that performance in the second social condition was improved relative to the first social condition.  Thus when collaborative search preceded individual search, the practice effect overshadowed the collaborative performance gains and when individual search preceded collaborative search the practice effect enhanced the collaborative performance gains.  Because test order was counterbalanced between participant pairs, I averaged the magnitude of collaborative performance gains across pairs of participants in the following analysis of the independence versus interaction of collaborative performance gains using Miller’s RMI. This approach is consistent with previous research involving test order effects in repeated measures designs (e.g., Wolfe, 1998). In the subsequent analysis of the association between the magnitude of collaborative performance gains and the features of the social interaction including Teamwork Quality, Communication Frequency, and Communication Similarity, I assert that the absolute measure of coactivation is not the important thing.  Because the absolute measure of coactivation varied as a function of test order, it   65 is not necessarily a reliable measure of the magnitude of collaborative performance gains for any given collaborative pair.  However, how the relative measure of collaborative performance gains varied with the factors of interest (e.g., the relative association between Teamwork Quality and coactivation across all collaborative pairs randomly assigned to complete the task in one of two test orders) provides important insight into the features of the social interaction that are associated with collaborative performance benefits.   The performance benefit of teamwork was the result of social interaction, not statistical facilitation. As shown in Figure 6, team performance by unfamiliar pairs surpassed the benchmark of the statistical facilitation model that denotes the fastest possible performance of two independent individuals. This violation of the RMI indicates that the benefit of collaboration results from social interaction and is not merely the statistical facilitation expected from independent individuals. This conclusion was supported by Bonferroni-corrected paired sample t-tests at the 10 CDF percentiles. Miller’s RMI was violated significantly at percentiles .15 through .55 (t(22) = 2.09, 2.11, 2.20, 2.24, and 2.10, respectively, all p’s < .05). This finding that social interaction underlies the benefit of collaborative cognition prefaces the following analysis of the social factors associated with effective team collaboration.   66 Figure 6. Mean cumulative density functions for correct RT by unaffiliated pairs in Chapter 3, averaged over 0- and 1-targets.  Faster responses are to the left, slower responses to the right.  Two-person teams (red solid line) were not only faster than the faster of two individuals working alone (blue solid line), their performance exceeded the prediction of statistical facilitation if the same two individuals worked independently (dashed black line).     67 Communication frequency is negatively associated with the benefit gained by collaborating with an unfamiliar other. The magnitude of the collaborative benefit for each team was calculated as the mean correct RT of the statistical facilitation model less the collaborative mean correct RT for each of the 10 CDF percentile bins. These values were entered as criterion variables to be regressed against the measures of Teamwork Quality, Communication Frequency, and Communication Similarity. As shown in Figure 7, I found that Communication Frequency [r = -.31, p = .03] was negatively associated with the magnitude of the collaborative benefit averaged across percentiles .15 through .55, where Miller’s RMI was significantly violated. This suggests that unfamiliar pairs who communicated more while working together in the team task benefitted less from their collaboration.   68 Figure 7. A scatter plot of the negative association between the frequency at which unfamiliar pairs verbally communicated and the interactive benefit gained by collaborating with another person. Unaffiliated pairs who verbally communicated more benefitted less from their collaboration.   69 Communication similarity associated with teamwork quality, but not the benefit of collaborating with an unfamiliar other. Participant-reported Teamwork Quality was correlated with Communication Similarity, such that pairs who reported working better together as a team communicated at a more similar frequency [r = .33, p = .02]. Despite the confirmation of this association, the results showed that both Teamwork Quality and Communication Similarity were unrelated to the benefit of collaborating with an unfamiliar other. The magnitude of the collaborative benefit averaged across percentiles .15 through .55 where Miller’s RMI was significantly violated was not significantly correlated with either Teamwork Quality [r = -.06, p = .67] or Communication Similarity [r = -.09, p = .56]. Figure 8 provides a diagram of the associations in Chapter 3.   70 Figure 8.  A diagram of the associations in Chapter 3.  The collaborative benefit in unaffiliated pairs was associated with Communication Frequency, such that pairs who communicated less showed larger collaborative benefits.  Although Communication Similarity was associated with Teamwork Quality, neither measure was associated with collaborative benefits in unaffiliated pairs.    71 Discussion In Chapter 3 the results showed that the performance of two collaborating individuals, unknown to one another at the beginning of the experiment, exceeded the fastest possible performance of the same two individuals working independently, as predicted by Miller’s (1982, Ulrich et al., 2007) RMI. This finding implies that social interaction, not statistical facilitation, underlies the benefit of two heads compared to one in enumeration visual search. Although previous research reported that two people outperform the post hoc combination of two random individuals (e.g., Neider et al., 2010), this is the first demonstration that such collaborative benefits result from social interaction, and are not merely the product of statistical facilitation expected from aggregated individual performances.  The results also showed that frequency with which unfamiliar pairs verbally communicated was negatively associated with the benefit gained by collaborating, such that pairs who interacted more while working together in the team task showed less of an interactive benefit of collaboration. This somewhat counterintuitive finding is important because it highlights the fact that Miller’s RMI captures both the benefits and costs of collaborative cognition. On the one hand, enumeration visual search was more efficient when two people divided the cognitive task load and successfully shared the attentional demands of the task. However, this coordination of cognition through verbal communication itself incurred a cost (e.g., Brennan, Chen, Dickinson, Neider, & Zelinsky, 2008). As such, teams   72 who were able to maximize the gains of joint attention while minimizing the costs of coordination showed the largest benefits of collaboration over and above the statistical facilitation benchmark of aggregated independent individual performance. This result is also important because it establishes a relationship between the social exchange of task-relevant information and the performance gain resulting from collaborative cognition. In other words, it demonstrates the association of verbal communication -- a social factor -- with the performance benefit of social cognition. Previous research has implicated the importance of verbal communication to successful joint attention in a low-level perceptual detection and decision-making task (Bahrami, Olsen, Bang, Roepstorff, Rees, & Frith, 2012; Fusaroli, Bahrami, Olsen, Roepstorff, Rees, Frith, & Tylen, 2012).  Verbal communication may have been negatively associated with the benefit gained by collaborating because verbal communication itself incurred a cost. Yet the possibility remains that this relationship was epiphenomenal. It is also plausible that pairs who experienced more difficulty completing the enumeration visual search task talked more to one another. Because these factors were correlated, not experimentally manipulated, both explanations remain tenable. Exploring the role of verbal communication in collaborative cognition in Chapter 3 was an important first step toward understanding the role of social factors in successful enumeration   73 visual search by two-person teams. I will explore the role of additional social factors in the coming chapters, including affiliation in Chapter 4 and nonverbal communication and behavioral synchrony in Chapter 5.      74 4 Independence versus interaction  during enumeration visual search by  affiliated pairs    Introduction In Chapter 3, the results from a study of team members unfamiliar to one another showed that collaborative performance on an enumeration visual search resulted from social interaction, and not merely from the statistical facilitation expected from of two individuals working independently. Using Miller’s (1982, 2007) RMI model, the results showed that performance by pairs of unacquainted participants exceeded the benchmark of the statistical facilitation model that denotes the fastest possible performance of two individuals working independently. The communication measures taken during these performances showed that the frequency with which pair members communicated with one another was associated with the magnitude of the benefit they gained from collaboration, such that pairs who communicated more benefitted less in their collaboration.   75 Given the main finding of Chapter 3, which was that interpersonal interaction underlies the benefit of collaboration in enumeration visual search, it is important to consider the mechanisms that might allow people to work together effectively on this cognitively demanding task. Previous research has shown that the social cohesion of groups influences group productivity, such that more cohesive groups (and their individual members) are generally more productive than (for reviews see Evans & Dion, 1991; Gully, Devine, & Whitney, 1995; Mullen & Copper, 1994). In Chapter 4 I will explore how the strength of affiliation between team members influences the magnitude of the collaborative performance benefit. In these two-person teams of people who are familiar with one another, I will assume that affiliation strength is a proxy for deeper mechanisms involved in successful joint attention and hypothesize that stronger affiliation between collaborators will result in larger collaborative performance benefits. Previous research suggests that friendship augments collaborative gains because friends more easily establish a shared problem-solving space (Barron, 2003; Brown, Collins, & Duduid, 1989).  One mechanism that may facilitate successful joint attention in affiliated teams is that friendship lessens the overall attentional load that is involved when individuals must work together to complete a cognitive task. With reference to the extensive literature on dual-task interference (for a review see Pashler, 1994), coordinating collaboration with another person acts as a secondary task, which could potentially interfere with the primary task of enumeration   76 visual search. There is considerable evidence for this hypothesis in a related literature, which has explored the consequences of self-focused attention on skilled task performance. For example, self-focused attention has been shown to interfere with the performance of sports and mental tests (for a review see Baumeister & Showers, 2006). Thinking along similar lines, I propose that when collaborators are known to one another, fewer attentional resources are directed to the secondary task of coordinating collaboration and therefore can be devoted to the primary cognitive task instead. Reports that familiarity with an activity lessens the attentional resources it requires support this idea (Beilock, Wierenga, & Carr, 2002).  It is also possible that affiliation is linked to successful joint attention by way of verbal communication. There are substantial costs associated with the coordination of verbal communication. For example, when participants performed a collaborative visual search task using verbal communication in addition to shared gaze information, the task was performed more slowly than when collaborative search was performed with shared gaze information alone (Brennan, Chen, Dickinson, Neider, & Zelinsky, 2008). Considering the finding in Chapter 3, that unaffiliated pairs who communicated less had larger interactive benefits of collaboration, it is plausible that stronger social ties result in less verbal communication and therefore larger benefits of collaborative performance. In support of this proposal is research demonstrating that an established history of discourse among friends allows for a   77 shared understanding to be reached with less conversation (Tannen, 2005).  While I propose that affiliation will facilitate successful joint attention, and therefore aid collaborative performance, research suggests the possibility that affiliation could conversely hinder collaborative performance. Group work with affiliated others lead to more off task distractions and disruptive behavior, stronger pressures to agree, and reluctance to be critical of ideas (Dutson, Todd, Magleby, & Sorenson, 1997; Zajac & Hartup, 1997). Similarly, teams with a high density of social connections amongst collaborators were less successful than teams with only an intermediate density of social connections (i.e., some, but not extensive experience working together; Uzzi & Spiro, 2005).  In this study, pairs of friends completed the same enumeration visual search task described in Chapter 3 for pairs of strangers. After completing individual and collaborative versions of the enumeration visual search, participants completed the Intimate Friendship Scale (IFS; Sharabany, 1974), a questionnaire measuring affiliation strength between friends. As in Chapter 3, I correlated the magnitude of the interactive benefit of collaboration for each team with the Communication Frequency and Communication Similarity measures of verbal communication. Social affiliation scores from the IFS replaced the Teamwork Quality measure used in Chapter 3. Together these measures allow an exploration of the associations   78 between affiliation, verbal communication, and the interactive benefit collaboration here in Chapter 4. Method Participants. Forty-four University of British Columbia students (26 female; age mean = 22.6) were recruited with advertisement asking them to participate in research with a friend (22 pairs of individuals). Participants were paid $10 per hour. RSEs resulting from coactivation have been reported with similar numbers of participants as tested in this experiment. They provided written informed consent and were debriefed in accordance with APA guidelines. I report all measures I collected in this experiment. Stimuli & Apparatus. The experimental displays depicted shelves containing 82 distractor objects commonly found in a home or office and 0, 1, or 2 of 4 possible targets. Distractors appeared in four different arrangements. The same target never appeared twice in a photograph. Each target appeared equally often in each display quadrant in one target displays and in one of the quadrants not already occupied by the first target in two target displays. This generated 356 search displays in total (4 0-target + 64 1-target, + 288 2-target displays). Displays subtended 40° x 32° visual angle on a 24-inch iMac computer. The experiment was controlled by Matlab 2010a software and Psychtoolbox3. An iSight webcam recorded participants’ head and torso during the collaborative task.   79 Procedure. Participants indicated as rapidly and accurately as possible the number of targets present. Before testing, they were familiarized with the displays and the four target objects, which remained continuously visible in four pictures placed underneath the computer screen. The reason initially given to participants for video-recording the session was for the purpose of ‘‘knowing where your eyes are looking when you search.’’ Upon debriefing, participants were informed that the real reason was ‘‘to be able to measure the verbal and nonverbal behavior associated with your performance as a team.’’ Participants were then given the option of declining to have their video data be used in the study, but all agreed. Half of the participants first performed the task individually before being tested as a team; the other half performed as a team before being tested individually. In each case, participants searched a subset of all of possible displays, and indicated the number of targets present by pressing keys labeled 0, 1, or 2. Participants completed 90 trials as individuals (30 each with 0, 1, and 2 targets) and 90 trials as a team. I used weighted random sampling of the 356 total search displays to ensure that 0, 1, and 2 target trials appeared with equal frequency. Participants received feedback on their percentage of correct responses every 15 trials. When searching as a team, participants were instructed to use whatever strategy they thought was best to work together. Because there was only one keyboard for response entry, each participant   80 took a turn entering the response, with the keyboard being exchanged after 45 trials.  Friendship Quality Measure. Following completion of the visual enumeration task, each participant completed the Intimate Friendship Scale (IFS; Sharabany, 1974). It consists of 32 questions such as ‘‘I can be sure that my friend will help me whenever I ask for it.’’ Responses were made using a 5-point Likert scale (see Appendix C for the full questionnaire). An IFS score for each pair was computed by summing individual participant scores. Communication Frequency Measure. A research assistant blind to the purpose of the study made a written transcript of all verbal communication during the team task. I then calculated the total number of distinct utterances made by each team member. Summing the individual utterance counts of each team member, created a Communication Frequency measure for each pair. The smaller this number, the less pairs spoke during the collaborative visual enumeration task.  Communication Similarity Measure. I created a measure of Communication Similarity by subtracting the number of utterances made by the less talkative member of each team from the number of utterances made by the more talkative member, and dividing this number by the Communication Frequency (i.e., the total number of team utterances). The smaller this score, the more similar the relative   81 contribution of each team member was to communication during the visual enumeration task. Analysis of independent versus interactive performance benefits. I tested for violations of Miller’s RMI in the distributions of the correct responses, using the algorithm and MATLAB routines provided in Ulrich et al., 2007. I substituted the cumulative density functions (CDFs) of correct response times (RTs) in the two single-signal conditions from Ulrich et al. (2007) with the CDFs of RTs by the faster and slower of the two individuals in each team, respectively. I also substituted the CDF of RT in the redundant signals condition of Ulrich et al. (2007) with the CDF of RTs by each team. Each of these three CDFs therefore contained a total of ninety correct RTs, less the small number errors of errors that were committed. To generate the model of statistical facilitation, the CDFs of RTs by the two independent individuals in each team were combined into a fourth CDF, which was truncated at the number of RTs in the two-person team CDF to eliminate the slower tail of the RT distribution. This statistical facilitation benchmark therefore represented the fastest performance that could be derived from the two individuals working independently. Miller’s RMI will be violated, and will therefore compel the interpretation of social interaction, if two-person teams are significantly faster than the statistical facilitation benchmark at any CDF percentile. Alternatively if two-person teams are no faster than this benchmark then the improved performance of two searchers compared to one can be interpreted as statistical   82 facilitation. Typically, the RMI is rejected if there is a significant violation at any percentile (Ulrich et al., 2007). Results Collaborative performance exceeded the performance of either individual considered alone. Consistent with previous research and the findings in Chapter 3, two people working together were faster and more accurate than either independent individual. Mean correct RT and accuracy are shown in Tables 6 and 7, respectively.   83 Table 6.  Mean correct RT (sec) with SE in parentheses for affiliated pairs in Chapter 4 by the factors: test order (individual before collaborative, collaborative before individual), number of targets in the display (0, 1, 2), and social condition (slower individual, faster individual, collaborative).   individual  collaborative Test order  collaborative  individual 0 target Slower individual 20.60 (1.54) 17.11 (1.72) Faster individual 18.07 (1.55) 14.36 (1.52) Collaborative  8.88 (0.52) 14.08 (1.37) 1 target Slower individual 20.15 (1.55) 16.06 (1.59) Faster individual 17.26 (1.67) 13.75 (1.27) Collaborative  9.05 (0.57) 13.35 (1.25) 2 targets Slower individual 11.45 (0.59) 8.72 (0.53) Faster individual 8.64 (0.72) 7.19 (0.32) Collaborative  5.40 (0.33) 8.13 (0.70)   84 Table 7.  Mean accuracy (%) with SE in parentheses for affiliated pairs in Chapter 4 by the factors: test order (individual before collaborative, collaborative before individual), number of targets in the display (0, 1, 2), and social condition (slower individual, faster individual, collaborative).   individual  collaborative Test order  collaborative  individual 0 target Slower individual 97.64 (1.19) 99.58 (0.31) Faster individual 97.52 (0.89) 99.09 (0.91) Collaborative  99.70 (0.30) 95.82 (3.05) 1 target Slower individual 92.12 (2.30) 93.94 (1.55) Faster individual 88.54 (2.77) 92.36 (3.83) Collaborative  95.09 (2.25) 88.18 (4.34) 2 targets Slower individual 82.24 (4.86) 82.85 (3.35) Faster individual 74.42 (7.50) 82.30 (4.81) Collaborative  90.61 (2.24) 86.48 (2.05)   85 Correct responses were made in an average of 12.92 sec (SD = 5.95). A mixed-design analysis of variance (ANOVA) tested for differences in correct RT, with social condition (slower individual, faster individual, collaborative) and target number (0, 1, 2) as repeated measures factors, and test order (individual before collaborative, collaborative before individual) as a between groups factor. This analysis revealed that team enumeration was faster than enumeration by either the slower or faster individual [F(2, 40) = 43.88, p < .01]. It also showed that enumerating 2 targets was faster than enumerating 1 or 0 targets [F(2, 40) = 105.52, p < .01]. There were significant 2- and 3-way interactions. The difference between team and individual RT was greater with 0 than 1 or 2 targets [F(4, 80) = 13.71, p < .01], and greater when collaborative search followed individual search [F(2, 40) = 22.23, p < .01]. These two effects were synergistic, such that team RTs maximally exceeded individual RTs with 0 targets when individuals were tested first [F(4, 80) = 3.86, p < .01].  Response accuracy was high at 91.03% (SD = 12.39). A mixed-design ANOVA with the same factors as correct RT revealed that accuracy declined as target number increased [F(2, 40) = 36.34, p < .01]. Social condition and target number significantly interacted, such that teams maximally exceeded individuals with 2 targets [F(4, 80) = 3.52, p = .01]. There was also an interaction between social condition and test order: team accuracy exceeded individual accuracy when teams were tested after individuals, while there was no difference in accuracy when teams preceded individuals [F(2, 40) = 3.35, p =   86 .05].Taken together these findings show that the benefit of teamwork was most evident in the RT data for 0 and 1 targets and in accuracy data for 2 targets. These data also reveal an overall speed-accuracy tradeoff, such that participants made relatively rapid responses to 2 targets at the expense of accuracy. Because of this, only the high accuracy RTs in the 0- and 1-target conditions were submitted to an RMI analysis of correct RT. Together these findings also demonstrate an effect of test order, such that collaborative performance exceeded individual performance more when individual search preceded collaborative search than when collaborative search preceded individual search. As described in Chapter 3, this can be attributed to the effect of practice with the enumeration visual search task.     The performance benefit of teamwork was the result of social interaction, not statistical facilitation. As shown in Figure 9, team performance surpassed the benchmark of the statistical facilitation model that denotes the fastest possible performance of two independent individuals (see Method: Data Analysis). This violation of the RMI indicates that the benefit of collaboration results from social interaction and is not merely the statistical facilitation expected from independent individuals. This conclusion was supported by Bonferroni-corrected paired sample t-tests at the 10 CDF percentiles. Miller’s RMI was violated significantly at percentiles .05 through .55 (t(21) = 1.90, 2.53, 2.43, 2.34, 2.21, and 1.92, respectively, all p’s < .05). This finding that social interaction underlies the benefit of collaborative cognition prefaces the   87 following analysis of the social factors associated with effective team collaboration. Figure 9. Mean cumulative density functions for correct RT by affiliated pairs in Chapter 4, averaged over 0- and 1-targets.  Faster responses are to the left, slower responses to the right.  Two-person teams (red solid line) were not only faster than the faster of two individuals working alone (blue solid line), their performance exceeded the prediction of statistical facilitation if the same two individuals worked independently (dashed black line).     88 Communication frequency associated with social affiliation, but not collaborative benefits. The magnitude of the collaborative benefit for each team was calculated as the mean correct RT of the statistical facilitation model less the collaborative mean correct RT for each of the 10 CDF percentile bins. These values denote the magnitude of the interactive benefit of collaboration (i.e., social coactivation) for each pair and show that some two-person teams benefited from their collaboration (i.e., social coactivation greater than 0 in Figure 10), while collaboration hindered the performance of other two-person teams relative to their performance as two independent individuals (i.e., social coactivation less than 0 in Figure 10). These values were entered as criterion variables to be regressed against the measures of communication frequency, communication similarity, and social affiliation. I found an association between Communication Frequency and Friendship Quality (r = -.417, p < .01), such that pairs who reported stronger affiliation verbally communicated less.      89  Figure 10. Communication similarity and Social affiliation scatter plots. Pairs who reported greater affiliation and communicated at a more similar frequency benefitted more from their collaboration. The partial correlation coefficients revealed that communication similarity and social affiliation were each independently associated with social coactivation.      90 Next, partial correlations (prs), which measure the degree of association between two variables controlling the effect of a set of other variables, were used to explore the unique association between Communication Frequency and the magnitude of the collaborative benefit and Friendship Quality and the magnitude of the collaborative benefit. I found that Communication Frequency was associated with Friendship Quality, but that it did not uniquely contribute to the magnitude of the collaborative benefit at the .05 percentile (pr = -.093, p > .10).  Communication similarity and social affiliation were each associated with collaborative benefits. The regression analyses showed that teams who communicated more equally (r = .428, p < .01), and reported stronger affiliation (r = .554, p < .01) showed larger collaborative benefits at the .05 percentile (see Figure 10). As further percentiles were added to the collaborative measure, the strength of these simple correlations declined slightly with each step. However both correlations remained significant (r = .339, p < .05 and r = .371, p < .05, respectively) when all 10 percentiles were included. This pattern can be expected since the distinction between the social interaction and statistical facilitation models is greatest for fast RTs in the first CDF percentiles. A multiple regression analysis predicting the collaborative benefit at the .05 percentile bin with two predictors showed that communication similarity (pr = .502, p < .01) and social affiliation (pr = .346, p < .03) each contributed uniquely to the collaborative   91 benefit; with the partial correlation between these predictors near zero (pr = .048, p > .50). Together these two predictors explained 39% of the total variance in the collaborative benefit between teams (R = .624, F(2, 39)= 12.44, p < .001). See Figure 11 for a diagram depicting the associations in Chapter 4.    92 Figure 11. A diagram of the associations in Chapter 4.  The collaborative benefit in affiliated pairs was independently associated with Friendship Quality and Communication Similarity, such that pairs reporting stronger affiliation and communicating more similarly showed larger collaborative benefits.  Although Communication Frequency was associated with Friendship Quality, such that more strongly affiliated pairs communicated less, it was unrelated to the collaborative benefit in affiliated pairs.    93 Discussion This study of affiliated partners in Chapter 4 replicated the main result of the stranger partners in Chapter 3, namely, that the performance of two collaborating individuals in enumeration visual search exceeded the fastest possible performance of the same two individuals working independently. According to Miller’s RMI (1982, Ulrich et al., 2007), this finding indicates that the benefit of two heads versus one results from social interaction, and that it is not merely the product of the statistical facilitation expected from aggregated individual performance. What was new in Chapter 4 was that the measures of verbal communication similarity and social affiliation were each independently associated positively with the collaborative performance benefit. Pairs of participants who communicated at a more equal frequency and reported stronger affiliation demonstrated larger interactive benefits of collaboration. While overall communication frequency was associated with social affiliation, such that pairs reporting stronger affiliation communicated less, it was not independently associated with the magnitude of the collaborative benefit.  The premise of this study was that the strength of social affiliation between team members was a proxy for deeper mechanisms involved in successful joint attention. I propose that affiliation lessens the overall attentional load that is involved when individuals must work together to complete a complex cognitive task. Coordinating collaboration with another person acts as a secondary task (e.g., Pashler, 1994) that interferes with the primary   94 enumeration visual search task. It follows that when collaborators are known to one another, less attentional resources must be directed to the secondary task of coordinating collaboration and therefore can be devoted to the primary search task instead. Reports that familiarity with an activity lessens the attentional resources it requires support this idea (Beilock, Wierenga, & Carr, 2002).  The study of collaborating friends in Chapter 4 has revealed that social affiliation is related to both the overall frequency of verbal communication and to the similarity of verbal communication between team members. This was important in light of the finding with stranger pairs in Chapter 3, showing that pairs who communicated less had larger interactive benefits of collaboration. In that study, I considered the possibility that social affiliation was linked to collaborative performance through verbal communication. For example, it was possible that more strongly affiliated pairs showed larger collaborative performance benefits because they communication at a lower frequency. If this were the case, I would expect to find that communication frequency was independently associated with collaborative performance benefits. Instead, following the same procedures as in Chapter 3, the results showed that stronger affiliated pairs communicated less. However, the results did not show that communication frequency was independently associated with the magnitude of the collaborative performance benefit. This is strong evidence that the relationship between social affiliation and the interactive benefit of collaboration was not mediated by the overall frequency of verbal   95 communication. In other words, there is more to the relationship between social affiliation and collaborative performance benefits than simply a difference in overall verbal communication frequency. Likewise, the relationship between social affiliation and collaborative performance benefits cannot be reduced to the degree to which team members are similar in their verbal communication frequency. Whereas the results showed that communicating at a more similar frequency was associated with larger collaborative benefits, this association was independent of the relationship between social affiliation and collaborative performance. This means the association between communication similarity and the collaborative benefit is interesting in its own right.  Previous research has shown that similarity in the content of verbal communication (i.e., alignment of task-relevant vocabularies communicating confidence in what was seen in a perceptual detection task) was correlated with the magnitude of collaborative performance gains (Fusaroli et al., 2012). Because the pairs of participants in the present study used verbal communication to share information about the identity and location of targets that each member found, members who spoke at a similar frequency presumably located similar numbers of targets. Thus, collaborative performance benefits were greatest when there was a near equitable division of cognitive labor between collaborators. Performance can be made more efficient when collaborators divide the cognitive load, so that together they process in parallel what an   96 individual would serially process (Houtkamp & Roelfsema, 2009). Presumably processing is most efficient with two equally efficient parallel processes compared to one more and one less efficient parallel process. This association between processing equality and collaborative performance benefits is consistent with research that has previously demonstrated enhanced low-level perceptual detection by collaborative pairs with equal visual sensitivities relative to pairs with disparate visual sensitivities (Bahrami et al., 2010). For a collaboration to benefit performance in this task, each individual had to enhance the detection ability of the pair. If one individual lacked sensitivity relative to their partner, this individual’s contribution did not improve collaborative performance beyond the detection ability of the more sensitive individual. As such, similar detection sensitivity in this experiment is akin to an equitable division of cognitive labor during enumeration search.  The study in Chapter 4 demonstrated that social affiliation was independently associated with the collaborative benefit in enumeration visual search, and this relationship was not mediated by either the overall frequency or the similarity of verbal communication among team members. However the possibility remains that social affiliation and collaborative performance benefits are linked by other factors that I have not considered. Research has shown that nonverbal communication through body language (i.e., eye contact, gesture, and posture) contributes to the   97 benefits afforded by collaboration (e.g., Brennan et al., 2008). There is also an extensive literature demonstrating that the mere presence of another person can enhance performance (i.e., social facilitation; Zajonc, 1965). Chapter 5 explores the role of nonverbal communication and social facilitation in the interactive benefit of collaboration during enumeration visual search.    98  5 Interpersonal interaction qualities  that facilitate collaborative benefits  in enumeration visual search   Introduction Chapters 3 and 4 showed that the performance of two collaborating individuals in an enumeration visual search task exceeded the fastest possible performance of the same two individuals working independently. According to Miller’s RMI (1982, Ulrich et al., 2007), this finding indicates that the benefit of two heads versus one results from social interaction, and that it is not merely the product of the statistical facilitation expected from aggregated individual performance. Chapter 3 showed that the frequency with which unfamiliar teams verbally communicated was negatively associated with the benefit they gained by collaborating: pairs who communicated more while working together in the team task benefitted less from their collaboration. In contrast to this, the study of affiliated teams in Chapter 4 revealed that the similarity in verbal communication between team members was positively associated with the magnitude of the collaborative benefit. Teams with members who communicated at similar frequencies demonstrated   99 larger interactive benefits of collaboration. The results also revealed that the strength of affiliation between team members was positively and independently associated with the collaborative performance benefit, such that teams who reported stronger affiliation demonstrated larger interactive benefits of collaboration.  Team member affiliation. In Chapter 5, I again considered the association between affiliation and the interactive benefits of collaboration, but this time I employed an experimental design where individuals were randomly assigned to participate with either a familiar or unfamiliar other. This design permits the conclusion that any differences found between groups of familiar versus unfamiliar teams were caused by the assignment to groups, rather than the weaker conclusions in previous chapters that were based on associations with existing group differences. The primary objective of the study in Chapter 5 is to directly examine the influence of team member affiliation on the collaborative visual enumeration search task with a true experimental design, leading to a causal inference rather than merely an association.  In addition to this main objective of testing the influence of affiliation on collaborative enumeration visual search with random assignment and an experimental design, I will pursue several secondary goals in this study. First, I will explore the relationship between verbal communication the collaborative performance benefit as I did in Chapters 3 and 4. This time the experimental design of Chapter 5 permits a test of differences in the frequency   100 and similarity of verbal communication between affiliated and unaffiliated collaborative teams. Next in the secondary goals of Chapter 5, I will explore the role of nonverbal communication (e.g., body movement, gesture, posture) and physiological arousal in collaborative enumeration visual search performance. These secondary goals are described in detail below.   Nonverbal communication. Considering that aspects of verbal communication between team members were associated with the interactive benefit of collaboration in Chapter 4, I hypothesize that nonverbal communication will also be an important feature of interpersonal interaction. Previous research has demonstrated that when pairs divided visual search displays between each of their two members, they were able to locate targets faster when they could only observe their partner’s eye gaze behavior through a gaze cursor that displayed eye position (i.e., no verbal communication; Brennan et al., 2008). The visibility of one’s conversation partner has also been shown to affect performance: pairs who were able to see one another during a way-finding map task were better able to share information and manage turn taking than pairs unable to see one another (Boyle, Anderson, & Newlands, 1994). Thus I will consider the role of nonverbal communication in collaborative cognition as one of the secondary aims of Chapter 5. To study the role of nonverbal communication during the collaborative enumeration visual search, I manipulated the pairs’ ability to communicate nonverbally using a barrier. One half of both   101 the affiliated and unaffiliated pairs were randomly assigned to complete the team portion of the task without any restriction to their nonverbal communication, identical to the procedure of Chapters 3 and 4. I refer to this as the no partition condition. The two members in the other half of the pairs in Chapter 5 were separated by a partition that prevented all nonverbal communication during the team task. The partition obstructed each team members’ sight of their partner, though it did not obstruct each person’s view of the visual display, nor did it interfere with each person’s ability to communicate verbally. The partition was simply designed to eliminate any communication involving eye contact, gesture, and posture during the team task. Physiological arousal. In addition to exploring the relationship between interpersonal bodily synchrony and the magnitude of the interactive benefit of collaboration, I will also explore the relationship between physiological arousal and collaborative performance benefits as one of the secondary aims in Chapter 5. While I interpreted the violations of Miller’s RMI in Chapters 3 and 4 as evidence that social interaction between team members produced the faster performance of two-person teams compared to two independent individuals, the possibility remains that social facilitation contributed to this effect. Here I explore the possibility that violations of Miller’s RMI in support of interaction between team members could be mediated by physiological arousal; this supposition is explored below.   102 Social facilitation is the tendency for simple task performance to improve when the task is completed in the presence of other people (for reviews see Bond & Titus, 1983; Sanders, 1981). Such social facilitation effects have been shown in hundreds of simple tasks, for example, people spun fishing reels faster with others than when alone (Triplett, 1898). However the mere presence of others has also been shown to impair performance of complex tasks, for example, people made more errors when learning nonsense words in the presence of others than when alone (Pessin, 1933). The vast literature on social facilitation classifies simple tasks as those in which the performer’s dominant tendency is to give the correct response and complex tasks as those in which the dominant tendency is to give an incorrect response. While enumeration visual search task has not been classified as either simple or complex, I reason that it is a simple task because of the high accuracy rates (upwards of 85%).  Drive theorists (e.g., Zajonc, 1980) have found that physiological arousal underlies the social facilitation of simple tasks, such that the mere presence of another person heightens an individual's arousal and improves performance. Therefore to test whether social facilitation contributed to collaborative performance gains, I tested for differences in participants’ levels of physiological arousal between social conditions.  I selected to use the physiological arousal measures of skin conductance response (SCR) and heart rate (HR). SCR indexes   103 sympathetic affective arousal and HR indexes more generalized arousal and bodily state Carlson (2013). Because HR is influenced by factors other than arousal (e.g., speech; Lynch, Thomas, Long, Malinow, Friedmann, Katcher, 1982), the results will be interpreted with caution and I will look for convergence in both measures of physiological arousal when drawing conclusions about their relationship to collaborative performance benefits. To test whether differences in physiological arousal are associated with collaborative performance gains I will compare average HR and SCR during individual search with average HR and SCR during collaborative search. I hypothesize that there will be no difference in HR and SCR between individual and collaborative task performance. The studies in Chapters 3 and 4 attempted to minimize differences in physiological arousal between social condition by positioning participants in close spatial proximity to one another while they searched independently. Because I posit there will be no difference in SCR and HR between search in the individual and collaborative conditions, I introduced an auditory startle to test the sensitivity of the physiological measures. This acts as a manipulation check, such that participants’ SCR and HR should increase as a result of the auditory startle, but participants’ SCR and HR should not differ between the individual and collaborative search conditions. Previous research has shown that high-decibel, unexpected auditory stimuli produce reliable increases in SCR and HR (Bradley, Cuthbert, & Lang, 2007). By contrasting the finding of no difference in physiological arousal between individual   104 and collaborative search conditions with the finding of increased physiological arousal following an auditory startle, I will demonstrate the sensitivity of the SCR and HR to changes in arousal and show that arousal does not differ between social search conditions. Method Participants. Ninety-two University of British Columbia students (60 female; age mean = 20.64) participated in exchange for extra course credit (46 pairs). Participants registered in the study using an online research participation system. Twenty-two participants were required to bring a friend when they came to the lab to participate (22 affiliated pairs). This friend needed to be registered in a UBC psychology course and wanting to earn extra course credit. Forty-eight participants registered to take part in a study about collaboration with another person during visual search. During registration, they were required to provide the name and contact information of a friend registered in a UBC psychology course and wanting to earn extra course credit. However they were not required to bring a friend along when they participated. These participants were grouped into 24 unaffiliated pairs based on the time at which they signed up to participate. In addition to these participants, data from 34 individuals (17 pairs) were excluded due to errors in the physiological or video recording equipment. RSEs resulting from coactivation have been reported with similar numbers of participants as tested in this experiment. All participants provided   105 written informed consent and were debriefed in accordance with APA guidelines. I report all measures I collected in this experiment. Stimuli & Apparatus. Identical to Chapters 3 and 4, the experimental displays depicted shelves containing 82 distractor objects commonly found in a home or office and 0, 1, or 2 of 4 possible targets. Distractors appeared in four different arrangements. The same target never appeared twice in a photograph. Each target appeared equally often in each display quadrant in one target displays and in one of the quadrants not already occupied by the first target in two target displays. This generated 356 search displays in total (4 displays with no target, 64 displays with one target, and 288 displays with two targets). Displays subtended 40° x 32° visual angle on a 24-inch iMac computer (screen resolution 1920 X 1200 pixels). The experiment was controlled by Matlab 2010a software and Psychtoolbox3.   An HD Pro Webcam C920 (1080p) recorded the head and torso of both participants during the team task. The camera was positioned above the computer and participants were positioned in front of the computer in such a way that each person was recorded in approximately one half of the video frame. This video recording was used to transcribe the verbal communication between pair members. Visibility partition. One half of affiliated and unaffiliated pairs were randomly assigned to complete the team portion of the task without   106 any restriction to their sight of their partner. The remaining one half of pairs completed the team task with a partition positioned between them that obstructed their sight of their partner and therefore prevented pairs from communicating nonverbally, for example, through eye contact, gesture, and posture. This partition was a standard office divider measuring 108 cm X 149 cm X 4 cm.  Although it fully obstructed team members’ sight of their partner, it did not alter the procedure of the enumeration visual search task or interfere with pairs’ ability to communicate verbally.  Physiological measurement. Two measurements of psychophysiology were recorded while participants searched individually and as members of two-person teams: heart rate (HR) and skin conductance response (SCR). The physiological measurement equipment, manufactured by Thought Technology Ltd, consisted of a BVP-Flex/Pro sensor (Model SA 9308M L5890) that recorded HR from blood volume pressure (BVP) and two SC-Flex/Pro sensors (Model SA 9309M) that recorded SCR. All three sensors recorded from participants’ left hand: the HR sensor from the palmar surface of the distal phalanx of the middle finger and the SCR sensors from the palmar surface of the distal phalanx of the index and ring fingers. Participants were instructed to minimize hand movements by keeping their left hand on the table or on their leg throughout the experiment in order to avoid movement artifacts in the psychophysiological recordings. The sampling rate was 2048 Hz and 256 Hz for the digitized signals of the HR sensors and SCR sensors, respectively.    107 Auditory startle. An auditory startle was played following trial 39 in both individual and team search. Because both individuals did not reach this point at the same time, the startle was played when a randomly selected individual completed trial 39. This sound was a 1000 Hz sine signal played for 40 msec through an external speaker. It measured between 110 and 115 dB (depending on where participants were positioned relative to the speakers) and was perceptually very loud, although the sound was not of a strong enough intensity to damage hearing. Participants were informed beforehand that a loud noise would be played at a random time during the experiment: once while they searched alone, and once while they search with their partner.  Procedure. Participants indicated as rapidly and accurately as possible the number of targets present. Before testing, they were familiarized with the displays and the four target objects, which remained continuously visible in four pictures placed underneath the computer screen. The reason initially given to participants for video-recording the session was for the purpose of ‘‘knowing where your eyes are looking when you search.’’ Upon debriefing, participants were informed that the real reason was ‘‘to be able to measure the verbal and nonverbal behavior associated with your performance as a team.’’ Participants were then given the option of declining to have their video data be used in the study, but all agreed.   108 Half of the participants first performed the task individually before being tested as a team; the other half performed as a team before being tested individually. In each case, participants searched a subset of all of possible displays, and indicated the number of targets present by pressing keys labeled 0, 1, or 2. Participants completed 60 trials as individuals (20 each with 0, 1, and 2 targets) and 60 trials as a team. Note that this differed from the 90 trials in each social condition in Chapters 3 and 4 due to time restraints imposed by the psychophysiological recording. I used weighted random sampling of the 356 total search displays to ensure that 0, 1, and 2 target trials appeared with equal frequency. Participants received feedback on their percentage of correct responses every 15 trials. When searching as a team, participants were instructed to use whatever strategy they thought was best to work together. Because there was only one keyboard for response entry, each participant took a turn entering the response, with the keyboard being exchanged after 30 trials. The auditory startle was played following trial 39 during both individual and team search. Communication Frequency Measure. A research assistant blind to the purpose of the study made a written transcript of all verbal communication during the team task. I then calculated the total number of distinct utterances made by each team member. Summing the individual utterance counts of each team member, I created a Communication Frequency measure for each pair. The   109 smaller this number, the less pairs spoke during the collaborative visual enumeration task.  Communication Similarity Measure. I created a measure of Communication Similarity by subtracting the number of utterances made by the less talkative member of each team from the number of utterances made by the more talkative member, and dividing this number by the Communication Frequency (i.e., the total number of team utterances). The smaller this score, the more similar the relative contribution of each team member was to communication during the visual enumeration task. Analysis of independent versus interactive performance benefits. I tested for violations of Miller’s RMI in the distributions of the correct responses, using the algorithm and MATLAB routines provided in Ulrich et al., 2007 and same procedure as Chapters 3 and 4. The CDFs of correct RTs in the two single-signal conditions from Ulrich et al. (2007) was substituted with the CDFs of RTs by the faster and slower of the two individuals in each team, respectively. The CDF of RT in the redundant signals condition of Ulrich et al. (2007) was also substituted the with the CDF of RTs by each team. Each of these three CDFs therefore contained a total of sixty correct RTs, less the small number errors of errors that were committed. To generate the model of statistical facilitation, the CDFs of RTs by the two independent individuals in each team were combined into a fourth CDF, which was truncated at the number of RTs in the two-person team CDF to eliminate the slower tail of the RT distribution. This   110 statistical facilitation benchmark therefore represented the fastest performance that could be derived from the two individuals working independently. Miller’s RMI will be violated, and will therefore compel the interpretation of social interaction, if two-person teams are significantly faster than the statistical facilitation benchmark. Alternatively if two-person teams are no faster than this benchmark, the improved performance of two searchers compared to one can be interpreted as statistical facilitation. Typically, the RMI is rejected if there is a significant violation at any percentile (Ulrich et al., 2007). Analysis of physiological data. The HR and SCR data were re-sampled offline to 32 Hz. No data filtering was used at any point during the analyses. An average HR in beats per minute (bpm) and SCR in microsiemens (µS) were obtained for both pair members during search in each social condition. I compared these four averages (average HR individual; average HR collaborative; average SCR individual; average SCR collaborative) to test whether searching together with another person produced a change in physiological arousal compared to searching independently while another person was present. During search by two-person teams, the recording for each individual team member began at the onset of the first trial and ended immediately following the last trial performed by the team. During individual search, the recording for each individual began at the onset of the first trial of the first person to begin searching and ended immediately following the last trial of the first individual to complete the search task.    111 To test the effect of the auditory startle on physiological arousal measures, the HR and SCR recordings were time stamped when the auditory startle was played following trial 39 in both social conditions. I computed baseline measures of HR and SCR by averaging these values during the 10 sec before the auditory startle was played in each social condition (pre-startle individual, pre-startle collaborative).  I examined the time course of changes in HR and SCR relative to these baselines in one-second discrete intervals to a maximum of 60 sec, finding that maximum values for HR and SCR were observed on average 2 sec and 15 sec after the auditory startle, respectively. I used these values (post-startle HR individual, post-startle HR collaborative, post-startle SCR individual, post-startle SCR collaborative) compared to the pre-startle baselines to test the effect of the auditory startle of psychophysiological arousal. Creating difference scores of these values, I then calculated the change in HR and SCR following the auditory startle in each social condition (HR individual change; HR collaborative change; SCR individual change; SCR collaborative change). I explored the influence of social search condition on the psychophysiological changes produced by the auditory startle using these values.  Results Following the analysis procedure employed in Chapters 3 and 4, the data were examined first for the effect of social condition (slower individual, faster individual, collaborative), target number (0, 1, 2), and test order (individual before collaborative, collaborative before individual) on correct RT and accuracy. The results of this first set of   112 analyses guided the selection of inputs to the subsequent analysis of the independent and interactive benefits of collaboration using Miller’s RMI. As in Chapters 3 and 4, I used Miller’s RMI to test whether the faster RTs during collaborative compared to individual enumeration visual search resulted from the statistical facilitation of independent individual responses or from interpersonal interaction between collaborators. Following the derivation of a summary measure of coactivation for each team pair in Chapter 5, I then tested the effect of social group (unaffiliated, affiliated) and partition condition (no partition, partition) on this measure.  The primary aim of this study that used an experimental design with random assignments was to test this influence of affiliation (unaffiliated, affiliated) and nonverbal communication  (i.e., nonverbal communication permitted without a partition, nonverbal communication prevented with a partition) on the interactive performance gains in collaborative enumeration visual search. In addition to this main analysis, I conducted several secondary analyses. This included exploring whether verbal communication (i.e., Communication Frequency and Communication Similarity) and psychophysiology arousal (i.e., HR, SCR) varied between the affiliation and partition conditions. In a final set of secondary analyses, I explored the relationship between the interactive benefit of collaboration and physiological arousal, testing whether physiological arousal differed between social search conditions (individual, collaborative). This included the manipulation check of   113 the sensitivity of the physiological measures to changes in arousal using the auditory startle. Collaborative performance exceeded the performance of either individual considered alone. Consistent with previous research and the findings in Chapters 3 and 4, two people working together were faster and more accurate than either individual member of the team working independently. Mean correct RT and accuracy are shown in Tables 8 and 9, respectively.  114 Table 8.  Mean correct RT (sec) with SE in parentheses for pairs in Chapter 5 by the factors: test order (individual before collaborative, collaborative before individual), number of targets in the display (0, 1, 2), and social condition (slower individual, faster individual, collaborative).   individual  collaborative Test order  collaborative  individual 0 target Slower individual 19.58 (0.87) 13.96 (0.79) Faster individual 17.73 (1.26) 12.67 (0.75) Collaborative  8.88 (0.45) 11.61 (0.70) 1 target Slower individual 18.81 (0.94) 13.26 (0.72) Faster individual 16.28 (0.93) 11.36 (0.55) Collaborative  8.83 (0.45) 11.94 (0.59) 2 targets Slower individual 12.65 (1.00) 7.60 (0.33) Faster individual 9.88 (0.51) 6.69 (0.38) Collaborative  5.91 (0.33) 8.66 (0.62)   115 Table 9.  Mean accuracy (%) with SE in parentheses for pairs in Chapter 5 by the factors: test order (individual before collaborative, collaborative before individual), number of targets in the display (0, 1, 2), and social condition (slower individual, faster individual, collaborative).   individual  collaborative Test order  collaborative  individual 0 target Slower individual 89.71 (2.86) 97.11 (1.29) Faster individual 82.65 (4.85) 96.75 (1.40) Collaborative  99.71 (0.29) 87.90 (3.23) 1 target Slower individual 83.53 (2.99) 88.89 (2.34) Faster individual 75.29 (4.65) 88.07 (2.54) Collaborative  94.41 (1.65) 82.28 (2.87) 2 targets Slower individual 82.06 (3.61) 79.56 (2.89) Faster individual 78.53 (3.40) 77.63 (3.69) Collaborative  86.47 (2.80) 80.96 (3.79)   116 Correct responses were made in an average of 11.95 sec (SD = 4.88). A mixed-design ANOVA tested for differences in correct RT, with social condition (slower individual, faster individual, collaborative) and target number (0, 1, 2) as repeated measures factors, and test order (individual before collaborative, collaborative before individual) as a between groups factor. This analysis revealed that team enumeration was faster than enumeration by either the slower or faster individual and that enumeration by the faster individual was faster than enumeration by the slower individual [F(2, 68) = 90.78, p < .01]. It also showed a main effect of test order, such that search was faster when individual search followed collaborative search than when collaborative search followed individual search [F(1, 34) = 11.14, p < .01]. It also showed a main effect of target number: enumerating 2 targets was faster than enumerating 1 or 0 targets [F(2, 68) = 125.57, p < .001], while there was no difference in enumeration speed between 0 and 1 targets.  This analysis revealed a significant 2-way interaction between test order and social condition [F(2, 68) = 79.41, p < .001], such that collaborative search was faster than search by both the slower and faster individual and the faster individual was faster than the slower individual when individual search preceded collaborative search, but when collaborative search preceded individual search, collaborative search was no faster than search by either the slower or faster individual. Search by the faster individual remained faster than search by the slower individual when collaborative search preceded individual search. There was also a significant 2-way   117 interaction between social condition and targets [F(4, 136) = 13.27, p < .001]. During search for 0 or 1 targets, collaborative teams were faster than the slower and faster individual and the faster individual was faster than the slower individual, compared to search for 2 targets, during which the slower individual was slower than both the faster individual and the collaborative team, but the collaborative team was not faster than the faster individual.  Overall response accuracy was 86.18% (SD = 14.13). A mixed-design ANOVA with the same factors as correct RT revealed a main effect of social condition [F(2, 68) = 3.86, p = .02], where collaborative search was more accurate than search by the faster individual, but there was no difference in accuracy between collaborative search and the slower individual or the faster and the slower individuals. There was also a main effect of target number [F(2, 68) = 28.76, p < .001]. Search for 0 targets was more accurate than 1 or 2 target search and search for 1 target was more accurate than search for 2 targets.  The analysis of accuracy also revealed significant 2- and 3-way interactions. Social condition and test order interacted such that collaborative search was more accurate than search by both the slower and faster individual when individual search preceded collaborative search, while there was no difference in accuracy between collaborative and individual search when collaborative search preceded individual search [F(2, 68) = 13.52, p < .001]. There was no accuracy difference between the slower and faster individual in either test order. In line with this 2-way interaction and the main   118 effect of target number, the factors of social condition, test order, and targets interacted [F(4, 136) = 3.31, p = .01] such that collaborative search was more accurate than search by both the slower and faster individual during search for 0 or 1 targets when individual search preceded collaborative search. There was no difference between collaborative and individual search accuracy during search for 2 targets when individual search preceded collaborative search, or during search for any number of targets when collaborative search preceded individual search.  Consistent with the results in Chapters 3 and 4, these findings show that the benefit of teamwork was most evident in the RT data for 0 and 1 targets. Here in Chapter 5 the benefit of teamwork was also revealed in the accuracy of search for 0 and 1 targets. Following the same method of analysis as I used in Chapters 3 and 4, the RTs in the 0- and 1-target conditions were submitted to Miller’s RMI analysis of correct RT. Together these findings also demonstrate an effect of test order, such that collaborative performance exceeded individual performance more when individual search preceded collaborative search than when collaborative search preceded individual search. As described in Chapter 3, this can be attributed to the effect of practice with the enumeration visual search task.     The performance benefit of teamwork was the result of social interaction, not statistical facilitation. Consistent with the findings in Chapters 3 and 4, team performance surpassed the benchmark of the statistical facilitation model that denotes the fastest possible   119 performance of two independent individuals (see Figure 12).  This violation of the RMI indicates that the benefit of collaboration results from social interaction and is not merely the statistical facilitation expected with two independent individuals. This conclusion was supported by Bonferroni-corrected paired sample t-tests at the 10 CDF percentiles. Miller’s RMI was violated significantly at percentiles .05 through .65 (t(35) = 3.83, 4.32, 4.08, 3.70, 3.48, 2.81, and 2.40, respectively, all ps < .02). This finding that social interaction underlies the benefit of collaborative cognition prefaces the following analysis of the factors associated with effective team collaboration.   120 Figure 12. Mean cumulative density functions for correct RT by pairs in Chapter 5, averaged over 0- and 1-targets.  Faster responses are to the left, slower responses to the right.  Two-person teams (red solid line) were not only faster than the faster of two individuals working alone (blue solid line), their performance exceeded the prediction of statistical facilitation if the same two individuals worked independently (dashed black line).   121 The magnitude of the collaborative benefit for each team was calculated as the mean correct RT of the statistical facilitation model less the collaborative mean correct RT for each of the 10 CDF percentile bins. I used these values in the following analysis of the factors associated with effective team collaboration. Specifically, I used the magnitude of the collaborative benefit at the .05 CDF percentile, since the distinction between the social interaction and statistical facilitation models is greatest for fast RTs in the first CDF percentiles. However, the same pattern of results as described below with the collaborative benefit at the .05 CDF percentile was observed when further percentiles were added to the collaborative measure; the strength of these effects simply declined.  The influence of social affiliation and non-verbal communication. I used a series of 2 X 2 between groups ANOVAs to test the effect of social group (unaffiliated, affiliated) and partition condition (no partition, partition) on the various dependent variables in Chapter 5. The primary goal of these analyses was to test the effect of affiliation and partition on the interactive benefit of collaboration (i.e., social coactivation). The secondary goal of these analyses was to test the effect of affiliation and partition on the measures of verbal communication (i.e., Communication Frequency and Communication Similarity) and physiological arousal (i.e., HR, SCR). The results of these analyses are reported below.  Social coactivation. The largest interactive benefit of collaboration occurred for affiliated team members able to see one another; a   122 partition reduced social coactivation to the level of unaffiliated pairs. As shown in Figure 13, this analysis revealed a main effect of partition condition [F(1, 60) = 6.51, p = .01], such that teams completing the enumeration visual search task with full sight of their partner (i.e., no partition) demonstrated a larger interactive benefit of collaboration than pairs unable to see one another. Partition condition and social group significantly interacted [F(1, 60) = 4.65, p < .04], such that affiliated pairs not separated by a partition had the highest levels of coactivation, but when separated by a partition, affiliated pairs did not differ in their level of coactivation from unaffiliated pairs.    123 Figure 13. The effects of social group (unaffiliated, affiliated) and partition condition (no partition, partition) on social coactivation. Affiliated pairs who were able to communicate nonverbally (no partition) demonstrated the largest interactive benefit of collaboration. The addition of a partition that prevented nonverbal communication reduced the interactive benefit of collaboration in affiliated pairs to the level of unaffiliated pairs. Error bars represent +/- one standard error of the mean.      124 Verbal Communication Frequency. Affiliated pairs separated by a partition used the most verbal communication. As shown in Figure 14, this analysis revealed significant main effects of social group [F(1, 52) = 4.14, p < .05] and partition condition [F(1, 52) = 10.91, p < .01], and a significant interaction between social group and partition condition [F(1, 52) = 8.40, p < .01]. Affiliated pairs verbally communicated more than unaffiliated pairs, and pairs used more verbal communication when separated by a partition that prevented non-verbal communication than when they were not separated by a partition.  These two main effects interacted synergistically, such that affiliated pairs separated by a partition used the most verbal communication.    125 Figure 14. The effects of social group (unaffiliated, affiliated) and partition condition (no partition, partition) on Verbal Communication Frequency. Affiliated pairs used more verbal communication than unaffiliated pairs, and pairs separated by a partition that prevented nonverbal communication used more verbal communication than pairs who were not separated by a partition. These two effects interacted synergistically such that affiliated pairs separated by a partition used the most verbal communication. Error bars represent +/- one standard error of the mean.      126 Verbal Communication Similarity. No differences in verbal communication similarity during the team task resulted from affiliation or partition. There were no significant main effects or interactions shown in this analysis [all ps > .30]. Verbal Communication Similarity did not differ between social groups or partition conditions.  Physiological arousal. Heart rate. No differences in average HR during the team task from affiliation or partition. There were no significant main effects or interactions shown in this analysis [all ps > .10]. HR did not differ between social groups or partition conditions.  Physiological arousal. Skin conductance response. Average SCR during the team task lower for affiliated than unaffiliated pairs, no effect of partition. This analysis revealed a significant main effect of social group [F(1, 48) = 5.39, p < .03], such that affiliated pairs had lower SCRs than unaffiliated pairs. The effect of partition condition and the interaction between social group and partition condition were not significant [ps > .20].  The relationship between the interactive benefit of collaboration (i.e., social coactivation), verbal communication and physiological arousal. In an exploration that was secondary to the main goal of the experiment in Chapter 5, I used a series of correlations to investigate the relationship between the interactive benefit of collaboration (i.e., social coactivation), verbal communication (i.e., Communication Frequency and Communication Similarity), and physiological arousal (i.e., HR, SCR). The results of these analyses are reported below.    127 Verbal Communication Frequency and Similarity. Communication frequency, but not similarity, negatively associated with social coactivation. I found that Communication Frequency was negatively associated with the magnitude of the interactive collaborative benefit [r = -.32, p = .01], suggesting that pairs who communicated more while working together in the team task benefitted less from their collaboration. This was consistent with the finding in Chapter 3. There was no relationship between communication similarity and the magnitude of the collaborative benefit [p = .24]. Physiological arousal. HR and SCR were not associated with social coactivation. I found no association between HR and the interactive benefit of collaboration [p = .84] or SCR and the interactive benefit of collaboration [p = .29]. Although there was no overall relationship between the level of physiological arousal of team members and the interactive benefit of their collaboration, I continued to explore the possibility that physiological arousal may contribute to collaborative performance gains in the following analyses. These analyses considered changes in physiological arousal that resulted from the social search condition (i.e., search by two individuals versus two-person teams) and as a manipulation check, the physiological changes that resulted from the auditory startle. Physiological arousal as a function of social search and auditory startle. A final set of analyses that were secondary to the main aim of Chapter 5 explored changes in physiological arousal as a function of social search condition (individual, collaborative) and the auditory   128 startle. These analyses explored the possibility that physiological arousal may contribute to the interactive performance gains in collaborative enumeration visual search. Multiple Bonferroni corrected paired sample t-tests were used to explore whether the physiological arousal measures of HR and SCR differed between individual and collaborative enumeration search performance. As a part of this analysis, I explored whether the auditory startle produced changes in psychophysiology in order to confirm the sensitivity of the measures.  Heart rate. HR was sensitive to the auditory startle and higher during collaborative than individual search. Average HR was higher during collaborative search [mean = 79.19] than during individual search [mean = 78.17; t(61) = -2.81, p < .01]. The auditory startle resulted in a significant change in HR during individual search [t(61) = -3.42, p < .01], such that HR increased following the auditory startle [mean = 85.14] from the before startle baseline [mean = 77.59]. The auditory startle also resulted in a significant change in HR during collaborative search [t(61) = -3.13, p < .01], such that HR increased following the auditory startle [mean = 84.43] from the before startle baseline [mean = 77.89]. There was no difference in the increase in HR resulting from the auditory startle between individual search [mean increase = 7.54] and collaborative search [mean increase = 6.53; t(61) = -0.84, p = .41]. Skin conductance response. SCR was sensitive to the auditory startle and no different during collaborative than individual search. Average   129 SCR was no different during individual search than collaborative search [t(61) = -0.84, p = .41]. The auditory startle resulted in a significant change in SCR during individual search [t(61) = -10.95 p < .001], such that SCR increased following the auditory startle [mean = 7.21] from the before startle baseline [mean = 5.30]. The auditory startle also resulted in a significant change in SCR during collaborative search [t(61) = -8.47, p < .001], such that SCR increased following the auditory startle [mean = 7.29] from the before startle baseline [mean = 6.03]. The increase in SCR from startle during individual search [mean increase = 1.90] was greater than the increase during collaborative search [mean increase = 1.26; t(61) = 4.10, p < .001]. Discussion The main finding of Chapter 5 replicated and extended the previous result of Chapters 3 and 4, that the performance of two collaborating individuals in enumeration visual search generally exceeded the fastest possible performance of the same two individuals working independently. According to Miller’s RMI (1982, Ulrich et al., 2007), this finding indicates that the benefit of two heads versus one results from social interaction, and that it is not merely the product of the statistical facilitation expected from aggregated individual performance. What was new in Chapter 5 was the finding that affiliation plays a causal positive role in the magnitude of this collaborative benefit, such that affiliated teams benefited more from their collaboration than unaffiliated teams.    130 Also new, was the demonstration that the benefits of affiliation are derived largely through nonverbal communication. Here the remarkable new finding was that the presence of a partition between collaborating team members that prevented nonverbal communication though body movement, gesture, and posture, reduced the performance of affiliated teams to the level performance by unaffiliated teams.  This experiment also explored the effect of social affiliation (unaffiliated, affiliated) and partition condition (no partition, partition) on several secondary dependent measures, including verbal communication (i.e., communication frequency and communication similarity) and physiological arousal (i.e., HR, SCR). The results showed that affiliated teams showed larger interactive benefit of collaboration than unaffiliated teams, and the teams separated by a partition benefitted less from their collaboration than pairs who were not separated by a partition. These effects were synergistic, such that affiliated team members who were able to see one another demonstrated the largest interactive benefit of collaboration and introducing a partition reduced coactivation to the level of unaffiliated pairs.  This finding mirrors the association between affiliation and the collaborative benefit I found in Chapter 4, and because I randomly assigned participants to take part in the enumeration search task in Chapter 5 with a familiar or unfamiliar partner, I now know that   131 affiliation causally influences the magnitude of collaborative performance gains. This finding also highlights the important influence of nonverbal communication in the collaborative benefit in enumeration visual search. Introducing a partition that prevented nonverbal communication between team members reduced the interactive benefit of collaboration.  Like previous research that demonstrated that team members could use their partners’ eye movements to coordinate collaboration during visual search in the absence of verbal communication (Brennan, Chen, Dickinson, Neider, & Zelinsky, 2008), I showed that aspects of nonverbal communication facilitated collaborative performance gains. I posit that this finding operates in tandem with the finding that affiliated pairs separated by a partition used the most verbal communication, such that there was a trading relationship between verbal and nonverbal communication. When nonverbal communication was prevented with the addition of a partition, affiliated pairs used verbal communication instead. In a final set of analyses I explored whether psychophysiological arousal, measured with HR and SCR, contributed the collaborative performance benefits I detected using Miller’s RMI. The mere presence of another person has been shown to improve performance of simple tasks (for reviews see Bond & Titus, 1983; Sanders, 1981), and increased psychophysiological arousal has been associated with this improvement (Zajonc, 1980). While I have interpreted the violations of Miller’s RMI as evidence that social   132 interaction between team members produced the faster performance of two-person teams compared to two independent individuals, it remains a possibility that social facilitation contributed to this effect. I found that HR, but not SCR, was on average higher during collaborative search than during individual search. The HR measure was not sensitive to the effects of affiliation level between team members or the introduction of a partition that obscured team members’ sight of their interaction partner, while the SCR measure was sensitive to the effect of affiliation level, demonstrating higher levels of psychophysiological arousal for unaffiliated teams than affiliated teams. The auditory startle that occurred unexpectedly on one occasion during both individual and team search resulted in a significant change in SCR and HR in both social conditions.  Considering these findings together, I conclude that physiological arousal did not contribute to the interactive benefit of teamwork found for enumeration visual search. Movements such as speech have been shown to cause increases in HR (Lynch, Thomas, Long, Malinow, Friedmann, Katcher, 1982), and collaborative teams but not individuals spoke during the search task. I believe that this difference was sufficient to account for the approximate 1 bpm increase in HR during collaborative compared to individual search performance. For arousal to underlie the interactive benefit of collaboration, I would have anticipated increases in both HR and SCR during collaborative performance, and both of these measures to have been sensitive to the effects of affiliation and nonverbal communication obscuring partition.    133 6 General discussion   I began in Chapter 1 by describing the dilemma currently facing the burgeoning field of research on collaborative cognition. While previous research demonstrated that two or more individuals can outperform one in several cognitive domains, it remained unclear whether such purported benefits of collaboration resulted from the communicative interaction among team members, or whether they simply reflected the statistical facilitation expected from aggregating independent responses. For example, two people who communicated their confidence in what they saw to one another outperformed one person in a perceptual detection task with noisy visual signals (Bahrami, Olsen, Latham, Roepstorff, Rees, & Frith, 2010). However, this same benefit was also obtained from non-communicating teams by selecting the response of the more confident independent individual (Koriat, 2012).  I proposed that Miller’s (1982; Ulrich, Miller & Schröter, 2007) race model inequality (RMI) was the ideal tool with which to test whether the benefit of two heads compared to one resulted from social interaction or the statistical facilitation of independent individual performance. Miller’s RMI was developed to test between two   134 different models of cognitive processing in an individual person that had both been proposed to account for the ‘redundant signal effect’ (RSE; Kinchla, 1974). By comparing response time distributions, it determined whether individual responses to two signals compared to one were especially fast because the observer could detect a signal in either of two ways (i.e., separate activation models) or because both signals contributed to a common pool of activation (i.e., coactivation models). Chapter 2 through 5 tested the applicability of Miller’s RMI to a collaborative cognition context. I began in Chapter 2 by replicating the classic finding of an RSE in the same simple visual search paradigm originally employed by Miller. Like the RSE in this seminal paper, the RSE reported here resulted from coactivation (i.e., interaction) between the target signals, not the statistical facilitation of independent target activation. In a second comparison, I extended Miller’s RMI to investigate whether two-person teams responded faster to two targets than one target during the simple visual search task. As predicted, teams sharing the task with respect to space (each searching at one location) or target (each searching for one target) responded no faster to two targets than one, demonstrating that dividing redundant target information between two individuals eliminates the RSE. In a third comparison in Chapter 2, I newly applied Miller’s RMI to test whether the overall performance of two-person interactive teams exceeded the performance of two independent individuals in the same simple   135 search task. Team performance was no faster than the performance of two independent individuals in this simple search task. I posited that team performance failed to exceed the performance two independent individuals in Chapter 2 because individuals already completed the simple visual search task efficiently. The reasoning behind this supposition was that two people speed search because collaboration allows the division of the cognitive task load between collaborators (e.g., Brennan, Chen, Dickinson, Neider, & Zelinsky, 2008), allowing two-people to process together in parallel what individuals would process serially. Therefore I selected to use a more complex enumeration visual search task to explore the independent versus interactive benefits of collaborative cognition in Chapter 3. With this task the performance of two collaborating individuals, unknown to one another at the beginning of the experiment, exceeded the fastest possible performance of the same two individuals working independently. This result, found by newly applying Miller’s RMI to the study of collaborative cognition, implies that social interaction, not statistical facilitation, underlies the benefit of two heads compared to one in enumeration visual search.  In Chapter 3 I also demonstrated that the frequency with which unfamiliar pairs verbally communicated was associated with the benefit gained by collaborating, such that pairs who interacted more while working together in the team task showed less of an interactive benefit of collaboration. This result was important   136 because it established a relationship between the social exchange of task-relevant information and the performance benefit that resulted from collaborative cognition. In other words, it demonstrated the association of verbal communication -- a social factor -- with the performance benefit of social cognition. Chapter 4 again employed this same enumeration visual search task, this time testing the independence versus interaction of collaborative performance gains in pairs of friends. Here I replicated the finding of Chapter 3 that social coactivation underlies the performance benefit of two person teams compared to two independent individuals. I also found further associations between social factors and the interactive benefits of collaborative cognition, such that teams of friends who communicated at a more equal frequency and reported stronger affiliation demonstrated larger interactive benefits of collaboration. This association between affiliation and social coactivationin Chapter 4 spurred an investigation of the causal influence of affiliation and the interactive benefit of collaboration using an experimental design with random assignment in Chapter 5. In Chapter 5 individuals were randomly assigned to participate with either a familiar or unfamiliar partner. This experimental design permitted causal inferences to be made concerning the role of affiliation in the interactive benefit of collaboration.  In this chapter pairs were also randomly assigned to complete the enumeration visual search task with or without a partition that prevented   137 nonverbal communication. I found that affiliation plays a causal positive role in the magnitude of this collaborative benefit, such that affiliated teams benefited more from their collaboration than unaffiliated teams. I also newly demonstrated that the benefits of affiliation are derived largely through nonverbal communication and body language. Here the remarkable new finding was that the presence of a partition between collaborating team members reduced the performance of affiliated teams to the level performance by unaffiliated teams. In a series of secondary analyses in Chapter 5, I showed that team members were synchronized in their overall body movement, posture, and gesture by using a frame differencing method (Paxton & Dale, 2013). Also that social facilitation, indexed with changes in the physiological arousal measures or heart rate and skin conductance response, did not contribute to the improved performance of two person teams relative to two independent individuals.  Limitations and future directions While making an important contribution to our understanding of collaborative cognition, the studies presented here are not without limitations. The correlation analyses in Chapters 3 through 5 were based on small sample sizes, and such the possibility remains that the associations are spurious. To assuage this limitation I showed scatter plots in Chapters 3 and 4 so that the overall trend in the data was apparent. In Chapter 5 I explored these associations further   138 using an experimental design that independently manipulated the factors of affiliation and nonverbal communication. The results of this experiment further substantiate the previously reported associations.   Apart from being interesting in their own right, the findings that affiliation, verbal, and nonverbal communication were linked to interactive performance gains highlights the fact that the measure of collaborative performance in Miller’s RMI captures both the benefits and costs of collaboration. As such, team performance could, in principle, fail to exceed the statistical facilitation model simply because the combined costs of collaboration outweigh the benefits, not because social interaction was not present. In the experiments presented here, teams who were able to maximize the gains of joint attention while minimizing the costs of coordination showed the largest benefits of collaboration. Although Miller’s RMI provided a more appropriate baseline measure of individual performance than used in previous research on collaborative cognition, I note that it is conservative in measuring interactive team performance because the social collaborative process inherently has both benefits and costs.  Miller’s RMI was the ideal tool with which to test the independence versus interaction of collaborative cognition performance gains in enumeration visual search because this task produced multiple correct response times, which are the required input to the RMI. However, it will also be important for research on the performance   139 gains of collaborative cognition to test the independence versus interaction of benefits in tasks that do not produce multiple correct response times. Examples of such tasks include those introduced in Chapter 1, such as creative problem solving (i.e., brainstorming) and the estimation of unknown values (i.e., the wisdom of the crowd). Additional research is required to explore whether Miller’s RMI can be expanded to test the independence versus interaction of collaborative cognition performance gains in tasks such as these that do not produce multiple correct response times.  Implications  This thesis serves as an important proof of concept of the applicability of Miller’s RMI to the study of collaborative cognition. They demonstrate that Miller’s RMI, developed to test the independence versus interaction between two signals in the mind of an individual person, can be harnessed to test the independence versus interaction of performance by two people. This model emphasizes the need and provides a method for a calculating a more appropriate baseline measure of individual performance against which to compare collaborative performance. Previous research on collaborative cognition has used one of two main individual performance baseline measures (for a review see Hill, 1982). They either compared performance by two-person teams to the performance of the better of two individuals (e.g., Bahrami et al., 2010), or to nominal pairs created by selecting the faster of two individual responses on each trial (e.g., Brennan et al., 2008). Collaborative cognition performance gains were purported when   140 the performance of two-person teams surpassed these individual performance benchmarks. However, I have demonstrated that these two methods underestimate the independent (i.e., statistical facilitation) benefit of two heads compared to one. In each experiment, I used Miller’s RMI to compare the fastest possible performance of two independent individuals against the performance of two-person collaborative teams. This allowed us to parse the interactive benefit of collaboration from the independent (i.e., statistical facilitation) benefit, in a way that previous research on collaborative cognition did not. In doing so, I demonstrated for the first time that the performance gain in enumeration visual search in Chapters 3 through 5 resulted from the interaction between two team members and not the statistical facilitation of performance by two independent individuals. This stands in contrast to the finding in Chapter 2, that two-person teams completing a simple search task were no faster than two independent individuals. Because an individual person was able of complete the simple visual search task very efficiently, a second person failed enhance team performance beyond the level of the individual. These contrasting findings between Chapters 2 and 3 through 5 suggest that the division of the cognitive task load between collaborators is important to collaborative cognition performance gains. Because individual attention has a limited capacity for representing visual space and search targets in working memory (Houtkamp & Roelfsema, 2009), social collaboration offsets the   141 limited capacity of individual attention by allowing collaborators to divide the cognitive task load. This allows a process executed serially by an individual to be executed more efficiently by a pair working in parallel. Past research has shown that collaborators employ different strategies to divide the cognitive load depending on the demands and constraints of the task. Pairs who searched for a visual target that differed from distractors by a single feature divided the display spatially, each person searching roughly half the items (Brennan et al., 2008). When searching for multiple targets, pairs adopted the strategy of dividing the task by target identity, each searching for half of the possible targets (Chen, 2007). Collaborators who were unable to communicate with one another did not divide the task (Brennan et al., 2008). I posit that affiliation, verbal, and nonverbal communication are linked to collaborative performance gains because these mechanisms facilitated the division of the cognitive task load and the transfer of information between collaborators. Presumably, greater affiliation lessened the overall attentional load that is involved when individuals worked together to complete a cognitive task. With reference to the extensive literature on dual-task interference (for a review see Pashler, 1994), coordinating collaboration with another person acted as a secondary task, which interfered with the primary task of enumeration visual search. When collaborators were known to one another, less attentional resources were directed to the secondary task of coordinating collaboration and therefore could be devoted to the primary search task instead.   142 Reports that familiarity with an activity lessens the attentional resources it requires (Beilock, Wierenga, & Carr, 2002) and that self-focused attention, like attention to another person, interferes with performance (for a review see Baumeister & Showers, 2006) support this idea. In addition to nonverbal behavior, teams used verbal communication to share information about the identity and location of targets. This important means of exchange allowed teams to divide the cognitive task load and their performance to improve as a result. Because verbal communication exchanged information about search targets, team members who spoke at a similar frequency presumably located similar numbers of targets. This more equal division of the cognitive task resulted in larger collaborative performance gains compared to teams with one member who spoke more frequently than the other team member. While verbal communication speeds performance by allowing teams to divide the cognitive task load between members, it also involves substantial coordination costs. For example, past research has shown that collaborative visual search with verbal communication in addition to shared gaze information was slower than collaborative search with shared gaze information alone (Brennan et al., 2008) and I demonstrated that pairs who communicated more showed less interactive collaborative performance gains.  While the experiments presented here demonstrate that Miller’s RMI can test the independence versus interaction of collaborative   143 performance benefits and also link these interactive gains to features of the social interaction, perhaps the most important contribution of this work is its emphasis on the study of human cognition in a social context. In the words of renowned biologist Steven Frank, ‘‘all of life is social,’’ and as such, we cognitive scientists should strive to study cognition using an experimental method that follows suit. It is my belief that in doing so, we should not toss to the wayside all that we have learned about cognition by studying the individual mind. Instead, we should harness what we have learned, and the methods and models we have used to learn it, as we ask and answer questions about social cognition. By newly applying Miller’s model of individual information processing to a collaborative context, we have shown that perhaps the best solution to a new problem facing research on collaborative cognition is with an old answer. We have aimed to show that there is often no need for a nascent field to reinvent the wheel when seeking a solution; we hope we have been successful.   144 References     Bahrami, B., Didino, D., Frith, C., Butterworth, B., & Rees, G. (2013). Collective enumeration. Journal of Experimental Psychology: Human Perception and Performance, 39(2), 338-347. Bahrami, B., Olsen, K., Bang, D., Roepstorff, A., Rees, G., & Frith, C.D. (2012a). Together, slowly, but surely: The role of social interaction and feedback on the build-up of benefit in collective decision-making. Journal of Experimental Psychology: Human Perception and Performance, 38(1), 3-8. Bahrami, B., Olsen, K., Bang, D., Roepstorff, A., Rees, G., & Frith, C. (2012b). What failure in collective decision-making tells us about meta- cognition. Philosophical Transactions of the Royal Society, B, 367, 1350--1365.  Bahrami, B., Olsen, K., Latham, P.E., Roepstorff, A., Rees, G., & Frith, C.D. (2010). Optimally interacting minds. Science, 329, 1081-1085. Barron, B. (2003). When smart groups fail. Journal of Learning Sciences, 12(3), 307-359.   145 Baumeister R.F., & Showers, C.J. (2006). A review of paradoxical performance effects: Choking under pressure in sports and mental tests. European Journal of Social Psychology, 16(4), 361-383. Beilock, S.L., Wierenga, S.A., & Carr, T.H. (2002). Expertise, attention, and memory in sensorimotor skill execution: Impact of novel task constraints on dual-task performance and episodic memory. Quarterly Journal of Experimental Psychology Section A -- Human Experimental Psychology, 55(4), 1211-1240. Bernieri, F. J. (1988). Coordinated movement and rapport in teacher-student interactions. Journal of Nonverbal Behavior, 12, 120-138.  Bond, C.F. Jr., & Titus, L.J. (1983). Social facilitation: A meta-analysis of 241 studies. Psychological Bulletin, 94(2), 265-292.  Boyle, E.A., Anderson, A.H., & Newlands, A. (1994). The effects if visibility on dialogue and performance in a cooperative problem solving task. Language and Speech, 37(1), 1-20.  Bradley, M.M., Cuthbert, B.N., & Lang, P.J. (2007). Startle reflex modification: Emotion or attention? Psychophysiology, 27(5), 513-522. Brennan, S.E., Chen, X., Dickinson, C.A., Neider, M.B., & Zelinsky, G.J. (2008). Coordinating cognition: The costs and benefits of shared gaze during collaborative search. Cognition, 106, 1465-1477.   146 Camacho, M.L., & Paulus, P.B. (1995). The role of social anxiousness in group brainstorming. Journal of Personality and Social Psychology, 68(6), 1071-1080. Carlson, N. (2013). Physiology of Behavior. New Jersey: Pearson Education, Inc. Chen, X. (2007). Collaborative visual search. (Doctoral dissertation). Retrieved from SUNY Digital Repository. (Accession Order No. 1951/44780)  Chomsky. N. (1968). Language and mind. New York: Harcourt, Brace & Jovanovich.  Clemen, R.T. (1989). Combining forecasts: A review and annotated bibliography. International Journal of Forecasting, 5, 559--583.  Collignon, O., Girard, S., Gosselin, F., Saint-Amour, D., Lepore, F., & Lassond, M. (2010). Women process multisensory emotion expressions more efficiently than men. Neuropsychologia, 48(1), 220-225. Cooper, W.H., Gallupe, R.B., Pollard, S., & Cadsby, J. (1998). Some liberating effects of anonymous electronic brainstorming. Small Group Research, 29(2), 147-178.  Corballis, M. C. (1998). Interhemispheric neural summation in the absence of the corpus callosum. Brain, 121, 1795--1807.    147 Dale, R., Fusaroli, R., Duran, N. D., & Richardson, D. C. (2013). The self-organization of human interaction. In B. Ross (Ed.), Psychology of Learning and Motivation, vol. 59 (pp. 43-95). Elsevier, Inc.: Academic Press. Dennis, A.R., & Valacich, J.S. (1994). Group, subgroup, and nominal group idea generation: New rules for new media? Journal of Management, 20(4), 723-736.  Diehl, M., & Stroebe, W. (1991). Productivity loss in idea-generating groups: tracking down the blocking effect. Journal of Personality and Social Psychology, 61(3), 392-403.  Evans, C.R., & Dion, K.L. (1991). Group cohesion and performance: A meta-analysis. Small Group Research, 22(2), 175-186. Frank, S.A. (2007). All of life is social. Current Biology, 17(16), R648-R650. Fusaroli, R., Bahrami, B., Olsen, K., Roepstorff, A., Rees, G., Frith, C., & Tylen, K. (2012). Coming to terms: Quantifying the benefits of linguistic coordination. Psychological Science, 23(8), 931-939.  Galton, F. 1907 Vox populi. Nature, 75, 450--451. Gallupe, R.B., Cooper, W.H., Grise, M.L., & Bastianutti, L.M. (1994). Blocking electronic brainstorms. Journal of Applied Psychology, 79(1), 77-86.  Gully, S.M., Devine, D.J., & Whitney, D.J. (1995). A meta-analysis of cohesion and performance: Effects of levels of analysis and task interdependence. Small Group Research, 26(4), 497-520.   148 Hershenson, M. (1962). Reaction time as a measure of intersensory facilitation. Journal of Experimental Psychology, 63, 289-293. Herzog, S.M., & Hertwig, R. (2009). The wisdom of many in one mind: Improving individual judgments with dialectical bootstrapping. Psychological Science, 20(2), 231-237. Hill, G.W. (1982). Group versus individual performance: Are N + 1 heads better than one? Psychological Bulletin, 91(3), 517-539. Hogarth, R.M. (1978). A note on aggregating opinions. Organizational Behavior and Human Performance, 21, 40--46. Houtkamp, R., & Roelfsema, P.R. (2009). Matching of visual input to only one item at any one time. Psychological Research, 73, 317-326. Hutchins, E. (1995). Cognition in the Wild. Cambridge, MA: MIT Press. Iacoboni, M., Ptito, A., Weekes, N. Y., & Zaidel, E. (2000). Parallel visuomotor processing in the split brain: Cortico--subcortical interactions. Brain, 123, 759--769.  Julien, D., Brault, M., Chartrand, E., & Begin, J. (2000). Immediacy behaviours and synchrony in satisfied and dissatisfied couples. Canadian Journal of Behavioral Science, 32, 84-90.  Kerr, N.L., & Tindale, R.S. (2004). Group performance and decision making. Annual Review of Psychology, 55, 625-655.    149 Knöblich, G., & Sebanz, N. (2006). The social nature of perception and action. Current Directions in Psychological Science, 15, 99-104. Koriat, A. (2012). When are two heads better than one and why? Science, 336, 360. Larrick, R.P., & Soll, J.B. (2006). Intuitions about combining opinions: Misappreciation of the averaging principle. Management Science, 52, 111--127.  Lorenz, J., Rauhut, H., Schweitzer, F., & Helbing, D. (2011). How social influence can undermine the wisdom of the crowd effect. Proceedings of the National Academy of Sciences, 108(22). 9020-9025. Lynch, J.J., Thomas, S.A., Long, J.M., Malinow, K.L., Friedmann, E., & Katcher, A.H. (1982). Blood pressure changes while talking. Israel Journal of Medical Sciences, 18(5), 575-579. Malcolmson, K.A., Reynolds, M.G., & Smilek, D. (2007). Collaboration during visual search. Psychonomic Bulletin & Review, 14(4), 704-709. Mathôt, S., Schreij, D., & Theeuwes, J. (2012). OpenSesame: An open-source, graphical experiment builder for the social sciences. Behavior Research Methods, 44(2), 314-324. doi:10.3758/s13428-011-0168-7    150 McGrath, J.E. (1984). Groups: Interaction and performance. Prentice-Hall. New Jersey.Miles, L.K., Griffiths, J.L., Richardson, M.J., & Macrae, C.N. (2010). Too late to coordinate: Contextual influences on behavioral synchrony. European Journal of Social Psychology, 40, 52--60. Miller, J. (1982). Divided attention: Evidence for coactivation with redundant signals. Cognitive Psychology, 14, 247-279. Miller, J. (1991). Channel interaction and the redundant-targets effect in bimodal divided attention. Journal of Experimental Psychology: Human Perception and Performance, 17(1), 160-169.  Miller, J. (2004). Exaggerated redundancy gain in the split brain: A hemispheric coactivation account. Cognitive Psychology, 49, 118-154. Mullen, B., & Copper, C. (1994). The relation between group cohesiveness and performance: An integration. Psychological Bulletin, 115(2), 210-227. Mullen, B., Johnson, C., & Salas, E. (1991). Productivity loss in brainstorming groups: A meta-analytic integration. Basic Applied Social Psychology, 12(1), 3-23.  Neider, M.B., Chen, X., Dickinson, C.A., Brennan, S.E., & Zelinsky, G.J. (2010). Coordinating spatial referencing using shared gaze. Psychonomic Bulletin & Review, 17(5), 718-724.    151 Nijstad, B.A., Stroebe, W., Lodewijkx, H.F.M. (2002). Cognitive stimulation and interference in groups: Exposure effects in an idea generation task. Journal of Experimental Social Psychology, 36(6), 535-544. Nummenmaa, L., & Calder, A.J. (2009). Neural mechanisms of social attention. Trends in Cognitive Sciences, 13(3), 135-143.  Osborn, A.F. (1948). Your Creative Power. New York: Scribner. Palermo, R., & Rhodes, G. (2007). The perception of emotion and social cues in faces. Neuropsychologia, 45(1), 75-92.  Pashler, H. (1994). Dual-task interference in simple tasks: Data and theory. Psychological Bulletin, 116(2), 220-244. Pennebaker, J.W., Mehl, M.R. & Niederhoffer, K.G. (2003). Psychological aspects of natural language use: Our words, our selves. Annual Review of Psychology, 54, 547-577. Pessin, J. (1933). The comparative effects of social and mechanical stimulation on memorizing. American Journal of Psychology, 45, 263-270. Pollmann, S., & Zaidel, E. (1999). Redundancy gains for visual search after complete commissurotomy. Neuropsychology, 13, 246--258.  Raab, D.H. (1962). Statistical facilitation of simple raction times. Transactions of the New York Academy of Sciences, 24. 574-590.   152 Reuter-Lorenz, P. A., Nozawa, G., Gazzaniga, M. S., & Hughes, H. C. (1995). Fate of neglected targets: A chronometric analysis of redundant target effects in the  bisected brain. Journal of Experimental Psychology: Human Perception and Performance, 21, 211--230.  Sanders, G.S. (1981). Driven by distraction: An integrative review of social facilitation theory and research. Journal of Experimental Social Psychology, 17(3), 227-251. Schmidt, R.C., Carello, C., & Turvey, M.T. (1990). Phase transitions and critical fluctuations in visual coordination of rhythmic movements between people. Journal of Experimental Psychology: Human Perception and Performance, 16, 227--247.  Sebanz, N., Knoblich, G., & Prinz, W. (2003). Representing others’ actions: Just like one’s own? Cognition, 88(3), B11-B21. Sebanz, N., Knoblich, G., & Prinz, W. (2005). How two share a task: Corepresenting stimulus-response mappings. Journal of Experimental Psychology: Human Perception and Performance, 31(6), 1234-1246. Sharabany, R. (1974). Intimate Friendship among Kibbutz and City Children and its Measurement. PhD dissertation, Cornell University, No. 74-17-682, Ann Arbor (MI 48106): University Microfilms International. Soll, J.B. (1999). Intuitive theories of information: Beliefs about the value of redundancy. Cognitive Psychology, 38, 317--346.    153 Steiner, I.D. (1972). Group process and productivity. New York: Academic Press. Sternberg, S. High-speed scanning in human memory. Science, 153(3736), 652-654. Tannen, D. (2005). Conversational Style: Analyzing Talk among Friends. Oxford University Press.  Tomasello, M. (1999). The Cultural Origins of Human Cognition. Cambridge, MA: Harvard University Press. Trick, L.M., & Pylyshyn, Z.W. (1994). Why are small and large numbers enumerated differently? A limited-capacity preattentive stage in vision. Psychological Review, 101(1), 80-102. Triplett, N. E. (1898). The dynamogenic factors in pacemaking and competition. American Journal of Psychology, 9, 507-533.  Ulrich, R., Miller, J., & Schröter, H. (2007). Testing the race model inequality: An algorithm and computer programs. Behavior Research Methods, 39(2), 291-302.  Vul, E., & Pashler, H. (2008). Measuring the crowd within: Probabilistic representations within individuals. Psychological Science, 19, 645--647. Wolfe, J.M. (1998). What can 1,000,000 trials tell us about visual search? Psychological Science, 9, 33--39.  Wolfers, J., & Zitzewitz, E. (2004). Prediction markets. Journal of Economic Perspectives, 18, 107--126.    154 Zajonc, R.B. (1965). Social facilitation. Science, 149(3681), 269-274. Zajonc, R.B. (1980). Feeling and thinking: Preferences need no inferences. American Psychologist, 35(2), 151-175.    155 Appendices    Appendix A.  The Teamwork Quality scale used in Chapter 3. This questionnaire asks about how you working with your partner during the collaborative portion of the experiment. Next to each statement, please put the number (1-5) that corresponds with your opinion.  Strongly Agree Agree Neutral Disagree Strongly Disagree 1 2 3 4 5  ___1. I felt at ease while working with my partner.  ___2. My partner was easy to talk to.   156 ___3. I enjoyed working with my partner.  ___4. Working with my partner allowed me to complete the search task faster and more accurately than when I worked by myself. ___5. I completed the task faster and more accurately when I worked by myself than when I worked with my partner. ___6. My partner was open to the search strategies I suggested. ___7. I was open to the search strategies that my partner suggested. ___8. My partner and I each contributed equally to the search task. ___9. My partner contributed more than me to completing the search task. ___10. I contributed more than my partner to complete the search task. ___11. I preferred to work with my partner compared to working alone. ___12. I preferred to work by myself compared to working with my partner.   157 ___13. My partner and I communicated effectively to complete the search task. ___14. My partner and I cooperated to complete the search task.   158 Appendix B.  A representative sample of verbal communication from a participant pair in Chapter 3 during ten trials. ‘‘L’’ denotes communication by the person seated to the left of the display and ‘‘R’’ communication by the person seated to the right. ‘‘---‘‘ denotes the beginning of a trial. --- L: Nothing. R: Penguin.  --- L: Nothing. R: Nothing. --- L: Apple. R: Nothing. --- L: Coffee. R: Yep. One. --- R: Don’t see any. L: Yep. --- L: Coffee. R: Baseball. --- R: Apple   159 L: Don’t see any. --- R: Penguin. R: Baseball. So two. --- L: Nothing. R: Nothing. --- R: Baseball. R: Apple. ---   160 Appendix C.  The Friendship Quality scale used in Chapter 4. This questionnaire asks about the relationship between you and your friend. Next to each statement, please put the number (1-5) that corresponds with your opinion of how well it describes your relationship with your friend. Remember, this is specifically about your friend that is with you, not your friends in general. Strongly Agree Agree Neutral Disagree Strongly Disagree 1 2 3 4 5  ___1. I stick with my friend when my friend wants to do something that other people don’t want to do. ___2. I feel free to talk to my friend about almost anything. ___3. The most exciting things happen when I am with my friend and nobody else is around. ___4. I feel close to my friend.   161 ___5. I know that whatever I tell my friend will be kept secret between us. ___6. I tell people nice things about my friend. ___7. Whenever you see me, you can be pretty sure that my friend is around, too. ___8. If my friend does something I don’t like, I can always talk to him/her about it. ___9. I know how my friend feels about his/her girlfriend/boyfriend.  ___10. I can tell when my friend is worried about something. ___11. I can tell my friend when I have done things that other people do not approve of. ___12. If my friend wants something, I let him/her have it, even if I want it too.  ___13. I work with my friend on some school or work projects. ___14. I do things with my friend that are quite different than what other people might do.   162 ___15. I can plan how we’ll spend our time without first having to check with my friend. ___16. I speak up to defend my friend when other people say bad things about him/her. ___17. I can use my friend’s things without asking permission. ___18. I talk to my friend about my hopes and plans for the future. ___19. I like to do things with my friend. ___20. When something nice happens to me, I share the experience with my friend. ___21. When my friend is not around, I keep wondering where he/she is and what he/she is doing. ___22. I work with my friend on some hobbies. ___23. I know how my friend feels about things without having to be told. ___24. I know what kind of books, hobbies and other activities my friend likes.   163 ___25. I will not go along with others to do anything against my friend. ___26. I offer my friend the use of my things (like clothes, possessions, food, etc.) ___27. It bothers me to have other people come around and join in when the two of us are doing something together. ___28. If I want my friend to do something for me, all I have to do is ask. ___29. Whenever my friend wants to tell me about a problem, I stop what I am doing and listen for as long as my friend wants. ___30. I like my friend. ___31. I can be sure that my friend will help me whenever I ask for it. ___32. When my friend is not around, I miss him/her.  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0166036/manifest

Comment

Related Items