272 HEERDINK.VAN KLEEF.HOMAN.AND FISCHER ng.Then we introduced the emotion inst n research (Ko Accepta cerejection scale. After completing the group task Sinaceur aching the ess happine h th in a rat th's a .73). the foca articipant we n ovided thes participants with in rationa which have be ound to be manipulation was checked using both self-report and peer-repor ns of did you show ality" told tha ruction (i.e. showing anger,happi were asked,"To what extent as the participant on your [left/right ud help all groupm explained tep-by-step how In order to get an estimate of the extent to which each were ins erbal behaviors they could ineda numb d d th of e ple r 87 88 andr )and happine( Cronbach's and o em ns ab nothe was t among pee abo hei 9 I sav c and ratings gated to th was chan er and happiness,and A and B agreed on both in ances).Fo roals of the g oup tasp mphas zing that only thebes ggregation.ICC()=0.10.F(98.9)=1.p=I5.Thus.o peer-reported manipulation check of rationality was possible E ide which Then,participants were given 10 min to complete Results The focal participants'influence in the group task he d vere analy☑ d with the Ime4 R package.with ar operati as contr ted by ipant alized as the d in groups.Follo ng th I th nd on the pling fro tion o f the eters (sec included on the group list appear on more than one ind e,if Group Me ers A and B bot rom the t statistie (which is explicitly discouraged)as there is well.group memb A and B both had a contribution of 1.Fo the articipant contrib age.as the tw https://stat.ethz.ch/pipermail/r-belp/2006-May/94765.hml To set the stage for the emotion manipulation, we emphasized the importance of critical evaluation for successful decision mak￾ing. Then we introduced the emotion instructions as “instructions on how to contribute to group decision making.” The instructions themselves were based on emotion instructions that have been used in negotiation research (Kopelman, Rosette, & Thompson, 2006; Sinaceur & Tiedens, 2006). Three versions of the instruc￾tions were used, coaching the participant to express anger, to express happiness, or to approach the group task in a rational, nonemotional manner. Rather than not providing any instructions to the focal participants, we provided these participants with in￾structions to “be rational,” which have been found to be effective in reducing the amount of emotion shown (Kopelman et al., 2006). Moreover, by providing the focal participants with instructions of more or less equal length to the emotion instructions, we made sure that they would not be suspicious about seeing the other two participants taking longer to read their instructions. The basic format for each of these instructions was the same. First, the participant was told that experts agree that adopting the strategy described in the instruction (i.e., showing anger, happi￾ness, or no emotion) would lead to optimal group performance, as it would help all group members to be open to critical ideas. Second, the instructions explained step-by-step how the strategy could be executed. For instance, in the emotion conditions, par￾ticipants were instructed on nonverbal behaviors they could use to express their emotion. It should be noted that these instructions also contained a number of example phrases that implicitly in￾structed participants to express emotions about other participants’ ideas, instead of about the person (e.g., “This kind of idea only makes me angry”). The focal participants read that they should control their emotions and think logically. Then, to ensure elabo￾ration of the instructions, participants were asked to summarize their strategy in their own words and write some sentences they could say during the group task to follow their strategy effectively. Group-wise selection of ideas. Directly after studying their instructions, the participants’ seating arrangement was changed so that they faced each other. The experimenter then rolled a die to determine who would keep track of the group ideas and re-iterated the goals of the group task, emphasizing that only the best ideas should end up on the list. All materials except the individual lists of ideas and the form on which the group ideas would be written were removed. Then, participants were given 10 min to complete the group task. Relative influence. The focal participants’ influence in the group task was operationalized as the ratio of ideas contributed by this participant to the number of ideas contributed by the other two group members. Contribution was operationalized as the number of items that appeared both on an individual group member’s list and on the final group list. In many cases, an idea (e.g., water) that was included on the group list appeared on more than one indi￾vidual list. In these cases, we counted the item as having been contributed by each group member who had this idea on his or her personal list. Thus, for instance, if Group Members A and B both had “water” on their list, and “water” appeared on the group list as well, Group Members A and B both had a contribution of 1. For this reason, the sum of these counts in each group could total more than the number of ideas generated by the group. If the focal participant contributed as many ideas, on average, as the two other group members, this ratio would be 0.5, indicating that the focal participant had an equal amount of influence as the other two group members. Ratios lower than 0.5 would reflect that the focal participant had relatively less influence than the other two group members (i.e., contributed fewer ideas) and ratios above 0.5 that the focal participant had relatively more influence. Acceptance/rejection scale. After completing the group task, participants were seated separately again and were given a ques￾tionnaire that contained the acceptance/rejection scale from Stud￾ies 1 and 2 (Cronbach’s
.73). Manipulation checks. After the participants had completed a second, unrelated group task, the effectiveness of the emotion manipulation was checked using both self-report and peer-report measures. The self-report measure was “During the group tasks, to what extent did you show . . .,” followed by two items for anger (“anger” and “irritation,” r
.81) and two items for happiness (“happiness” and “enthusiasm,” r
.79). One more item was included to check for differences in rationality (“rationality”). For the peer-report measure, participants filled out two ques￾tionnaires, one for each of their two fellow group members. They were asked, “To what extent was the participant on your [left/right] [angry/happy/rational].” We also asked, “To what extent did the participant on your [left/right] show [anger/happiness/enthusi￾asm].” In order to get an estimate of the extent to which each participant had shown anger, happiness, and rationality, we first estimated the reliabilities of the respective scales. For both anger (r
.88 and r
.93) and happiness (Cronbach’s
.87 and
.87), the relevant items could be combined to form a scale. Then, the extent to which two peers agreed about another participant’s emotional expressions was determined before we aggregated the participant level by averaging the two observations. There was significant agreement among peers about each partici￾pant’s level of anger, intraclass correlation (ICC)(1)
0.37, F(98, 99)
2.16, p  .001, and happiness, ICC(1)
0.16, F(98, 99)
1.39, p
.050, and ratings could therefore be aggregated to the participant level by averaging the peer ratings (e.g., Participants A and B each rated the extent to which Participant C had shown anger and happiness, and A and B agreed on both instances). For rationality, there was not enough agreement among peers to justify aggregation, ICC(1)
0.10, F(98, 99)
1.23, p
.15. Thus, no peer-reported manipulation check of rationality was possible. Results The data were analyzed with the lme4 R package, with a random intercept for group (unless otherwise stated), because our focal participants were nested in groups. Following the recommendation of the package author,6 we determined the significance of the results of the multilevel analyses by Markov chain Monte Carlo sampling from the posterior distribution of the parameters (see, e.g., Gilks, Richardson, & Spiegelhalter, 1996; 10,000 samples), which results in a confidence interval for the regression coefficient that can be interpreted analogous to bootstrapping results. This yields a more reliable hypothesis test than calculating p values from the t statistic (which is explicitly discouraged), as there is little agreement about how to correctly estimate the relevant num￾ber of degrees of freedom. 6 https://stat.ethz.ch/pipermail/r-help/2006-May/094765.html This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 272 HEERDINK, VAN KLEEF, HOMAN, AND FISCHER