CHAPTER II
Original Research
The combined needs to reduce uncertainty through identification with groups we are in at the moment or are relevant (Hogg, 2012) and to arrive at consensus in the most informed and socially efficient way in preference combining groups (Isenberg, 1992; Stasser & Titus, 1985) may lead to a greater willingness to work with such a group when it has a clear preference norm. In plain language, when we feel uncertain, we should be more willing to work with groups that already individually mostly agrees on a course of action, as well as are similar, the latter as proven by other research (see 2018 Hogg & Kim meta-analysis). What is novel about the present study is agreement and similarity being viewed -as separate variables. Self-uncertainty, or being unsure of how one should think, act, or feel, leads people to identify with salient (relevant in the moment or self-relevant) groups, especially groups that are (seen as) more coherent, similar or homogeneous, interdependent, or in general less diverse and more likely perceived as a single unitary entity. This reduction of uncertainty happens because identification with social categories leads people to take on the traits they see as prototypical for these groups, hence an easy-to-ascertain prototypical opinion would also be preferred by self-uncertain people. Several lines of evidence suggest that self-uncertainty leads people to do more in the way of agreeing with, upholding, and identifying more with groups with a clear prototype (i.e., set of ideal traits) or prototypical position on specific topics. In the context of decision-making groups, there are several group norms at play, including the average attributes of the group that can be more or less (un)related to the decision to be made, the rules both explicit and implicit for arriving at a collective decision and interacting in general, and the overall preferred option of individuals in the group before they discuss.
These kinds of groups can simultaneously appear highly entitative by being homogeneous on a particular trait and can clearly prefer one course of action over another in a decision-making scenario. If the trait dimension that is more or less “diverse” is unrelated to the decision-making scenario, some research on work group functioning implies that similarity on one trait can leave people open to addressing disagreement over choice preferences or otherwise emphasizing group-wide similarities so that functional differences relevant to finding consensus or otherwise completing a task with a group are not perceived as a threat to group cohesion or functioning. The question that the present research addresses is: whether self-uncertainty will lead people to identify and be more willing to work with decision-making groups when similarity of personality makes the group appear entitative but the group has no obviously favored choice and, if when dissimilar, participants will identify more with a group that clearly favors one option over another. If groups with both a clear norm and similarity are preferred over groups with one or the other, but similarity is expected to also leave people open at least to working with disagreeing groups, then one must differentiate between risk-leaning, caution-leaning and ambivalent or disagreeing groups and the assignment of participants to low and high self-uncertainty conditions to allow for both of these possibilities: ambivalence or disagreement and skewed preference norms both predicting higher identification. If the two are valued together more, similar agreed groups should boast more highly identified members than similar disagreed and dissimilar agreed groups and if not, similar disagreed groups should boast members equally identified as members of dissimilar agreed groups. Risky norms should be valued in groups by self-uncertain participants over disagreement and cautiously skewed norms and every pairwise comparison between similar and dissimilar groups within each preference norm condition should show more identification and cohesion with similar groups. If not additive, and one or the other is needed to reduce uncertainty (via identification) not both, there should be no differences between preference norm conditions if the group is described as similar.
A three-way between-subjects study was implemented to assess whether the findings that self-uncertain individuals (or individuals otherwise motivated to affirm the positive distinctiveness or clarity of their self-concept) identify less with non-entitative (including dissimilar or heterogeneous) groups and take steps to clarify and uphold ingroup norms and prototypical traits, translates into the small group decision-making context. In this context, groups can be socially and demographically diverse as well as in terms of the information available to group members or the individual preferred choices this information suggests to individuals before they share their opinions or preferences with each other to reach a collective response. Pre-discussion preferences have not been manipulated in the face of uncertainty to test whether the lack of a clear preference norm also independently or only in dissimilar groups predicts lower identification. The reported research made use of noninteractive potential discussion groups and, before priming self-uncertainty, manipulated perceived similarity of the group first before manipulating perceptions of the group as having a clear preference norm, then measuring participant identification with these groups. This study (personality diversity: high, low) x 2 (uncertainty: high, low) x 3 (clear preference norm: yes–skewed risky, yes–skewed cautious, no), design assessed whether uncertain participants would identify more with non-diverse groups without a clear preference norm, and when diverse, more so with groups with a clear rather than unclear norm, and whether these differences in the case of personality diversity are mediated by the perception of dissimilar groups as less entitative, with two measures of identification, one related to group turnover, as dependent measures.
Hypotheses
H1 Ignoring self-uncertainty and the presence or lack of a clear preference norm, participants will identify more and be more willing to work with personality-homogeneous groups, than heterogenous groups.
Participants viewing the responses, to personality questions they personally answered, of other eleven supposed concurrent survey takers, scores that participants could not possibly have calculated themselves in the time given, presumably covering a range that could contain participants’ own scores, will identify more with these other participants if these data points are graphed closer together (i.e., if the group is similarly described in terms of personality).
H2 Ignoring self-uncertainty and intra-group similarity, participants will identify more and be more willing to work with groups clearly mostly preferring one course of action over another.
If the distribution of responses to the question of whether participants would advise someone to choose a higher payoff/higher risk or lower payoff/lower risk behavior is clearly skewed, meaning the majority of eleven other respondents were shown to have opted for one or the other route and not indicated they were unsure, participants will identify with them more.
H3 Participants will identify and be more willing to work with either consensus having groups skewed toward the risky option (than disagreed or skewed cautious groups) and personality homogeneous over heterogeneous groups when self-uncertain.
H4 The relationship between agreement, homogeneity, and identification (and willingness to work with the group) in self-uncertain participants, will be mediated be changes in perceived entitativity.
H5 When uncertain, participants will identify more with similar groups with no clear preference norm. If a group is described as being similar then people will be more willing to deal with a ambivalent or disagreeing group because presumably this group will come to a consensus easier being similar in any way and this in turn will reduce uncertainty.
This hypothesis entails two predictions: that similar, disagreed groups will boast higher identification and cohesion than dissimilar, disagreed groups, and that people in similar, disagreed groups will be more identified and cohesive than similar agreed cautious groups.
H6 When uncertain and in dissimilar groups, participants will identify and be more willing to work with groups with a clear (at least, skewed risky) preference norm.
This hypothesis entails a prediction that people will prefer groups with a risk-skewed preference norm when feeling uncertain, implies that this difference will be more pronounced when groups are dissimilar and does not preclude the possibility that similar, risk-agreed groups will be preferred the most and more than with similar, disagreed groups. The main prediction of H6 is that comparing preference distribution conditions within high uncertainty and similar group conditions will show more identification and cohesion with skewed, especially if not only risk-skewed norms.
H7 These differences (hypothesized in H3 through H6 excepting H4) will be nonexistent or smaller among participants primed to feel certain.
No predictions regarding whether both or one or the other between similarity and preference norm need be present for self-uncertainty to boost identification were made before the results were analyzed so none are described here.
Materials and Measures
Personality (Dis)similarity
The measure of personality used in the manipulation of perceived intra-group similarity was a combination of items from Rosenberg’s (1965) General Self-Esteem Scale, the short-form of the Intolerance of Uncertainty Scale (Carleton, Norton, & Asmundson, 2007), and the BASIS-24 (Allen, Fowler, & Fruch, 2013) sociability scale. After responding to items from these three measures, participants viewed a three-dimensional graph supposedly representing the aggregate of eleven other concurrent survey-takers’ scores on these three scales, with the hopes that participants would have noticed three different topics addressed by the personality measure(s). This graphic was output from a discriminant analysis run using SPSS/PASW on completely made-up data. In the high personality similarity, homogeneity or non-diverse condition, these eleven points were highly concentrated around the centroid of the three axes. In the low similarity, dissimilarity, or personality-heterogeneous or -diverse condition, the points were scattered more widely.
Pre-discussion Agreement
The presence of a clear preference norm was manipulated in much the same way personality similarity was, with a previously generated graphical representation of eleven other supposed responses to a question adapted from the Choice Dilemma Questionnaire (CDQ; Kogan & Wallach, 1964) that all participants were asked to respond to. Each CDQ item presents a hypothetical situation in which a protagonist must decide between two courses of action, one risky or more rewarding but less likely to succeed or occur and one cautious that is relatively less rewarding but more likely to pan out. In this study, the scenario involved a one Mr. L who is employed as an academic and deciding whether to take a 5-year appointment at another major private university to research more difficult problems that could yield groundbreaking results but would preclude him from returning to his current position, or stay with his colleagues and maintain job security working on short-term problems where he is more likely to find solutions but the questions are of less scientific importance. Participants were asked to indicate the advice they would give Mr. L as he tries to choose between the risky or high risk/higher reward option (to take the 5-year appointment) and the cautious or low risk/lower reward option (and stick with his current job). Participants responded on a five-point bipolar scale with -2 meaning they would tell Mr. L to make the riskier choice, +2 meaning they would recommend the cautious option, and 0 indicating being unsure which route they would recommend he take.
After responding, participants viewed one of three made-up CDQ-response distributions, in the form of histograms representing the frequency of eleven concurrent participants’ responses to the same question about Mr. L. The skewed-risky distribution showed that four people indicated they would say Mr. L should ‘definitely’ choose the risky option, four said he should consider it, two were unsure, and one would have him consider the cautious option. These frequencies were reversed in the skewed-cautious condition and for the full disagreement, ambivalent, or preference-diverse condition, three people presumably were unsure of what they would suggest, and two people chose each of the remaining options on either side of ambivalence.
Social Identification
An eight-item social identification measure, assessing participants’ feelings of attachment and similarity with eleven other supposed concurrent survey takers, was adapted from scales validated across many social identity-relevant studies (e.g., Grant, Hogg, & Crano, 2015; Hogg et al., 2007; Rast, Hackett, Alabastro, & Hogg, 2015). This scale, based on a three-factor conceptualization (Cameron, 2004) of the construct, and employing a nine-point Likert-type response scale (1 Extremely disagree, 9 Extremely agree). In addition, participants also completed a behavioral or social distance-type measure of what is more cohesion than identification, asking how strongly or not they would prefer to discuss their CDQ responses with their participant group or switch to a new group about which they had no information, indicating their preference using a five-point scale (1 I would actually strongly prefer to switch, 3 I am indifferent about switching, 5 I would actually strongly prefer to switch).
Entitativity
A three-item measure of entitativity was employed after both aspects of the groups heterogeneity or lack thereof had been viewed, as a check on whether dissimilar and disagreeing groups were seen as less a coherent and unified entity and to test the hypothesis that changes in identification would belie changes in the perception of the group as a single unit. The scale used items from different measures using different response scales, and the actual response distribution of participants was transformed to fall on a nine-point scale. Participants rated, among two other characteristics, how like one another the eleven other supposed concurrent participants appeared (1 Not at all, 9 Very much).
Self-Uncertainty Manipulation
The self-uncertainty prime entailed listing three aspects of the participant’s life that made him or her feel uncertain (or “certain” in the low uncertainty condition) about his or her self, “present situation,” or future and was modeled after the materials described in several uncertainty-identity studies (e.g., Hogg et al., 2007; 2010; Hohman, Hogg, & Bligh, 2010; Sherman et al., 2009).
Procedure
Participants were recruited through an advertisement of a survey study on Amazon’s Mechanical Turk web-based service for crowdsourcing mostly consumer ratings studies and menial tasks like checking for broken links or typos in online listings. All were paid somewhere between ten to fifty cents for their time and effort save a few who were compensated up to one US dollar as the number of respondents dwindled to at most one per day and deadlines approached. No data was collected on who was paid what amount. After giving informed consent to participate in what was only described as ,“a study on decision-making tendencies in individuals and groups,” in which they “may have an opportunity to chat online with other current survey takers about a decision-making scenario,” participants were told that, in order for their participant group’s survey responses to be aggregated “in real time” as they complete the study, each page (after the page showing the initial instructions and demographic items described below) of the survey must be displayed for a set amount of time and cannot be advanced until the timer runs down. Page display times were indeed regulated this way, but some deception was used to hide the fact that participants did not actually view any other participants’ scores.
After reading these instructions, they first responded to four demographic items at their own pace. Advancing the page, participants were given one hundred and fifty seconds to complete a “personality test” comprised of the three scales referenced above. Once the timer on this page expired, the SurveyGizmo web-app randomly assigned participants to view one of the two graphics described above depending on if they were assigned to the personality-homogeneous or -heterogeneous condition; they were given fifty seconds to view these graphics. Next, participants were given three minutes to read the instructions for answering and then to answer the single Choice Dilemma Questionnaire item. After thirty seconds to supposedly allow the computer time to aggregate survey group responses, participants viewed one of three graphics (bar graphs), also described above, showing a different distribution of the rest of the group’s answers to the CDQ item or pre-discussion preferences, depending on if they were assigned to one of the two clear preference norm or agreement conditions or the disagreement condition where roughly equal numbers of people were shown to prefer each response (the average being in-between the cautious or risky response). Participants were next given twenty seconds to complete the three-item entitativity measure before completing the uncertainty priming task followed by the two dependent measures of identification and desire to interact with their participant group. Lastly, participants were debriefed and compensated for their time.
CHAPTER III
Results and Implications
Inferential statistics were utilized to test whether the average response to identification and cohesion measures differed between experimental conditions more than measures of variance around these means. Multivariate analysis of variance is indicated when comparing responses between more than two discrete experimental conditions and looking at multiple dependent measures that are conceptually related. Analysis of covariance assesses these differences collapsing across variables like demographic traits expected or observed to have an effect, on the dependent variables, that cannot be differentiated from the effect of being assigned to different experimental conditions. The results of inferential statistical tests such as these are not the only valid type of evidence for assessing the abovementioned predictions and do not eliminate alternative possible explanations (discussed below) of the social psychological underpinnings of the results, nor do they guarantee a true randomized and controlled experiment and the assertion of causation over correlation. That can only be determined by judging the rigor, logic, and realism of the experimental design and the amount of support in the literature and common human experience for the interpretation of the following inferential statistical analysis.
The results give strong evidence that people reminded of self-uncertainty are more willing to work with seemingly undecided groups when these groups are described as more similar than different in terms of personality, and when dissimilar, people will work more with a group preferring a (risky) choice over a group undecided (or one that prefers a cautious choice pre-discussion). Theory suggests what matters is not the riskiness of the choice but that most people either know the risky choice is wrong or would choose it in a hypothetical scenario that differentiates it from the cautious choice that need not be discussed or would not be chosen.
Statistical Analysis
Sample Characteristics and Exclusions
Four hundred forty-eight people, at least two hundred twenty-two women and one hundred and sixty men (with the rest not responding or checking “other” in response to being asked their gender) responded to the advertisement for a study on decision-making posted online. Self-identifying men and women were evenly assigned to each of twelve experimental conditions, Χ2(3, N = 437) = 2.99, ns (p = .39).The majority of respondents who indicated their ethnic background were “‘White’/Caucasian” (65%) with the remaining people relatively evenly distributed between “Asian/Pacific Islander” (15%), “’Black’/African-American” (7%), and “Latino/Hispanic/Chicano” (5%). Most respondents were evenly distributed between three age groups, 18 to 25 (21%), 26 to 31 (21%), and 32 to 40 (23%) years old. In the case where multiple responses came from the same IP address, all but the first complete response are excluded from analysis. Those who did not complete any of the dependent measure questions were also excluded, along with responses coming from outside of the United States.
Reliabilities and Covariates
The 8-item continuous measure of social identification was highly internally consistent (Cronbach’s α = .92; 418 responding). A factor analysis of the dependent measures including entitativity showed the similarity-relevant item from the three-item entitativity measure loaded more strongly on the factor containing the classic identification items so this was included in the aggregate identification variable before analysis, reducing the inter-item reliability of this now nine-item measure to α = .78 and reducing the number responding to n = 339. The correlation between the remaining two entitativity items was significant (r = .97, p < .001) and much larger than the reliability coefficient of the three-item measure (α = .50, n = 339). Entitativity was marginally negatively related to gender, with men (M = 5.61) perceiving the group as less entitative than women (M = 5.90) on average, t(347) = 1.84, p = .07, so gender was entered as a covariate in the main analysis.
Manipulation Check
Only the manipulation of preference norm distribution (i.e., agreement versus disagreement) marginally affected perceived entitativity, F(2, 314) = 2.63, p = .074, which was not predicted over the personality diversity manipulation having an effect on such perceptions, which it did not. Groups skewed toward the cautious choice (M = 6.62) were seen as marginally more entitative than disagreeing or ambivalent (M = 5.32) groups.
Main Analysis
To test the hypotheses described above, a multiple analysis of covariance with self-uncertainty, personality composition, and preference norm as three independent factors, gender as a covariate, and identification, cohesion, and entitativity as dependent variables was run using a statistical software package (IBM’s SPSS/PASW). The omnibus F-test showed evidence of two main effects. of personality F(3, 312) = 6.25, p < .001 and opinion diversity F(6, 626) = 3.30, p < .01 qualified at least when looking at willingness to stay in the group, not identification, by a three-way interaction between all independent variables, F(6, 626) = 2.65, p = .015. That is identification was affected by both group characteristic manipulations but not necessarily differently in self-uncertain participants. The level of cohesion on the other hand depends on the combination of all three independent variables. Unpacking which specific combinations of all three manipulations differed provides possible confirmation of almost all hypotheses.
In confirmation of Hypotheses 1, that people would identify and want to work more with personality homogeneous groups, a test of the univariate effect of personality “diversity” on the 8-item traditional identification measure showed participants in groups described as more ‘clumped’ together on a three-dimensional measure of personality identified with their participant groups more (M = 4.69) by about half a point than those viewing a personality-diffuse group (M = 4.03), t(314) = 3.57, p < .001. As for the desire to stay with (for the supposed interactive portion of the study) and not switch participant groups, participants were more willing to interact with homogeneous groups (M = 5.78) than with heterogeneous groups (M = 5.20), t(314) = 3.57, p < .001.
Hypothesis 2 that people would identify and be more willing to work with agreeing groups was only partially confirmed. There was a main or single-variable effect of preference distribution on identification, F(2, 314) = 5.08, p = .007–participants viewing risk-skewed groups (M = 4.75) identified more than with disagreeing groups (M = 4.3), t(367) = 2.13, p = .034, and skewed cautious groups (M = 4.04), t(314) = 3.15, p = .002. But, there was only a marginal difference in willingness to work with the group between participants in skewed risky (M = 5.75) and un-skewed (M = 5.32) preference groups, t(3144) = 1.62, p = .095, and the overall effect of preference distribution on this variable was nonsignificant, F(2, 314) = 1.57, ns.
Hypotheses 3 and 4 were not confirmed, there were no (H3) two-way interaction effects indicating that independently either distribution of personality or preference affected identification, behavioral or otherwise, in highly self-uncertain participants, let alone (H4) these differences being mediated by changes in entitativity. This essentially means that one cannot consider the effect (on identification or cohesion among self-uncertain people) of preference distribution without considering personality and vise-versa. Sobel’s test of mediation comparing the indirect effect of entitativity on either continuous or single item measures to the effect of homogeneity or agreement on these measures controlling for entitativity shows no significant decrease in the magnitude of the (dis)similarity-identification or (dis)agreement-identification relationship in self-uncertain participants.
There is, however, strong evidence supporting Hypothesis 5, 6, and 7, that (H7) self-uncertain participants more than “self-certain” participants would (H5) identify more with groups with no skewed preference norm if the group is also similar in personality rather than dissimilar. In all but one case, the differences between cell means were nonsignificant when comparing participants in low uncertainty conditions, confirming Hypothesis 7 for the cohesion-type measure. Assessing Hypothesis 5, univariate analysis showed that intention to stay with the group was higher among self-uncertain participants (ignoring those in the “certainty” conditions) viewing similar, disagreeing groups over dissimilar, disagreeing groups. People intended to stay in personality similar versus dissimilar groups when the group was not clearly agreed, t(314) = 4.44, p < .001, and this difference was less pronounced among low uncertainty participants, t(314) = 2.05, p = .042. This pattern did not hold for the traditional continuous measure of identification, suggesting that the experimental manipulations only interacted to affect participants’ willingness to work with a (dis)similar or (dis)agreeing group rather than how agreed or personally similar participants themselves were in relation to the group. Furthermore, amongst people viewing similar groups neither skewed risky or cautious norms were not identified with more-or-less or more-or-less cohesive with these over disagreeing groups.
Hypothesis 6 was that personality dissimilar groups would be identified with more under uncertainty when the distribution of pre-discussion CDQ responses from eleven other people were skewed in favor of at least the risky option if not when skewed either way. This prediction was confirmed: self-uncertain participants in personality diverse groups identified more, F(2, 314) = 7.20, p = .001, respectively with groups with a preference distribution skewed risky (M = 6.05), followed by groups slightly in favor of the ambivalent response option (i.e., unsure of which route to take) but almost equally representing all options (M = 4.9), t(314) = 2.17, p = .03, followed by cautiously skewed groups (M = 3.75), t(314) = 2.08, p = .038, with skewed risky and skewed cautious preference norm groups also being significantly different, t(314) = 3.79, p < .001 . Furthermore, these differences were nonexistent among participants asked to think of times they were not uncertain, F(2, 314) = 1.19, ns, further corroborating Hypothesis 7.
Again, there were no differences between (dis)agreement conditions among self-uncertain participants viewing personality-similar groups. Results show overall that participants in the high uncertainty condition viewing personality diffuse groups more often indicated they would stay with their survey group when the rest of the group leaned in favor of the risky CDQ response, with no meaningful differences among self-uncertain people viewing closely clustered groups and smaller differences among participants primed to feel “certain.”
Alex Ash Creative 