Experimental Design
British participants whose stated nationality is either English or Scottish will be recruited from a large Internet panel. Participants will be asked to privately make a series of binary choices, and in particular to state their opinions on a variety of policy suggestions, and to perform incentivized assessment tasks. Tasks will be of two types: (1) tasks where both nationalities are likely to be perceived as equally capable of task performance (e.g., assessing whether there are more dots or more triangles in pictures); (2) tasks where one nationality is likely to be perceived as having more task-related knowledge, and therefore as more capable of task performance (e.g., indicating whether two Scottish Tartans are associated with the same Scottish family/clan or with different Scottish families/clans). [Importantly, tasks of the second type will be selected such that although one nationality is likely to be perceived as having more knowledge, most members of this nationality are not much more knowledgeable].
To allow social learning, choice items will ideally involve uncertainty about their intrinsic value. Choice items will be selected based on a preliminary survey, and will ideally reduce preexisting group dispositions and beliefs about group norms.
Survey participants will be randomly assigned to receive or not receive social information. Those who receive social information will be matched with a random sample of each nationality group participating in the preliminary survey. Before making each choice, they will be informed of the descriptive norms prevailing in the samples of nationality groups assigned to them. The information about norms will include a description of the distribution of choices in the sample of each social group (e.g., a% of Group A chose option x, and (1-a)% of Group A chose option y; b% of Group B chose option x, and (1-b)% of Group B chose option y), alongside a graphic illustration of the distribution. Ideally there should be four main social-information conditions (in addition to the no-information condition), where participants will discover that a majority or minority of their in-group, and a majority or minority of their out-group chose an option, leading to a 2 (in-group majority, in-group minority) X 2 (out-group majority, out-group minority) + 1 (no social information) design.
To create the distribution of choices for the different social-information conditions, while avoiding deception, each survey participant in the social-information conditions will be matched with an odd-sized small sample from each nationality group participating in the preliminary survey. The choice items presented to survey participants will be those for which the distribution of choices in the preliminary survey is the closest to uniform. Due to this procedure, we may not have enough identifying variation for some choice items.
Comparing the choices of survey participants who observe different (or no) combinations of in-group and out-group norms, and who have different levels of social identification, will allow us to disentangle the three main channels of social influence. Particularly, let us denote the size of the majority by j% (j>50), and one of the choice options by x. Then participants who receive social information will discover that option x was chosen by either (1) j% of Group A and (1-j)% of Group B, (2) (1-j)% of both groups, (3) (1-j)% of Group A and j% of Group B; or (4) j% of both groups. [Note that the first three conditions, or alternatively conditions (1), (3) and (4), allow separating the main channels].
Social learning alone would predict that in the absence of perceived differences in group expertise, the likelihood of choosing option x will be the smallest in condition (2)—where the total relative number of such choices is the smallest—and similarly larger in conditions (1) and (3) that share a similar total relative number of such choices. Yet preferences for in-group conformity and out-group differentiation—whereby resemblance to the typical in-group member and distinction from the typical out-group member, respectively, are sought after—would predict otherwise: For members of Group A, they would predict that the likelihood of choosing x will be the largest in condition (1), smaller in condition (2) and smaller still in condition (3). For members of Group B, they would predict that the likelihood of choosing x will be the largest in condition (3), smaller in condition (2) and smaller still in condition (1). Finally, they would predict these differences in the likelihood of choosing x to intensify with the strength of social identity.
Thus, in principle comparing the likelihood of choosing x in conditions (1) and (2) will reveal the combined effect of social learning and preferences for in-group conformity for members of Group A [the rival effects of social learning and preferences for out-group differentiation for members of Group B]. Comparing the likelihood of choosing x in conditions (2) and (3) will reveal the rival effects of social learning and preferences for out-group differentiation for members of Group A [the combined effect of social learning and preferences for in-group conformity for members of Group B]. Comparing the likelihood of choosing x in conditions (1) and (3) will reveal the combined effect of preferences for in-group conformity and out-group differentiation for members of both groups. Moreover, replacing condition (2) with condition (4) will allow us to test the generalizability of the results, as well as derive the combined effect of social learning and preferences for in-group conformity [the rival effects of social learning and preferences for out-group differentiation] while controlling for preferences for out-group differentiation [preferences for in-group conformity]. Finally, the likelihood of choosing x in the no-information condition will capture the groups' tastes and private information about intrinsic value.
When one group is perceived to have task-related expertise, social learning alone would predict that more weight will be placed on that group's task choices. Hence, comparing choice likelihoods in the different social-information conditions across tasks of different types, will further substantiate the separation of the main channels of social influence.
This discussion shows how the experimental design allows us to separate different theoretical channels. We will analyze the results using a regression framework where we examine how individual choices are affected by the observed descriptive norms of the ingroup and the outgroup, that is, the proportion in each group making a particular choice. Again, we will separate the effects by the extent of group identification, measured by our survey items.
After choices are made, one of the participant's task choices will be randomly selected. If their answer to the selected task is correct, the participant will receive payment in addition to a participation fee. Each participant will also be asked to complete a brief survey, which will include questions about their reasoning during the experiment, standard psychological questionnaires on identity, demographics, and feedback to the researchers. We are interested in also exploring differences in in-group conformity, out-group differentiation and social learning by risk attitudes, education, and gender.