Let Me Be the Judge: Ideology, Identity, and Judicial Selection

Abstract A substantial body of research has found biased recruitment in a variety of societal spheres. We study selection in the judiciary, a domain that has received less attention than the economic and political spheres. Our field experiment took place in the midst of a Swedish government campaign encouraging ordinary citizens to contact local parties, which are responsible for recruiting lay judges (jurors) and put themselves forward as lay judge candidates. Parties’ responsiveness to citizen requests does not seem to favor their own sympathizers, does not vary at all with signals of gender, and is only marginally affected by ethnicity and age. Given the potential importance of ideology and identity in judicial decision-making, the finding that there is little bias with respect to these factors at this first stage of the recruitment process is reassuring from the perspective of impartiality.

particular, we focus on the Swedish process of selecting lay judges (citizens who have only minimal legal training and who participate in the judicial decision-making of courts). In this process, private citizens are encouraged by the government to put themselves forward as candidates by contacting local party officials, who control the nominations. This provides an excellent opportunity to test the impact of demographic factors, as well as partisan cues, at this first, and important, step towards becoming a lay judge. If party officials are systematically more responsive to candidates with certain ideological or identitarian traits, this would be troubling from a normative perspective. This is not only a matter of fairness in selection but also a concern for the potential for political bias in court rulings, as previous research shows that judicial outcomes are substantially affected by the ideology and demography of decision-makers (Anwar, Bayer, and Hjalmarsson 2012, 2014, 2019Martén 2015;Boyd 2016;Harris and Sen 2019).
The experiment took place in the midst of a government communication campaign that was carried out by the Swedish National Courts Administration in the spring of 2019. The main aim of the campaign was to encourage citizens from all backgrounds and of all ages above 18 to contact local party branches and put themselves forward as candidates for the next 4-year term of office (Ridal Ceder and Yngve 2020). During this campaign, we sent e-mails from fictitious individuals to local party officials expressing an interest in becoming a lay judge and asking for more information about the position. We employed a factorial design, where the content of the e-mails sent to party officials was randomly varied along a number of dimensions. First, half of the e-mails contained a signal that suggested that the writer was a long-time party sympathizer. This allowed us to test the extent to which political allegiances play a role in recruitment for these supposedly apolitical positions. Second, by randomly varying the name of the e-mailer, we varied signals of supposed gender (male/female-sounding names) and supposed ethnic origin (Arabic/Nordic-sounding names). Third, we included signals of the applicant's age (25/35/45/55/65). Having sent out these e-mails, we then recorded whether local party branches' likelihood of responding was affected by these ideological and demographic signals. We also recorded the informativeness of responses, and the degree to which they used a warm and welcoming tone.
The response rate was 42%. While contacting the local party organization through their official e-mail is not the only channel for ordinary citizens to signal their interest in becoming a lay judge, it is the one suggested by the government communications campaign, so this response rate must be considered less than ideal. However, we found that signals of political allegiance had little impact on the likelihood that local party officials would respond, give informative answers, and use a warm and welcoming tone. The same was true for signals of gender. Turning to ethnicity, the results were mixed and inconclusive; for some of the outcomes studied, there was small bias against e-mailers with Arabic-sounding names. The same is true of the results for age; for some outcomes, there appeared to be a slight preference for candidates in their 30s and 40s over both older and younger candidates.
We make several contributions to the literature. First, while there have been experimental studies of bias in jury selection, for example through the use of peremptory challenges Norton, Sommers, and Brauner 2007), our study is the first field experimental study of ideological and identitarian bias in the selection of laymen jurors.
Second, our study shows that recruitment in the judiciary by officials of political parties does not always result in demographic and ideological bias. In the Swedish case, the aim of the selection is to generate a demographically diverse set of lay judges with regard to age, gender, and ethnic background who are not necessarily active in political parties (Fritz 2016). From this perspective, our results are reassuring, as there appears to be limited bias with respect to most of these factors at this stage of the selection process. Moreover, and although we only study the first step in the process of becoming a lay judge, our results suggest that political favoritism toward party members or specific age and ethnic groups may not explain the existing discrepancy between societal demographics and the overall composition of lay judges in Sweden. 1 Instead, alternative explanations should be considered, including the possibility that representational gaps begin already at the supply stage.
Third, we contribute to the broader literature on how descriptive representation manifests itself in different societal spheres. While research in the Swedish context has found age, gender, and ethnic bias in the labor market, as well as in the recruitment of party activists (Ahmed, Andersson, and Hammarstedt 2012;Eriksson and Vernby 2021;Vernby and Dancygier 2019), we find little such evidence in our study. A general lesson from these results, then, is that biased recruitment in one societal sphere does not necessarily spill over to biased recruitment in other spheres.

Case and methodology
The Swedish case In most legal systems, laymen participate in judicial decision-making. In countries that follow the English legal tradition, such as the USA, England, Canada, and Australia, among others, this is institutionalized through the jury system. Other countries, such as Germany, Finland, Sweden, and most of continental Europe, use lay judges. Both types of systems attempt to increase the legitimacy of the judicial system by recruiting an impartial and demographically representative body of laymen as judges. Whereas juries in systems that follow the English tradition are made up of a random selection of citizens, lay judges are randomly drawn from a pool of candidates established by local elections or political nominations (Anwar, Bayer, and Hjalmarsson 2019). 2 In Sweden, lay judges are citizens with only minimal legal training who participate in the judicial decision-making of courts. These lay judges are supposed to be apolitical and, like in a traditional jury system, broadly representative of the population (Ridal Ceder and Yngve 2020). 3 In principle, any citizen can volunteer to 1 In 2020, the Swedish National Courts Administration surveyed the body of lay judges and found that, among them, 18% were between 18 and 44 years old (while this group constituted 44% of society) and that the share of foreign born were only 11% (22%). The age group of 65 years and older remained overrepresented at 47% (25%) (the Swedish National Courts Administration, 2020). become a lay judge by contacting one of the political parties, which control the selection process. There is no formal requirement for membership in any specific political party to become a lay judge, and the parties are supposed to recruit broadly from outside of their membership (Nilsson, Ahlstrand, and Lyckman 2007).
Recent evaluations show that in 2015 almost all (95%) Swedish lay judges were party members, as well as disproportionately old and native-born compared to the general population (Fritz 2016). 4 To address these discrepancies, in 2016 the Swedish National Courts Administration was commissioned by the government to carry out a communications campaign, prior to the election of new lay judges in 2019. The aim of the campaign was to reach more people outside of party membership circles, as well as more young people (ages 18-44), and to inform them of the opportunity of becoming lay judges. To this end, the communications were carried out continuously between 2016 and 2019 in dialogue with both the nominating and electing bodies. In addition, an extensive campaign encouraging members of the public to put themselves forward as candidates for the position of lay judge was carried out between March and June of 2019 (Ridal Ceder, and Yngve 2020). 5

Experimental design
We employed an e-mail correspondence study with a factorial design, where we varied the party sympathy, signals of supposed ethnic background and gender, and the age of fictitious e-mail senders. 6 Because of the crucial function of political parties as gatekeepers for the selection of lay judges on Swedish courts, all local branches of the eight political parties represented in the Swedish parliament were contacted. 7 In total, 2,104 e-mail addresses were collected using the parties' web pages. 8 Since our sample consisted of the universe of local party branches, its size could not be further increased without sending multiple e-mails to each addressee. 9 On April 17, 2019, in the midst of the Swedish National Courts Administration's public campaign (March-June) described above, we launched our experiment: Each local party branch was contacted by a fictitious e-mailer expressing interest in becoming a lay judge, asking about the possibility of becoming nominated by 4 In 2013, only 3% of all lay judges were between 18 and 25 years old (this group constituted 14% of the Swedish population), 17% (compared to 32% of the general population) were between 26 and 45 years old, 38% (compared to 31% of the general population) were between 46 and 65 years old, and 43% (compared to 23% in the general population) were 65 years or older (Gustafsson and Hagström 2015). Moreover, only 11% (compared to 16% of the general population) of all lay judges were foreign born. The proportion of female lay judges in 2015 was not reported (Fritz 2016), but was found to be close to 50% both in 2013 (Gustafsson and Hagström 2015) and in 2020 (the Swedish National Courts Administration, 2020). 5 See Appendix A for a outline detailing the 2016 communications campaign. 6 Our study has been registered prior to realization of outcomes in the Evidence in Governance and Politics (EGAP) registry on April 10, 2019 and was gated until November 01, 2019: ID 20190410AA.  Sending multiple e-mails to each addressee was ruled out because of the increased risk of detection as well for ethical reasons (see Appendix C). As can be seen in our pre-registration, this sample still leaves us powered to detect effect sizes of 10 %-points with near certainty, and effects of 5 %-points with high likelihood. the party and the requirements for becoming a lay judge. Each addressee received only one e-mail. For an example e-mail, see Figure 1. 10 Of the e-mail addressees, half were randomly selected to receive an e-mail signed with an Arabic-sounding name. The other half of the e-mails were signed with a traditional Nordic-sounding name. The main reason for focusing on Arabicsounding names was the possibility of observing discriminatory behavior. Research on 'ethnic hierarchies' in other societal spheres shows that people belonging to this group is among the most discriminated against in Sweden (e.g., Vernby and Dancygier 2019), suggesting that if we failed to uncover discrimination against this group when contacting parties about becoming a lay judge, people with names that sounds common to other ethnic minority groups are also unlikely to meet discrimination when doing the same. A second consideration when choosing to focus on Arabic-sounding names was the size of the group. Individuals born in Syria, Iran, Iraq, Somalia, and Afghanistan, and who typically speak Arabic and have Arabicsounding names, while also belonging to different ethnic groups, account for over a quarter of the foreign-born population in Sweden and 6% of the entire population (Eriksson and Vernby 2021). That said, we are aware that having an Arabicsounding name by no means necessarily entails being born outside of Sweden but what we are after is a strong signal of an ethnic out-group. Moreover, we randomized whether the letter was signed by a female-or a male-sounding name. 11 To test if age matters, an age between 25 and 75 was also randomly assigned to the letter, further increasing the number of variations of the letter sender.
Finally, to test the extent to which party political allegiances play a role in recruitment to these supposedly apolitical positions, half of the e-mails contained a signal suggesting that the writer of the e-mail is a long-time party sympathizer. Thus, our factorial design also varied the content of the e-mails by appealing to party sympathy. More precisely, apart from the content shown in Figure 1, some e-mails also contained an additional opening in the second sentence where the e-mailer appealed to party sympathy. To illustrate, e-mails with this appeal had a second sentence that instead of the second sentence in Figure 1 read, "I have sympathized with [party name] a long time Figure 1 Example of e-mail sent to the political parties.

10
All types of letters that were sent are available in Appendix B, both in original language (Swedish) and translated versions (English).

11
To avoid detection, three first and last names for each supposed ethnicity and gender were used. We chose the most common Nordic-and Arabic-sounding names from the Swedish population registry (See Appendix B). and I write because I am interested in being a lay judge" (see Appendix B for more details on the variations of the e-mails).
We analyze a number of outcomes. First, we have coded whether our alias received a reply. Second, to measure the informativeness of the response, we counted the number of words in each response, how many of our alias' questions were answered and whether the response provided additional information. Third, we coded the tone of the response by noting whether it used the name of our alias and whether it included a phrase that welcomed the e-mail. Finally, we also coded if the response proposed to follow up with our alias, either by invitation to party activities or by inviting continued personal e-mail communication or suggesting a meeting in the future 12

Results
In Table 1, we show the results for testing whether party officials' responsivity depended on the supposed ethnicity and gender, and age of the e-mailer, as well as signals of co-partisanship. The results come from regressions where we controlled for all other treatments, the party of the recipient, as well as the municipality of the local party branch. We have coded non-responses as zero rather than missing in order to avoid selection bias (Coppock 2019). 13 In the table, we have included the mean for each outcome as a point of reference.
The overall response rate is 42% and while the results are mixed, there is no strong indication that party officials responded more favorably to e-mailers with Nordic-sounding names when it comes to conveying relevant information about how to become a lay judge. Beginning with the response rate, Arabic-sounding aliases did have a 4.3 percentage-point lower response rate. However, the effect is only barely statistically significant at conventional levels (p < .10), and its size is small to moderate in light of the overall response rate (42%). Turning to the informativeness of the replies, while the results suggest that the replies to Arabic-sounding aliases contained significantly fewer words (p < .05), those aliases were not less likely to receive answers to their questions or receive additional information. It cannot therefore be argued that our Arabic-sounding aliases received less meaningful information than the Nordic-sounding e-mailers regarding how to become a lay judge.
Moving to the tone of e-mail replies, the results show that local party officials were 6.4 percentage-points less likely to use the name of the Arabic-sounding e-mailer in their responses (p < .01), while they were not less likely to welcome the e-mail. Regarding invitations from the party officials to follow up with our e-mailers, both personal and party contact invitations were less likely to occur when the e-mailer had an Arabicsounding name. More precisely, a future personal contact was 6 percentage-points less likely to be offered to our Arabic-sounding aliases compared to our Nordic-sounding e-mailers, a result which is significant at a 99% level of confidence (p < .01). Furthermore, our Arabic-sounding e-mailers were 2 percentage-points less likely to receive a 12 Additional details regarding the coding (including examples and information on inter-coder reliability), as well as descriptive statistics for the outcome variables are given in Appendix B. 13 When the analysis excludes non-responses, results are similar (see Appendix D). general offer to connect with the party in the future compared to our Nordic-sounding senders (p < .05).
Turning to the supposed gender of the sender, results consistently showed no differences in the treatment of female-and male-sounding e-mailers. When it comes to age, the results were mixed and inconclusive. The only inclination of favoritism found was toward our e-mailers of age 35. In fact, this group was 10.1 percentage-points more likely to receive a response (p < .01) compared to e-mailers in the baseline category (age 25). The responses were also more likely to contain more words (p < .01), answer all questions (p < .01), and address the alias by name (p < 0.05).
Perhaps surprisingly, the signal of co-partisanship did not have a consistent impact across outcomes. One the one hand, we found that it was more likely for the party official to use more words in their response (p < 0.01), welcome the e-mail (p < .01), or to extend a general invitation to become politically engaged within the local party branch (p < .01), when co-partisanship is signaled. For all other outcomes in Table 1, however, differences were not significant at conventional levels. While not part of the pre-registered design, we explored this issue further by coding the prevalence of questions posed by the party representative to our alias regarding ideology and party membership. Of the e-mails sent, 9% received a response that asked whether the e-mailer was a party member and 10% were asked about their ideology. In addition, 17% received a response that claimed that party membership was required and 9% were told that one has to share the party's ideology. Given that the position of lay judge is supposed to be non-partisan, such answers are problematic. However, in light of the reality that almost all lay judges are party members, these figures must be considered surprisingly low. 14 Robustness and extensions A number of robustness checks and extensions of the analysis were performed. The results of which are available in Appendix D. First, we show that our main results are similar when we include no or limited sets of controls. Second, we also test whether many small biases add up to more significant bias and find no evidence of such an effect. Third, we utilize the factorial design to perform heterogeneity analyses. In particular, we show that there are no systematic interactions between treatments. Fourth, we analyze whether response rates differ and react differently to treatments across different types of municipalities. We find that they do not.
Since previous research has highlighted the potential connection between party ideology on the one hand, and gender and ethnic bias on the other (e.g. Eriksson and Vernby 2021), we have also performed an analysis where we interact each treatment with the left-right position of the party to which the local official being contacted belongs. The left-right positions are taken from the Chapel Hill Expert Survey of 2019. As can be seen in Appendix D, virtually none of the interaction effects are statistically significant, suggesting that ideology is not an important factor in this particular setting. In addition, we estimate the treatment effects separately for each party. As in the main analysis, most effects are substantively small, and very few are statistically significant. Rather than underscoring the importance of partisanship, our findings suggest that these local party officials are guided by a different set of concerns when responding to our requests (cf. Alizade, Dancygier, and Ditlmann 2021). 14 We have also tested whether any of these outcomes were significantly affected by our treatments and found no such effects. See Appendix D.

Conclusion
Prior research on recruitment bias in the economic and political spheres provides a relatively clear picture: Identitarian and ideological traits matter and can often result in discrimination. Our study, by contrast, focused on recruitment in the judiciary and found little evidence of such discrimination. The absence of discrimination against female candidates and lack of favorable treatment of co-partisans challenges notions that biased recruitment follows the same logic across different societal spheres. While there was some suggestive evidence of ethnic discrimination in our study of judicial selection, it has often been found to be more pervasive in studies of the economic and political spheres.
We also test whether there is a slight preference for candidates in their 30s and 40s. This behavior would be in line with the expressed desires of the Swedish Government and the National Courts Administration, which have declared that parties should strive to rejuvenate the pool of lay judges. Previous research shows that bureaucrats can be induced to treat underrepresented group favorably when provided with clear administrative guidelines to do so (Arai et al., 2021), and we find some evidence, albeit inconclusive, that party officials may also be responsive to such government signals.