Ethnical discrimination in Europe: Field evidence from the finance industry

The integration of ethnical minorities has been a hotly discussed topic in the political, societal, and economic debate. Persistent discrimination of ethnical minorities can hinder successful integration. Given that unequal access to investment and financing opportunities can cause social and economic disparities due to inferior economic prospects, we conducted a field experiment on ethnical discrimination in the finance sector with 1,218 banks in seven European countries. We contacted banks via e-mail, either with domestic or Arabic sounding names, asking for contact details only. We find pronounced discrimination in terms of a substantially lower response rate to e-mails from Arabic senders. Remarkably, the observed discrimination effect is robust for loan- and investment-related requests, across rural and urban locations of banks, and across countries.

The e-mail template has been written in German with the help of Austrian bankers. In the next step, for each country the e-mail has been translated into the official national language with the help of two independent native speakers. If a country has more than one official language, individual institutions were approached in the language that is spoken primarily in the respective region. Institutions were contacted via their official information addresses. To ensure that e-mails are not forwarded to a headquarter, we only approached institutions and subsidiaries with a unique contact e-mail address.
All e-mails have been sent out at the same day. The time frame of the data collection was five days (a working week), ranging from May 9, 2016 to May 13, 2016. We sent a second e-mail to those banks that replied to the query, with a note that the issue has been resolved and that their service is very much appreciated but no longer needed.

B. Ethics
This study was approved by the Ethics Committee of the University of Innsbruck. Furthermore, we took all measures to ensure full confidentiality. However, there are two arguments that might be raised against the ethical justification of the study, which we want to discuss in what follows. For further details on ethical issues in field studies on discrimination see [1].
The first argument is that banks were contacted and, thereby, their time was consumed in answering our e-mails. Even though we fully agree that it is important to cautiously consider possible costs and effects that field studies incur on the subjects under investigation, we still consider our study to be ethically valid for the following reasons: First and foremost, it is our opinion that the possible costs for the banks are outweighed by the benefit of the study for the society as a whole. It is important for societies to acquire a accurate view on discrimination based on objective data to acquire informed policies, since the costs stemming from failed integration and discrimination can be severe for individuals and societies as a whole. Second, we consider our e-mail inquiry to incur only little cost and effort to the respective banks' employees. The e-mail inquiry can be answered straightforwardly without requiring to gather further information. To reduce additional effort by checking back whether senders need more information or further help, we sent out the second e-mail described in Section A that the issue has been resolved and no further information is necessary.
The second argument is that banks were contacted with fake names and inquiries that are specifically designed for this study. As experimental economists we are very concerned with deception of subjects. However, field studies on discrimination are impossible to conduct without a minimal form of misinformation. Moreover, utilizing deceptive elements in our experimental design does not force subjects into situations they are not facing in their everyday environment. Rather, this kind of information is used to observe common behavior in the workplace. Again, we argue that the costs incurred with this practice are justified by the high information content of our study for society and policy makers. with regards to area refer to a threshold of 10,000 inhabitants (different thresholds are discussed below). We find a pronounced and significant discrimination effect in terms of lower response rates to e-mails with Arabic sounding names, irrespective of the type of request and the banks' locations.

C. Supplementary Analyses
Furthermore, we find no significant differences in the discrimination rates between the two types of requests and urban and rural areas, respectively.
Although it is not the primary focus of our paper, two additional results may be noteworthy: First, the difference in response rates between the loan-(47.5%, sem = 2.0%) and investment-related requests (39.4%, sem = 2.0%) of 8.1pp is statistically significant (χ 2 (1) = 8.110, p = 0.004). Furthermore, the difference in response rates between the loan-and investment-related request for domestic sounding names (9.3pp) turns out to be statistically significant (χ 2 (1) = 5.355, p = 0.021), although the difference for Arabic sounding names (6.9pp) does not significantly differ from zero (χ 2 (1) = 3.376, p = 0.066). The lower response rates in the investment domain might be driven by higher potential revenues from the loan business or by banks' current excess liquidity in the Euro area, both potentially rendering the investment business less attractive for banks. Second, we report level effects in response rates for the variable area. The response rates of 46.0% (sem = 1.6%) in urban areas is significantly larger than the response rate of 30.9% (sem = 3.2%) in rural areas (χ 2 (1) = 15.896, p < 0.001). The This might be due to the fact that e-mail inquiries are more common in urban areas. However, both explanations, for the differences in response rates over types as well as areas, are only speculative and our data set does not allow to infer where these level effects stem from.
In addition, the observed discrimination effect also persists for the interaction of both variables, type and area, as can be seen in Fig B. The response rate for loan-related requests in urban areas is 60.2% (sem = 3.0%) for domestic sounding names, compared to 39.2% (sem = 3.1%) for Arabic sounding names (χ 2 (1) = 22.508, p < 0.001). For investment-related requests in urban areas, response rates for domestic and Arabic sounding names are 53.7% (sem = 3.2%) and 30.9% (sem = 2.9%), respectively (χ 2 (1) = 26.763, p < 0.001). In rural areas, loan-related requests sent with a domestic sounding name trigger a response rate of 57.4% (sem = 7.3%) as compared to a response rate of 16.4% (sem = 5.0%) for Arabic sounding names (χ 2 (1) = 18.730, p < 0.001). For investment-related requests in rural areas, response rates for domestic and Arabic sounding names equal 37.3% (sem = 6.3%) and 13.0% (sem = 5.0%), respectively (χ 2 (1) = 7.769, p = 0.005). Thus, the observed discrimination effect reported in the main text is also robust to the interaction of the factorial combinations, which further corroborates our findings. Results reported with respect to the rurality of the banks' location are based on a threshold of 10,000 inhabitants. Accordingly, a bank is considered being located in an urban area whenever the number of inhabitants of the location has at least 10,000 inhabitants; otherwise, the bank is considered being located in a rural area. Although a threshold of 10,000 inhabitants appears to be reasonable for differentiating between urban and rural areas in Europe, the classification has to remain somewhat arbitrary. For this reason, we test for the robustness of the reported results by varying the threshold between 5,000 and 25,000 inhabitants in steps of 2,500 and re-estimating all results reported for any of these levels as reported in Table C. It is reassuring that the effect reported in the main text as well as above are robust for this range of thresholds, i.e. discrimination effects remain highly significant and at similar levels, while differences in discrimination rates remain insignificant. Notes: Standard errors of the mean are reported in parenthesis. Significance indications are based on χ 2 (1) statistics for column and row differences and permutation tests (on differences in frequencies; 10,000 permutations) for differences-in-differences, respectively.  Logit regressions of the treatment effects reported in the main text and the supporting information are reported in Table E. For all model specifications, the dependent variable is an indicator variable, which is equal to one if the request triggered a response and zero otherwise. Models 2-5 include country controls. The difference in response rates for domestic sounding and Arabic sounding names is highly significant for the pooled data (Models 1 and 2), separated for the type of request (Model 3), separated for the rurality of the banks' location (Model 4), and controlling for both the type of request and the rurality of bank's location (Model 5). Marginal effects of discrimination rates are highly robust, varying between −22.7% and −23.7%, and highly significant for all model specifications. Differences in discrimination rates, for the type of request, and for the rurality of the banks' locations are statistically insignificant (as captured by the interaction terms Ethnicity#Type and Ethnicity#Area). Notes: Logit regressions. Estimates are reported as odds ratios; marginal effects are reported in brackets. Standard errors are provided in parentheses. * , * * , and * * * refer to the 5%, 1%, and 0.1% significance level, respectively. # indicates an interaction term. Correct predictions report percentage of correctly predicted occurrences of the variable Response by the model. The model degrees of freedom, df , of likelihood ratio tests are 1, 7, 9, 9, and 11 for models 1-5, respectively.