This rejoinder to the response by Adida, Laitin, and Valfort briefly addresses some of their specific arguments. Like my original comments, it is not meant to detract from the merits of a book that sets a new standard for empirical excellence, but to continue to show how difficult it is for even excellent scholars to produce policy recommendations on the basis of rigorous empirical research.

Regarding the broadcasting of findings, Adida, Laitin, and Valfort are likely correct that advertising the effects or irrationality of discrimination (particularly if it is condemned with the moral authority of the state) could nudge corporations to discriminate less against Muslims. However, the decision to include religion in state statistics and to give weight to findings on religious discrimination is a political, not an administrative, one in France. Resistance to it is often couched in a discourse that is so anti-religious in its laïcité that it puts French Islam and Muslims outside the scope of the republic. There is a dearth of indications that important French political parties are interested in reinterpreting their understanding of republican ideals to make room for French Muslims who refuse to eliminate all signs of their faith, a prerequisite for the state to start advertising research findings about discrimination against Muslims qua Muslims, instead of residents of “quartiers défavorisés.”

Adida, Laitin, and Valfort believe that their own research provides evidence to support their advice to Muslims to give their newborns neutral or French names in the hope of improving their lifetime incomes. Their finding that “Muslim Senegalese subjects with African (but not identifiably Muslim) first names receive more generous treatment from rooted French players than those with identifiable Muslim first names” is undoubtedly correct. The problem, however, is one of external validity. Both treatments in the Adida, Laitin, and Valfort experiment represent realistic personas in popular French consciousness (African Muslims and African non-Muslims). The majority of French Muslims, however, are North Africans from exclusively Muslim countries. While a West-African with an African, French, or Christian name would not raise eyebrows, someone with a French first name and a Maghrebin last name could inspire a more discriminatory attitude, particularly in the context of a popular far-right narrative that Muslims are an insidious threat to France. The Adida, Laitin, and Valfort findings | by themselves | therefore provide no guarantees that the effect of naming on treatment by French natives | they find in the case of the Senegalese | would extend to the majority of French Muslims.

In their book, Adida, Laitin, and Valfort lean on an article by Duguet et al. (Reference Duguet, Leandri, L’Horty and Petit2010) to make their policy recommendation because this study controls for Moroccan versus French first and last name effects on callback rates (p. 152). Duguet et al. find that applicants with Moroccan names have to submit ten times more applications than applicants with French names to get an interview. More relevantly, it concludes with the key finding that “indicating a Moroccan nationality on a résumé or having a Moroccan forename is less of a handicap than having a Moroccan-sounding surname” (p. 207). This finding suggests the opposite of what Adida, Laitin, and Valfort conclude: while some biased employers may accept a French first name as a sign of assimilation, or be distracted by it enough not to notice that an applicant is likely Moroccan, most recruiters who discriminate would not call back any Moroccan, regardless of the effort the applicant makes to assimilate. This finding suggests instead that Moroccans, and likely other French Muslims, might as well proudly stick to their culture and naming practices because the majority of discriminatory recruiters would not be mollified by their costly efforts to assimilate.

What Adida, Laitin, and Valfort show very well is that the French favor non-Muslims over Muslims, but they do not show that they are more generous to Muslims with a French first name than they are to Muslims with a Muslim first name. Their need to selectively rely on a secondary and partial finding from someone else’s research confirms the point I made in the original commentary—there is no avoiding stretching findings and relying on others’ scholarship to make policy recommendations when rigorous experimental research becomes narrow and costly.

Another way Adida, Laitin, and Valfort address criticism of their advice to Muslims to give their children neutral or native French names is to engage in a cursory ad hoc meta-analysis of other scholars’ work. As one would expect, the studies were conducted using different methods and focused on minorities in other contexts. This is a sensible way to proceed, and it also suggests that perhaps, just as medical researchers sometimes do when formulating health-related advice, it may be best to root policy recommendations in meta-analyses instead of a single, rigorous, narrowly-focused study, unless the study manages to directly and comprehensively research all the issues and considerations relevant to the policy being formulated. In this case, Adida, Laitin, and Valfort find the aggregate evidence to go both ways. Perhaps that is not enough to advise members of a marginalized minority to make costly cultural and economic bets on the future.

Regarding the use of correspondence tests for auditing compliance with anti-discrimination laws, my issue is not with the capabilities of the French bureaucracy. I am questioning whether correspondence experiments are useful to test with confidence whether a single corporation is systematically discriminating, as opposed to the existence of measurable discrimination within the labor market or within an industry. It would likely take thousands of paired applications submitted to a corporation with strategic recruiters who only call back a tiny proportion of applicants to prove statistically that the corporation is highly likely to be discriminating against a particular minority. To produce legally meaningful findings, such correspondence tests would be prohibitively costly and would have to be conducted over years for each corporation. They would only be useful, if at all, to audit discrimination by very large firms with large numbers of job openings.

Adida, Laitin, and Valfort are simply following a prevalent disciplinary practice by concluding their study with policy recommendations. Theirs also happen to be usefully thought-provoking, reasonable, and well-intentioned. Simultaneously, and also reflecting a common practice in the discipline, Adida, Laitin, and Valfort’s policy recommendations are not directly and rigorously rooted in their research because the demands of empirical rigor limit the range of the findings necessary to develop useful policy recommendations. Many within the discipline have to negotiate this tradeoff between rigor and relevance, and may not be aware of its pitfalls. We therefore may need to start a conversation about how to convincingly produce policy recommendations based on our research; how to ascribe a level of confidence to our policy recommendations based on the strength of their ties to robust research; whether probabilistic social science tools can be used to meet the needs of auditors, the courts, or policy makers beyond their ability to reveal broad trends and patterns; and how to develop research designs that better balance the requirements of rigor with the need for well-supported policy recommendations.