The actor who dominated Hindi cinema for two decades was born Muhammad Yusuf Khan in Peshawar, into a Pashtun fruit-merchant’s family. He took the screen name Dilip Kumar in 1944, and across a career of more than fifty years, he played a Muslim character exactly once. He was not an outlier. Madhubala was Mumtaz Jehan Dehlavi; Meena Kumari was Mahjabeen Bano; Nargis was Fatima Rashid. In the 1940s and 50s, the actors who topped the Indian box office were Hindu or Parsi, and the industry took for granted that a Muslim name could not carry a film.
Half a world away, the same decade produced an almost identical roster. Issur Danielovitch became Kirk Douglas; Emanuel Goldenberg became Edward G. Robinson; Bernard Schwartz became Tony Curtis; Betty Joan Perske became Lauren Bacall. Jewish performers anglicized their names to get past the antisemitism of the era and onto a marquee. The cleanest single case comes from Britain: Krishna Pandit Bhanji, son of a Gujarati father, became Ben Kingsley. He has described the mechanism in one sentence. “As soon as I changed my name, I got the jobs,” he told Radio Times, recalling auditions where directors admired the work but could not place a Krishna Bhanji.
I come at this as a labor economist. A name change is a revealed-preference statement. Each of these people judged that the cost of giving up their name was smaller than the cost of keeping it. That choice only makes sense if a name carries enough information, or triggers enough of a reaction, to move outcomes before any of the talent underneath it is observed. The anecdotes give us the mechanism.
The mechanism
Start with why a name would matter at all. A casting director working through a stack of submissions, or an employer working through a pile of résumés, is making fast decisions on thin information. The name sits at the top of the page. It is often the first thing seen and sometimes the only thing processed before a candidate is set aside. If the name, in the gatekeeper’s mind, is correlated with anything they are screening for, it does its work before the applicant gets a hearing. Changing it degrades the signal others use to exclude you. In a screening environment like that, the response is hard to call anything but rational. There is direct experimental evidence that the screening is this shallow. Bartoš, Bauer, Chytilová, and Matějka (2016) embedded tools to monitor how much attention employers actually paid to each application, and found that a negatively stereotyped minority name reduced the effort employers spent even reading the résumé. The name does not merely tilt a careful evaluation; it suppresses the evaluation, which is why better credentials buried inside the page so often fail to rescue the application.
The harder question is the size of the penalty being evaded, and here the anecdotes hit a wall. We observe Dilip Kumar and Ben Kingsley because they succeeded. We do not observe the equally talented people who kept their names and never broke through, because nothing records them. Inferring the strength of discrimination from the success of those who changed their names is selection on the outcome: reading the size of an effect off a sample that was chosen because the effect went a particular way. To get a real number, you need a design that holds talent fixed and varies only the name. That design exists.
What the audit studies measure
The tool is the correspondence study. Send employers fictitious résumés that are identical in every respect except the name, randomize which name lands on which résumé, and measure the callback gap. Because quality is held fixed by construction, any difference in callbacks is caused by the name and nothing else.
The canonical result is Bertrand and Mullainathan (2004). Résumés with white-sounding names like Emily and Greg drew callbacks 9.65 percent of the time; identical résumés with Black-sounding names like Lakisha and Jamal drew them 6.45 percent of the time. That is a 50 percent gap, produced by the name alone. In units that an economist can feel, the authors estimate that a white name was worth about as much as eight additional years of experience. Two further findings matter for what follows. There was little sign that employers were reading social class off the names. And a stronger résumé led to more callbacks for white names than for Black ones, so better credentials widened the gap rather than narrowing it.
The pattern is not confined to the United States or to the Black-white margin. Oreopoulos (2011) sent roughly thirteen thousand résumés to employers in Toronto. Applicants with common English names were contacted about 16 percent of the time; those with Indian, Chinese, or Pakistani names, on otherwise identical Canadian résumés, about 11 percent, a gap nearly as large as the American one. Listing language fluency, a degree from a selective school, or active extracurriculars did not close it. For résumés showing several years of experience, the applicant’s name mattered more for landing an interview than additional education or extra languages did.
The same design has been run inside the subcontinent itself, where the marked categories are not immigrant origin but caste and creed. The clearest evidence comes from India. Thorat and Attewell (2007) sent matched résumés to private firms and found that, for an equivalent qualification, a Dalit applicant’s odds of a positive response were about two-thirds those of a high-caste Hindu, and a Muslim applicant’s odds were only about a third. Banerjee, Bertrand, Datta, and Mullainathan (2009) ran the same exercise across more than three thousand applications to software and call-center jobs in Delhi, and found the penalty concentrated where screening is fast and impersonal: low-caste applicants to call centers had to send roughly a fifth more résumés to draw the same callback, a gap that all but vanished in software, where productivity is easier to observe.
Pakistan has nothing as systematic. No one has run the matched-résumé test on a Shia, an Ahmadi, or a Christian name there, so the formal evidence is thinner than India’s, even as the anecdotes abound. And where the Western experiments must hide the marked trait inside a name and infer the penalty from a callback rate, Pakistan sometimes prints the screen in the advertisement. Amnesty International’s 2025 investigation of sanitation work documents recruitment notices that specify “non-Muslims only,” and found that around eighty percent of the country’s sanitation workers come from religious minorities, overwhelmingly Christian, with more than half of those surveyed naming their caste or religious identity as decisive in how they were hired. The penalty here is not a quiet thumb on the scale; it is the eligibility rule, stated in advance. Late in 2025, the Islamabad High Court barred the phrase from such advertisements, which tells you it was common enough to need a ruling.
The gap has also been stubborn. Quillian, Pager, Hexel, and Midtbøen (2017) pooled all available hiring field experiments, including 28 studies covering more than 55,000 applications. Across studies since 1989, white applicants received on average 36 percent more callbacks than otherwise identical Black applicants, with no decline over twenty-five years. At the point of hire, a name that signals a minority race or origin carries a large and persistent penalty. Against a measured penalty of that size, changing the name is the optimal individual response.
Does shedding the name pay?
The audit studies measure the penalty attached to the marked name. The complementary question is whether the people who dropped it did better, and that has been estimated directly.
Biavaschi, Giulietti, and Siddique (2017) built a panel from early-twentieth-century US naturalization records and tracked immigrants’ naming choices over time. Those who Americanized their names climbed further up the occupational ladder than those who did not, and the ones who adopted very common American names like John or William gained at least 14 percent in occupation-based earnings.
Arai and Skogman Thoursie (2009) found the same in Sweden, with a cleaner natural experiment. Among immigrants from Asian, African, and Slavic countries who changed to a Swedish-sounding or neutral surname during the 1990s, annual earnings rose substantially after the change. Two features make the result persuasive. There was no earnings trend before the change, which rules out the story that already-rising people change names. And there was no gain for those who switched from one foreign name to another equally foreign one, which rules out the story that changing a name is merely correlated with effort. What moved earnings was shedding the name specifically.
The historical extreme is passing. Nix and Qian (2015), using the full count of US censuses from 1880 to 1940, document that more than 19 percent of Black men passed for white at some point in their lives, with passing accompanied by relocation to whiter communities, concentrated in the North, and associated with better economic opportunity. Later work by the same authors, using two-sided record linking, revises the rate downward, so treat 19 percent as an upper bound rather than a settled figure. People paid extraordinary personal costs, cutting ties with family and leaving home, to escape a penalty attached to a category. A surname change is the same move at a far lower price.
Why the famous cases understate the cost
Now back to selection. The intuitive reading of Dilip Kumar and Ben Kingsley is that they became enormous after changing their names, so the discrimination they evaded must have been enormous. The logic is shaky for the reason already given: they are visible precisely because they won. The counterfactual we want is the same man, with the same talent, in the same universe, the only thing altered being the name on the marquee. History does not issue a spare universe to check against, which is the fundamental problem of causal inference in its barest form: one run per person, and in this one, he changed the name. Kingsley’s testimony is unusually good evidence, since it is a within-person before-and-after with talent roughly held constant, but it is still one person, reporting after the fact.
The sharper point is that the stars are the wrong place to look for the cost. Suppose getting cast, or hired, is a threshold rule: you clear the bar if your perceived quality, which is your true talent minus whatever the name subtracts, exceeds some cutoff. Then the careers that the name penalty decides are not the superstars. They are the marginal people, the ones who would have just cleared the bar under a different name and fallen just short under their own. A Dilip Kumar clears the bar either way; the penalty costs him some delay and some roles, not his career. The people the penalty truly sinks sit near the margin, and those are exactly the people we never hear about, because failing to get cast produces no record. The visible name-changers, therefore, understate the damage. The real cost sits in the talent that stayed on the wrong side of the cutoff and never surfaced.
That cost can be measured in the aggregate. Hsieh, Hurst, Jones, and Klenow (2019) begin from a striking fact: in 1960, 94 percent of American doctors and lawyers were white men; by 2010, the figure was 62 percent. Since the innate distribution of talent for those professions is not plausibly different across groups, the convergence implies that in 1960, a large pool of talented women and Black men were not in the jobs that fit their abilities. Running that reallocation through an occupational choice model, the authors attribute between 20 and 40 percent of the growth in US output per person since 1960 to the improved allocation of talent as frictions fell. The talent was there the whole time. Discrimination held it in the wrong place, at a price of a fifth to two-fifths of half a century of growth.
Whose taste?
Economics splits discrimination along two axes, and both matter here. The first axis is where the distaste sits. Following Becker, it can live with the employer, with coworkers, or with customers. The second axis is what kind of thing it is. Taste-based discrimination is animus: the decider, or whoever the decider answers to, simply prefers not to deal with the group and will give up money or quality to avoid them. Statistical discrimination carries no animus; the name is a noisy proxy for something the decider cares about but cannot observe, so, under uncertainty, they fall back on the group average, and an individual is sorted by the mean of people who share their name.
I think both film cases are cases of customer discrimination. The studio is a profit-maximizer. It does not rename Issur Danielovitch because it dislikes him; it renames him because it believes ticket-buyers will pay less to see that name on a marquee. The distaste lives with the audience, and the studio is a conduit for it. The giveaway, in the Hollywood case, is who ran the studios. The men who built the industry were overwhelmingly Jewish immigrants, and they were the ones changing Jewish actors’ names. That rules out employer animus almost by construction: a Jewish studio boss renaming a Jewish star is not acting on his own prejudice, he is betting on his gentile customers’. Bollywood has the same shape. The premise that a Muslim name could not headline a film was a claim about the audience, transmitted by studios responding to what they assumed the audience wanted.
Where does statistical discrimination enter, then? Mostly as the form of the decision, not its substance. A studio uncertain about any given film’s box office uses the lead’s name as a proxy for expected revenue, which looks like statistical discrimination. But the thing the proxy stands in for is the size of the audience’s taste-based aversion. The inference is statistical in form and taste-based in substance, with the taste sitting one step downstream, in the consumer. So the live question is not taste versus statistics. It is whether the studios were reading a real customer taste or enforcing a stereotype about their own audience. If moviegoers really had stayed home, the studios were tracking a genuine preference, ugly but real. If not, they were policing a belief about their customers that no one ever tested, because everyone changed names, and the experiment was never run. The welfare verdict turns on which it was: a real third-party distaste the actor paid to evade, or a phantom one the industry manufactured and then obeyed.
The résumé experiments speak to a different channel, employer screening rather than customer demand, but two of their findings travel. The name penalty is real and causal where information is thin. And it does not behave the way statistical discrimination about productivity should: in Bertrand and Mullainathan, a better résumé widened the racial gap instead of closing it, and in Oreopoulos, fluency and an elite degree failed to move the penalty on an ethnic name. If the gap were due to rational updating under uncertainty, more information should have narrowed it. That it did not is the signature of taste rather than statistics, which is what the customer-driven reading of the film cases would predict.
That thin-information gate is now being rebuilt in software, and AI makes the rebuild cut both ways. Most large employers already run applications through automated screens before a human sees them, and the next generation of these tools is built on models trained on past hiring decisions. A system that learns from a history in which Lakisha and the Dalit applicant were filtered out will reproduce the filter at machine speed and machine scale, with the added hazard that it can rediscover a forbidden name through proxies, a postal code, a school, a turn of phrase, long after the name itself is redacted. The same technology, pointed the other way, is the cleanest blind audition ever available: strip the name, the address, and the photograph, score only the work, and the gate that punished Krishna Bhanji need never see what it used to screen on. Which way it cuts is not a property of the machine but a choice about what we ask it to optimize and what we are willing to measure afterward.
I would resist reading these stories of name-shedding as triumphs over adversity. They are better read as data. Every one of them is a person stating that the penalty on their name was great enough to be worth the real cost of escape. The people we remember are the surviving tail of a distribution whose left side leaves no trace. Dilip Kumar and Ben Kingsley got through. The quantity a policymaker should care about is not them. It is everyone with their talent and their original name who did not, and whom the design of a résumé screen, then and now, was built not to see.