The World Bank Economic Review, 38(2), 2024, 209–228 https://doi.org10.1093/wber/lhad039 Article Removing Barriers to Entry in Medicine: Evidence Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 from Pakistan Fatima Aqeel Abstract In 1992, Pakistan equalized admissions criteria for women and men applying to medical schools, causing a rapid increase in the female share of medical graduates. Using birth cohort variation, I find that equalizing admissions criteria increased employment among female doctors by 21 percentage points and among doctors overall by 9 percentage points, even though female doctors are less likely to be employed than male doctors. Earnings for male medical graduates increased as lower ability males were crowded out. The 1992 reform led to increased gender diversification in a wide range of medical specialties, but it also concentrated doctors in urban districts where women prefer to practice. JEL classification: O15, J16, I2 Keywords: development, gender, education 1. Introduction In contexts where there is a shortage of working professionals, institutions and states face a trade-off. On the one hand, educating women is an important objective linked to the Millennium Development Goals (United Nations 2004). On the other, if women are less likely to work than men, there may be efforts to restrict women’s access to school. Medical schools (and sometimes other professional schools) are at the forefront of this trade-off. As recently as 2006, Singapore had a quota on the share of women admitted to medical school. Prior to 1992, Pakistan had a combination of a quota and a higher test- score requirement for female applicants. China allegedly adjusted enrollment ratios in higher education to favor men in 2017. In Japan, women’s test scores were artificially lowered after 2006 in an effort to exclude them from medical school.1 One motivation for entry restrictions is the concern that women exit the labor force at a higher rate than men do. In China, a government official explained, “In view of considerations of national interest, to meet personnel training needs in some job areas or specialties, Fatima Aqeel is a assistant professor in the Department of Economics, Colgate University, Hamilton, NY. Her email is faqeel@colgate.edu. The author thanks her advisors Samuel Bazzi, Dilip Mookherjee, and Daniele Paserman for their con- tinuous guidance and support; Sohail Javed, Ambreen Fatima, and the Applied Economics Research Center, Karachi, for generously sharing the administrative and labor-force survey data used in the project; seminar discussants at Boston Univer- sity, Colgate University, GDPC, NEUDC, and LAGDEV; and David N. Weil, Martin Fiszbein, Kevin Lang, Yuhei Miyauchi, Carolina Castilla, Jesse Bruhn, Undral Byambadalai, Thea How Choon, Giovanna Marcolongo, Nicole Simpson, and Patrick Power for helpful discussions. Samay Gupta and Zixing Guo provided expert research assistance. A supplementary online appendix is available with this article at The World Bank Economic Review website. 1 Singapore Straits Times, 2017; Pakistan, BBC News, 2014; China, New York Times, 2017; BBC News, 2018. C The Author(s) 2024. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 210 Aqeel a few colleges may appropriately adjust the enrollment ratios of men and women.”2 And in Pakistan, the context of this paper, a doctor proponent of reinstating the women’s quota in 2014 argued, “It’s not discrimination.... This shortage of doctors is the biggest challenge to Pakistan’s health system.”3 Only 53 percent of Pakistani female medical graduates (MG) who attended medical school before 1992 reported working, whereas 86 percent of male MG did (LFS 2012). Despite concerns over women’s dropout rates, Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 entry restrictions are sometimes abandoned due to their controversial and discriminatory nature. In this paper, I examine the labor-market effects of removing admissions restrictions on women in Pakistani medical schools. In particular, I study how equal opportunity affects MG employment, earnings, and the distribution of medical practice to understand some of the trade-offs of the reform. Before 1992, the proportion of women in Pakistani medical schools was capped at 20 percent. Within this quota, women were required to score 8 percent more points than men in the overall admissions score to be admitted.4 The policy was changed to equal opportunity in 1992, following a successful court case by a group of women who alleged they were denied admission to medical school despite scoring higher than men.5 In my analysis, I show first that the new admissions policy abruptly increased the female share of MG by 10 percentage points in a single year, with a steady increase thereafter. Next, using labor-force survey (LFS 2012) and medical registration data sets, I find that the admissions reform increased employment among female MG by 21 percentage points. The increase was higher for female doctors in their late twenties and thirties, who are traditionally the most likely to quit. This effect was also large enough to weakly boost overall MG’s employment by 9 percentage points relative to other college graduates, as male MG’s employment was not significantly affected. Though women are less likely to work than men on average, the policy change did not reduce the overall supply of medical doctors. Meanwhile, male MG who were not crowded out of medical school experienced a 42 percent increase in earnings. Female MG earnings did not experience a significant change, though overall MG earnings increased by 29 percent. The increase in earnings is consistent with increased average ability as the ad- missions process became more competitive after the reform. In addition to changes in the Pakistani medical labor market, the reform caused changes in medical specialization and the spatial distribution of doctors. Historically, female doctors dominated gynecology but were rare in other specialties. The reform led to greater diversity in several medical specialties which could positively affect patient care downstream (Alsan, Garrick, and Graziani 2019; Powell et al. 2019; Greenwood, Carnahan, and Huang 2018). In contrast, individuals residing in remote and rural areas may have been hurt by the policy. Women are substantially more likely to practice in urban areas than men and new female doctors registered almost exclusively in cities during this period, while the rural medical labor force did not experience substantial changes. Due partly to the rapidly rising population, the ratio of doctors per person decreased over this time period, and the reduction was twice as large in rural areas than in urban areas. These outcomes point to policies that encourage a more equal distribution of doctors and further retention of female MG, instead of restrictive admissions quotas in schools. My paper contributes to the literature on encouraging women to enter the medical field and improving health care in developing countries. Patients can benefit from greater diversity among doctors. For exam- ple, Alsan, Garrick, and Graziani (2019), Greenwood et al. (2020), and Greenwood et al. (2018) find that patients have improved health outcomes when treated by a doctor more similar to themselves. Female 2 “Women in China Face Rising University Entry Barriers,” New York Times, 2017. 3 “Are Pakistan’s Female Medical Students to Be Doctors or Wives?,” BBC News, 2015. 4 The score cutoffs were 820 for women in their seat quota, and 731 for men in their seat quota, out of a total of score of about 1,100. 5 The old policy was instituted in 1948 for non-discriminatory reasons meant to ensure some seats for women in medical school. Over time, it became a ceiling as women’s demand for medical education exceeded the seats allocated to them The World Bank Economic Review 211 patients also prefer consulting with female doctors in several contexts (Waller 1988; Kerssens, Bensing, and Andela 1997). Moreover, due to different work styles, female doctors can outperform male doctors in certain cases. Wallis et al. (2017) and Tsugawa et al. (2017) find that patients treated by female doctors have lower 30-day mortality rates and readmission rates than those treated by male physicians. Despite substantial benefits to engaging women in medicine, there are valid concerns that female doc- Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 tors may be more difficult to retain. Female doctors often quit or shift into part-time work within six years of graduating from medical school (Frank et al. 2019). Policies to attract female doctors have in- cluded introducing flexible work times and capping residency hours (Wasserman 2023). But the higher quit rates and need for retention policies can be seen as costs of employing female doctors. An alternative type of policy is to limit women’s admissions to medical school. To the best of my knowledge, restrictive policies have not previously been explored in this context. My paper analyzes the labor-market effects of removing one such policy, and shows that restricting admissions can have unintended consequences. Improving the quantity and quality of doctors is a second major health-care goal among developing countries. Doctors moving to more lucrative markets abroad is a major reason why developing countries are left with fewer doctors (Docquier, Lohest, and Marfouk 2007).6 The value of a seat in medical school is especially high in this context, and serves as an impetus for implementing gender quotas in admissions, since female MG are less likely to work than male MG. This paper shows that removing gender quotas leads to small increases in the supply of doctors in the short run. Doctor retention is most challenging in rural areas, where low wages, poor working conditions, and the lack of support, equipment, and infrastructure detract doctors (Lehmann, Dieleman, and Martineau 2008). As a result, health-provision disparities persist between urban and rural areas (Das et al. 2022; Strasser, Kam, and Regalado 2016). This paper shows that adding female MG to the doctor labor force can exacerbate the urban–rural doctor disparity. The remaining paper is structured as follows. In the Background section, I provide contextual back- ground and institutional details. I discuss the identification strategy and relevant data sets in the Empirical Strategy section, followed by a section on the Results, Effects on the Distribution of Doctors, Robustness Checks, and Conclusion. 2. Background 2.1. Medical Schools and Health Care in Pakistan Health care in Pakistan represents approximately 3 percent of GDP annually (World Bank 2018) and is provided by both the public and private sectors. Though the private health-care sector is small, it caters to a large proportion of the urban population. In 2016, 35 percent of all physicians worked in the private sector and were consulted by 75 percent of the urban population at least once (World Health Organization 2016). Medicine has historically been a lucrative profession in Pakistan. Students can apply to medical school or another college at the end of high school after taking the appropriate pre-college subjects. According to the Pakistan Medical and Dental Council (PMDC), medical-school admissions are determined by an entrance exam score (which accounts for 50 percent of the overall candidate score), high-school grades (40 percent of the overall score), and middle-school grades (10 percent of the overall score). Before the year 2000, a candidate’s score was determined by high-school and middle-school grades alone (Andrabi and Singh 2015). After the standard five years of medical school, MG can register to practice with their basic M.B.B.S. degree,7 or pursue a residency and fellowship in a specialty. Meanwhile, other bachelor’s- degree-granting colleges typically take four years to complete. Of the 19 medical schools, 17 were publicly 6 https://www.nytimes.com/2022/01/24/health/covid- health- worker- immigration.html. Other reasons for doctor scarcity include doctor selection and burnout (Al-Shamsi 2017). 7 M.B.B.S. stands for Bachelor of Medicine, Bachelor of Surgery. It is equivalent to being a Doctor of Medicine. 212 Aqeel funded in 1992. The typical cost of a private school is two to three times that of a public school, and tuition for public medical schools is especially subsidized.8 2.2. The Policy Change Due to a shortage of female doctors after Pakistan’s independence in 1947, a women’s medical college Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 with a class size of 36 was initially established in 1948 (Fatima Jinnah Medical College). The shortage of female doctors persisted, however, and 20 percent of seats in all other public and private medical colleges were subsequently reserved for female students. The remaining 80 percent of seats were allocated to male students. Eventually, the demand for female medical education increased because medicine is regarded as a female-appropriate profession, and the number of female applicants outgrew the number of reserved female seats (Masood 2019). Additionally, in the early 1980s, concerns about the quality of medical education led to a reduction in the total number of medical seats. Within the women’s quota, the entry threshold of scores increased and became 8 percent higher for women than for men (Andrabi and Singh 2015). The turning point for women’s admissions into medical school was in 1989 when a group of women led by Shrin Munir filed a case against the existing admissions policy, alleging that they were denied ad- mission despite scoring higher than several men. A court ruling decreed that the old admissions policy was unconstitutional, stating that “the selection procedures violated [sections of] the Pakistani Constitu- tion prohibiting discrimination...on the basis of sex” (Shrin Munir and others v. Government of Punjab through Secretary Health, Lahore and another (PLD 1990 SC 295).9 A few months later, a similar case was won in the province of Sindh.10 Subsequently, the admissions policy was changed in 1991 to being based purely on merit. Women and men had an equal opportunity of being admitted to medical school for all cohorts starting in 1992. 3. Empirical Strategy 3.1. Identification Strategy The admissions reform in 1992 substantially decreased entry barriers for women who applied to medical school, and its benefit varied exogenously based on a woman’s birth cohort. I do not have data on when an individual attended medical school. Instead, I exploit the fact that individuals’ exposure to the reform varied exogenously with their age. For example, if a woman was aged 16 or less at the time of the policy in 1992 (born in or after 1976), she had a greater opportunity to prepare for medical-school applications than previous cohorts did. This includes selecting the appropriate pre-medical subjects in high school and studying for and taking the entrance exam. To account for trends over birth cohorts, I estimate changes in outcomes relative to the outcomes of other college-educated individuals (women and men where relevant). This is a group of educated individ- uals that closely resembles MG. I therefore use a standard difference-in-differences strategy that exploits variation across birth cohorts and between two or more groups as in Lucas (2010) and Bleakley (2010). My main estimation examines how MG outcomes evolve relative to other college-graduate outcomes across birth cohorts using event studies. For the main analysis, I divide birth cohorts between 1970 and 1980 into six roughly equal-sized groups so that there are three pre-reform cohorts and three post-reform 8 Sources: Prospectus’s of the Aga Khan Medical College, Karachi Medical and Dental College, Lahore University of Management Sciences, Karachi University. 9 Available from https://pubmed.ncbi.nlm.nih.gov/12344101/. 10 Available from https://sys.lhc.gov.pk/appjudgments/2014LHC6198.pdf. The World Bank Economic Review 213 cohorts.11 I estimate coefficients for each birth cohort group using the following specification : 2 Yict = β0 + βk D k ict ∗ Deg.Medict + β3 Deg.Medict + γ Xict + θt + θc + uict , (1) −3 where Yict is an outcome for a medical graduate or college graduate i, born in birth cohort c, and observed Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 in survey year t. The outcomes examined are the likelihood of employment for women, the likelihood of employment overall, and monthly earnings. The Dk ict are birth cohort group indicators which are interacted with an indicator for having a medical degree (“Deg.Med.”). The effects of being born in a birth cohort are captured by the fixed effects θ c and of being surveyed in a specific year by the fixed effects θ t . The matrix Xict includes individual level controls for age, age-squared, and city fixed effects to account for differences between cities, such as in labor markets and culture. Note that age effects, birth cohort fixed effects, and survey year fixed effects can simultaneously be estimated because multiple survey rounds are used, so the three are not perfectly collinear.12 Finally, uict is a disturbance term that captures unobserved variables that affect the outcome. The event study estimates are presented in the Results section. In equation (2), I estimate the difference-in-differences effect for all affected cohorts. I define the term “Treated” as equal to 1 if an individual was born in or after 1976. This term is interacted with an indicator for having a medical degree in the regression. I estimate the following regression and present the main findings in the Results section: Yict = β0 + β1 (Treatedc ∗ Deg.Med.ict ) + β2 Deg.Med.ict + γ Xict + θt + θc + uict . (2) A central concern of my identification strategy is that other college graduates could indirectly be af- fected by the policy change as well. Some women who pursue medicine after the policy may plausibly have pursued another college degree if the admissions process had not changed. To address this, first I argue that MG are a very small fraction of other college graduates13 . The movement of women into medicine would substantially change the MG average, but not change the other college average by much. In fact, if every woman who attended medical school post policy was sourced from another college program, the total would comprise only 3.6 percent of post-policy cohorts of other college women. This is an up- per bound. Second, I plot event studies of MG without any control group that could confound results (fig. S1.9), and I compare MG separately to engineers, other bachelor’s graduates, and postgraduates in table S2.2. While I cannot rule out the fact that other college graduates are an imperfect control group, these strategies serve to mitigate concerns about using other college graduates as a control group. A second concern is that delayed decision making in medical-school applications, or grade repetition, would imply that some individuals currently in pre-policy-change cohorts could be exposed to the policy. This would underestimate the main effects. Nevertheless, in a second robustness exercise I repeat my anal- ysis after omitting birth cohorts that could be partially treated. The results of this exercise are presented in table S2.2 and are consistent with my main findings in the Results section. Notably, I study the immediate effects of the reform which are estimated using birth cohorts just around the reform. Given the short time horizon, it is likely that treatment was uniform and did not change signif- icantly over this period. Moreover, other colleges also did not experience a similar reform and treatment was not staggered for any group. The coefficient of interest therefore is plausibly not biased by effects from other periods as may sometimes occur in a difference-in-differences specification (De Chaisemartin and d’Haultfoeuille 2020). 11 The grouped birth cohorts are 1970, 1971–1972, 1973–1974, 1975–1976, 1977–1978, and 1989–1980. The grouping was generated by statistical software. Each individual birth cohort has a small number of doctors, leading to unstable coefficients that are reported in fig. S1.10. 12 For example, a 30-year-old in the data could have been surveyed in 1990 and born in 1960, or be surveyed in 2010 and be born in 1980. 13 Only 3 percent of women with greater than a high-school degree are MG in the data, as I describe below. 214 Aqeel Finally, although the court’s decision was anticipated to some extent by medical schools, it was seem- ingly unanticipated by the general public (Andrabi and Singh 2015) and does not appear to have received wide coverage in newspapers at the time.14 The level of anticipation by the general public would affect how individuals invest in their education, and would make post-reform individuals different from pre- reform individuals. To some extent, this concern is mitigated by cohort fixed effects, which remove the Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 effect of fixed unobservable variables within cohorts, such as higher average preparation. Since public awareness of the policy likely grew over time, it is also mitigated by analyzing birth cohorts immediately around the reform in the short run. In the Results section, I present how these individuals were affected by the admissions reform. All sensitivity checks are reported in the Robustness Checks section. 3.2. Data Labor-Force Data. I measure labor-market outcomes using 13 repeated cross sections of LFS (2012) data between 1990 and 2012. I restrict my sample to individuals with at least a high-school education that reside in a city at the time of the survey.15 The reform would have affected individuals born in and after 1976. I use birth cohorts immediately around the reform to avoid comparing cohorts that are far apart in time and are therefore dissimilar.16 For the pre- and post-reform averages, I use the years between 1972 and 1978. For the event studies, which tend to require more observations, I use the years between 1970 and 1980. I restrict my sample to cities because these were most affected by the reform, and because urban and rural areas experience different health trends. There were major investments made in the rural health- care system around the time of the reform, such as the community-based Lady Health Worker program in 1994 and the Family Health Project in 1993 (World Health Organization 2016). To the extent that these programs substitute for medical doctors, they are likely to affect the medical labor market. Moreover, only 1.6 percent of rural women had greater than a secondary education at the time of the reform (DHS 1991). To isolate the effects of the medical-school reform and examine the population that is most affected by it, I focus on cities.17 Finally, I include individuals who are at least 20 years old, as they are more likely to have completed their education decisions.18 The resulting sample consists of 6,539 individuals in 13 cities. Table 1 provides sample summary statistics. Even in this urban sample between 1972 and 1978, only 3 percent of individuals have a medical degree, highlighting the elite nature of medical degrees. Most college-educated individuals obtain a non-medical or non-engineering degree. This includes every other bachelor’s degree, from biology and history to business administration.19 The surveys do not provide further information on the specific field of study. Additional Medical Doctors’ Data. I complement labor-force survey data with the database of doctor registrations from the Pakistan Medical and Dental Council (PMDC). A medical doctor is required to register with the PMDC if they intend to practice in any capacity in the country. The data therefore provide a census of approximately 246,000 registering doctors between the years 1952 and 2018. The data include the city and school from which a medical degree was obtained, as well as specialization field. The advantages of these data are the large number of observations, the lack of sampling variation 14 A search in the archives of the major national newspaper at the time, Dawn News, revealed no coverage in 1989–1991. 15 Cities are defined by the Pakistan Bureau of Statistics (PBS) using the administrative or municipal status of a locality (PBS 2023; Ullah 2022). 16 Based on speculation that the reform may have been anticipated by medical schools (Andrabi and Singh 2015) and the birth cohorts between 1974 and 1975 may be partially treated themselves, I extend the pre-reform years to 1972. 17 The main results hold with the inclusion of rural areas, but there are greater differences in pre-trends. 18 At age 20, 26 percent of individuals report working, including doing part-time work. 19 In the LFS (2012), 84 percent of women and 79 percent of men have less than a college degree. The World Bank Economic Review 215 Table 1. Sample Summary Statistics Mean Standard deviation Obs. Min Max Engineering degree 0.05 0.22 6,539 0 1 Medical degree 0.03 0.18 6,539 0 1 Other college degree 0.68 0.47 6,539 0 1 Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 Postgraduate degree 0.23 0.42 6,539 0 1 Age 31.33 4.99 6,539 20 40 Married 0.71 0.46 6,539 0 1 Employed 0.59 0.49 6,539 0 1 Earnings in 1998 USD 376 266 2,903 0 814 Women Pr(engineering degree) 0.03 0.16 2,786 0 1 Pr(medical degree) 0.03 0.17 2,786 0 1 Pr(other college degree) 0.73 0.44 2,786 0 1 Pr(postgraduate degree) 0.21 0.41 2,786 0 1 Age 30.89 5.18 2,786 20 40 Pr(married) 0.73 0.44 2,786 0 1 Pr(employed) 0.20 0.40 2,786 0 1 Earnings in 1998 USD 268 258 595 0 814 Men Pr(engineering degree) 0.07 0.26 3,753 0 1 Pr(medical degree) 0.04 0.19 3,753 0 1 Pr(other college degree) 0.64 0.48 3,753 0 1 Pr(postgraduate degree) 0.24 0.43 3,753 0 1 Age 31.66 4.83 3,753 20 40 Pr(married) 0.69 0.46 3,753 0 1 Pr(employed) 0.88 0.32 3,753 0 1 Earnings in 1998 USD 404 261 2,308 0 814 Source: Data are from repeated cross sections of the Labor Force Surveys of Pakistan, 1990–2012. Note: The table presents summary statistics of means, standard deviations, range, and the number of observations for de- pendent and independent variables used in the paper and supplementary online appendix. The sample consists of individuals aged 20–47 residing in cities, and with at least a college degree. The definition of being employed includes working for a wage or family gain for at least an hour last week, having a job even if the individual did not work last week, and working for a family business or farm. Earnings are nominal monthly earnings expressed in 1998 USD using an exchange rate of 1 USD = 43 PKR (US Department of State Archive 1998). because they cover all registrations, and the details of registration patterns over a long period of time and locations. 4. Results This section presents reform effects on the composition of MG and their labor-market outcomes. I start by highlighting the direct effects of the reform on the share of female MG. Then I report the effects of the reform on the following labor-market outcomes: (a) female MG’s employment (b) overall MG’s employment, and (c)MG earnings. 4.1. Reform Effects on Women’s Entry in Medical Schools After the change in admissions policy, the proportion of women in medical school rapidly increased as observed in fig. 1 and in the lower-left panel of fig. S1.1. Women born in or after 1976 were young enough at the time of the policy removal to be affected by the equal opportunity it provided. At that age they would still have the time to select pre-medical subjects, study, and take the entrance exam. Women born 216 Aqeel Figure 1. Policy Effect on the Proportion of Female Medical Students Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 Source: Data are from the Pakistan Statistical Yearbook, various years. Note: The graph shows the effect of the medical-admissions policy change on the share of women admitted to medical school. Points represent the female share of medical students enrolled in a given birth year, lines are local polynomial regressions fitted to the data. The medical-school-admissions policy change was effective for individuals born in or after 1976. before 1976 are more likely to have completed their educational decisions when the policy was changed and are less likely to be affected. Notably, the proportion of women in medical school before the policy change exceeds 20 percent (fig. 1). In earlier years, this is likely because of a women-only medical school which is included in the data, though I cannot rule out imperfect compliance. In the years around the reform, the share of women may also represent women who attended medical school at later ages. It may also represent the response of medical schools who may have anticipated the court’s ruling and made ad hoc adjustments to seats (Andrabi and Singh 2015). Without additional data, it may be difficult to disentangle the effects. In the Robustness Checks section, however, I explore how delayed educational decisions affect my main results. Despite the constraint not being exactly at 20 percent, it was binding. There is a noticeable jump in the female proportion of MG for birth cohorts starting in 1976, as evident in fig. 1. The magnitude is approximately 10 percentage points, after which the proportion rises steadily. Before the policy, on average 28 percent of MG are women, and after the policy change the proportion rises to 55 percent. One-third of the increase occurs in the single year of the policy change. This suggests that the old admissions policy meaningfully restricted women’s access to medical school. As fig. 1 suggests, some male MG were crowded out by female MG under the new, competitive admis- sions policy. Medical schools also concur with the fact that some male MG were replaced by female MG. For example, in 2014 one of the largest medical schools in Karachi reported that 70 percent of its students were now women.20 In the Results section , I show that the gender composition change in medical schools is mirrored in the composition of registering doctors as well. 20 Dawn News, 2014. The World Bank Economic Review 217 Figure 2. Probability of Employment—Women Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 Source: Data are from repeated cross sections of the Labor Force Surveys of Pakistan, 1990–2012. Note: The figure shows event study estimates based on equation (1) for the probability of employment for women in cities, aged 20–47 with at least a college degree. Each coefficient (circle) represents the employment probability of female MG relative to other female college graduates for a given birth cohort interval, and bars represent 95 percent confidence intervals. The base year is t = −1. Mappings from years since the reform to birth cohort groups are −3 = 1970, −2 = 1971–1972, −1 = 1973–1974, 0 = 1975–1976, +1 = 1977–1978, +2 = 1979–1980. Figure S1.1 plots the female share of graduates by birth cohort in comparable educational fields. The only other post-secondary-school fields available in the LFS (2012) are general bachelor’s degrees, engi- neering degrees, and postgraduate degrees (such as a master’s degree). The lower-right panel is an analo- gous graph for MG in the LFS (2012). The graphs show an increasing trend in female higher education across fields; however, only in medicine is there a discrete jump in 1976, as other fields did not experience a similar admissions policy change. 4.2. Employment for Female Medical Graduates Figure 2 presents an event study of female MG’s employment around the reform. Each coefficient repre- sents the policy’s effect on female MG born in a birth cohort relative to the same cohort of other female college graduates. Prior to the policy change, coefficients hover around 0, suggesting a lack of strong trend differences between the two groups. In contrast, there is a sharp and consistent increase in employment for female MG after the reform. The results suggest that female MG were more likely to be employed relative to other college females after the reform, despite being on similar trends prior to the reform. In fig. S1.10, I plot event studies using individual birth cohorts. The coefficients are more erratic, potentially due to small sample sizes, but point to a similar increase in female MG’s employment. In table 2, I report average reform effects for female MG. The estimation without any control variables (column 1) and with the full set of controls (column 2) finds that female MG’s employment increases by about 20 to 21 percentage points after the reform, relative to other women college graduates. This effect is statistically significant at the 1 percent level, even when using a wild bootstrap t-statistic to account for the small number of clusters (Cameron, Gelbach, and Miller 2008). 218 Aqeel Table 2. Probability of Employment for Female Medical Graduates Pr(employed) (1) (2) (3) (4) Treated × med. grad. 0.20∗∗∗ 0.21∗∗∗ — 0.74∗∗∗ Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 (0.02) (0.02) (0.011) [0.00] [0.00] [0.00] Treated age [20–25] × med. grad. — — 0.18 — (0.20) [0.58] Treated age [26–30] × med. grad. — — 0.26∗ — (0.13) [0.016] Treated age [31–36] × med. grad. — — 0.20∗∗∗ — (0.02) [0.00] Observations 2,786 2,786 2,786 2,786 R2 0.04 0.07 0.07 0.07 District FE N Y Y Y Cohort FE N Y Y Y Survey year FE N Y Y Y Mean DV 0.20 0.20 0.20 0.20 Source: Data are from repeated cross sections of the Labor Force Surveys of Pakistan, 1990–2012. Note: The table reports the reform’s effect on the employment of female MG relative to other female college graduates. Column (1) presents estimates from a basic difference-in-differences regression. Column (2) presents results from the differences in differences in equation (1) with all controls and fixed effects. Column (3) interacts the interaction term with indicators for three age groups: 20–25, 26–30 and 31–36. Column (4) reports marginal effects from a probit regression with the full set of controls. The coefficients on each of the own terms are included in the regression but not reported. The p-values associated with the wild cluster bootstrap-T method (Cameron, Gelbach, and Miller 2008) are reported in square brackets. Standard errors in parentheses are clustered at the birth cohort level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Column 3 separates the treated group into three different age groups: ages 20–25, 26–30, and 31–36.21 Typically, women’s employment in Pakistan peaks in their mid-twenties and declines in the late twenties and thirties as women marry (DHS 1991). After the reform, I find that female MG in their late twenties and thirties are significantly more likely to be employed than female MG in the same age groups pre- reform. This result suggests that single women are delaying marriage and working, or married women were returning to work, both of which are interesting cultural shifts. The female MG who were more likely to drop out, therefore, were now more motivated to be employed. Finally, in column 4 I estimate a probit version of equation (2) to vary the functional form of the estimation. The marginal effects reported in column 4 are much larger when using the probit functional form, but point to the same direction of results as my main estimates. Since the results in table 2 are for female MG relative to other college graduates, changes in other fields could be at play. I mitigate these concerns using two approaches. First, I estimate an event study of female MG employment without any control group (fig. S1.9). My results remain similar to those in the main analysis. Second, I use the PMDC database to plot the number and share of female MG who 21 The estimation used for age groups A1 , A2 , and A3 , corresponding to each of the age groups above, is Yict = β0 + β1 A1 ∗ Treated ∗ Deg.Med.ict + β2 A2 ∗ Treated ∗ Deg.Med.ict (3) +β3 A3 ∗ Treated ∗ Deg.Med.ict + β4 Deg.Med.ict + γ Xict + αt + αc + uict , where Xict includes city fixed effects and controls for age squared. The World Bank Economic Review 219 registered to practice before and after the reform. In fig. S1.2, the number of female M.B.B.S. doctors (panel a) and specialists (panel b) increased by nearly 50 percent. The female shares of employed MG show similar patterns using LFS (2012) data in panel (c) and registration data in panel (d). The increase was also experienced across most of Pakistan’s largest cities (fig. S1.4) and by public and private schools (fig. S1.5). Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 How Did the Reform Affect Labor-Market Prospects for Women? Although the share of women with college and postgraduate degrees has steadily increased in Pakistan, the fraction of female medical or engineering graduates is dwarfed by shares in other fields (fig. S1.6). The figure motivates why overall female employment would not rise as more females attend medical school. Despite this, the medical- school reform opened doors to a higher-earning profession for many women, especially in cities. The average pay for female MG over this time period (493 USD) is almost twice that of non-MG females with a college or higher degree (244 USD) and engineers (235 USD). To shed light on what female MG may have done had they not gone to medical school, I plot the female share of students in three different types of colleges by year (in fig. S1.7). These categories are those with a pre-existing high share of women, those with a pre-existing low share of women, and those that are female only, based on administrative data from Punjab province, which is the most populous province in Pakistan. The most visible changes are in fields that are male dominated (panel b), for example, in technology polytechnic institutes, engineering, and computer science (CS). Meanwhile, fields with high female shares largely remained on their pre-reform trajectory. In table S2.7, I examine the characteristics of female engineers around 1992. In general, I do not find large changes in characteristics among female engineers, which suggests that new female MG are not redirected exclusively from engineering only. Instead, they were likely drawn from a range of other colleges. Taken together, my results imply that women who would have entered male-dominated fields found an alternative in medicine, which is equally high earning and is considered more female appropriate (Masood 2019). Characterizing Post-Reform Female Medical Graduates Female MG’s higher employment may be due to productivity increases as they worked alongside more women, or because they felt rewarded for being admitted to medical school (e.g., a treatment effect). Or it could be that more motivated women attended medical school post-reform (e.g., a selection effect). While the policy’s structure makes distinguishing be- tween the two effects challenging, in this section I characterize how post-reform female MG were different from earlier cohorts along two dimensions: their ability and average household characteristics. The entry score threshold for women prior to the policy change was 820 out of 1,100 (74 percent), which was 8 percentage points higher than the male threshold of 731 (66 percent).22 Assuming that the reform equalized the threshold to 731, the marginal women who entered medical school after 1992 were likely those who scored between 66 percent and 74 percent. In general, female MG tend to score higher than their male counterparts even after the equalizing of entry thresholds, as suggested by the earliest test score data that I have from a medical school in 2001 (fig. S1.8). Merit-based admissions therefore likely raised the competitiveness and ability of medical-school applicants. Further, test scores are also positively linked to the likelihood of registering to practice in this context (table S2.5). It is therefore surprising that lowering entry barriers for women in medical school led to higher employment among female MG. I can also use the LFS (2012) to characterize certain household characteristics of MG. Medical gradu- ates are typically from urban, middle-, and high-income backgrounds (Zaidi 1986). Before the reform, 56 percent of female MG are high income and 35 percent are middle income (table S2.6). After the reform, female MG were slightly more middle income and had fewer doctors in the family, though most observed differences between pre- and post-reform female MG are not statistically significant. Overall, the charac- teristics of female MG do not appear to have changed drastically. Rather, the reform potentially allowed 22 This was the combined result of a 20 percent quota on women in medicine, and seat reductions in medical colleges which differently affected men and women. 220 Aqeel entry for women who were previously at the cusp of having the resources (and the connections) needed to attend medical school. This could be one motivating factor that led to higher female MG employment. Interestingly, there is suggestive evidence that female MG also married later: after the reform fewer MG are married by age 25, but more are married by age 35. These women may therefore have simultaneously delayed marriage and prolonged their careers.23 Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 An important caveat to these statistics is that individuals are observed only in their current household in the LFS (2012), and neighborhood type reflects their current neighborhood. Traditionally, most married women live with their spouses and single women with their parents. I cannot rule out the possibility that income background reflects marriage matching patterns, for example if low-income women married up. However, this might be a greater concern if I observed an increase in high-income women after the reform, instead of more middle-income women as I do. 4.3. Employment of Medical Graduates Overall Two effects could be at play in determining the employment of doctors overall. On the one hand, female MG were 33 percentage points less likely to be employed than male MG pre-reform, and overall doctor employment could decrease as more women enter medicine. On the other hand, female MG’s employment increased post-reform, while male MG employment was not significantly affected (table S2.4). This second effect could pull up the overall MG employment. Tracing MG’s employment over birth cohorts relative to other college graduates (fig. 3) shows a positive effect for the birth cohorts 1977–1978 (t = +1). Meanwhile, the coefficients for time periods t = −2 to t = 0 are all similar in size and not significantly different from zero, suggesting similar trends between MG and other college graduates right before the policy change. Table 3 examines the overall probability of employment using LFS (2012) data. In column 2 with the full set of controls, overall employment increased by 9 percentage points after the reform relative to other college graduates. This coefficient is significant at the 10 percent level, which is weak but plausible given the two opposing factors at play. Column (3) shows that the increase was larger for individuals in their late twenties, which is similar to the findings for women. Together, the results suggest that the increase in overall employment stemmed from female MG’s employment increase after the reform. To mitigate the concern that the changes observed in table 3 are driven by changes in other fields, I reinforce my results by plotting an event study using MG only (fig. S1.9). Overall, I find that equalizing admissions restrictions did not worsen doctor employment, and the increase in female MG employment was important. 4.4. Earnings The altered average ability and gender composition in medical school after the reform could be reflected in MG earnings. I examine how MG earnings were affected by the reform using self-reported monthly earnings in the LFS (2012). Wages for doctors in public institutions are predetermined and there is no known institutional difference in pay by gender. There is scope for a gender gap in earnings for those in private practice, however, in addition to differences across medical practices, such as the amount of time spent with each patient and the number of procedures performed. For my main earnings analysis I subset the LFS (2012) to the years between 1994 and 2012 and to individuals who work for a positive income.24 In table S2.1, I present reform effects using three other data 23 The LFS (2012) provides information on the number of unmarried children in the house, and not on each woman’s children within the house. This can be a concern in joint families. Since the number of female MG and engineers is small, the data was manually checked to verify that the number of children likely corresponds to a female MG’s own children. 24 This is the preferred sample for my analysis, though I repeat my analysis with different samples. Surveys from 1990– 1994 made changes in the way that earnings were coded. For example, they code earnings for unemployed individuals The World Bank Economic Review 221 Figure 3. Overall Employment of MG Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 Source: Data are from repeated cross sections of the Labor Force Surveys of Pakistan, 1990–2012. Note: The figure shows event study estimates based on equation (2) for the probability of employment for men and women in cities, aged 20–47, with at least a college degree. Each coefficient (circle) represents the employment probability of medical graduates relative to other college graduates for a given birth cohort, and bars represent 95 percent confidence intervals. The base year is t = −1. Mappings from years since the reform to birth cohort groups are −3 = 1970, −2 = 1971–1972, −1 = 1973–1974, 0 = 1975–1976, +1 = 1977–1978, +2 = 1979–1980. samples: (a) individuals who work with positive earnings in the survey years between 1990 and 2012, (b) all individuals who work (including those who earn a 0 income) in the survey years between 1994 and 2012, and (c) all individuals in the survey years between 1990 and 2012, with earnings for unemployed individuals coded as missing before 1994 to match their coding after 1994. In fig. 4, I report the effects of the reform on log earnings of male, female, and overall MG. For each group, the coefficients for pre-1976 birth cohorts are similar in size and are not statistically different from 0. For men, earnings increase for the t = 0 and more strongly for the t = +1 birth cohorts. Meanwhile, there are no statistically significant effects for female MG post-reform. Table 4 confirms these findings. Post-reform, male medical earnings increase by 23 percent relative to other college graduates (column 4). This implies an additional seven days of earnings for male MG. Overall, earnings for MG increase by 16 percent, or about five days’ worth of additional earnings. The results do not change substantially when hours worked are controlled for (table S2.1). The increase in male MG earnings is consistent with the average ability of male MG improving, as weaker male candidates are replaced by female candidates. Cognitive ability is linked to better labor- market outcomes (Lindqvist and Vestman 2011; Blau and Kahn 2005; Cawley et al. 2001). Academic success could be a precursor to placement at a more lucrative institution and having higher earnings post-graduation. While male earnings could also be affected by the smaller proportion of male MG post- as 0, mixing them with individuals who are employed and not earning positive amounts. Meanwhile, surveys after 1994 code earnings for unemployed individuals as missing. These largely unexplained coding standards make the surveys before 1994 less reliable. Further, 14 men in this urban college or higher-educated sample reported working with no positive income, mostly due to non-standard work arrangements such as working without a formal contract or being a casual wage employee. These have been excluded from the preferred sample because they are plausibly outliers for this group. 222 Aqeel Table 3. Probability of Employment for Medical Graduates Pr(employed) (1) (2) (3) (4) Treated × med. grad. 0.10∗∗ 0.09∗ — 0.83∗∗∗ Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 (0.03) (0.05) (0.03) [0.02] [0.17] [0.05] Treated age [20–25] × med. grad. — — 0.13 — (0.14) [0.73] Treated age [26–30] × med. grad. — — 0.16∗ — (0.07) [0.00] Treated age [31–36] × med. grad. — — 0.06 — (0.04) [0.11] Observations 6,539 6,539 6,539 6,539 R2 0.01 0.09 0.09 0.09 District FE N Y Y Y Cohort FE N Y Y Y Survey year FE N Y Y Y Mean DV 0.59 0.59 0.59 0.59 Source: Data are from repeated cross sections of the Labor Force Surveys of Pakistan, 1990–2012. Note: The table reports the reform’s effect on the employment of MG relative to other college graduates. Column (1) presents estimates from a basic difference-in-differences regression. Column (2) presents results from the differences in differences in equation (1) with all controls and fixed effects. Column (3) interacts the interaction term with indicators for three age groups: 20–25, 26–30, and 31–36. Column (4) reports marginal effects from a probit regression. The coefficients on each of the own terms are included in the regression but not reported. The p-values associated with the wild cluster bootstrap-T method (Cameron, Gelbach, and Miller 2008) are reported in square brackets. Standard errors are in parentheses, clustered at the birth cohort level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Table 4. Monthly Earnings Monthly log(earnings) Women Men Total (1) (2) (3) (4) (5) (6) Treated × med. grad. 0.18 0.18 0.23∗∗∗ 0.23∗∗ 0.20∗∗∗ 0.16∗∗ (0.15) (0.19) (0.05) (0.08) (0.04) (0.06) [0.29] [0.38] [0.02] [0.00] [0.00] [0.00] Observations 472 472 2,207 2,207 2,679 2,679 R2 0.05 0.41 0.01 0.34 0.02 0.34 District FE N Y N Y N Y Cohort FE N Y N Y N Y Survey year FE N Y N Y N Y Mean DV (USD) 268 268 404 404 376 376 Source: Data are from repeated cross sections of the Labor Force Surveys of Pakistan, 1995–2012. Note: The table reports the reform’s effect on the logarithm of monthly earnings for MG relative to other college graduates. The sample is restricted to individuals in cities, aged 20–47, with a college or higher degree who worked for a positive income (the years 1990–1994 are omitted because the earnings of individuals not in the labor force are coded as 0). Odd-numbered columns present estimates of a basic differences in differences without any controls or fixed effects. Even-numbered columns present results from the differences in differences in equation (1) with all controls and fixed effects. The coefficients on each of the own terms are included in the regression but not reported here. The p-values associated with the wild cluster bootstrap-T method (Cameron, Gelbach, and Miller 2008) are reported in square brackets. Standard errors are in parentheses, clustered at the birth cohort level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. The World Bank Economic Review 223 Figure 4. Logarithm of Earnings for Medical Graduates Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 Source: Data are from repeated cross sections of the Labor Force Surveys of Pakistan, 1990–2012. Note: The figure shows event study estimates based on equation (2) for the logarithm of monthly earnings. The sample is of individuals earning a non-zero income with at least a college degree who live in cities. The years 1990–1994 are omitted because they code the earnings of individuals not in the labor force as 0. Each coefficient (circle) represents the employment probability of medical graduates relative to other college graduates for a given birth cohort interval, and bars represent 95 percent confidence intervals. The base year is t = −1. Mappings from years since the reform to birth cohort groups are −3 = 1970, −2 = 1971–1972, −1 = 1973–1974, 0 = 1975–1976, +1 = 1977–1978, +2 = 1979–1980. reform, this is less likely to be the case since there are no distinct markets for male and female doctors among specialists and non-specialists. Moreover, the number of medical students admitted (fig. S1.3) and the overall supply of doctors (fig. 3) both increased slightly after the reform. Yet overall MG earnings improved, and are more consistent with higher MG quality. Surprisingly, female MG earnings did not decrease in response to lower average ability or increased competition. I find this consistently across various exercises, such as controlling for hours worked (table S2.1). The result could be due to other counteracting effects at play. These include women’s entry into higher-paying specializations as I show in the section on the Effects on the Distribution of Doctors, or due to a greater demand for female doctors through female specialists referring more patients to one another (Zeltzer 2020). Finally, the LFS (2012) has a smaller number of income observations for females compared to males, and the drop in sample size may be one reason that no statistically significant effect is discernible. 224 Aqeel Figure 5. Increase in Doctors in Urban and Rural Districts Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 Source: Data are from the Pakistan Medical and Dental Council Registration Database, 1952–2018. Note: The graphs depict the number of doctor registrations in urban and rural areas in Pakistan. The left panel is for male doctors and the right for female doctors. Dots represent the number of doctors registering in urban or in rural areas for a given birth cohort. Lines are fitted polynomials. 5. Effects on the Distribution of Doctors In this section, I explore changes in the gender composition of specialists and the spatial distribution of doctors due to the reform. Changing gender distributions in medical specialties can have implications for patient care. Meanwhile, the spatial distribution of doctors is tightly linked to the equity of health care availability across regions. Specialization Distribution. Figure S1.11 plots the female proportion of medical specialties and the relative popularity of the specialty for six of the most popular medical specialties. Almost all major fields experienced an increase in the female proportion of specialists, apart from gynecology, which was already saturated by female specialists. Table S2.8 reports the magnitudes of trend breaks in the female share of specialists caused by the reform, which supports the graphs. In pediatrics and anesthesia the reform reversed a negative trend in the female share of specialists. In surgery and internal medicine (IM), the share of female specialists increased faster (by 16 percentage points and 25 percentage points respectively) after the reform. To generalize from the six largest specialties and quantify how gender segregation across all medical specialties changed, I compute an index of specialization segregation by gender for all specialties based on Cortes and Pan (2018).25 Figure S1.12 plots the gender segregation index over 28 medical specialties by birth cohort. Before the reform, the index shows a moderately unequal distribution of women and men across specialties. The trend of the index changed slightly immediately after 1976, and the gender distribution of specialists became more even within 10 years of the reform. Geographic Distribution. In fig. 5, I plot the number of male and female doctors by urban and rural region of registration, where urban regions are the 10 largest cities in Pakistan.26 This is a conservative measure of urbanity, since smaller cities are grouped with remote villages. The number of female doctors 25 The index is defined as D = 0.5 |M j − F j |, (4) j where Mj (Fj ) is the proportion of registered male (female) doctors in specialty j ∈ J. If the distribution of women and men across specialties was equal, the index would be 0, and if all specialties were dominated by one gender, the index would equal 1. 26 Karachi, Hyderabad, Rawalpindi, Lahore, Faisalabad, Gujranwala, Sialkot, Peshawar, Quetta, and Islamabad. The World Bank Economic Review 225 rapidly increased by 80 percent in urban areas after the reform. Rural areas experienced an increase in flows, but a comparatively modest one. I show that the largest trend break was for female urban doctors, which grew about five times faster after the reform (table S2.9). Growth in urban male doctors marginally declined, while the number of rural male and female doctors increased at a much slower rate. The increase in urban doctors is consistent with the fact that female non-specialists are 10 percentage Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 points more likely to register in urban areas than men are, and 90 percent of female specialists are located in the 10 largest cities, due to the lack of basic facilities and transportation as impediments to practicing in rural areas (Jaffry et al. 2017). Consequently, when the reform caused an influx of women relative to men in medical school, newer cohorts of medical doctors practiced more exclusively in urban areas (fig. S1.13). Given the rapid rate of population growth in Pakistan, the doctor to population ratio has declined in both urban and rural areas. Using the 1998 census and doctor registrations, I find that the ratio declined from 45 to 42 doctors per 100,000 individuals in urban areas (a 6.7 percent decline). In rural areas, the percentage decline was twice as large, from 16 to 14 doctors per 100,000 individuals (12.5 percent). The difference in doctor registrations in urban and rural areas will likely contribute to how fast the supply of doctors can keep pace with population growth. 6. Robustness In this section, I conduct robustness tests for effects on the labor market, namely employment and male MG earnings (table S2.2). First, some individuals older than 16 in 1992 could be affected by the reform, either because of delayed or repeated decision making or due to medical schools anticipating the policy change (Andrabi and Singh 2015). To account for partially treated individuals, I drop the birth cohorts born between 1974 and 1975, which correspond to individuals aged 17 and 18 in 1992 (column 1). Second, I expand the time frame of the analysis to birth cohorts five years before and five years after 1976, using the cohorts 1970–1981. Reassuringly, the results remain consistent in sign and magnitude with my main results. Third, the results could be driven by changing overall trends in outcomes over the birth cohorts 1972 and 1978. I test this prediction by expanding the sample to cohorts between 1965 and 1985 and using three different placebo birth years from which reform effects could start: 1965, 1970, and 1980 (columns 3–5). For all three placebo years, the majority of coefficients become statistically insignificant and sometimes negative. Fourth, changes in the control group may affect the difference-in-differences results. In columns (6)–(8), I repeat the analysis for the three main outcomes but compare MG separately with engineering graduates, other bachelor’s degree graduates (OB), and postgraduates (PG). Fifth, to examine whether the results are dependent on one definition of working, I use an alternative definition of full-time workers only. The results are generally similar to my main results. This is especially true in column (7) which compares MG with other bachelor’s degrees, as the latter comprises the bulk of other college degrees. Finally, other changes in medicine, such as medical-school enrollments, could have affected employ- ment. Medical-graduate enrollment data demonstrate that one year after the reform (corresponding to the 1977 birth cohort), there was a small increase in national medical enrollments of about 15 percent (fig. S1.3). In table S2.3, I directly control for the total number of medical enrollments at the national level in my main analysis. This is a valid control to the extent that the change in total enrollments is not a response to the reform itself. The estimates remain similar to my main results. 7. Conclusion This paper establishes the effects of removing a gender-restrictive admissions policy in Pakistan in favor of admissions that are more competitive and provide equal opportunities for women. The reform’s effects on 226 Aqeel the medical labor market in Pakistan inform the policy debate on imposing quotas on women in medical schools. Critics of the 1992 reform worried that allowing more women into medical school would attract less motivated women, who subsequently exit the labor force and waste a medical seat that could otherwise have gone to a male. Counter to critics’ concerns, female MG’s employment increased after the reform, Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 without adverse effects on the overall employment among MG. The increase in labor-force participation was larger for female MG in their twenties and thirties. Male MG experienced higher earnings, which is consistent with increased average ability as admissions became more competitive after the reform. Finally, the reform gradually decreased gender segregation in medical specialties, but increased inequality in health-care availability between urban and rural areas. Together, my results show that the admissions policy positively affected employment and earnings for MG, but had a nuanced effect on the distribution of doctors. Medicine is an elite field in Pakistan. While the reform did not change the labor market for Pakistani women as a whole, it allowed women access to an important, high-paying profession. It also allowed Pakistani women to be more engaged in the medical labor force at a time when there are global efforts to retain women in medicine. My results suggest that the policy debate should be redirected from restricting admissions toward encouraging doctors to work in rural areas, motivating individuals from rural areas to practice in their rural community, and training community members in rural areas in health-care provision. Declarations of interest None. Data availability statement 1) Labor Force Surveys (LFS): All LFS data can be requested from the Pakistan Bureau of Statistics, Government of Pakistan. Available at https://www.pbs.gov.pk/content/data-dissemination. It is also avail- able by request from the Applied Economics Research Center, KU Circular Road, University of Karachi, Karachi, Sindh, Pakistan. 2) Administrative enrollment data: Data from the Pakistan Statistical Yearbook and the Punjab District Development Reports can be requested from the Pakistan Bureau of Statistics, Government of Pakistan, available at https://www.pbs.gov.pk/content/data-dissemination. It is also available by request from the Applied Economics Research Center, KU Circular Road, University of Karachi, Karachi, Sindh, Pakistan. 3) Medical school data: Available through the Pakistan Medical and Dental Council’s website at https: //www.pmc.gov.pk/. In the past year, some of the data has been removed from the website and must be requested. References Al-Shamsi, M. 2017. “Addressing the Physicians’ Shortage in Developing Countries by Accelerating and Reforming the Medical Education: Is it Possible?” Journal of Advances in Medical Education & Professionalism 5(4): 210. Alsan, M., O. Garrick, and G. Graziani. 2019. “Does Diversity Matter for Health? Experimental Evidence from Oak- land.” American Economic Review 109(12): 4071–111. Andrabi, T., and N. Singh. 2015. “Almost Breaking the Glass Ceiling: Open Merit Admissions in Medical Education in Pakistan.” Mimeo.https://studylib.net/doc/9903856/breaking- the- glass- ceiling- - open- merit- admissions- in. Blau, F. D., and L. M. Kahn. 2005. “Do Cognitive Test Scores Explain Higher US Wage Inequality?” Review of Eco- nomics and Statistics 87(1): 184–93. Bleakley, H. 2010. “Malaria Eradication in the Americas: A Retrospective Analysis of Childhood Exposure.” American Economic Journal: Applied Economics 2(2): 1–45. The World Bank Economic Review 227 Cameron, A. C., J. B. Gelbach, and D. L. Miller. 2008. “Bootstrap-Based Improvements for Inference with Clustered Errors.” Review of Economics and Statistics 90(3): 414–27. Cawley, J., J. Heckman, and E. Vytlacil. 2001. “Three Observations on Wages and Measured Cognitive Ability.” Labour Economics 8(4): 419–42. Cortes, P., and J. Pan. 2018. “Occupation and Gender.” The Oxford Handbook of Women and the Economy: 425–52. Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 Das, J., B. Daniels, M. Ashok, E.-Y. Shim, and K. Muralidharan. 2022. “Two Indias: The Structure of Primary Health Care Markets in Rural Indian Villages with Implications for Policy.” Social Science & Medicine 301: 112799. https://doi.org/10.1016/j.socscimed.2020.112799. De Chaisemartin, C., and X. d’Haultfoeuille. 2020. “Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects.” American Economic Review 110(9): 2964–96. DHS. 1991. “Pakistan Demographic and Health Survey 1990-1991.” Islamabad, Pakistan and Maryland, US, National Institute of Population Studies (NIPS) and ICF. http://dhsprogram.com/pubs/pdf/FR29/FR29.pdf. Docquier, F., O. Lohest, and A. Marfouk. 2007. “Brain Drain in Developing Countries.” World Bank Economic Review 21(2): 193–218. Frank, E., Z. Zhao, S. Sen, and C. Guille. 2019. “Gender Disparities in Work and Parental Status among Early Career Physicians.” JAMA Network Open 2(8): e198340–e198340. Greenwood, B. N., S. Carnahan, and L. Huang. 2018. “Patient-Physician Gender Concordance and Increased Mortality among Female Heart Attack Patients.” PNAS 115(34): 8569–74. Greenwood, B. N., R. R. Hardeman, L. Huang, and S. Aaron. 2020. “Physician-Patient Racial Concordance and Disparities in Birthing Mortality for Newborns.” PNAS 117(35): 21194–200. Jaffry, T. N., S. Mirza, S. Farheen, and S. Khalid. 2017. “Reluctance to Serve in Rural Areas: Doctors’ Perspective.” Pakistan Journal of Public Health 7(3): 157–62. Kerssens, J. J., J. M. Bensing, and M. G. Andela. 1997. “Patient Preference for Genders of Health Professionals.” Social Science & Medicine 44(10): 1531–40. Lehmann, U., M. Dieleman, and T. Martineau. 2008. “Staffing Remote Rural Areas in Middle- and Low-Income Countries: A Literature Review of Attraction and Retention.” BMC Health Services Research 8(1): 1–10. LFS. 2012. “Labor Force Surveys, Former Federal Bureau of Statistics, Government of Pakistan.” http://www.pbs.go v.pk. Lindqvist, E., and R. Vestman. 2011. “The Labor Market Returns to Cognitive and Noncognitive Ability: Evidence from the Swedish Enlistment.” American Economic Journal: Applied Economics 3(1): 101–28. M Lucas, A.. 2010. “Malaria Eradication and Educational Attainment: Evidence from Paraguay and Sri Lanka.” American Economic Journal: Applied Economics 2(2): 46–71. Masood, A. 2019. “The Influence of Marriage on Women’s Participation in Medicine: The Case of Doctor Brides of Pakistan.” Sex Roles 80(1): 1–18. PBS. 2023. “Census Field Operation Plan.” Ministry of Planning, Development, and Special Initiatives, Pakistanhttps: //www.pbs.gov.pk/sites/default/files/population/2023/FOP_final_26- 2- 23_final.pdf. Powell, W., J. Richmond, D. Mohottige, I. Yen, A. Joslyn, and G. Corbie-Smith. 2019. “Medical Mistrust, Racism, and Delays in Preventive Health Screening among African-American Men.” Behavioral Medicine 45(2): 102–117. Strasser, R., S. M. Kam, and S. M. Regalado. 2016. “Rural Health Care Access and Policy in Developing Countries.” Annual Review of Public Health 37: 395–412. Tsugawa, Y., A. Jena, J. Figueroa, E. Orav, D. Blumenthal, and A. Jha. 2017. “Comparison of Hospital Mortality and Readmission Rates for Medicare Patients Treated by Male vs Female Physicians.” JAMA Intern Med. 177(2): 206–13. Ullah, I. 2022. “Re-identifying the Rural/Urban: A Case Study of Pakistan.” Espaço e Economica. Revista brasileira de geografia econômica Ano XI( 23): 1–12. United Nations. 2004. “Millenium Development Goals: Progress Report.” UN Dept. of Economic and Social Affairs : UN Dept. of Public Information US Department of State Archive. 1998. “1998 Country Report on Economic Policy and Trade Practices: Pakistan.” U.S. Department of Commerce, Bureau of Economic Analysis state.gov. Waller, K. 1988. “Women Doctors for Women Patients.” British Journal of Medical Psychology 61(2): 125–35. 228 Aqeel Wallis, C. J., B. Ravi, N. Coburn, R. K. Nam, A. S. Detsky, and R. Satkunasivam. 2017. “Comparison of Postoperative Outcomes among Patients Treated by Male and Female Surgeons: A Population Based Matched Cohort Study.” BMJ 359: j4366. Wasserman, M. 2023. “Hours Constraints, Occupational Choice, and Gender: Evidence from Medical Residents.” The Review of Economic Studies 90(3): 1535–1568. Downloaded from https://academic.oup.com/wber/article/38/2/209/7505292 by Sectoral Library Rm MC-C3-220 user on 01 May 2024 World Bank. 2018. “World Health Organization Global Health Expenditure Database.”https://apps.who.int/nha/dat abase/Home/Index/en. World Health Organization. 2016. “Primary Care Systems Profiles and Performance (Primasys).” Zaidi, S. A. 1986. “Class Composition of Medical Students: Some Indications from Sind, Pakistan.” Economic and Political Weekly 21(40): 1756–59. Zeltzer, D. 2020. “Gender Homophily in Referral Networks: Consequences for the Medicare Physician Earnings Gap.” American Economic Journal: Applied Economics 12(2): 169–97.