Medical education reforms and the origins of the rural physician shortage

Despite an increased production of the total number of physicians, a persistent geographic maldistribution of physicians has characterized the past 70 to 80 years...The opportunity for medical education in this century is to recapture the diversity and relevance of distributed training, even as patient care, education, and research are further improved. Distributed medical education that is uniquely adapted and responsive to the needs of rural underserved communities has the potential to reclaim medicine’s social contract with the public. (American Academy of Family Physicians 2013).

Abstract

In the first two decades of the twentieth century, medical schools increased standards for admission and added basic science to their curricula. During this time period, the probability a new medical school graduate located in a rural area declined by 40%. Using novel data from the American Medical Directories, we find that physicians trained in more rigorous programs with higher admission standards were less likely to set up practice in rural areas. While all physicians were being drawn to metropolitan areas during this period, the pull was stronger for graduates of the higher quality schools. We also find some evidence that physicians trained in the more scientifically and clinically based programs were more strongly attracted to places with more hospitals. These findings suggest that the medical education reforms of the early twentieth century contributed to the urban–rural disparity in access to physician care.

1 Introduction

An estimated 20% of the US population lives in rural areas, yet only 11.4% of physicians practice in these regions, a fact that concerns policymakers who worry the that the “persistent geographic maldistribution of physicians” limits the ability of rural people to access health care (Burrows et al. 2012; Rosenblatt and Hart 2000; Rosenblatt 2010). These concerns are not new; as early as the 1920s, public health officials sounded alarms about the dwindling number of practitioners in rural areas (van Bibber 1929). In this paper, we argue that the decline of the rural doctor has its origins in the changes that occurred in medical education in the late nineteenth and early twentieth centuries. American medical schools adopted more rigorous admission standards, extended periods of instruction, and added basic science and clinical studies to their curricula in an effort to catch up to the standards of European medical schools. These changes, initially implemented by a subset of schools, were accelerated by the 1910 publication of Abraham Flexner’s scathing report on the state of medical education in North America (Flexner 1910) and subsequent changes in physician licensing by the states.

While these changes improved the quality of medical education and encouraged uniformity among medical schools, they may have done so at the expense of making medical school less accessible to potential physicians from rural areas (Page and Baranchuk 2010; Dhalla et al. 2002; Kwong et al. 2002, 2005). The reforms increased the costs of going to medical school (both direct dollar costs and lost wages from years spent in school), thus influencing the types and numbers of students willing and able to enter the medical profession, as well as the locations and types of positions they were willing to accept after graduation. Raising the bar for admission to medical school was thought to prevent the “poor boy” and the “country boy” from entering the profession. Because such individuals were believed to be more willing than those from an urban background to set up practice in rural areas, their shrinking numbers in medical school likely meant a decrease in the flow of new physicians to the countryside (Mayers and Harrison 1924, p. 138).

In addition, reforms to medical education may have changed physicians’ preferences over practice location characteristics. The more scientifically and clinically based training may have encouraged new physicians to seek out places with better access to cutting edge medical technology, larger communities of health professionals, and modern hospitals (Commission on Medical Education 1932, p. 113).

Although the origins of the urban–rural disparity in physician access have long been linked to the transformation of American medical education of the early twentieth century, this conjecture has never been subject to careful empirical examination. The evidence supporting the link has been the timing alone; the decline in the country doctor coincided with the changes in medical education. However, in one of the earliest studies of the geographical distribution of physicians, Mayers and Harrison (1924) argued that the reforms of medical education could have been no “more than a secondary influence in leading recent graduates to avoid rural locations.” They claimed that even physicians who had “received their medical education under a regime far removed from that of today” were moving away from rural areas during this period, and concluded that the primary factors leading doctors to more urban areas were “the increasingly superior financial, social, and professional advantages of the larger places” (Mayers and Harrison 1924, p. 149).

We examine the connections between medical training and practice location choice using a rich new dataset on physicians constructed from the American Medical Directories of 1909 to 1921, and reports of the American Medical Association’s Committee on Medical Education. We exploit differences across schools in the timing of curricular changes, using as a summary statistic of these changes the year in which a school adopted requirements of a year or more of college coursework for admission. We find that physicians’ location decisions varied systematically with the quality of their medical training. Graduates of schools with pre-medical school education requirements were more likely than other graduates to locate in metropolitan areas and in areas with greater professional amenities, such as more hospital beds per capita.

Because of data limitations, we are unable to identify why physicians trained in more modern schools were more likely to locate in urban areas. It could be because the only applicants who could meet the higher standards were already predisposed to settle in an urban location, or because some feature of training persuaded physicians to move to urban regions. In other words, reforms might have changed who selects into becoming a physician, or changes in curriculum had a treatment effect on location decisions. As such, the majority of our results should be interpreted as including both the selection and treatment effects of the reforms. Nevertheless, modern research also suggests that physicians born in rural areas may be more likely to return to rural areas (Rosenblatt and Hart 2000). Thus, we also use a cross-section of data from North Carolina where we match physicians to their birthplaces to determine the relative impact of medical training compared to birthplace. As with the full sample results, being trained in a more rigorous school reduced the likelihood of setting up practice in a rural area. However, this effect was half the magnitude, and opposite in sign, to that of being born in a rural area. Physicians born in rural areas were about 45 percentage points more likely than those born in urban areas to be practicing in a rural area, even after controlling for their medical school characteristics. This result indicates that the medical education reforms of the early twentieth century contributed to the early development of the urban–rural gap in access to physicians by reducing the number of “country boy” and “poor boy” medical school graduates.

Our study makes a number of important contributions. First, it adds to the growing body of work documenting educational differences in migration patterns. More highly educated workers are, on average, more mobile and attracted to different types of places than workers with less education. These differences are linked to self-selection into education and to differences in information sets or preferences over location-specific amenities (Wozniak 2010; Gottlieb and Joseph 2006). What makes the arguments about physician location choices in the early twentieth-century distinct is that the educational differences of interest are not differences in levels, but rather in curricula studied. All medical students received an M.D. and intended to practice as physicians upon graduation, but in the early twentieth century, medical students were subjected to very different courses of study based on school and cohort. Whether by changing the selection into medical school or changing the preferences for place characteristics, the transformation of medical education led to a shift in the geographical distribution of medical practice.

Our findings also provide insights into the changing mortality patterns of the period. At the turn of the twentieth century, urban areas had much higher mortality rates than rural areas. This urban disadvantage, however, diminished over the first half of the century. The early decline was due to improvements in public infrastructure that led to cleaner water and reduced exposure to sewage, but Haines (2001) has attributed the later declines to more rigorous medical training and stricter licensing of physicians and other health care providers (p. 45). Our results provide support for Haines’ conjecture by demonstrating that the urban advantage in medical care was not due to just the higher numbers of physicians per capita in urban areas but also to the better training of physicians who established practices in urban areas. In an era when medical technology changed rapidly, the urban preference of more rigorously trained physicians may have set the stage for later gaps in access between rural and urban areas that persist today.¹

2 The transformation of American Medical Education

Although the 1910 publication of Abraham Flexner’s “Medical Education in the USA and Canada” is often credited with ushering in a new era of standards in medical education, the changes that led to improved standards in medical education were initiated by the schools themselves beginning in the 1870s. During the 1800s, numerous medical schools opened to supplement the apprenticeship system that produced physicians during the colonial period. Medical schools were inexpensive to operate and profitable for faculty.² As the number of medical schools increased, they competed heavily for students, and competition among medical schools led to shorter terms and more lax requirements for graduation (Rothstein 1972, p. 97). Would-be physicians faced virtually no entrance standards, not even literacy (Ludmerer 1985, pp. 12–13). By the time Charles Eliot took over as President of Harvard in 1869, Harvard Medical School admitted any fee-paying student. Only 20% of the students held undergraduate degrees, and one faculty member suggested that over half the students could not write. Medical educators recognized this as a fundamental barrier to improving the quality of medical education. Said one, “It ought not to be necessary to teach elementary chemistry or elementary biology in the medical school any more than it ought to be necessary to teach elementary English in the law school” (Ludmerer 1985, pp. 114–115). Term lengths were short: Harvard’s curriculum consisted of two, 4-month terms of lectures. Students did not engage in laboratory or clinical work, and needed to pass only five of nine five-minute oral quizzes to pass their studies (Ludmerer 1985, pp. 49–50). The quality of physicians was low enough that during the Civil War, the United States “...imposed compulsory examinations for a physician to be admitted to the army or navy’s medical service,” an exam that only 25% passed (Ludmerer 1985, pp. 15–16).

Physicians who wanted a rigorous medical education could not rely on US medical schools, but those with resources could train in Europe, where the focus had moved beyond apprenticeship to clinical and experimental research.³ France, with its reliance on acute clinical observation in hospitals, had been the European leader of medical training until the mid-nineteenth century, when scientists in Germany embraced experimental methods. Major discoveries were made in cell theory, pathology, and physiology. The use of experimental methods in the laboratory “...allowed an epistemological shift of revolutionary proportions. It became clear to knowledgeable physicians that experimental methods could be applied to the study of disease and therapeutics as well as to the study of the healthy state. Scientific information ...began to represent the core of what a modern doctor needed to know” (Ludmerer 1985, pp. 30–31). Around 15,000 American physicians studied in Germany (or at German-speaking universities in Austria and Switzerland) between 1870 and 1914, with the peak of the migration occurring in the 1870s and 1880s (Bonner 1963, p. 23). Many returned and went on to become leading physicians, and to join the faculties at schools such as Harvard, Cornell, Michigan and others. Notably, these foreign-trained doctors were the strongest advocates for more rigorous entrance requirements and curricula, transforming medical education in the USA into the European model (Ludmerer 1985, p. 33). In this way, “European medical science provided and European medical schools imparted to Americans the knowledge and procedures which led to reform of American medical education” (Field 1970, p. 504).

The first movers started instituting reforms in the late nineteenth century. By the 1870s, Harvard had implemented a series of reforms at the medical school, which included changes in both administration and curriculum. The university took over the finances of the medical school, and no longer allowed professors to divide fees among themselves. The program was extended to 3 years, with 9-month terms, and the curriculum emphasized science, laboratory work, and clinical instruction. By 1880, the University of Pennsylvania followed Harvard’s lead, as did the University of Michigan. Johns Hopkins opened in 1893 with the strictest requirements of all: requiring students to have a bachelor’s degree for admission (including evidence of coursework in Latin, French, German, physics, biology, and chemistry), to receive rigorous training in the sciences and laboratory work, and to have 2 years of clinical instruction (Ludmerer 1985, pp. 50–51).⁴

As major scientific advances occurred—such as the germ theory of disease in the late nineteenth century—more schools began to emulate these early pioneers in increasing term lengths, use of laboratories, and pre-medical requirements. Although an early attempt to organize schools with 3-year courses failed in 1876, the Association of American Medical Colleges (AAMC) reformed in 1889, and by 1891, 71 of 115 regular (as opposed to homeopathic, Thomsonian, or eclectic) colleges were members (Rothstein 1972, p. 288). The AMA established its Council on Medical Education (CME) in 1904, and in 1905, the AMA recommended that all entering medical school students have a high school education, and that medical education be completed over 5 years (with 4 years of “pure” medical work and at least 1 year of study devoted to preliminary work in biology, chemistry, and physics (American Medical Association 1906).

The Council on Medical Education conducted a survey of the nation’s medical schools in 1906. Given that its members were mainly academic physicians, the Council used the rigorous university model as the yardstick against which it compared all schools. Not surprisingly, the Council drew largely the same conclusions as Abraham Flexner ultimately would in 1910: that most medical schools did not come close to meeting these standards (Ludmerer 1985, p. 170). Beginning in 1907, the AMA published ratings of medical schools. Schools rated “A” were acceptable, schools rated “B” were in need of improvement, but redeemable, and schools rated “C” were in need of complete reorganization. In response, more schools began imposing more rigorous prerequisites; by 1908, 57 schools required applicants to have completed at least 1 year of college before starting medical school. Nevertheless, since the AMA could not be openly critical of medical schools, it invited Abraham Flexner of the Carnegie Foundation for the Advancement of Teaching to perform a similar study. In his report, Flexner evaluated medical schools based on entrance requirements, the number of students enrolled, the number and characteristics of teaching staff, the financial resources of the school, and the laboratory and clinical facilities available to students. He suggested that the vast majority of medical schools were of very low quality and that there was gross oversupply in the number of physicians.

While reforms in medical education were well underway when Flexner published his report in 1910, the report drew public attention to the issue and may have accelerated the speed at which some schools closed and others improved their standards. Despite the changes happening in medical education prior to the publication of Flexner’s report, some medical schools were slow to upgrade their standards, if at all, partially because state licensing laws remained weak. By 1904, only 4 out of 162 schools then in operation even required two years of college work to be admitted; most required a high school diploma or less (US Department of the Interior 1921, p. 72). Most state licensing boards were ineffective; as late as 1906, 13 states granted medical licenses even to people who had never graduated from medical school (Ludmerer 1985, p. 235). The licensing requirements “...served only as a minimal restraint to the worst excesses of the low-grade schools” (Rothstein 1972, p. 291). Many schools had improved standards of their own accord, but Flexner recognized that “...legal enactments on the subject of medical education and practice will be required before the medical schools either give up or relate themselves soundly to the educational resources of their respective states.” (Flexner 1910, p. 49). Medical educators had pushed for more restrictive licensing laws for years, thus supporting the economic theory that licensing laws benefit existing firms by limiting competition, but it was not until after the publication of Flexner’s report in 1910 that many states began to implement higher standards (Ludmerer 1985, p. 237). Figure 1 shows that while only a handful of states required pre-medical college study for licensure in 1910, the vast majority required it by 1920. In addition, state licensing boards decided whether to recognize a school’s diploma based on its AMA rating, which incorporated the rigorous criteria suggested by Flexner. As early as 1914, 31 states denied recognition to graduates of medical schools with a “C” rating (Ludmerer 1985, p. 241). States also began to require physicians to have graduated from a school with college coursework admissions requirements.

Fig. 1

Source: “Medical Education in the United States,” Journal of American Medical Association, Aug. 27, 1932: p. 746

States with pre-medical college requirements for physician licensure.

Fig. 2

Source: Pre-medical college requirements are from American Medical Association Council on Medical Education (1919, 1923). Medical schools and graduates from American Medical Association Council on Medical Education (1905–1910; 1911–1914; 1915–1920)

Number of medical schools and graduates by college requirements.

Fig. 3

Source: Pre-medical college requirements are from American Medical Association Council on Medical Education (1919, 1923). Fees are from American Medical Association Council on Medical Education (1905–1910; 1911–1914; 1915–1920)

Average total fees for a 4-year course by pre-medical requirements. Notes: Figures are in current dollars.

Schools responded quickly to these changes. An increasing number of schools raised their admission requirements. Low quality and independent schools were driven out of the market and the total number of medical schools in the USA fell from over 150 in 1900 to around 80 by 1923 (Mayers and Harrison 1924, p. 16). As the overall number of medical schools declined in the first decades of the twentieth century, an increasing number of schools adopted more rigorous requirements, as shown in Panel A of Fig. 2. By 1915, 50 out of 95 medical schools required at least 1 year of college; 40 required two or more years.⁵ As the number of medical schools fell, so too did the number of medical school graduates, as shown in Panel B. At the turn of the century, over 5000 physicians graduated from medical school each year. This number declined sharply during the 1910s, and rebounded somewhat only after 1920. Over these same years, the population grew by 39%. Together, these trends caused a decline in the number of physicians per capita from 17 to 14 per 10,000 residents at the national level. In their analysis of the rise of occupational licensing in the Progressive Era, Law and Kim (2005) affirm that state laws requiring more pre-medical education were associated with a decline in the number of physicians.⁶

3 How changes in physician training impact location decisions

As entrance requirements increased and medical education became more rigorous, the cost of becoming a physician increased. Would-be medical students had to bear the tuition—and time—costs of doing undergraduate college coursework prior to applying for admission, and the more rigorous medical schools also tended to charge higher tuition and fees. Figure 3 shows medical school fees by whether a school adopted a pre-medical college requirement. Schools with more extensive admissions requirements charged more in fees than schools without such requirements. Physicians trained in these more expensive programs may have been more drawn to practice in cities after graduation than in rural areas by the prospect of greater demand for their services and higher earnings. In their 1924 study of the geographic distribution of doctors, Mayers and Harrison (1924) note that the growth of roads and the rise of specialty practice increased incomes for physicians in urban areas at the expense of rural practitioners:

The larger financial possibilities of the successful town practitioner have of course always been a feature in favor of town location; but the possibilities of the successful town specialist rise much higher still, creating in reality a wholly new level of medical earnings, and holding out to the rural practitioner a lure to the town far more powerful than formerly existed (Mayers and Harrison 1924, p. 27).

Another report published by the Commission on Medical Education (CME) in 1932 summarized several studies that suggested “... that the greater opportunity in the cities for financial rewards and better opportunities for practice are the oustanding (sic) causes of the tendency of physicians to locate in the cities” (Commission on Medical Education 1932, p. 113). The Commission’s report also indicated that changes in medical training pulled students to urban areas because the changes “...accentuate the dependence of the doctor upon hospitals, laboratory technicians, nurses, consultants, and specialists, and ...have a very distinct effect upon the attitude of students toward practice, to some extent handicapping them in assuming the responsibility of individual practices in the smaller communities” (Commission on Medical Education 1932, p. 114).⁷

The increase in costs, both direct and indirect, may have also made it more difficult for individuals from rural areas to attend medical school. Research from the current period indicates that rural-born medical students are more likely to locate in rural areas to set up practice (Rosenblatt and Hart 2000). In the early twentieth century, the rhetoric was that the “country boy” could no longer aspire to become a doctor because of the higher admissions standards and the increased time and tuition costs of attending medical school. These “country boys” were believed to be the most likely to become country doctors (Mayers and Harrison 1924, p. 138). If medical school graduates were less likely to be rural-born as schools improved, this would shift the distribution of graduates away from rural areas.

To test whether doctors trained in more rigorous (and expensive) programs preferred urban areas to their less rigorously trained counterparts, we construct a new data set on physicians, including information on their medical training and where they chose to practice. Our empirical analysis is based on a simple theoretical framework for a physician’s location choice decision based on comparing the relative rewards across potential localities.⁸ We also provide narrative evidence from the period to support our choice of variables.

In our model, a physician locates a practice by choosing the location $i \in I$ with the highest expected utility.

$$\begin{aligned} \underset{i\in I}{\text {arg max }} U(\omega _{i})= \underset{i\in I}{\text {arg max }}\left\{ \sum _{t} \delta ^{t} \left[ \frac{E\left( w(s)_{it}\right) }{p_{it}} - c(s)_{it} \right] \right\} \end{aligned}$$

(1)

where $\omega$ is consumption, t is year, s indexes skill group (graduate of medical school with or without pre-medical college requirements), w is the nominal wage, p is the price level, and $\delta$ is a discount factor. Consumption amenities and costs tied to a particular location are captured in c, which is allowed to differ by skill group. Preferences over rural or urban living, or other location-specific attributes, such as proximity to family, are included here. Positive productive amenities increase the nominal wage. Examples include the presence of hospitals and laboratories, the size of the market, extent of road system, and any agglomeration effects of being located near other highly trained professionals or a medical school. Nominal wages are determined in a spatial equilibrium of the type developed by Roback (1982), Rosen (1979), and Moretti (2011). There is somewhat of a debate in the literature on whether wages in a local market are determined by supply and demand or if physicians are discriminating monopolists (Kessel 1958; Arrow 1963; Andersen and Anderson 1967; Newhouse 1970; Sloan and Feldman 1978; McCarthy 1985).⁹ In a sense, the type of competition at the local market level is irrelevant for our application; we are interested in spatial equilibrium outcomes where the marginal physician’s utility is equilibrated across local markets.

A physician chooses to locate in $i^{*}$ if the expected benefits exceed those of all other locations. The probability of choosing location $i^{*}$ increases in the real wage, and decreases in consumption disamenities, and can be written as:

$$\begin{aligned} Pr({\textit{choose }} i^{*}) = Pr\left[ \sum _{t} \delta ^{t} \frac{E\left( w(s)_{i^{*}t}\right) }{p_{i^{*}t}} - c(s)_{i^{*}t} > \sum _{t} \delta ^{t} \frac{E\left( w(s)_{it}\right) }{p_{i^{}t}} - c(s)_{it} \right] \qquad \forall i, i\ne i^{*} \end{aligned}$$

(2)

The formulation of the problem is now a standard discrete choice problem considered in McFadden (1974), and for which the conditional logit model is appropriate to study the factors that increase the likelihood of a physician to choose a particular location in which to start a practice. In some analyses, we further simplify the model to the binary discrete choice between a rural and urban location, effectively reframing the choice as a comparison between the expected utility of the top-ranked urban location and the top-ranked rural location.¹⁰

Our primary measure of medical school quality is an indicator for the presence of a 1- or 2-year pre-medical college education requirement. This variable captures both the increase in admission standards and introduction of rigorous scientific curriculum, as well as the increased opportunity costs of attending medical school. In this paper, we do not attempt, nor do the data allow us, to separately estimate the impact of these components on physicians’ location decisions. Instead, we treat pre-medical college requirements as a sufficient statistic for the various quality measures available, and their associated opportunity costs. In support of this view, the AMA and AAMC used pre-medical college requirements as the primary tool to increase the academic quality of applicants, while simultaneously reducing the supply of newly minted physicians. State licensing reform also focused on the requirements for pre-med college education, which Law and Kim (2005) have shown had an important negative impact on the supply of physicians during the period.¹¹

4 Data

Data to examine the relationship between the changes in medical education and the geographic distribution of physicians come from the 1909, 1914, 1918, and 1921 American Medical Directories (AMDs).¹² The AMDs provide information (by location) on all practicing physicians, the medical schools they attended, their graduation dates, and for the later years, their specialties, if any. We supplement the AMD data with data on medical school characteristics from the Flexner report and AMA publications. The medical school-specific information is crucial for examining how education reforms affected physician’s location choices. As discussed above, the reform movement began with a handful of schools and then spread in fits and starts to other schools. The timing of the adoption of higher admission standards varied greatly across schools. This variation was present even within states, as many schools adopted higher standards before state licensing laws required them to do so. Schools not only implemented pre-medical college requirements; they also lengthened terms and upgraded curricula. We focus on the implementation of pre-medical college requirements because over the period we examine, most schools had already imposed 4-year terms and modified curricula.¹³ In our probit models, the inclusion of fees interacted with the state of practice does not substantively change the magnitude of the coefficient on pre-medical college requirements, but does increase the uncertainty of the estimates. See Appendix Table 17 for results. Our lists of physicians were manually digitized and represent a novel data set for understanding the economic history of the physician labor market in the early twentieth-century USA.

The amount of information contained in the AMDs is vast: 135,000 physicians listed in 1909, with similar numbers in each of the other 3 years we include.¹⁴ Given the constraints in digitizing the entirety of the AMD, we limit our sample to four states: California, Mississippi, New York, and North Carolina. We focus on these states for geographic and socioeconomic reasons. California and New York both contained major metropolitan areas, as well as sizable rural population, and are located in very different regions. California is also of interest because of the high level of in-migration during this period and the dramatic growth of Los Angeles and other metropolitan areas. We also entered Mississippi and North Carolina to look at states with predominantly rural populations and large populations of African Americans.

The AMD data allow us to consider how the geographic distribution of physicians varied by the choice of medical school and timing of graduation. The overall US population became more urban during the first decades of the twentieth century.¹⁵ Nevertheless, the rate at which physicians and surgeons moved to urban areas exceeded that of the general population. Between 1900 and 1920, the share of doctors living in cities with a population greater than 100,000 increased by 9.2 percentage points, compared to an increase of 7.2 percentage points for the population as a whole. Similarly, the fraction of physicians living in areas with a population under 10,000 declined by 13.5 percentage points, compared to a decline of 10.4 percentage points for the overall population. There is also evidence that not all physicians and surgeons were equally like to move to urban areas. Table 1 presents the distribution by location for our four-state sample of AMD listings, for all physicians, established physicians who have been in practice for over 5 years, and recent medical school graduates (five or fewer years since graduation). Since our sample includes New York and California, two populous states with several large cities, it has a much higher fraction in metropolitan areas than found in census data for the nation as a whole. Nonetheless, the AMD data show the same overall pattern of movement of physicians from rural areas into areas with population greater than 10,000. The comparison of panels B and C, however, reveals that this movement was much more pronounced for newly minted physicians. In 1909, about one-third of recent graduates had set up practice in cities with fewer than 10,000 people, compared to only 16.4% by 1921. There is also evidence that graduates of schools with more rigorous requirements differentially selected into large urban areas. Panels A and B of Table 2 show the distribution of new graduates by whether their school required at least 1 year of college prior to admission to medical school to those with no college prerequisite. Graduates of schools without college requirements were three times more likely to locate in smaller areas with less than 10,000 population in 1909 than graduates of schools with more rigorous requirements. Although the gap narrowed over time, the likelihood that graduates of schools without college prerequisites settled in rural areas was nearly two times greater in 1924 than that of physicians graduating from schools that required at least 1 year of college before admittance.

5 Empirical analysis

5.1 Probit model of rural location choice

We first estimate probit models of rural location choice to test the hypothesis that graduates of schools with more rigorous admission and curricular requirements were less likely to locate in rural areas than graduates of lower-quality schools. In these models, we consider how individual physician characteristics and the characteristics of his medical school are correlated with the likelihood that he located in an area with less than 10,000 population.¹⁶ In this model, physician i chooses a location in time t, based on the characteristics of the medical school attended, j, in the year he graduated, g, and his own characteristics. We include as explanatory variables indicators for whether the school had adopted a requirement of one or more years of college as a prerequisite, and whether the physician attended a medical school located in a rural non-metropolitan county.¹⁷ To gauge whether the effects of medical school characteristics were changing over time, we estimate models separately by the year of publication of the AMD list of physicians.

We separate the sample by years since graduation, and first estimate the model for physicians in practice for more than 5 years (established physicians), and then for those in practice for 5 or less years (new graduates). Many “established” physicians attended medical school in the pre-Flexner era and their medical education may have significantly differed from that of later graduates. Splitting the sample into “established” and “new” physicians helps capture these differences. Moreover, this grouping allows for the migration of established physicians into or away from rural areas. The new doctors' sample focuses on the initial practice location decision immediately after graduation from medical school. For the model including only established physicians practicing over 5 years, we include the number of years a physician had been in practice—defined as the number of years since he graduated from medical school—and its square.

A number of medical schools closed during the period under study, so we include an indicator equal to one if the physician’s alma mater closed before 1923, which can be interpreted as a further measure of medical school quality. In addition, North Carolina enacted a law in 1918, and Mississippi in 1919, that required physicians to attend schools with pre-medical college requirements; we include a variable to indicate if a physician faced one of these laws during the year of their graduation. We also include state fixed effects to capture the persistent differences across states in the sizes of their rural population. Table 3 provides summary statistics for our sample broken down by new physicians and established physicians.

Table 4 presents estimated marginal effects from probit models on the sample of established physicians (Panel A), and then separately for new doctors (those who had graduated from medical school within the past 5 years—Panel B).¹⁸ After controlling for years in practice, we see strong differences in location choices by medical school characteristics. Among new physicians, graduates of schools that required one or more years of college before admission were more likely (20.6 percentage points in 1909 and 2.9 percentage points in 1921) than graduates of other schools to locate in metropolitan areas with over 10,000 population. The magnitude of the correlation decreases over time, likely because of the changing selection of schools into the group requiring pre-med college education. Only 2.6% of the new doctors in our sample graduated from these schools in 1909 (97% of the graduates were from Harvard or Johns Hopkins). By 1921, 61% of new doctors graduated from school with pre-med education requirements. We see the same patterns for established physicians, with even larger negative correlations than observed for new physicians. Again, this is likely because of the change in composition. Physicians who attended medical schools with pre-med college requirements prior to 1910 were even more positively selected in terms of school quality than physicians from medical schools that adopted the requirements in later years. The early adopters were the most prestigious medical schools.

Table 1 Distribution of physicians American Medical Directory samples by location, all doctors and recent medical school graduates.

Table 2 Distribution of recent medical school graduates by location, by pre-med college requirements.

A particularly striking finding is that even after controlling for medical school quality based on requirements for admission, new graduates of medical schools located in rural counties were roughly 3 to 8 percentage points more likely to set up practice in rural areas than their colleagues who had attended medical school in, or close to, a metropolitan area. For established physicians, the relationship is even stronger: physicians who attended a rural medical school were 7 to 11 percentage points more likely to locate in rural areas than their counterparts trained in non-rural schools. A straightforward interpretation of this result is that physicians preferred to set up practice near where they went to medical school (a hypothesis we will test directly in the conditional logit models we present later in the paper). An alternative, albeit more speculative, interpretation is that this effect captures the “country boy” story. If attending a rural medical school is a proxy for being from a rural area, then the greater probability of graduates of these schools of setting up practice in rural areas can be interpreted as supporting the idea that country doctors were mostly country boys returning home to serve their communities.

5.2 Location characteristics and physician location decisions

An important element of most explanations connecting the movement of physicians to more urban areas as the medical education reforms of the 1910s progressed is that the more science and technology-based curricula—and their higher costs—made doctors trained in such programs more likely to be attracted to places with larger patient bases and more professional amenities such as hospitals and laboratories. To determine which place characteristics attracted or repelled physicians, we estimate conditional logit models (McFadden 1974). In these models, we include factors expected to have particular influence on physicians such as the numbers of hospital beds and established doctors per capita. In order to test the hypothesis that location choices differed by type of medical training, we examine whether graduates of different quality medical schools responded differently to location characteristics.

McFadden’s conditional logit model assumes each individual faces a set of J unordered alternatives and chooses the alternative that provides him with the highest level of utility (McFadden 1974). We assume that the choice set faced by a recent medical school graduate is the set of counties in the state in which he takes the medical board examinations. In essence, this amounts to assuming a sequential decision-making process in which a physician first chooses the state in which to locate and then at the next step, chooses the county within that state. This assumption is necessitated by the nature of the data. In reality, most new medical school graduates first decided in which town or city they wished to locate, and that dictated the state in which to take the licensing exam. However, given the structure of the data, estimating models with extended choice sets of cities or counties in multiple states is intractable. Moreover, once a physician was licensed in a particular state, the costs of interstate moves increased greatly. The county-based classification scheme we use is based on the concept of metropolitan areas, where a metropolitan area is a county, or set of counties, with a large city as its economic center. We divide non-metropolitan counties into rural counties that are adjacent to metropolitan areas (since the rapid diffusion of automobiles and the expansion of road networks expanded the influence of metropolitan areas on their neighboring areas), and rural counties not adjacent to metropolitan areas.¹⁹

Table 3 Summary statistics of AMD physician sample.

Table 4 Determinants of rural practice location choice: probit marginal effects.

In the choice model, we include county characteristics that potentially influenced physicians’ location choices: the number of hospital beds and the number of established physicians (having graduated from medical school more than 5 years earlier) per 10,000 persons in the population, and the natural log of the number of miles from the physician’s medical school to the county seat. These variables capture productive amenities. Physicians may have found it useful to locate near other physicians, and nearly all physicians practiced independently but needed hospitals to admit patients. Ideally, we could directly include measures of physician income to capture spatial differences in the economic return of choosing a particular location. Unfortunately, systematic income data for physicians does not exist for our sample period. We choose to use proxies instead, such as market size and the presence of productive amenities.

Given the findings above, we are also interested in how physicians’ location choices were influenced by the characteristics of their medical schools. In the conditional logit framework, the characteristics of the individual decision-maker are incorporated through interactions with the characteristics of the elements of the choice sets. In other words, an individual trait like the type of medical school one attended is allowed to affect the degree to which that individual is drawn to, or repelled by, certain county characteristics. We use the measure of medical school quality used above, an indicator equal to one if the school adopted an admissions requirement of at least 1 year of college work, similar to the methodology used by Wozniak (2010), Gottlieb and Joseph (2006) and Polsky et al. (2002).

We estimate conditional logit models separately by year and state, a choice driven by the practical concern that county characteristics vary dramatically across these two dimensions. Moreover, estimating separate models allows us to consider how new physicians’ location decisions varied over time and across states. Because we are interested in the decisions to choose metropolitan versus rural counties, we limit the sample to New York and California, since North Carolina and Mississippi contained no metropolitan areas during the sample period. By choosing to set up practice in those states, physicians effectively limited their choice set to rural counties. The sample is also limited to graduates trained in American medical schools to allow estimation of the importance of distance from medical school for location choices.

Table 5 presents the estimated odds ratios from the conditional logit models for all doctors in New York and California in 1909 and 1921. Odds ratios indicate how one-unit changes in the county characteristics affect the odds a physician chooses to locate in a particular county. An odds ratio greater than one indicates that an increase in that characteristic increases the probability a physician chooses a county whereas an odds ratio less than one indicates that an increase in that characteristic decreases the probability.

Not surprisingly, physicians were drawn to counties in metropolitan areas and, to a lesser extent, counties adjacent to metropolitan areas. By 1921, the odds a new doctor settled in a metropolitan county rather than a remote rural county was almost 8 to 1 in California and nearly 2 to 1 in New York. In both states, physicians had a preference of being closer to where they went to medical school: the odds of locating in a county were decreasing as the distance from that county to a physician’s medical school increased. This result supports the finding of Chen et al. (2013) that the location of graduate medical education programs affects the practice locations of physicians.

Table 5 Odds ratios from conditional logit models of physicians choices of counties.

Table 6 Nonlinear Oaxaca–Blinder decomposition of changes in probability of rural location of new doctors.

In both states, physicians were attracted to areas with greater numbers of established physicians, with the exception of New York in 1921. The number of established doctors can be interpreted as picking up any number of unobservable factors that draw physicians but are not included in our model (e.g., productive amenities such as public laboratories, or consumption amenities not captured by the metro/rural county categories). We find conflicting evidence for the number of hospitals beds per 1000 population. In New York in 1909, both established and new physicians were attracted to counties with more beds—1.37 and 1.48 respectively—and this attraction grew by 1921 to 1.77 and 2.65. In contrast, physicians in California were only slightly more likely to locate in counties with more hospital beds in 1909—1.04 for established and 1.09 for new doctors—, and if anything this decreases by 1921 to 0.88 and 0.96, respectively.

Of particular interest, though, is whether the effects of location characteristics on location choices differed by type of medical school training, which we measure by interacting the location characteristics with an indicator variable equal to one if the medical school attended required one or more years of college work. As previously noted, we cannot distinguish between whether this effect is one of self-selection into the more rigorous programs or the effect of the program in shaping location preferences. Our intent here is simply to estimate the relationship between medical training and later location choice. To interpret the impact of an interaction term, the odds ratio of the interaction term is multiplied with the main effect. For instance, the effect of being a metropolitan county on recent graduates of medical schools requiring at least 1 year of college is $13.77*6.23 = 85.8$ for California in 1909. Thus, new physicians trained under the more rigorous standards were more likely than other new physicians to set up practice in metropolitan counties in 1909 in California, and in both 1909 and 1921 in New York. However, in California, the attraction to metropolitan areas for new physicians did not differ by medical school type by 1921. Additionally, in New York this gap decreases between 1909 and 1921. These results suggest the growing convergence in the location choices of recent medical school graduates, partially due to the dramatic increase in the proportion of our sample who had graduated from medical schools that required pre-medical college education: from 2.6% in 1909 to 61% in 1921.

Table 7 Characteristics of recent medical school graduates by birthplace, North Carolina, 1918.

Table 8 Marginal effects from probit model of practice in rural area, North Carolina 1918.

The geographic distribution of older physicians also reflected strong differences in location choices by medical school characteristics. Graduates of schools with higher admissions standards were generally drawn to places with larger medical communities and were much more likely than their colleagues from less rigorous schools to set up practice in metropolitan areas. Interestingly, in New York, the stronger attraction of graduates of medical schools with more strict requirements for admission to metropolitan areas did not decrease between 1909 and 1921. This likely reflects the very strong pull in this state of New York City and its surrounding areas. Established physicians in both states in 1921 who attended more rigorous schools were more likely to set up practice farther from their medical schools than physicians who attended schools of lower quality, either because the applicants to more rigorous schools were more likely to move to urban areas than the applicants to programs with lower standards, or because the rigorous programs affected location decisions.²⁰

5.3 Decomposition of changes in physician location choice

In this section, we turn to trying to explain the causes of the changes in physician location choice during this transformative time for medical education in the USA. We present results from a decomposition of the change in the likelihood of rural practice for new doctors between 1909 and later years into the share due to changes in the characteristics of physicians and the share due to changes in the coefficients. The decomposition follows Fairlie (2005) and Fairlie (2017), an extension of the Oaxaca-Blinder decomposition to nonlinear models.²¹ The average difference in rural location between 2 years $t_{1}$ and $t_{2}$ can be expressed as:

$$\begin{aligned} {\bar{Y}}^{t_{2}} - {\bar{Y}}^{t_{1}}= & {} \left[ \sum ^{N^{t_{2}}}_{i=1} \frac{F(X^{t_{2}}_{i}{\hat{\beta }}^{t_{2}})}{N^{t_{2}}} - \sum ^{N^{t_{1}}}_{i=1} \frac{F(X^{t_{1}}_{i}{\hat{\beta }}^{t_{2}})}{N^{t_{1}}} \right] \nonumber \\&+\,\left[ \sum ^{N^{t_{1}}}_{i=1} \frac{F(X^{t_{1}}_{i}{\hat{\beta }}^{t_{2}})}{N^{t_{1}}} - \sum ^{N^{t_{1}}}_{i=1} \frac{F(X^{t_{1}}_{i}{\hat{\beta }}^{t_{1}})}{N^{t_{1}}} \right] . \end{aligned}$$

(3)

${\bar{Y}}^{t}$ is the proportion of doctors choosing a rural practice location in our sample in year t. ${\hat{\beta }}^{t}$ is the vector of coefficients from a probit regression of an indicator for rural location on the vector of physician characteristics in year t, $X^{t}_{i}$, including an intercept. Finally, $N^{t}$ is the total number of physicians in our sample in year t. The first term on the right-hand side of Eq. 3 isolates the difference between year $t_{1}$ and $t_{2}$ in the proportion of physicians choosing to practice in a rural location that is due to changes in characteristics, holding the coefficients constant at those estimated for year $t_{2}$, and is referred to as the “explained” portion. In our context, it is an estimate of how large the change would have been if the 1921 physicians were assigned the characteristics of the 1909 physicians, but the behaviors given those characteristics were “benchmarked” at the 1921 levels. We can just as easily apply the 1909 characteristics to the 1921 physicians, holding the 1909 coefficients constant. We report both estimates. The first term can be further decomposed into the individual contribution of each covariate. Fairlie (2017) presents a procedure to solve the path dependency problem in nonlinear decompositions because of the arbitrarily selected ordering of variables by randomly ordering the variables in each of many replications; we run 5000 replications for each decomposition.

Table 6 presents decompositions of the change over time in the proportion of new doctors choosing to settle in rural practice locations. Relative to a 1909 base-level of 33%, there was a decline of 1.08 percentage points to 1914, 5.79 percentage points to 1918, and 12.97 percentage points to 1921. Fixing the coefficients at the 1909 estimates, changes in physician characteristics are responsible for the majority of the observed change in doctors choosing rural locations: 61% in 1914, 91% in 1918, and 87% in 1921.

The bottom half of Table 6 lists the explanatory power of each individual physician characteristic. Graduation from a medical school with pre-med requirements is the most important, accounting for over 100% of the full “explained” portion. New graduates from these schools in 1909 were much less likely to locate in rural areas. The coefficient on pre-medical college requirements is negative in all years, but the largest in magnitude in 1909. The set of early schools with requirements was highly selected in 1909; 97% of these graduates were from Harvard or John Hopkins. Moreover, the proportion of the sample that attended medical schools with pre-med requirements increased dramatically over our period: from 2.6% in 1909 to 61% in 1921. Given these two facts—a large negative 1909 coefficient and a rapid increase in schools with pre-med requirements - it is no surprise that this single variable provides so much explanatory power.

The only other variable that provides some explanatory power is attendance at an out-of-state medical school, but its contribution in the decomposition is dependent on the choice of base year. In 1909, physicians trained out-of-state were more likely to practice in rural areas, all else equal [see coefficient (0.119) from Table 4]. By 1914, attending medical school out-of-state is no longer associated with rural practice. Relative to 1909, the number of physicians trained out of state in our sample increases by about 7 percentage points in all years, after partialling out all other variables. Combined with the strong positive association of out-of-state training with rural practice in the 1909 sample, these changes in proportion of out-of-state trained physicians imply an increase in rural practice relative to 1909 (i.e., the economically and statistically significant negative contribution “Attended out-of-state med school” in the first three columns of Table 6). However, when later years are used as the base, the contribution is dramatically attenuated, because out-of-state training is not associated with rural practice in these years.

Using the coefficients from the final year as the reference group attenuates the explanatory power of observable characteristics. As medical schools choosing to require pre-medical college education became less positively selected over time, the large negative correlation between college requirements and rural practice moved toward zero. Interpreting the final column of Table 6, the proportion of physicians practicing in rural areas in 1921 would have been 4.01 percentage points lower than in 1909 based on differences in their observable characteristics. This is a much smaller proportion of the observed change than that calculated using the 1909 coefficients, but it is still sizable at 31%.

We interpret these results as suggesting that changing characteristics of physicians, especially characteristics of the medical school curriculum, cost, and admission requirements, account for an important part of the total decline in the number of rural physicians over the early twentieth century. Although sensitive to the choice of reference group, the results behave in a way that is consistent with the positive selection into early adoption of the reforms with lower-ranked schools eventually following the leaders. The question remains as to why the behaviors of physicians with specific characteristics changed so dramatically over an 11-year period. Two factors likely play an important role. First, the publication of Flexner’s review and subsequent passage of state legislation requiring pre-medical college requirements caused close to half of all medical schools to close their doors. The schools that went on to close or to be subsumed into other schools by 1924 accounted for almost a third of the total medical school enrollments in 1910. Closures were almost entirely of smaller faculty-owned lower-ranked lower-cost schools. The AMA’s desire had been achieved; the so-called “diploma mills” were shuttered, and the supply of physicians contracted. In a sense, the lower half of the physician “skill” distribution was removed. With the current data available it is difficult to account for this change in the composition of physicians in our decomposition, and part of it is loaded onto the “unexplained” portion when using 1921 as the base year. What we do we capture is the second factor, the change in pre-med college requirements for the remaining schools that did not close. As the lower-ranked schools added pre-med college requirements, this category included more physicians that were inherently more likely to practice in an urban area.

6 The importance of physician birthplace

Our results show that better-educated physicians differentially moved to urban areas compared to physicians who went to schools with lower standards, which suggests one reason why a rural shortage of physicians occurred over time as medical education requirements increased. Another reason underlying the current shortage of rural physicians may be that, as contemporaries of Flexner noted, these reforms made it more difficult for people from rural areas to get into medical school. While we cannot test this directly with our data, we determine whether it was indeed true that rural-born physicians were more likely to set up practice in rural areas by linking our AMD data on physician practice locations to data on physicians’ birthplaces. Linking the AMD data to the contemporaneous population censuses would not be useful because the census only reports the state of birth, not town. Therefore, we take advantage of the digitized draft cards for World War I available on Ancestry.com, which do report town of birth. Draft registration for WWI took place in three waves. The first wave took place on June 5, 1917 and included men born between June 6, 1886 and June 5, 1896; the second wave took place on June 5, 1918, and included men born between June 6, 1896 and June 5, 1897; and the final wave took place on September 12, 1918 and included men born between September 11, 1872 and September 2, 1900. The registration cards for the first two waves included a question on the place of birth that specifically asked for town as well as state and nation. Since we are interested in learning more about physicians who set up practice in rural areas, we chose to focus on physicians practicing in North Carolina, a state with a predominantly rural population in the early twentieth century.²² We started with all North Carolina physicians listed in the 1918 AMD who were born between 1886 and 1897, the group required to register in the first two waves of the draft. This gave us a population of 422 physicians. We were able to find the draft cards for 360 of these doctors, a match rate of 85%. However, for seven of these, the draft card was from the third wave of registration and did not contain detailed birthplace information. To increase our sample size, we searched Ancestry.com for any records (e.g, other military records, death records) for the doctors for whom we were not able to get information on birthplace in the draft cards, including those matched but having blank and illegible entries for town of birth. In the end, we were able to find the town or county of birth for all but 56 physicians out of the 422 in the original sample, but restrict our analysis to the 302 physicians for whom we have the town or county of birth and who were born in North Carolina.²³

Table 9 Oaxaca–Blinder decomposition of difference in rural practice location by birthplace, North Carolina, 1918

Table 7 provides summary statistics of this sample broken down by rural v. urban birthplace. Because North Carolina is such a rural state with few large cities, we follow the US Census guidelines of 1910, by classifying a birthplace as rural if it has fewer than 2500 inhabitants.²⁴ As shown in Table  7, 73.8% of physicians born in rural areas chose to practice in rural areas, while only 24.2% of physicians born in non-rural locations chose to practice in rural areas. In fact, almost a third of those born in rural areas (73 out of 240) were practicing in the town in which they were born. Those with non-rural birthplaces were also more likely to have attended a “better” medical school—one that required at least 1 year of college work and one that survived past 1923. However, these “quality” measures did not translate into a higher pass rate on state boards, calculated at the school level. Finally, differences are reported by race. Draft cards report the race of the physician, whereas the AMD lists do not. The vast majority of black physicians in North Carolina were born in urban areas, potentially leading to a significant urban–rural gap in physician access for black patients in Jim Crow era when racial segregation in services was legal and access to medical care differed for blacks and whites.

Table 8 presents marginal effects from a probit model of rural practice choice of North Carolina physicians, controlling for rural nativity, medical school location and quality, and whether the physician was black. All else equal, a physician born in a rural area was roughly 45 percentage points more likely to settle in a rural area than one who was not, regardless of the quality of the medical school attended. The entire sample of North Carolina doctors practiced in rural areas 63.6% of the time. Clearly, being from a rural area made it much more likely a physician would set up practice in a rural area, and this single factor made a significant contribution to overall rural access to physicians. However, there is still evidence that medical school characteristics mattered. Even after controlling for graduation year fixed effects, having attended a medical school with one or two years of college required for admission reduces the likelihood of practicing in a rural area by around 22 percentage points. This effect is present even for physicians born in rural areas. In the final column, the indicator for pre-med college requirements interacts with an indicator for rural birth. We find no evidence suggestive of heterogeneous effects of college requirements based on physician nativity.

Next, we conduct a nonlinear Oaxaca-Blinder decomposition on the large difference in the likelihood of rural practice between the urban-born and rural-born North Carolina physicians. Results are reported in Table 9. Rows (1) and (2) show that 73.75% of rural-born physicians chose to practice in rural locations compared to only 24.19% of urban-born physicians. This large difference in the likelihood of rural practice location between the two groups is not explained by differences in observable characteristics of the physicians or the medical schools from which they graduated. Holding the coefficients constant at those of the urban-born sample, rural-born physicians would have been 15.6 percentage points less likely to practice in rural areas if they had the same observable characteristics as the urban-born physicians, which explains only 31.5% of the total difference. Repeating the same exercise, but using their own coefficient estimates, rural-born physicians would have been 9.16 percentage points less likely to practice in rural areas, explaining only 18.4% of the total difference. “Unexplained” factors account for the vast majority of the difference in rural practice, which may be interpreted as reflecting differences in preferences, information, attitudes, or omitted variables. Had a public health authority wanted to increase rural access to physicians in the 1920s – or at least prevent the decline of the country doctor – the policy prescription from our results would suggest reducing the barriers faced by rural-born applicants to medical school. In reality, the American Medical Association and Association for American Medical Colleges, in conjunction with the state-based physician licensure boards, worked to increase admission requirements that would heighten the barriers for “country boys” to become doctors.

Although the results are from a single year and do not allow us to examine the impact of changing medical education over time, they suggest that the early twentieth-century medical education reforms did in fact reduce the supply of physicians in rural areas through two separate effects. First, physicians trained in “modern” medical schools tended to locate in urban areas. Second, rural-born doctors provided the majority of rural-practicing physicians, and the medical reforms reduced the supply of physicians by limiting the opportunities of the “country boys” to attend medical school. However, they did so by reducing the supply of poorly trained physicians. We note, however, that we have no evidence that “poorly” trained meant worse outcomes for the patients.

7 Conclusions

During the first decades of the twentieth century, the overall population of the USA became more urbanized, and physicians and surgeons moved to urban areas at greater rates than the population as a whole. Using novel data from the American Medical Directories, we examine whether the location choices of physicians graduating in the early twentieth century were related to the massive changes in medical education that occurred during the period. We find that physicians’ location decisions did vary systematically with their vintage and the quality of their medical school training. Recent medical school graduates, particularly those from the higher quality medical programs, were more strongly drawn to urban areas than were other physicians. While our data do not allow us to determine whether these results are driven by self-selection of students or the impact of more rigorous programs on students, we do find evidence supporting the “country boy” story: that those doctors most likely to return to rural areas were born in rural areas. Using conditional logit models, we confirm that graduates of better schools were disproportionately more likely to be drawn to metropolitan areas, with some evidence they were also drawn toward areas with more hospital beds per capita. Taken together, these results suggest that the modern-day “shortage” of physicians in rural areas may have had its antecedents in changes occurring in medical education over 100 years ago. They also indicate that policies that attract students from rural areas to medical school, and policies that increase the ability of rural physicians to connect with other physicians and professional amenities, may help to lure more physicians to less populated areas.

Moreover, the results indicate that a skills gap between rural and urban physicians may have developed over time. While Flexner believed that the graduates of more rigorous programs would be just as ready to settle in rural areas as graduates of schools with lower standards, we find this was not the case. More recently trained physicians and those trained in more science- and lab-based curricula were increasingly drawn to urban areas. Many of the physicians who remained in the countryside and small towns had been trained in the nineteenth century or at schools with less rigorous courses of study. These doctors may have been less able or willing to adopt the new medical technologies that were being developed during this period. Hence, rural populations potentially benefited less than urban populations from the dramatic advances in medical technology of the mid-twentieth century.

Appendix

See Fig. 4 and Tables 10, 11, 12, 13, 14, 15, 16 and 17.

Fig. 4

Example page from 1906 American Medical Directory

Table 10 Determinants of rural $(< 10,000)$ practice location choice: linear probability model.

Table 11 Determinants of rural $(< 2500)$ practice location choice: probit marginal effects.

Table 12 Rural practice location choice of new doctors: probit marginal effects (including AMA ratings as controls).

Table 13 Linear Oaxaca–Blinder decomposition of changes in probability of rural location of new doctors.

Table 14 Characteristics of recent medical school graduates by birthplace (rural defined as pop < 10,000), North Carolina, 1918.

Table 15 Results from linear probability model of practice in rural area, North Carolina 1918.

Table 16 Marginal effects from probit model of practice in rural area (defined as pop < 10,000), North Carolina 1918.

Table 17 Determinants of rural practice location choice for new doctors: probit marginal effects.