Medical Workforce

3. Labor Supply Topics

3.1. Rationale for Government Intervention in Medical Labor Markets

In practice, medical markets have not been allowed to operate as described above, so we have no knowledge of whether they would in fact do so. Most medical labor markets are heavily regulated because policy makers do not believe the unaided market will produce the optimal number, quality, specialty mix, and/or geographic location of health professionals. The principal concern is that because of asymmetric information consumers will not be able to determine the quality of services provided by medical labor (Arrow, 1963). One response is to try to assure consumers that the inputs into the health production function exceed a minimum acceptable quality level. The policies consist of some combination of: (1) requiring workers to be licensed before they may legally practice; (2) certifying workers who pass an exam to distinguish them from lower-quality labor; and (3) accrediting schools and programs to ensure that graduates face a rigorous curriculum. In the United States, for example, physicians must graduate from an accredited medical school and complete at least one year of residency training at an accredited program before being eligible for a medical license. In Japan, an aspiring physician must graduate from an accredited medical school, complete a two-year internship, and pass a national licensing exam before being allowed to treat patients.

Another way to increase the quality of the medical workforce is to subsidize medical education, which creates rents for successful applicants and produces an excess of applicants relative to available positions. High-ability students are more likely to be admitted when medical school positions are rationed. Rationed entry is likely to make licensing irrelevant for physician services; medical schools will reject applicants at the lower part of the quality or ability distribution (Arrow, 1963). Licensing can still play an important quality assurance role for non-physician personnel whose education is not subsidized and rationed, and for graduates of international medical schools who may not have been subject to as rigorous an admission screen as domestically trained physicians (Cooper and Aiken, 2001).

There are costs associated with medical licensing generally and restrictions on entering medical school. Licensing places constraints on the health production function by making it illegal to substitute between certain types of labor inputs for certain tasks. Licensing fees and the education requirements necessary to receive a license increase the cost of entering a profession and reduce labor supply at all wages, and the rationing of medical school slots truncates the supply of physician services. This will increase costs in output markets (e.g. physician services), increase the wages of licensed professions, and restrict product variety available to consumers. As Arrow (1963) pointed out almost 50 years ago, “Both the licensing laws and the standards of medical-school training have limited the possibilities of alternative qualities of medical care. The declining ratio of physicians to total employees in the medical care industry shows that the substitution of less trained personnel, technicians, and the like, is not prevented completely, but the central role of the highly trained physician is not affected at all” (Arrow, 1963, p. 953).

The regulation of medical labor markets creates rents, which in turn creates incentives for providers to organize to capture (and expand) these rents. In fact, governments often allow medical professions to define and control the nature of the regulation by allowing professional associations to define the requirements for obtaining a license, to administer licenses, or, in some cases, to explicitly determine the number of people allowed to enter the profession or a specialty within the profession. This creates the appearance that medical professions are restricting entry to create rents, and is supported by theoretical predictions that a professional group allowed to define the minimum quality threshold will set a standard that exceeds the socially optimal benchmark (Leland, 1979).

In the United States the Liaison Committee on Medical Education, which is formed by the Association of American Medical Colleges (AAMC) and the American Medical Association (AMA), accredits US medical schools. Because it is difficult for a student who attends a non-accredited medical school to practice medicine in the US, the number of US medical schools is essentially determined by physician organizations. As described in section 3.5, specialty associations in the US determine how many residents can train in each specialty, and therefore determine the flow of new entrants.

Self-regulation by non-governmental medical labor organizations is common in many other countries. For example, the Medical Council of India, the Korean Institute of Medical Education, the General Medical Council (UK), the Netherlands Flemish Accreditation Organization, and the Japan University Accreditation Organization approve curricula and accredit medical schools in their respective countries.

3.2. Licensing, Certification, and Accreditation: Objectives and Implications

In this section we examine theoretical arguments for how the government can address the asymmetric information problem, the extent of occupational regulation that actually occurs in practice, and the implications of occupational regulation on output quality, the quantity of licensed and unlicensed labor, prices, and the earnings of regulated professions.

If producing information is costless, the ideal solution is to eliminate information asymmetries directly by providing consumers with information about the attributes and quality of medical services (Leland, 1979). Consumers are primarily interested in the expected health outcomes that result from a production process, not the quality of inputs that enter into the production process. That is, they want information on the efficacy of bypass operations performed by a particular physician at a particular hospital, not information on the quantity and type of education and training received by the physician, nurses, operating room technicians, pharmacists, and therapists who work at that hospital.

Dranove and Jin (2010) define quality disclosure “as an effort by a certification agency to systematically measure and report product quality for a nontrivial percentage of products in a market.” Such disclosure could be self-reported voluntarily by sellers, or disclosed by a third party. The US health care market is moving in this direction, particularly with third-party reports by organizations such as US News & World Report, Medicare, the Leapfrog group, and New York and Pennsylvania for

Cardiovascular Care. Although most empirical studies find that highly ranked providers benefit after the dissemination of information, the effects tend to be small (Dranove and Jin, 2010). Two possible explanations are that it is difficult to measure a provider’s value added given the difficulty of measuring a patient’s health, both before and after treatment, and consumers and referring providers already knew the information in the report cards.

A second way to address information asymmetry would be to make sellers liable for poor quality (Leland, 1979), as is currently done with medical malpractice. If quality problems are easy to detect and providers bear the cost of poor quality, malpractice might be an effective policy. Unfortunately there is little evidence that the medical malpractice system deters negligent behavior because few of the patients who experience malpractice actually sue, these patients are as likely to win a lawsuit or settlement as patients who did not in fact experience negligence, and physicians’ malpractice premiums are not experience rated (Danzon, 2000).

A third alternative is to have health insurers, acting as consumers’ agents, write quality-contingent reimbursement contracts with providers (Shapiro, 1986). If health insurers can observe the quality of medical services or the patient’s health outcome, high-ability providers and providers who invest in quality-enhancing skills will be rewarded. As with provider report cards, this is the direction in which the market is headed. Primary care physicians in the United Kingdom, for example, received average incremental payments of about $40,000 per year (32 percent of their income) for providing services deemed to be high quality. In most countries, however, pay-for-performance payments constitute a small percentage of a provider’s income or profits. A US physician who received the maximum pay-for-performance payments, for example, would have received an estimated 5 percent of her base income as a bonus, and a hospital only an estimated 2 percent of their profit (Nicholson, 2008).

A potentially less expensive alternative to the above quality assurance approaches is to regulate certain inputs into the health production function and thereby assure consumers indirectly about output quality. This indirect approach is not ideal, but it may be the most feasible approach, particularly because licensing enforcement is done one time only when a professional is trained, rather than perpetually in the case of output regulation (Shapiro, 1986). Kleiner and Krueger (2009) highlight three different types of occupational regulation. Registration, the least restrictive form, requires individuals to provide their name, address, and qualifications to a government agency before practicing. Certification allows any person to provide a service, but the government or a private, non-profit agency distinguishes providers who have passed an exam. Under licensing, the strictest form of occupational regulation, it is illegal for a person to work in an occupation for compensation without first meeting the state or federal licensing standard. A licensing board typically examines a candidate’s credentials, determines whether the schools and degrees satisfy the minimum training standards, and then sets the pass rate of the licensing exam (Kleiner, 2006).

According to a 2008 survey, an estimated 29 percent of all US workers indicated that they were required to have a government-issued license to perform their job (Kleiner and Krueger, 2009). In the United Kingdom, by contrast, about 13.5 percent of the general workforce required a license in 2008. Licensing is much more common in health-related occupations. Among non-physician US health care workers, about 76 percent require a license (Humphris et al., 2011), and all physicians are required to be licensed. In the UK, an estimated 73 percent of the medical workforce (including physicians) requires a license.

There are two prominent theoretical models describing the motivation for and likely impact of occupational licensing. Leland (1979) focuses on how licensing can ameliorate the adverse selection problem stemming from information asymmetries. As in Akerlof’s (1970) model of used cars, if physicians know their own abilities but consumers do not, physician fees will not vary by quality, and prospective physicians of relatively high ability will decide not to enter the profession. This withdrawal reduces the average quality of physician services, reduces prices, and results in a profession of quacks only. Social welfare can be increased if a licensing authority establishes a minimum quality level, but if a professional group is allowed to define the quality threshold necessary for a license, the minimum quality level will exceed the socially optimal level. Licensing will be most beneficial in markets where: (1) consumers value quality highly; (2) demand is inelastic; and (3) the marginal cost of providing quality services is relatively low. Although the second criterion is likely to be true for most medical services and probably the first as well, the marginal cost of providing quality services is likely to be high.

In Figure 1, licensure increases the cost of entering an occupation and shifts upward the supply of labor. Entry costs will increase if workers have to pay license fees or, more importantly, if the licensing authority requires substantial education/ training or a long residency requirement. If consumers believe that licensing improves output quality, demand for medical services and the prices of those services may rise, thereby increasing the licensed profession’s marginal revenue product and shifting out labor demand. Both of these phenomena would cause the wage of the licensed occupation to rise.

The impact of licensing on health outcomes is unclear theoretically. Licensing screens out low-quality providers, restricts supply, creates rents, and increases the expected returns to quality-enhancing training. This will increase the quality of labor inputs and thus the quality of medical services provided. However, the higher licensed wage will encourage firms to substitute capital and non-licensed labor for licensed labor where possible. Licensure may also place restrictions on the health production function, such as forbidding nurse assistants from administering drugs or requiring that certified registered nurse anesthetists be supervised by an anesthesiologist. These constraints, along with higher wages for licensed labor, will increase output costs, output prices, and reduce the quantity consumed. If licensed labor represents a large share of the production costs, the marginal rate of technical substitution between licensed labor and other inputs is small, and output demand is inelastic, which all would appear to be true empirically for medical services in general, then production costs and prices may increase substantially as a result of licensing. Higher prices will reduce the amount of medical services received. Therefore the quality of medical services actually provided should increase due to licensing, but the amount of services provided should decrease; the net effect on population health is ambiguous (Kleiner and Kudrle, 2000).

In the Leland model, quality is innate and licensing prevents low-quality workers from entering an occupation, truncates the ability/quality distribution, and thereby improves average worker (and output) quality. Shapiro (1986), on the other hand, argues that quality derives in part from endogenous human capital investment. In Shapiro’s model all providers receive the same price for services rendered during an initial “introductory” period. Once consumers observe reputations, however, providers who invested substantially in human capital receive a higher price than providers who accumulated less human capital. Licensing requires workers to receive a minimum level of human capital. This constraint increases the price of low-quality services (to encourage entry) but reduces the price of high-quality services based on the critical assumption that an increase in training reduces the marginal cost of quality.

In Shapiro’s model licensing benefits consumers who value quality highly (due to the lower price of high-quality services) but harms consumers who do not (due to the higher price of low-quality services). The social cost of licensing is the excessive training received by low-quality service providers—training that is not appreciated by consumers who do not value quality highly. Although it is difficult to determine whether the benefits of enhanced quality outweigh the excess training costs, Shapiro highlights that strict licensing is likely to reduce welfare when: (1) consumers and/or referring physicians can observe a provider’s reputation very soon after the training period; (2) when the marginal cost of quality is large; and (3) when few consumers value quality highly.

Are provider reputations likely to provide sufficiently strong incentives for workers to invest in human capital to the point of rendering licensing irrelevant? Recent empirical work on bypass surgery would indicate that the answer is “no.” Johnson (2010) finds that relatively low-quality cardiovascular surgeons are 10 percent more likely to stop performing surgeries or to relocate relative to higher-quality surgeons. She finds no evidence, however, that high-quality cardiovascular surgeons perform more procedures or generate more charges than low-quality surgeons. This weak relationship between provider quality and earnings could be due, in part, to poor information. Kolstad (2009) presents evidence that when Pennsylvania publicly reported the quality of cardiac surgeons, health outcomes improved substantially due to physician’s intrinsic motivation. That is, once a physician realized where she stood in the quality distribution, she invested in additional human capital for reasons other than increasing her earnings.

The existing empirical studies generally provide a pessimistic assessment on the welfare effects of licensing in medical labor markets. These studies conclude that licensing is associated with restricted labor supply, an increased wage of the licensed occupation, rents, increased output prices, and no measurable effect on output quality. Kleiner (2006) summarizes the literature on the effect of licensing on earnings as follows: “For the higher-education and higher-income occupations working mainly in the quasi-private sector, like physicians, dentists, and lawyers, licensing appears to have large effects (on earnings) through either limiting entry or restricting movement to the state.” For other medical occupations the results generally show no effect on wages, such as for nurses (White, 1980), or a small effect on wages, such as for radiologic technicians (Timmons and Thornton, 2008) and clinical laboratory personnel (White, 1978). Kleiner (2006) concludes that occupations that work independently and interact directly with customers, such as physicians and dentists, receive larger benefits from licensing than paraprofessionals who are supervised by others, such as nurses.

However, no study has yet examined the effect of licensing on the demand for physician services and the health outcomes produced by physician services, which is arguably the most important licensing policy question given the central role played by physicians. This gap in the literature exists not for lack of interest in the result, but for lack of a research design that produces causal effect of licensing. The non-licensing counterfactual is difficult to predict when physicians in all developed countries are licensed. Furthermore, rationing of medical school slots may screen out low-ability physicians such that the licensing requirements are not binding. Arrow (1963), among others, believes that there are likely to be substantial positive welfare gains associated with at least some degree of physician licensing.

The simplest way to estimate the wage effect of licensing is to compare the wages of licensed occupations to unlicensed occupations that have similar skill and education requirements. More sophisticated studies also control for individual-level human capital and demographic control variables that might have independent wage effects. Using the latter method and data from the 1990 and 2000 US censuses, Kleiner (2006) finds that physicians and dentists earn 50 percent and 90 percent more, respectively, than biological and life scientists (an unlicensed occupation), controlling for other individual characteristics.

Other studies exploit differences across states in whether licensing is required, or the degree of licensing (e.g. the percent of applicants who pass a dental licensing exam, or whether a state accepts a dental license from another state) when licensing is required in all states. Anderson et al. (2000) find that US physicians who practice in states with strict regulations on the practice of alternative medicine earn higher incomes. Kleiner and Kudrle (2000) find that US dentists practicing in states with the strictest licensing requirements earn 12 percent more than those practicing in less regulated states, and White (1978) finds that clinical laboratory personnel in licensed states earn 17 percent more than those in unlicensed states.

The above results are at odds with the studies reviewed later in section 3.4, which find that dentists and physicians earn similar (or lower) rates of return relative to “similar” occupations. The studies in the previous paragraph assume that selection on unobservables is unimportant within an occupation between states (e.g. productive dentists do not locate disproportionately in states with strict licensing requirements); the rate of return studies assume that selection on unobservables is unimportant between occupations (e.g. productive college graduates do not choose an MBA degree disproportionately versus a medical degree). Although one of these selection effects is likely to be more important than the other, it is difficult to determine which one.

Two studies use instrumental variables or a regression discontinuity design to estimate the effect of licensing on wages while accounting for the possibility that individuals with relatively high or low unobserved ability may select into licensed occupations (or states requiring a license within the occupation). Kugler and Sauer (2005) find that physicians emigrating to Israel from the Soviet Union who had slightly more than 20 years of prior experience, and thus were granted a license to practice in Israel, earned more than three times as much as physicians with slightly less than 20 years of experience, who faced a more onerous re-licensing process. Interestingly, the OLS estimate is 50 percent smaller, indicating that immigrant physicians who acquired a license had lower earnings potential than those who did not. That is, relatively low-ability workers may seek the economic protection provided by licensing. Timmons and Thornton (2008) find similar negative selection for radiologic technicians. The estimated effect of licensing on wages using instrumental variables (7 percent) is twice as large as the OLS estimate.

Licensing also appears to increase wages in the European Union. In the United Kingdom, wages in the licensed occupations of medical practitioners, pharmacists, pharmacologists, and dental practitioners are an estimated 6 to 65 percent higher than otherwise similar workers in unlicensed occupations (Kleiner, 2006). Physicians and dentists in France earn an estimated 8 to 21 percent more than their unlicensed colleagues, whereas workers in those professions in Germany have similar wages relative to unlicensed occupations.

Empirical studies confirm that licensing increases wages by restricting labor supply. Kleiner (2006) finds that the supply of respiratory therapists and dieticians grew faster in US states that did not require a license for these occupations versus states that did. Kleiner and Kudrle (2000) find that the number of dentists per capita grew more slowly between 1980 and 1990 in US states with stricter dental licensing regulations. Finally, Wanchek (2009) finds that the supply of dental hygienists per capita is higher in US states with fewer entry restrictions.

Although there are many studies documenting a link between licensing and rents to the regulated occupation, there are no known studies identifying a positive effect of licensing on the quality of medical services or patients’ health outcomes—the rationale for licensing. Kleiner and Kudrle (2000) surveyed new enlistees in the Air Force to collect information on where they grew up, whether their family had dental insurance, and the other household characteristics. They combined this information with data on the strictness of dental licensing requirements in each US state, and a comprehensive dental exam performed at the time of enlistment. They find that strict dental licensing is not associated with better dental care, fewer complaints to a state licensing board, or lower dental malpractice premiums. Strict licensing is associated, however, with higher dental prices, suggesting that the reduced demand associated with licensing could offset any improvement in input quality (although they do not detect differences in input quality in this study). Wanchek (2009) reaches a similar conclusion with regards to dental hygienist licensing: higher wages for dental hygienists due to strict licensing rules are associated with a lower demand for dental care. It is worth repeating, though, the empirical difficulty of measuring the benefits of physician licensing, as mentioned above.

One policy recommendation is for the government to certify rather than license medical labor. Under this scheme, a government agency administers an exam and certifies individuals who pass, while still allowing individuals who do not pass to legally practice the profession. This would allow consumers (and/or health insurers) to decide whether to pay higher wages for workers deemed to be high quality without barring low-quality workers altogether (Kleiner, 2000). This policy, which would still provide incentives for human capital investment, would seek to eliminate or reduce excessive training received by low-quality service providers—the source of the social costs of licensing in Shapiro’s (1986) model. Shapiro points out, however, that certification can still lead to overinvestment by high-quality workers in an attempt to signal their quality.

Humphris et al. (2011) examine the welfare effects if the US federal government no longer paid a wage premium on non-physician licensed labor. They estimate that such a policy would shift $102 billion (in 2008) from medical labor to consumers, thereby reducing medical spending by about 5 percent.

Given the negative empirical findings, why is licensing so prevalent in the medical labor market? One option would be to license professionals who interact directly with consumers, such as physicians and dentists, and allow the quality of paraprofessionals (e.g. nurses, technologists) who work under the supervision of professionals to be monitored by the professionals. White and Marmor (1982) consider such a structure, where professionals have an incentive to use paraprofessionals as long as the productivity gains exceed the costs. They argue, however, that medical professionals in the US have avoided this monitoring role because they lack administrative training and the opportunity cost of their time is high. Another explanation might be that in a fee-for-service system such as the US, no single professional is responsible for a consumer’s health. When many providers work independently (e.g. home care nurse or physical therapist), there is a greater need to license paraprofessionals. This raises the possibility that in countries or health systems that rely more heavily on capitation or explicitly place physicians in charge (e.g. UK), licensing may be less valuable. Cooper and Aiken (2001) argue that the main impact of the proliferation of paraprofessional (or non-physician clinician) licensing is to empower chiropractors, optometrists, psychologists, physician assistants, nurse practitioners, and nurse anesthetists to provide services that substitute for physician services. Under this interpretation paraprofessional licensing increases the marginal rate of technical substitution between physicians and non-physician clinicians, and reduces physician rents associated with licensing and entry barriers generally.

3.3. Subsidizing Medical Education

In 1906 the Council on Medical Education of the American Medical Association (AMA) inspected the 160 medical schools that were operating in the United States and fully approved only 82. In 1910 the Carnegie Foundation issued the Flexner Report, which recommended fewer medical schools and controls to assure a quality educational curriculum. The number of medical schools fell to 85 in 1920, 73 in 1930, and 69 in 1944, before rising to its current level of 126 (Kessel, 1958). Early economic studies concluded that the AMA controlled the number of medical school positions and set the number to generate rents (Friedman and Kuznets, 1945; Kessel, 1958). Kessel likened the AMA’s position to “giving the American Iron and Steel Institute the power to determine the output of steel” (Kessel, 1958, p. 29).

The ratio of applicants to medical (and dental) school positions seems to support the hypothesis that a cartel restricts entry to generate rents. There are about ten times as many applicants as available positions in the UK and China, two to three times as many in the US and the Netherlands, and about twice as many dental school applications as positions in the UK (Jetha, 2002). As McGuire (2000) points out, if medical school tuition were set equal to cost, the surplus of applicants would indeed constitute strong evidence for the cartel theory.

But medical school is heavily subsidized, so of course the demand for slots will be higher than if tuition was set equal to cost. The complexity of medical school finances makes it difficult to measure the percentage of a medical school education actually paid by students. In 2006 tuition and student fees accounted for only 3.4 percent of US medical schools’ revenues. The largest funding sources were medical services provided to patients (38.0 percent), federal research grants and contracts (20.0 percent), transfers from universities and teaching hospitals (20.0 percent), federal, state, and local government appropriations (6.1 percent), and gifts and endowment income (4.5 percent). Although this probably overstates the magnitude of the subsidy, government transfers relax medical schools’ budget constraints and allow them to set tuition below the cost of education. That said, medical schools choose to subsidize tuition, possibly to satisfy donors’ preferences; they are not mandated to do so. An older study that collected detailed spending data concluded that in 1993 US medical school tuition covered only 16 percent of instruction costs in 1993 (Ganem et al., 1995).

There are several implications of subsidized medical education. First, medical schools have great control over the medical workforce; their decisions regarding how many students to accept determine the flow of domestically trained physicians. Second, determining whether medical schools are restricting entry to create physician rents rather than fulfilling other more noble objectives requires a model of non-profit behavior, which several authors have articulated (Hall and Lindsay, 1980; Eckstein et al., 1988). Third, the rationing of medical school positions may make licensing irrelevant or less important because schools will weed out (some) low-quality applicants (Arrow, 1963). Fourth, it is unclear whether there would be fewer or more physicians trained if medical schools set tuition equal to cost (McGuire, 2000). Fewer students would demand a position once the subsidy is removed, but positions would now be determined by supply and demand rather than by fiat.

3.4. Rates of Return to Medical Training and Specialization

Entry barriers can create rents. Above we reviewed the literature on the effect of licensing on wages, but licensing is only one possible entry barrier. In this section we review the literature on the rate of return to medical training and to specialization within medicine, which should capture the complete effects of all entry barriers. Persistently high rates of return provide evidence for the existence of entry barriers. Calculating rates of return requires data on expected lifetime earnings in a medical profession and an alternative profession, the costs of education, and the length of training. Ideally the alternative profession accounts for any possible selection on unobserved characteristics.

Nicholson (2008) summarizes eight studies that estimate the rate of return to a medical school education in the United States between 1929 and 1990. The general conclusion is that the financial returns from entering medicine are generally in line with the returns experienced by similar occupations. Using earnings data from 1990, Weeks et al. (1994) estimate that physicians can expect a 16 percent return on their investment relative to high school graduates. The same study reported that dentists (21 percent), lawyers (25 percent), and MBAs (28 percent) could expect greater rates of return than US physicians. Weeks and Wallace (2002) updated their analysis using 1997 rather than 1990 income data, and estimated similar rates of return. Using crosssection earnings data from 2008 and assuming a 3 percent real discount rate, Vaughn et al. (2010) report that the average primary care physician could expect to earn $2.5 million over his lifetime, net of education costs, versus $1.7 million for an MBA graduate and $340,000 for the average college graduate. This study does not report rates of return that account for different training lengths.

The three studies above assume that the current earningsexperience profile will remain the same in the future and they do not allow occupation-specific earnings to depend on a person’s observed or unobserved abilities. The Weeks et al. (1994) 16 percent rate of return estimate is likely to be too high because physicians could certainly expect to earn more than the average high school graduate if they decided to forego medicine. Using earnings data between 1979 and 2004 from the 1979 National Longitudinal Survey of Youth (NLSY), Glied et al. (2009) model the earnings physicians would receive if they decided not to attend medical school based on their observed ability (e.g. grade point average, AFQT test score). They estimate a relatively low 7 to 9 percent rate of return for US primary care physicians.

Few rate of return studies have been performed for other health professions in the US, or for other countries. Using 1996 earnings data, Stark (2007) finds that physicians, dentists, optometrists, and veterinarians in Canada experience rates of returns similar to professions such as engineering, business, and college graduates generally. Morris and McGuire (2002) estimate that nurses in the United Kingdom experience rates of return of 8 to 13 percent in the early 1990s, and Mott et al. (1995) present evidence that pharmacists in the US in the 1980s and early 1990s had rates of return similar to high school graduates. Taken as a whole, the literature generally does not find excessive rates of return to a health education.

By contrast, most studies find that physicians who specialize in non-primary care earn substantial returns relative to general practitioners or family practitioners. Nicholson (2008) summarizes four studies that estimate the returns to medical specialization in the US between 1951 and 1998. The results of these studies show that the returns to specializing in a particular area within medicine (e.g. surgery, radiology, obstetrics/gynecology) are high and have increased sharply over time. Between 1987 and 1998, for example, the rate of return in radiology (relative to family practice) has ranged from 47 to 105 percent. The persistence of these high rates of return, combined with an excess of applicants relative to available positions, indicates that there are entry barriers. In the next section we discuss whether specialty associations functioning as cartels create these barriers, or whether they are due to more benign factors. Cordes et al. (2002) also find large rates of return to specializing within dentistry in the United States: 16.6 percent for orthodontics and 26.8 percent for oral and maxillofacial surgeons. Finally, Vaughn et al. (2010) estimate that a cardiologist could expect to earn twice as much as a primary care physician ($5.2 million versus $2.5 million) in 2008 dollars, net of income taxes, living expenses, and education expenses.

3.5. Economics of Professional and Specialty Associations

In this section we explore why there are persistent high rates of return to physician specialization. Medical school graduates must receive at least one year of residency training at an accredited residency program in order to practice medicine in the United States. Therefore, the market for residents functions as an intermediate market that largely determines the number and specialty distribution of physicians in the United States. In 2008, a total of 22,000 first-year residency positions were offered by over 1,200 hospitals in 26 different specialties. Sixty-nine percent of these positions were filled by students who graduated from US medical schools, with most of the remainder filled by international medical graduates (IMGs). Because there is routinely an excess supply of residents, any increase in the number of US medical school positions (as is currently occurring) without a concomitant increase in the demand by hospitals for residents is likely to crowd out IMGs without affecting the flow of newly licensed physicians.

In the United States a computer algorithm allocates prospective residents to residency programs each spring in the National Resident Matching Program, or the “Match.” Highly desirable specialties, usually those with high rates of return, tend to have an excess supply of residents. Between 1991 and 2009, for example, the ratio of the supply of residents (i.e. the number of applicants ranking a specialty as their first choice) to the demand for residents (i.e. the number of available first-year residency positions) exceeded 1.40 in orthopedic surgery in all but one year, and between 1997 and 2009 the ratio exceeded 1.60 in dermatology in all but one year.

In a well-functioning labor market, the excess supply of residents should reduce the residents’ wage in oversubscribed specialties, encourage hospitals to open more positions in those specialties, eventually increase the number of licensed physicians entering the market, and equalize rates of return across specialties. The fact that this does not happen confirms that there are barriers restricting entry to certain non-primary care specialties. These barriers exist in other countries as well. It is common for medical school graduates in Greece to wait several years for a non-primary care residency position opening (Mariolis et al., 2007), and only 3 percent of Greek medical school graduates enter the relatively low-paying family practice specialty.

The high rates of return to specialization and the entry barriers have created a tournament of sorts among US medical students. The four most difficult specialties to enter in the US in Match in recent years have been plastic surgery, dermatology, orthopedic surgery, and otolaryngology (i.e. ear, nose, and throat surgery). Not surprisingly, these specialties have high mean incomes and/or desirable work schedules.

Between 18 and 45 percent of US medical school graduates who ranked one of these four specialties as their first choice in the 2007 Match were not able to secure a firstyear residency position in that specialty. Students need to perform well in medical school if they want to obtain a residency position in one of these competitive specialties. After the second year of medical school, all US medical students take the National Board of Medical Examiners Step 1 exam. The mean Step 1 score for US medical students who successfully entered these four specialties ranged from 236 (orthopedics) to 246 (plastic surgery), well above the overall mean score of 220.

There also appears to be a tournament in Japan for entrance into certain specialties. The Japanese government keeps its health care costs relatively low by setting prices. But prices are not regulated in services that are excluded from national health insurance, such as cosmetic surgery. Ramseyer (2008) finds that cosmetic surgeons in Japan earn more than other specialists, are more talented, have superior training, and are more likely to have attended a prestigious medical school.

The entry barriers at the specialty level do not appear to be justified on the basis of asymmetric information between providers and consumers regarding the quality of medical services. All physicians must be licensed, regardless of specialty, and all prospective residents have already run the medical school gauntlet that screens out low-ability students. Nicholson (2003) explores four possible explanations for the persistently large rates of return to specialization in the US: cartel behavior by professional associations, a shortage of teaching material, wage rigidity, and entry barriers to deter supply-induced demand. Physicians exert great control over specialty entry, so the question is not whether physicians restrict entry, but whether their motives are purely self-serving.

One possible explanation is that Residency Review Committees (RRCs), private organizations in the US that consist primarily of physicians from a particular specialty, restrict the flow of new physicians to that specialty to create rents. The Accreditation Council of Graduate Medical Education (ACGME), a private organization responsible for overseeing residency training, is sponsored by five provider organizations (Accreditation Council of Graduate Medical Education, 1996). The ACGME sets overall policies and allows a separate RRC to review and accredit residency programs in each of the 26 specialties. Each of the five sponsoring organizations appoints four representatives to an RRC and the government appoints one non-voting representative.

A teaching hospital that wants to open a new residency program or increase the number of residents in an existing program must apply to the relevant RRC. Therefore an RRC essentially has complete control over the flow of physicians into a specialty because medical students who attend residency programs that are not certified by the ACGME are not eligible to take the licensing exam, and thus are not able to practice in the United States. Entry into specialties in the Netherlands is also strongly influenced by the physicians themselves: “medical school graduates can only register as general practitioners or medical specialists after completing specific training programs. The capacity of the training programs and the criteria for registration are determined largely by professional associations” (Schut, 1995, p. 644).

An RRC has the same power as a labor union that is the sole source of a certain type of labor and negotiates collectively with a firm or set of firms (Nicholson, 2003). Economists have modeled labor unions as maximizing a utility function whose arguments are the quantity of union members employed and either the workers’ rents or the total wage bill, subject to the firms’ labor demand (Pencavel, 1984; MaCurdy and Pencavel, 1986). The union chooses the wage and firms respond by hiring a quantity of union workers such that the marginal product equals the wage.

The relevant constraint for an RRC is consumers’ aggregate demand for physician services. Whereas a union chooses the wage and allows a firm to choose the profitmaximizing quantity of labor conditional on that wage, an RRC chooses the flow of residents that will in the long run produce the desired, utility-maximizing quantity of physician services (and therefore the quantity of physicians) and the price of those services (and therefore the rents). One problem with this argument is that it fails to explain why RRCs appear to differ in their ability or desire to generate rents, with non-primary care RRCs apparently more effective at erecting barriers to entry than primary care RRCs. A simple, but difficult to prove, explanation is that preferences for rents versus the quantity of physicians in the specialty may differ across RRCs. Or perhaps smaller specialties (e.g. non-primary care specialties) are better able to restrict entry across a relatively small number of residency programs than more populous specialties (e.g. primary care). That said, Bhattacharya (2005) does conclude that entry barriers differ between specialties in the US, as will be discussed below.

A second possible explanation for persistently high rates of return to various nonprimary care specialties is that there may be a shortage of patients in certain specialties that constrains the number of residents who can develop competencies. An RRC considers a number of factors when deciding whether to approve a program and how many residents may be trained at each program: “Those numbers (of residents who may be trained by a program) will be based primarily on the number, qualifications, and commitment of the faculty; the volume and variety of the patient population available for educational purposes; the quality of the educational offering; and the totality of the institutional resources” (Accreditation Council of Graduate Medical Education, 1996).

A third possible explanation is that teaching hospitals may not be willing, due to equity concerns, or able to adjust resident wages to allow the market to clear (Nicholson, 2003). The ACGME used to require teaching hospitals to pay all residents the same wage, regardless of specialty. Although they changed that policy, they still require that residents be paid an undefined minimum wage, which presumably is positive. This restriction might prevent the wage from adjusting to clear the excess supply of residents to certain non-primary care specialties.

A fourth possible explanation is that RRCs may restrict entry not to create rents for that specialty but to control medical spending by deterring physician-induced demand behavior (Cooper and Aiken, 2001). According to this line of argument, if the dermatology RRC allowed programs to expand, practicing dermatologists would respond by exploiting their information advantage to promote low-value medical services to patients.

3.6. How Elastic Is Medical Workforce Labor Supply?

We now turn to the question of whether money exerts a strong influence on six different decisions made by the medical workforce: occupational choice by prospective physicians, specialty choice by physicians, geographic location by physicians, labor force participation by nurses, and hours of labor supplied by nurses and physicians. If money is important for these decisions and wages/prices are determined by market forces, then labor shortages and surpluses will be short-lived: labor will relocate, enter/leave occupations, shift between specialties, and respond to higher wages by working more.

3.7. Physician Specialty Choice

Studies on physician specialty choice can be categorized according to whether they account for barriers to entering certain specialties; whether they allow specialty choice to be endogenous when forecasting a medical student’s expected earnings in each specialty; and the source of the variation in expected earnings between specialties that identifies the occupational choice expected earnings elasticity. Three studies (Sloan, 1970; Bazzoli, 1985; Gagne and Leger, 2005) examine multiple cohorts of graduating medical students and identify an elasticity based on changes over time in specialty-specific expected earnings or, in one case, changes over time in province-specific fees. The two key assumptions in these studies are that changes in specialty-specific nonmonetary attributes (e.g. prestige, work schedule, psychological costs associated with malpractice suits) are uncorrelated with changes in specialty-specific expected earnings, and all students graduating in the same year expect to earn the same amount over their lifetime in the same specialties (i.e. unobserved ability does not affect expected earnings, or specialty choice is assumed to be exogenous when forecasting expected earnings).

Sloan (1970) reports specialty-specific elasticity estimates (the percentage change in residents associated with a 1 percent increase in the net present value (NPV) of expected earnings in that specialty) that range from zero to 0.28. Bazzoli (1985) finds that medical students are more likely to choose primary care when the expected earnings are relatively large, but this effect is small. Specifically, a $10,000 increase in the expected earnings of primary versus non-primary care (about a 20 percent increase in the mean value) is associated with a 1.4 percentage point increase in the probability of choosing primary care. Gagne and Leger (2005) predict that a 10 percent increase in the fee-per-consultation for a specialist in Quebec, for example, would lead to a 0.4 percent reduction in the proportion of medical students entering general practice and a 2.5 percent increase in the proportion entering surgical specialties. Thus all three of these studies conclude that although expected earnings do influence a medical student’s specialty choice, the effect is rather small; medical students place great importance on non-monetary attributes when selecting a specialty.

The papers reviewed above assume that medical students can enter whatever specialty they like. That is, these studies do not account for the possibility that some medical students who want to enter a specialty will be unsuccessful. As discussed earlier, residency positions in high-income specialties are rationed and many medical students fail to enter their preferred specialty. Nicholson’s (2002) main contribution is to examine how differences in expected earnings affect the number of students who desire to enter a specialty rather than the number who actually enter a specialty. He finds that medical students’ specialty choices are quite responsive to expected earnings: the average income elasticity across seven specialties is 1.42. Medical students appear to be willing to incur the risk of not receiving any residency position (by ranking a competitive specialty as their first choice) in order to secure large rents in certain non-primary care specialties.

Bhattacharya (2005) is the only study that has jointly modeled specialty choice and expected earnings and allowed unobserved factors to affect both. He presents some evidence that is consistent with the first three studies discussed above—money is not the most important determinant of specialty choice. Bhattacharya predicts, for example, that physicians who decided to become family practitioners would actually earn more than surgeons had they decided (and been able) to enter surgery. His results also provide some justification for the typical assumption that all medical students in a cohort could expect to earn the same amount in the same specialty. In 17 of 20 comparisons, a physician’s predicted lifetime earnings if he entered a specialty other than the one he actually entered would be within 4 percent of the observed earnings in that specialty. Bhattacharya finds that only half of the earnings difference between primary and non-primary care physicians can be explained by differences in work effort, the required training period, and observed and unobserved ability. He concludes that differences between specialties in entry barriers are the most likely explanation for the residual earnings difference.

3.8. Physicians’ Geographic Location Decisions

A number of papers in the 1970s and early 1980s recognized that in the US the physician-to-population ratio is urban areas was much higher than in rural areas, and the difference was growing over time. The conventional explanation was that physicians preferred living in cities and could continue locating there in spite of the high existing supply by inducing demand for their services (Newhouse et al., 1982). A number of policies were enacted to encourage physicians to locate in rural areas, such as the National Health Services Corps program that would pay off a medical school debt for physicians willing to practice in underserved areas.

Newhouse et al. (1982) criticized these studies for failing to apply standard economic location theory to the physician market. They argued that areas above a critical size would be able to attract and support a physician within a particular specialty, whereas areas below this threshold would not. Furthermore, the specialty-specific growth over time would depend on whether or not an area initially had a physician in that specialty. As predicted, Newhouse et al. (1982) find that general/family practitioners locate disproportionately in counties with small populations, and the number of general practitioners decreased more rapidly in small versus large towns between 1970 and 1979 whereas the reverse pattern occurred for specialists. Frank (1985) concludes that the fee a psychiatrist can expect to receive in a market is not a particularly important determinant of location decisions. He also finds that the stock of psychiatrists adjusts slowly in response to a change in the fee: between 20 and 40 percent of the gap between the current and long-run equilibrium number of psychiatrists is filled each year. This is consistent with substantial transactions costs of moving a practice for experienced physicians and the fact that the flow of newly trained physicians is small relative to the stock.

Polsky et al. (2000) also conclude that changes in market conditions have a minor effect on the location decisions of practicing physicians, which is consistent with amenities being important relative to income and with large transactions costs. An increase in the HMO penetration rate in a market did not affect the probability that primary care physicians, hospital-based specialists, or late-career medical/surgical specialists would stop providing patient care in that market, and the effect on early-career medical/surgical specialists was minor. Expected income appears to be more important, however, among new physicians when transactions costs are the same across markets. Physicians in all specialties who completed residency training in 1994 were less likely to locate in markets with high HMO penetration, presumably because expected earnings were relatively low in these markets (Escarce et al., 1998).

3.9. Nurses’ Labor Supply

The early literature on nurse labor supply is summarized nicely by Shields (2004): “With respect to the likely impact of increasing the RN (registered nurse) wage rate, although there are considerable differences and inconsistencies across the studies reviewed, the main conclusion is that the wage elasticity is unresponsive (or inelastic) and that very large increases in wages would be needed to induce even moderate increases in nurse labor supply” (p. F493). Shields (2004) also concludes that nurse wages have a stronger effect on labor force participation than the number of hours worked conditional on working. It should be noted, however, that the wage elasticities from the early nursing literature, which average about 0.30, are in fact consistent with the general (medical and non-medical professions) literature on female workers (Borjas, 2000).

Many of the older studies do not address in a convincing way the key empirical challenges inherent in estimating labor supply: identifying plausibly exogenous variation in wages across workers and/or across time; measuring wages and hours accurately; accounting for self-selection into the workforce; and controlling for individual unobserved heterogeneity that should affect wages, such as differences in motivation, ability, and job characteristics (e.g. shift work).

Several recent studies use natural experiments, panel data, comprehensive administrative data sets, and/or more sophisticated econometric methods. Holma˚s (2002) follows 5,300 Norwegian nurses over a five-year period. He finds that a 1 percent increase in the nurse wage is associated with a 4.9 percent decrease in the hazard of exiting the labor force. This implies that a permanent 10 percent increase in the nurse wage would reduce the exit rate by approximately 1 percentage point per year for five years, from baseline exit rate of about 2 percent per year. When Holma˚s omits an indicator for whether a nurse is required to work off-hour shifts, the wage estimate is 50 percent smaller, which highlights the importance of modeling individual heterogeneity. Frijters et al. (2007) follow 28,000 nurses in the United Kingdom over a one-year period. They instrument for a nurse’s wage with her pre-nurse educational qualifications, and find that a 10 percent increase in a nurse’s wage would reduce the annual exit rate by 0.7 percent, a much smaller effect than in the Holma˚s (2002) study.

Askildsen et al. (2003) use the same Norwegian panel data as Holma˚s (2002). After instrumenting for the wage using the financial status of the local municipality, the lagged wage of auxiliary nurses working in the same municipality as a particular nurse, and a nurse’s work experience, they estimate an uncompensated wage elasticity of 0.21. As above, this low elasticity indicates that it would be expensive to eliminate nurse vacancies by increasing the wage. Specifically, they estimate that the wage would have to rise by 43 percent to eliminate 4,000 nurse vacancies in Norway, and thereby increase overall health expenditures by about 4 percent. Staiger et al. (1999) exploit a 1991 policy that changed nurse wages in the Veterans Affairs (VA) hospitals exogenously by a different amount across different US markets. They estimate a wage elasticity ranging from zero to 0.2, consistent with the Askildsen et al. (2003) study.

In 2004, 66 percent of registered nurses in the United States were married (Buerhaus et al., 2007). Most nurse labor supply studies conclude that an increase in the wage of a nurse’s spouse and an increase in non-labor income reduces her hours worked (Antonazzo et al., 2003; Buerhaus et al., 2007; Holma˚s, 2002). Furthermore, these effects are stronger for the labor force participation decision than on hours worked among nurses who are already working. As we discuss in the next section, these effects are likely to aggravate the tendency for the medical labor market to oscillate between situations of shortages and surpluses. Wages and non-labor income tend to increase when the economy is strong. The demand for medical labor is pro-cyclical due to a positive income elasticity of the demand for health, whereas the supply of nurse labor may be counter-cyclical due to the strong effects of spouse-wage and non-labor income.

3.10. Physicians’ Labor Supply

Staiger et al. (2010) report that mean hours worked per week by physicians practicing in the US decreased by 7 percent between 1996 and 2008. Between 1981 and 2001, Canadian physicians worked an average of five fewer hours per week (Crossley et al., 2006). These trends are causing concern among policy makers who believe developed countries have, or will soon have, a shortage of physicians (Cooper et al., 2002). The most likely explanation for this decreasing work effort is that in many countries physician fees have been falling and/or preferences for leisure have changed. Between 1995 and 2006 in the United States, for example, physician fees decreased by 25 percent in real terms (Staiger et al., 2010). In this section we review the empirical literature regarding the magnitude of the income and substitution effects of a change in the physician wage, and whether the physician labor supply curve is in fact backward bending. As with nurses, most empirical studies conclude that physicians’ labor supply is not more responsive to the wage than other professions. Furthermore, recent studies using micro data also conclude that changes in income do not have a strong effect on physician work effort—the income effect is small.

Most early studies used aggregate time series data on physician services and fees to infer the shape of the physician labor supply curve. Almost all of these studies concluded that physicians’ labor supply functions were backward bending—the negative uncompensated wage elasticity implied that the income effect of a wage increase was stronger than the substitution effect (Feldstein, 1970; Vahovich, 1977; Brown and Lapan, 1979; Hu and Yang, 1988; Brown, 1989). The one exception is Sloan (1975), who found that physicians’ hours worked were not responsive to wages or non-labor income. A backward-bending labor supply curve could hamper a policy that tries to control medical expenditures by constraining physician fees. If physicians’ marginal utility of income is large, they may respond to fee cuts by inducing demand for their own services, such that expenditures may not fall (McGuire and Pauly, 1991). However, studies using aggregate data were not able to separately measure the income and substitution effects, and early studies that looked strictly at income effects generally found the income elasticity to be zero or small (McGuire and Pauly, 1991).

The results of more recent studies, which use micro data to estimate traditional labor supply equations, generally find small positive uncompensated wage elasticities and no or small income effects. Using US data from the mid-1980s, Thornton (1998) first estimates a production function for physician services and then derives the physicians’ marginal shadow wage. He then estimates an uncompensated wage elasticity of 0.06 and an income elasticity of 20.09, both consistent with Sloan’s (1975) conclusion that physicians are not particularly responsive to wage and non-labor income. Rizzo and Blumenthal (1994) use experience to instrument for a physician’s wage. Their key assumption is that experience affects labor supply only through its effect on a physician’s wage, which seems reasonable given that they examine physicians under the age of 40 rather than those close to retirement. They find that the substitution effect exceeds the income effect. Specifically, they estimate an uncompensated wage elasticity of 0.27, an income elasticity of 20.17, and a compensated wage elasticity estimate of 0.44. Rizzo and Blumenthal also find that the labor supply decisions of female physicians are more responsive than those of male physicians.

Thornton and Eakin (1997) use market-level demand variables such as per capita income and degree of urbanization to instrument for a physician’s wage. They estimate a small negative uncompensated wage elasticity (20.02, and not significantly different from zero) and a small negative income elasticity (20.03) for solo practicing physicians. Although they conclude that the labor supply curve is backward bending, it is nearly vertical and consistent with the results in Sloan (1975) and Thornton (1998).

Showalter and Thurston (1997) examine how US physicians’ labor supply behavior in the mid-1980s was affected by variation between states in the maximum marginal tax rate. They estimate an uncompensated (net of the tax) wage elasticity of 0.30 for selfemployed physicians but find that employee physicians are not responsive to differences in the wage. Baltagi et al. (2005) estimate a dynamic labor supply model using panel data on Norwegian physicians. The wage effect is identified in part by a wage settlement in 1996 that increased physician wages differently across markets. They report shortand long-run uncompensated wage elasticity estimates of 0.30 and 0.55, respectively, and no income effect. Finally, Sæther (2005) uses data on Norwegian physicians to estimate a static labor supply model that allows a wage change in one sector of a physician’s practice (e.g. hospital, private office) to affect labor supply in the other sector. He reports an uncompensated wage elasticity estimate of 0.18 across both sectors, with a greater response in the office than the hospital setting.

To summarize, the six studies discussed above that use micro data report uncompensated wage elasticity estimates ranging from zero to 0.30, with four of the estimates between 0.18 and 0.30. The four studies that examine income effects report elasticity estimates ranging from zero to 20.17. Given the different methods, data sources, and countries studied, there appears to be general agreement that as with most occupations, physicians are not particularly responsive to wage changes. And income elasticities are small.

3.11. Shortages and Surpluses of Medical Labor: Cyclicality and Persistence

In this section we examine why medical labor markets in many countries cycle between surpluses and shortages, and why shortages/surpluses often persist for many years. In Canada, for example, “the speed with which the shortage has arisen is even more surprising (than the unprecedented shortage of family practitioners in relatively large cities): in less than a decade, the supply of physicians in Canada reversed from a perceived surplus of physicians to a perceived shortage” (Crossley et al., 2006, p. 1). In 1991 the BarerStoddart report, which was commissioned by the Canadian Ministry of Health, recommended reducing medical school enrollment by about 10 percent and reducing Canada’s dependence on internationally trained physicians in order to maintain the physician-to-population ratio into the future (Barer and Stoddart, 1991). The Canadian government accepted these recommendations the following year (Tyrell and Dauphinee, 1999), and the policy worked; after peaking in 1993 at 2.2 physicians per 1,000 people, the ratio has stabilized at 2.1 since then. The Canadian government has now reversed course. Since 1998, the number of medical school positions has increased by 39 percent (Esmail, 2005).

Japan also reversed its physician workforce policy over a short time period. Japan reduced the number of medical school positions by 8 percent between 1986 and 2006 based on a perceived impending surplus (Toyabe, 2009). The education ministry recently reversed its policy and decided to consider accrediting new medical schools to expand the flow of physicians. Medical school positions in 2010 are now at an all-time high. In the US a shortage of registered nurses began in 1998 and lasted for 10 years, well beyond what would be predicted by standard economic models. At the peak in 2001, 13 percent of hospital RN positions were vacant (Buerhaus et al., 2009). Norway, the United Kingdom, Canada, Australia, and South Africa all experienced nurse shortages during the 2000s (Shields, 2004).

How would one know there is in fact a shortage or surplus of a particular type of medical labor? In most markets a shortage would generate rising wages, rising earnings, an increase in the flow of new entrants, longer patient waiting times for an appointment (in the case of physicians/dentists), shorter appointments with a provider, and an increased use of substitute labor to the extent the production function allows it. With administered prices, several of these mechanisms might not be triggered unless public and private insurers respond as the market would. Furthermore, entry barriers may delay or prevent the supply response. These challenges make it difficult to determine whether a shortage or surplus exists. Since 2000 there have been many studies concluding that the US has or soon will have a physician shortage—18 reports from states, medical societies, and hospital associations, and 19 reports from medical organizations since 2000 (Iglehart, 2008). However, these studies usually reach that conclusion by applying demographic projections to current treatment levels rather than documenting market evidence of a shortage.

Long et al. (2008) argue that one should not base a nursing shortage on vacancy rates because hospitals increasingly use temporary or agency nurses as a strategic response to volatile demand conditions; costs can be minimized by hiring fewer fulltime nurses and using part-time nurses to respond to demand shocks. Governments may also relax immigration policy to respond to physician shortages. Most developed countries are relying increasingly on physicians trained abroad. For example, internationally trained physicians account for a larger proportion of the physician workforce in 2007 than in 2000 in the following countries: Ireland (33 percent of physicians in 2007 were trained abroad), New Zealand (31 percent), US (25 percent), and Switzerland (21 percent) (OECD Health Data, 2009).

We provide several explanations for the persistence and cyclical nature of medical labor surpluses and shortages. First, the demand for health is reasonably income elastic. Using changes in oil prices interacted with oil reserves in a locality to measure the causal effect of income on hospital spending, Acemoglu et al. 2009 estimate an income elasticity of 0.72. This indicates that as a country’s income rises and falls, the demand for health by its citizens will likewise rise and fall, although not to such an extent that health care is a luxury good. Because the demand for medical labor is derived from the demand for health, the demand for labor will be pro-cyclical. Labor supply, however, will respond slowly to the income-induced demand changes for several reasons: training periods are often long, the flow of newly trained labor is usually small relative to the stock of labor, and administered prices may prevent the market signals from reaching eligible labor.

Another explanation for the cyclicality of shortages and surpluses is that governments are not particularly good at forecasting future supply and demand and are heavy-handed when they act. Nicholson (2009) describes the US government’s long history of trying to forecast the future supply of and demand for physician services, beginning in the early 1900s. Government physician manpower policy responds to, and perhaps accentuates, the oscillation from a perceived (and perhaps real) surplus to perceived shortage of physicians. The Canadian and Japanese experiences described at the beginning of this section are similar to the US experience. Perhaps there is a political explanation why attitudes about the adequacy of professional occupations are cyclical.

A final explanation for the persistence of physician shortages (real or perceived) is that policy makers and academics have different normative perspectives regarding the role of physicians in the health care system. Nicholson (2009) presents a simplified argument where people can be divided into two camps depending on how they answer two questions: (1) are policy makers willing and able to reform the health care system and improve its efficiency?; and (2) if more physicians begin practicing, will the value of their incremental services exceed their cost? People such as Cooper et al. (2002) who are skeptical that policy makers (or the market) can reform the health care system to improve physician productivity, and who believe that physicians cannot or do not induce demand for their own services, are likely to support expanding the physician workforce in anticipation of growth in demand for physician services due to the growth and aging of the population and rising real income. People such as Goodman and Fisher (2008), who are optimistic that policy makers (or the market) can reform payment systems to improve physician productivity, who believe that adding physicians will take pressure off policy makers and make reform less likely, and who believe incremental physician services would be low value, are likely to favor constraining the growth of the physician workforce. Policies enacted by one camp will trigger cries of “surplus” or “shortage” by the other camp.

As Blumenthal (2004) points out, “The physician-supply debate is therefore now enmeshed in and inseparable from a larger discussion about the value of the services physicians provide and the future of the health care system—how big it should be, how to organize it, and whether its trajectory can be controlled. Proponents of the deficit theory (i.e., U.S. has a physician shortage) argue that ignoring or resisting inevitable increases in the demand for physicians’ services will only lead to ‘public discontent’ and invite other health care professionals to take over the roles traditionally played by physicians. Proponents of the surplus theory (i.e., no physician shortage) seem to believe that constraining the supply of physicians is one way to begin restructuring our health care system in order to improve its rationality and efficiency.”