Printer Friendly

An analysis of the location decision of young primary care physicians.


Rapid changes in the health care sector have had a profound impact on the market for physician services. Nearly gone are the days when an appointment with the doctor meant a visit to a male physician who owned a solo, fee-for-service practice. Today, more than 1 in 6 physicians are female and only about one-third of all physicians are self-employed and operate a solo practice. Multiphysician practices are the norm, and any physician who wishes to prosper is now forced to deal with rapidly expanding integrated health care delivery systems.

These changes have had a profound impact on the labor market for physician services and have sparked a renewed interest in the physician location decision. The major objective of this study is to determine the factors influencing the first-time location decisions of young primary care physicians under the age of 35.

Review of the Literature

The literature on the physician location decision is rich and spans over three decades. Early studies generally regressed some measure of the stock of physicians on a host of economic and noneconomic variables thought to influence the physician location decision. The most notable study by Newhouse et al. [1982] developed an economic model of physician behavior based on standard location theory. The purpose was to test if there was a market failure creating a geographic maldistribution of physicians. Any indication of market failure was presumed to result from the ability of physicians to induce the demand for their services thereby creating an incentive for them to locate in larger communities. The strength of their results leads them to reject the notion of market failure and to conclude that standard location theory can be useful in predicting the geographical distribution of physicians. Building on this approach and utilizing an alternative data set, Dionne et al. [1987] found that average income and the presence of a hospital, among other variables, increase the probability that at least one physician practices in a given community.

Two more recent studies [Leonardson et al., 1985; Holmes and Miller, 1986] utilized surveys of recent medical school graduates to provide descriptive looks at the physician location decision. Together they found that the size of the community, where the physician grew up, opportunities for further education and professional development, and personal and career goals of the spouse weigh heavily on the location decision.

Langwell et al. [1987] focused exclusively on the location decision of young physicians who elected to practice in rural communities. They found the probability that a young physician locating in a rural county is positively correlated with the population, the number of physicians in the county, the percentage of white-collar employment, and whether or not a college was located in the county. They also uncovered the likelihood that a young physician practicing medicine in a rural county is negatively related to the size of the farm population.

A recent study by Morrisey et al. [1991] found that the recent reduction in hospitals and hospital beds in rural counties had only a marginal impact on the number of physicians practicing in rural counties. They also found that population growth had a positive and significant impact on the number of physicians practicing in rural communities.

The present study contributes to the literature in two important ways. First, it relies on a more contemporary data set which includes both rural and nonrural counties. Second, because the study deals exclusively with physicians under the age of 35, it allows for an analysis of the first-time location decision of physicians who have completed their residency.(2)

Theoretical Model

The model employed is the typical location model used by Morrisey et al. [1991] and others. Since the purpose here is to enhance the empirical understanding of the literature rather than extend the theoretical frontiers, a rigorous development of the theory is forgone. The model follows traditional methodology and is developed in two steps. First, the demand and supply of primary care physicians are identified. Then, by imposing market clearance, the reduced form equilibrium equation is generated and estimated.

The decision to focus on young primary care physicians (less than 35 years old) is essentially two-fold.(3) First, at this point in their careers, location decisions are perhaps as unbiased as they can ever be. While very adequately educated, the human capital obtained in medical school and a residency program can be seen as a rather general form of training. That is, young physicians are not so specialized as to be excluded from large parts of the labor market across the country. Furthermore, being young, it is more likely they will consider the entire spectrum of personal amenities offered by a particular area. It would seem reasonable that these physicians are more susceptible to the nonpecuniary and nonprofessional attributes of an area than their older counterparts. This is entirely consistent with Johnson [1978] who argues that younger labor market participants are not as risk-averse as older, more established workers.

The Demand for Primary Care

Individual consumers are assumed to be price-takers of health care services who express, via their demands, a utility-maximizing, decisionmaking process toward health care. Once the overall demand for health care is expressed, it is reasonable to assume that the demand for physicians is proportional. The amount demanded is dependent not only upon the price, but also upon a myriad of demographic and infrastructure influences which shape and shift a consumer's demand for primary health care and, therefore, physicians.

Consider for a moment the demographic influences. The very young and the very old have special needs which require more health care than do other age groups. Thus, if an area's population has a significantly young or old segment, then that area is expected to demand more health care. Also, populations which have an above-average education or income might be more likely to demand preventative services. As a result, the demand for primary care doctors would be increased.

Equally important is the existing health care industry in the area which typically reflects the community's historic tastes and preferences toward health care. Localities with a number of hospitals and a large number of hospital beds need a large staff of doctors. It is expected that the larger the health care industry (as expressed by the employment in health care services, for example), the larger the demand for physicians. Yet the impact of individual components of the health care labor market are not so straightforward. On one hand, a greater number of older physicians might seem to indicate a scale of operation which supports a large number of new hires. Yet it could be argued that this typifies a stagnant environment where new hires are closed out. Similarly, the way in which young physicians are compared vis a vis residents is critical. If the existing health care industry views residents and doctors as complements in production, then a positive relationship could be seen. However, if their relationship is viewed more as substitutes in production, then an inverse relation is anticipated.

In total then, an area's demand for primary health care can be expressed as:

[Q.sub.d] = [Q.sub.d] ([P.sub.d], [D.sub.d], [HC.sub.d]), (1)

where [P.sub.d] is the demand price of primary health care, [D.sub.d] is a vector of demographic variables which affect demand, and [HC.sub.d] is a vector of local health care industry characteristics which affect demand. Since the number of doctors demanded is taken as being proportional to the demand for health care services, it is clearly the case that:

[PC.sub.d] = [PC.sub.d] ([P.sub.d], [D.sub.d], [HC.sub.d]), (2)

where [PC.sub.d] is the demand for primary care physicians under 35 and the other variables are defined as in (1).

Equivalently, this demand relationship can be inverted so the demand price is a function of the number of primary care physicians, the vector of demographic variables, and the vector of health care industry characteristics. The demand price function is then seen as:

[P.sub.d] = [P.sub.d] ([PC.sub.d], [D.sub.d], [HC.sub.d]). (3)

The Supply of Primary Care

Young physicians can be viewed as being price-taking participants in the labor market. As a result, the professional environment that a community offers their physicians is very important in the location decision.(4) Is there a strong population base? Is the health care industry large? Is the hospital one of many? Does the hospital support residents? Is there a teaching faculty? Is research offered? An affirmative answer is assumed to enhance the professional atmosphere of a locality and, therefore, its desirability.

Amenities offered by the community certainly have an impact on anyone's location decision. Observation of an area's unemployment rate, job growth, income, poverty rates, housing values, and tax rates provides a searcher with important information concerning the local economic environment. Similarly, characteristics such as education levels, crime statistics, and population growth describe the general level of pleasantries associated with a community.

As a result, a general supply function of primary health care services can be expressed as:

[Q.sub.s] = [Q.sub.s] ([P.sub.s], [HC.sub.s], [E.sub.s], [D.sub.s]), (4)

where [P.sub.s] is the supply price of primary health care, [HC.sub.s] is a vector of health care industry characteristics, [E.sub.s] is a vector of economic attributes which affect the supply, and [D.sub.s] is a vector of demographic attributes affecting the supply. Once again, the number of primary care physicians is assumed to be proportional to the level of service provided so that:

[PC.sub.S] = [PC.sub.S] ([P.sub.S], [HC.sub.S], [E.sub.S], [D.sub.S]), (5)

where PC is the supply of young primary care physicians. As with demand, the supply function is inverted so the supply price is dependent upon the number of primary care physicians, the health care industry characteristics, and demographic and economic attributes. The resulting supply price function is then expressed as:

[P.sub.s] = [P.sub.s] ([PC.sub.s], [HC.sub.s], [E.sub.s], [D.sub.s]). (6)


The result of a short-run equilibrium process within the labor market for young physicians is observed as a function of all supply and demand characteristics. Market clearance, of course, implies [P.sub.d] = [P.sub.s]. If (3) is set equal to (6), and like terms are collected, then the reduced-form equilibrium equation results:

PC = PC(HC, E,D), (7)

where PC is the equilibrium quantity of young primary care physicians, HC is a vector of health care industry characteristics, E is a vector of economic characteristics, and D is a vector of demographic characteristics. This model is estimated below.

Data Set

The data set encompasses the continental U.S. only for two principal reasons. First, the data for Alaska and Hawaii are often incomplete or nonexistent, so these states have been omitted. Logically, an individual is attracted to a particular community for the work environment and amenities offered by that community. While this level of disaggregation might be optimal, it is not at all practical. Prior empirical investigations faced the same dilemma and have typically relied on a county as the unit of measurement [Newhouse et al., 1982; Morrisey et al., 1991]. Therefore, county-level data have been used for convenience and comparability to previous studies.

One of the distinguishing characteristics of the present analysis is the emphasis on the entire continental U.S. To be complete, the information contained in the 44 independent cities (that is, cities that are not in any county and serve as county equivalents) of Virginia, Maryland, and Nevada must be incorporated. Consistent data for all of these independent cities are available, so their autonomy in this study has been preserved. Therefore, the data set contains observations for the continental U.S. on 3,066 counties, 44 independent cities, and the District of Columbia for a total of 3,111 observations.

All of the physician data have been obtained from the Physician Data by County: 1993, published by the American Medical Association [1993]. For each county and independent city, this data source compiles a complete count of the number of physicians, the number of physicians under 35, and the number of residents for the specialties of general practice, family practice, internal medicine, and pediatrics. The definition of primary care physicians is taken to be the total of general practice, family practice, internal medicine, and pediatrics. Also, the original data identifies total physicians as being composed of both residents and nonresidents. Given the number of residents, a simple subtraction of residents from total physicians yields nonresident physicians. However, the American Medical Association does not identify residents under 35 separately from residents over 35 and for good reason - the overwhelming majority of residents are under 35. Therefore, it is assumed that all residents are under 35 years of age. The remaining quantitative variables (for example, the number of hospital beds and population growth) come from the 1994 County City Data Book [U.S. Census Bureau, 1994].

Empirical Model(5)

The Appendix contains the precise definition of each variable in the model. Equation 7 is estimated using ordinary least squares (OLS) where the dependent variable is the number of primary care physicians under 35 in each county less residents (PC35NONRES). Using the absolute magnitude of physicians and not a relative measure, as is often done in literature, underscores a primary motivation of this study: the understanding and explanation of the short-run market equilibrium process which determines the location of young physicians. If the model relies on relative measures, then the independent influence of supply and demand functions for physicians is lost to a mixture of these market processes with the overall economic and demographic influences in the county. What is explained in those models is not the pure physician market equilibrium processes, but a mixture of a number of processes.

Health Care Industry Characteristics

One of the primary care physician-related independent variables is the number of primary care physicians over 35 (PCGT35). If the number of older physicians reflects the scale of primary care in the county, and thus acts as a magnet for younger doctors, then this should exert a positive effect on the number of younger doctors locating in the county. On the other hand, if the existing stock of physicians acts in a fashion to close out entry by younger doctors, then this would be negatively related.

The last primary care physician variable is PCRES, the number of primary care residents. Once again, it is impossible to determine a priori the relationship between residents and young physicians. Traditional arguments arise mostly from a supply perspective. This view holds that a young physician finds an area with plenty of residents desirable since the residents can perform many of the more mundane, support activities, thereby freeing the physician from these objectionable duties. Yet these models, being reduced-form equilibrium models, are a mixture of demand and supply effects and the traditional view neglects the demand side completely. From the demand side, it is important to gauge whether physicians and residents are substitutes or complements in provisioning primary health care. If demanders view physicians and residents as complements or even mild substitutes, then the equilibrium relationship between them will be positive. However, if they are viewed as strong substitutes, then a negative relationship is entirely possible.

It is conceivable that a young physician would find an area professionally attractive the more closely connected the area is to academic pursuits. To explore this possibility, three additional variables have been included. TEACHER is the number of medical teaching faculty in a county, RESEARCH is the number of physicians active in research in a county, and FELLOWS is the number of fellows affiliated with hospitals in a county. These variables are expected to exert a positive influence on the number of young primary care physicians.

The number of hospitals (NUMHOSP), hospital beds (BEDS), people in nursing homes (NUMINNH), and the proportion of the labor force in the health care industry (LFINHS) all address various aspects of the importance of the medical community within the county and its existing capital stock. At first glance, NUMHOSP and BEDS appear to be purely demand-side attributes (as either of these variables increases, the greater the need for all physicians, including young ones). Yet there might be supply effects as well especially if a physician is particularly averse to or attracted to largeness. Similarly, it seems that NUMINNH attests to the scale and, therefore, the need for physicians. This positive demand effect might be enhanced by physicians' attraction to geriatrics or dampened by physicians' aversion to geriatrics. At the market level, these impacts would most likely be minor at best. On the whole however, these variables are expected to exert positive influences on the location decision.

The overall size of the health care industry is intriguing. The immediate impression is that it again addresses the scale of operation which clearly implies a positive relationship. As with residents, the issues of complementarity and substitutability arise. If the size of the health care industry is due in part to a large number of alternative care providers and these providers are strong substitutes for physician services, then it is quite possible for this variable to be negatively associated with the location decision of young physicians. In effect then, there are no expectations as to the sign of this variable.

Demographic Attributes

Three different population strata are included in this study: percent of the population less than 5 years old (POPLT5), between the ages of 5 and 17 (POP5TO17), and older than 65 years (POPGT65). The intent here is to identify the special needs of the young and old and the impacts they have on the physician location decision. While there might be some contradictory supply-side effects, it seems reasonable to expect that all these variables will show clear, positive demand-side influences. BIRTHS, while not a population strata per se, is certainly closely related. This sign is also expected to be positive.

The remaining population characteristics are captured in total population (POP92 and POP92SQ), the population density (POPDEN), and the growth rate of population from 1980 to 1992 (POPGROW). As witnessed above with other variables, the population variables capture demand-side as well as supply-side influences. The demand-side influences are clear. The larger and more dense a population, the greater the need for physicians. Furthermore, a higher population growth rate implies a greater future need for physicians and, thus, a greater current demand for physicians to address that need.

The supply-side influences are not quite as direct. If the Johnson [1978] job-shopping model and Marder's [1990] earnings risk analysis are in effect, then young physicians are drawn into the high earnings risk (that is, metropolitan areas) early in their career. In this instance, the positive supply effect reinforces the demand effect and population exerts an overall positive influence. On the other hand, a young physician's aversion to high population areas or perception of job market saturation in these populated areas might create disincentives to locate in highly populated areas. Furthermore, rapid growth in population might serve to warn entrants of future urban problems thereby creating a disincentive to locate in high-growth areas. These adverse supply-side effects dampen the direct demand-side effects resulting in, perhaps, insignificance or negative relations. The squared population term allows for asymmetries in population influences. If, for instance, the sign of population is positive and the sign on squared population is negative, then population increases exert a positive but diminishing influence and larger communities eventually become a negative influence. Overall, no expectations regarding these population variables can be posited.

The median value of a home (MEDVALHOME) can be used to address two separate issues: the obvious expense of housing in an area and an input cost for those physicians not joining large group practices. In the first instance, an inverse relationship is expected. As homes are more costly, the desire to locate in that area diminishes. Likewise, if a young physician is not considering a large group practice, then he faces the need to obtain office space. As the cost of office space rises, with all else being the same, the desire to locate in that area is diminished. Certainly this input cost effect is not a consideration for those joining large group practices. As these two influences reinforce one another, the expected sign is negative.

Economic Attributes and Amenities

A county's general economic environment is mirrored in the median family income (MFY), the growth of median family income (MFYGROW), the percent of the population below the poverty line (POVPERCENT), the growth of the labor force (LFGROW), and the county's unemployment rate (UNEMP). MFY and MFYGROW simultaneously proxy the current and expected future cost of living and earnings potential in an area. As a cost of living indicator, the influence should be negative. As an indicator of future earnings potential, the influence would most likely be positive. Taken together, the net impact cannot be predetermined. It is anticipated that both POVPERCENT and UNEMP have negative influences on the location decision as lower values would indicate a more dynamic, prosperous economic climate. For the same reason, LFGROW is expected to be positively associated with the location decision.

General county amenities make up the last group of quantitative variables. COLLGRADS is the percent of the population with at least a bachelor's degree. With all else being constant, an area with a high proportion of college graduates would provide the social and cultural infrastructure consistent with that found desirable by a young physician. Thus, this variable is expected to positively influence the location decision. Crime, on the other hand, serves as a deterrent to the location decision. The number of serious crimes reported to the police (CRIME) should have a negative impact on the location decision. Finally, the more taxes an individual must pay, the less desirable a county is. Therefore, per capita taxes paid (PERCAPTAX) is predicted to be inversely related to the location decision.

One dummy variable (RURAL) is included in the model to address the rural/urban controversy. According to the 1990 census, if a county is not included in a metropolitan statistical area, then it is designated as rural, and RURAL takes on the value of 1. Otherwise, RURAL takes the value of 0. A priori, the expectations with respect to RURAL's sign are nonexistent. Much of the existing literature suggests this variable should be negative. Yet this literature either focuses primarily on supply-side events or obtains results which could be adversely biased by conflicting activities.

Empirical Results

The results of the initial OLS estimation are presented in the uncorrected column of Table 1. Overall, the fit is extremely strong with adjusted [R.sup.2] equal to .94 and an F-statistic of 1,728. As a group, the physician variables are impressive. The parameter estimate on PCGT35 is positive and significant at the 5 percent level. This implies that young physicians are not closed out of markets with a large existing stock of older physicians but are, instead, drawn in. On the other hand, PCRES is somewhat of a surprise. It is found to be significantly negative at the 1 percent level, suggesting that a strong degree of substitutability from the demand side of the market is overwhelming the positive supply effects. As anticipated, all of the academic variables are positive and significant at the 1 percent level.


For the most part, the variables describing the existing health care industry behaved as expected. Both NUMHOSP and NUMINNH are positive and significant at the 1 percent level, while BEDS is positive but insignificant. A priori, there were no expectations regarding the sign of LFINHS. As with residents, this uncertainty comes from how physicians and alternate care providers are viewed from the demand side. The estimated coefficient is negative and significant at the 1 percent level. Once again, the data suggest that physicians have strong substitutes challenging the traditional role of the doctor as the primary provider of health care.

On the whole, the population characteristic variables are disappointing. BIRTHS is positive, as expected, and significant at the 5 percent level. However, all of the population strata variables are of the wrong sign, with POP5TO17 significantly so. Furthermore, contradictory results come from the general population variables. Population density is negative and significant at the 5 percent level, suggesting an overall aversion to locating in urban areas.

Initially population and population squared do not seem to be in opposition to the density effect. The estimated coefficient for population is positive while that for population squared is negative. Both coefficients are significant at the 1 percent level. This pattern is consistent with a quadratic relationship that rises for low populations (implying an aversion to being overly rural) but then falls off for higher population levels. Closer inspection reveals the county population must exceed 4.6 million people before the negative impacts of population on the location decision are felt, which is applicable in only two counties. Clearly this contradicts the population density effects.

The estimated coefficient for median home value is negative and significant at the 1 percent level as expected. However, neither median family income nor the growth of median family income are significant. As stated earlier, it was argued that income serves as a proxy for the cost of living in a county and also as a proxy for income potential. This insignificance might therefore be due to conflicting considerations associated with the income variables.

All of the parameter estimates for the amenities and general economic variables are of the anticipated signs. However, only parameter estimates for the proportion of college graduates and per capita level of taxes are significant. Finally, a physician's decision to locate is not influenced by the county being rural or urban according to the RURAL dummy variable's insignificance.

A residual analysis indicates that the error structure is heteroskedastic.(6) based on the OLS residuals, White's test for heteroskedasticity with the variables and their squares (but no cross-product terms) was performed. The [[Chi].sup.2] test statistic of 1,052 is significant at the .5 percent level of significance, strongly indicating a heteroskedastic error structure. An application of the Breusch-Pagan test gives a [[Chi].sup.2] test statistic of 37,899 which is also significant at the .5 percent level of significance. Both tests lead to the rejection of the null hypothesis that the error structure is homoskedastic.

The Spearman rank correlation coefficients were calculated in an attempt to isolate the cause of the error variance problem. The results found that most variables had a significant rank correlation. From this result it seems that the heteroskedastic nature of the error structure cannot be isolated to a single variable. Indeed, the Spearman rank correlation between the fitted value of PC35NONRES and the absolute residuals yields an estimate of .43 and a t-test statistic of 26.2, significantly larger than the critical value at the 1 percent level. Further tests verified these results and suggested a quadratic relationship between the error variance and the variables. based on these results, a weighted least squares estimation with the weights equal to the inverse of the fitted values of PC35NONRES was employed. The results are found in the corrected column of Table 1.

The general fit is every bit as significant as the original model, even if there has been a reduction in the explanatory power. Also, the F-statistic has been reduced considerably but is still well above the critical value at the 1 percent level. Two striking changes are evident in the physician variables. First, PCRES is now significantly positive rather than significantly negative, which seems to vindicate the usual view of residents. This result does not refute the possibility of residents and physicians being substitutes for they can still be mild substitutes in demand. It does suggest that the supply-side impacts from the young physician's point of view dominates. Oddly enough, FELLOWS changed signs as well but is not significantly negative.

Similarly, the interpretation of NUMOFHOSP has changed. It now inexplicably has a negative sign and is significant. The number of beds is now significantly positive as naturally expected.(7) The magnitude of changes brought on by a marginal change in the proportion of the labor force in health care has been softened considerably. However, it is still significantly negative. This dampening of substitutability is quite consistent with the finding in residents above. No real changes are observed in the number of people in nursing homes.

The population variables have changed considerably and have been brought more in line with respect to prior expectations. BIRTHS remain significantly positive but POPLT5 and POPGT65 have both changed to being significantly positive. POP5TO17 remains negative but insignificantly so.

Population density, population, and population squared are unchanged except for their magnitudes. However, a somewhat more consistent relationship now emerges. The quadratic function implied by POP92 and POP92SQ suggests that population exerts a positive influence on a physician's location decision so long as the population is less than approximately 340,000. From this, 151 counties or independent cities are large enough to be adversely impacted in the location decision process. Now the interaction between population and population density is more palatable. On the whole, young physicians would rather locate in a moderately populated area. As in the original OLS estimation, the evidence suggests that young physicians are somewhat more risk-averse than prior studies have found. Notice that young physicians are drawn to growing areas as the population growth estimate is now positive and significant at the 1 percent level.

The income variables and the expense of housing are now all negative and significant at the 1 percent level. Clearly, the cost of living aspect of location is more important than income potential. The remaining economic and amenity variables are essentially unchanged with the exception of the unemployment rate. Inexplicably, UNEMP is now positive and significant at the 1 percent level.


Overall, this study lends support to much of the existing location literature. First of all, young primary care physicians seem to be drawn to counties where there is already a large number of practicing primary care physicians, where there are a large number of beds, and where there is a larger young and old population. Simply put, young physicians are more likely to locate where the demand for their services is strong, with all else being the same.

As with any equilibrium process, supply considerations cannot be neglected. The findings here suggest that a young physician's willingness to locate in a particular county is enhanced by professional as well as economic and amenity considerations. Where there is a strong academic presence, there is a strong supply response. Similarly, areas free of crime, poverty, and excessive taxation are preferred. Finally, a moderate cost of living is very important even if it means a reduced, nominal income.

Yet, this study has also extended the traditional orthodoxy as well. The findings strongly suggest the presence of substitutes for physician services, and they are becoming an important consideration. As the health care industry changes in the future, this substitution effect is expected to become more pronounced. The empirical results also call into question the commonly held belief that physicians prefer to live in large metropolitan areas. The asymmetric impact that population has on the physician location decision suggests that young physicians have an affinity for small to medium-sized cities and prefer not to locate in either sparsely populated communities or major metropolitan areas. From the perspective of a young physician, medium-sized cities offer the economic opportunities and amenities compatible with launching a career in primary care medicine and possibly raising a family.

Variable Description

Variable Name                       Definition

BEDS                   Number of hospitals beds per 100,000
                       population in the county for 1991.

BIRTHS                 Number of births per 100,000 population in
                       the county for 1988.

COLLGRADS              Percent of the population with at least a
                       bachelor's degree in the county for 1990.

CRIME                  Number of violent crimes for 1991.

FELLOW                 Number of fellows in the county for 1992.

LFGROW                 Growth in labor from 1980 to 1990.

LFINHS                 Percent of the work force in the health care
                       industry for 1990.

MEDVALHOME             Median value of a house in the county for

MFY                    Median value income in the county for 1989.

MFYGROW                Growth in median income from 1979 to 1989.

NUMHOSP                Number of hospitals in the county for 1991.

NUMINNH                Number of persons in nursing homes in the
                       county for 1990.

PC35NONRES             Number of primary care physicians under the
                       age of 35 who are nonresidents for 1992.

PCGT35                 Number of primary care physicians whose age
                       exceeds 35 in the country for 1992.

PCRES                  Number of primary care residents in the
                       county for 1992.

PERCAPTAX              Per capita tax in the county for 1986-87.

POPLT5                 Percent of the population whose age was less
                       than 5 in the county for 1990.

POP5TO17               Percent of the population whose age was
                       between 5 and 17 in the county for 1990.

POPGT65                Percent of the population whose age exceeds
                       65 in the county for 1990.

POP92                  Total population in the county for 1992.

POP92SQ                Total population in the county squared for

POPDEN                 Total population in the county for 1992
                       divided total square miles.

POPGROW                Population growth in the county for 1980 to

POVPERCENT             Percent of the county's population below the
                       poverty line in 1989.

RESEARCH               Number of physicians engaged in research in
                       the county for 1992.

RURAL                  Rural/urban dummy that equaled 1 if the
                       county was rural and 0 if it was urban.

TEACHER                Number of physicians engaged in teaching in
                       the county for 1992.

UNEMP                  Countywide unemployment for 1991.


1. For a review of the earlier literature, consult Feldman [1979] or Rosko and Broyles [1988].

2. Although Langwell et al. [1987] focus on young physicians, they confine their analysis exclusively to rural communities.

3. While not universally accepted, this study looks upon a primary care physician as one whose specialty is family practice, general practice, pediatrics, or internal medicine.

4. It is assumed that physicians reside in the county where they practice.

5. EViews [1998] has been used for all estimations.

6. Any regression model is likely to encounter some degree of multicolinearity and the present model is no exception. The correlation matrix and auxiliary regressions show a moderate degree of intercorrelation between a number of the variables. Interestingly enough, there is very little evidence of the classic consequences associated with multicollinearity. For instance, only 3 of the 28 estimated coefficients are clearly of the wrong sign, and 16 coefficients are significant. Also, alternative models yielded remarkably consistent results with respect to incorrect signs and significance. Most likely, the size of the data set accounts for the stability.

7. It must be remembered that hospitals provide an array of medical services other than inpatient services which rely heavily on hospital beds. For example, most hospitals provide a variety of outpatient clinic services along with emergency care. It may be the case that inpatient services (hospital beds) and physician services are complements in the production of medical services while many other hospital services, such as outpatient clinic services, are substitutes for private physician services. This may explain why there is a negative relationship between the number of hospitals and primary care physicians and a positive relationship between the number of hospital beds and primary care physicians.


American Medical Association. Physician Data by County: 1993, vols. 1 and 2, Chicago, IL: AMA, 1993.

Dionne, Georges; Langlois, Alain; Lemire, Nicole. "More on the Geographic Distribution of Physicians," Journal of Health Economics, 6, 4, 1987, pp. 365-75.

EViews. Quantitative Micro Software, 1998.

Feldman, Roger. "A Model of Physician Location and Pricing Behavior," in R. M. Sheffler, ed., Research in Health Economics, Greenwich, CT: JAI Press, 1979.

Holmes, June E.; Miller, Deborah A. "Factors Affecting Decisions on Practice Locations," Journal of Medical Education, 61, 9, September 1986, pp. 721-26.

Johnson, W. R. "A Theory of Job Shopping," Quarterly Journal of Economics, 92, 2, May 1978, pp. 261-77.

Langwell, Kathryn M.; Drabek, John; Nelson, Shelly L.; Lenk, Edward. "Effects of Community Characteristics on Young Physicians' Decisions Regarding Rural Practice," Public Health Reports, 102, 3, May-June 1987, pp. 317-28.

Leonardson, Gary; Lapierre, Rene; Hollingsworth, Doneen. "Factors Predictive of Physician Location," Journal of Medical Education, 60, 1, January 1985, pp. 37-43.

Marder, William D. "Physician Mobility: The Role of Income Differentials," Ph.D. dissertation, University of Chicago, 1990.

Morrisey, Michael A.; Kletke, Phillip R.; Marder, William D. "The Role of Local Hospitals in Physician Rural Location Decisions," Inquiry, 28, Spring 1991, pp. 7-18.

Newhouse, Joseph P.; Williams, Albert P.; Bennett, Bruce W.; Schwartz, William B. "Does the Geographical Distribution of Physicians Reflect Market Failure," Bell Journal of Economics, 13, 2, 1982, pp. 493-505.

Rosko, Michael; Broyles, Robert W. The Economics of Health Care: A Reference Handbook, Westport, CT: Greenwood Press, 1988.

U.S. Census Bureau. 1994 County City Data Book, U.S. Department of Commerce, Economics and Statistics Administration, Washington, DC: USCB, 1994.
COPYRIGHT 1999 Atlantic Economic Society
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1999 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Carpenter, Bruce E.; Neun, Stephen P.
Publication:Atlantic Economic Journal
Date:Jun 1, 1999
Previous Article:Cross product subsidization in the health insurance market with managed care: a model and issues.
Next Article:The impact of removing corn subsidies in Mexico: a general equilibrium assessment.

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters