Distance to hospital and children's use of preventive care: is being closer better, and for whom?

Currie, Janet ; Reagan, Patricia B.

I. INTRODUCTION

This article examines the effect of distance to hospital on the utilization of preventive care among children. Many poor children, lacking alternative providers, rely on hospitals and clinics for preventive care. Moreover, in many poor neighborhoods, the majority of private doctors' offices are located in buildings adjacent to a hospital. Thus, hospitals indirectly serve to attract physician services that would otherwise be lacking in the neighborhood. This pattern of service provision raises several concerns. First, it is inefficient for children to be receiving preventive care directly from hospitals because preventive care can be delivered more cheaply in doctors' offices, and doctors' offices also provide greater continuity of care (an index of quality). Second, even when children are receiving care in a doctor's office adjacent to a hospital, continuation of that service may be jeopardized if the hospital were to close.

In this study, we use distance to hospital as a proxy for access to medical services to examine the effect of distance to hospital on the use of preventive care services. If the use of these services falls with distance, other things being equal, then we interpret this as evidence of lack of access to alternative providers. Most previous studies of the effects of distance have concentrated on specific geographical areas and/or hospitalizations for specific procedures. We are therefore unaware of any previous studies that have focused on the effects of distance on the utilization of preventive care among children. (1)

We use a national sample of children created by matching records from the National Longitudinal Survey of Youth's Child-Mother (NLSCM) file with the American Hospital Association's 1990 Hospital Survey. We allow the effects of distance to vary with race, ethnicity, insurance status, and degree of urbanicity. We control for other factors that might affect utilization of preventive care by including a rich set of control variables and by estimating models that include either city dummy variables or mother fixed effects. In particular, city dummies control for unobserved differences across cities in public transportation services and population density. Mother fixed effects control for unobserved differences in preferences for preventive health care. Robust estimates across models allow us to rule out these alternative explanations for why distance to hospital might affect different groups differentially. These innovations address the important question of whether children who rely on hospitals for preventive ca re do so because they lack access to other providers.

We find that distance to hospital has significant effects on access to preventive care only among central-city black children. For these children, each additional mile from the hospital is associated with a 3% decline in the probability of having had a checkup (from a mean baseline of 74%). This effect is comparable to the 3% increase in the probability of having a checkup, which is associated with having private health insurance coverage rather than being uninsured. A striking result is that among these children, the size of the distance effect is similar for both the privately insured and those with Medicaid, suggesting that even black urban children with private health insurance have difficulty obtaining access to preventive care outside the area around hospitals. Thus, for this group questions of access to providers may be as important as insurance coverage in predicting use of preventive care.

II. CHILDREN WHO RELY ON HOSPITALS FOR PRIMARY CARE

Prior research suggests that children who are members of minorities, uninsured, covered by Medicaid, or residing in rural areas are all more likely to rely on hospitals for preventive care. This section discusses each of these groups in turn.

Bloom (1990) demonstrates that, nationally, black children are twice as likely as white children to receive care in an institutional setting, such as a clinic or emergency room, and that they are more likely to be attended by residents than by staff physicians. Moreover, the Centers for Disease Control and Prevention (CDC) (1997) find that most children have a usual source of care (93% of black children and 95% of white children). However, 92% of white children with a usual source of care rely on a doctor's office, whereas only 67% of black children do. Weigers et al. (1996) use data from the 1996 Medical Expenditure Panel Survey to show that most of this discrepancy is accounted for by a greater reliance on hospitals for primary care among black children.

These differences by race in patterns of utilization of services may be explained by differences in the spatial distribution of physician offices between predominantly black and non-black neighborhoods. Fossett et al. (1992) find that in Chicago there are twice as many children per office-based pediatrician in the inner city compared to the best-served neighborhoods. In addition there are 60% more children per child health care provider in the inner city. This spatial distribution of physicians holds in most American cities. Figure 1 shows a map of Columbus, Ohio, displaying the locations of hospitals, pediatricians, and family doctors overlayed on the fraction of the population by census tract in 1990 that was black for tracts with at least 50% black population. The map illustrates the point that children in predominantly black neighborhoods are served by pediatricians and family doctors located near hospitals to a far greater extent than is found in predominantly nonblack neighborhoods.

Although there has been less investigation of access to health care among Hispanic children, the CDC (1997) data suggest that they are less likely than other children to have a usual source of care (86%). Moreover, only 70% of Hispanic children with a usual source use a doctor's office for primary care. (2) Denton and Massey (1989) show that in many areas recent Hispanic immigrants are even more residentially segregated than blacks, so that spatial factors may account for this pattern of usage. Furthermore, language may create a barrier between these families and some doctors in private practice, whereas many hospitals have translation services available.

Lack of insurance coverage is a second risk factor for inadequate access to care. Despite the existence of public programs to provide insurance for the poor, the Children's Defense Fund (1997) finds that 10 million children (14%) have no health insurance coverage at all. Children without health insurance are less likely to have a regular source of care and are five times more likely than other children to use an emergency room as a regular source of care. One in four uninsured children use an emergency room as his or her regular source of care or has no regular provider. There is a great deal of literature showing that these children are less likely to seek routine preventive care and receive less care when they are sick, for example, Currie and Thomas (1995), Weissman and Epstein (1994), and Kogan et al. (1995).

Children with Medicaid (the main source of public health insurance for poor children) may also have problems finding private doctors who will accept this coverage. Rowland and Salganicoff (1994) show that Medicaid pays about half as much as private insurers for the same pediatric services. Twenty percent of U.S. pediatricians refuse to see Medicaid patients at all, and 40% limit the number of Medicaid patients in their practices. Moreover, both percentages have been growing over time as more physicians opt out of the Medicaid program. Yudkowsky et al. (1990) show that in 1977 only 15% of physicians refused Medicaid patients and only 26% limited the number of their Medicaid patients.

The result is that although children on Medicaid are more likely than uninsured children to have a usual source of care and to receive routine care on an appropriate time frame, they are less likely than privately insured children to be seen in doctors' offices. They are also more likely than privately insured children to lack continuity between usual sources of routine and sick care because they typically receive routine care at a clinic and sick care in a hospital emergency room, as documented by St. Peter et al. (1992).

Degree of urbanicity is a third risk factor for lack of access to care. Many rural children face shortages of providers in private practice, and hence are forced to rely on hospitals for primary care. In 1990, the U.S. Department of Health and Human Services (1990) surveyed the states and found that virtually all of them cited general shortages of primary care physicians, particularly in rural areas, as a serious concern. Halfon et al. (1998) find that children living in counties in which the supply of primary care physicians was in the top 20% had half the odds of reporting emergency departments as their usual source of care as other children.

III. DATA

Our main source of data is the NLSCM. This data set is based on the National Longitudinal Survey of Youth (NLSY) 1979, which began in 1979 with 6,283 young women aged 14 to 21 (and a similar number of young men). These women were followed annually until 1994 and biennially thereafter. In 1986, the NSLCM began biennially surveying the children born to the female respondents. As of the 1994 survey, 10,042 children have been included in the NLSCM. Thus, a good deal of information is available about both mothers and children.

Our sample is composed of children drawn from the five waves of the survey conducted between 1986 and 1994. It consists of all children under 12 years old, excluding those whose mothers were in the military sample or in the poor white oversample. The latter two groups were dropped from the survey partway through our sample period. Although the NLSCM is based on a nationally representative sample of young women who were in the United States in 1978, their children may not form a representative sample of U.S. children for several reasons. First, the youngest women in the NLSCM were 30 in 1994 and may not have completed their childbearing. Thus, the children of the NLSCM tend to be born to young mothers on average. Second, the composition of some groups, such as Hispanics, has changed dramatically since 1978 due to immigration. Our calculations using Current Population Survey data suggest that about 8% of the 1957 to 1964 birth cohort currently living in the United States have immigrated since 1979. These limita tions of the data should be kept in mind.

Our dependent variable is whether or not the child has had a routine checkup in the past year. The question asked in the survey is: "When did [name] last see a doctor for a routine health checkup?" The American Academy of Pediatrics recommends that all children in this age group receive at least one routine checkup per year, so children who lack this minimum contact with the medical establishment are going without recommended preventive care. For some children, this may have little consequence. For others, lack of preventive care will mean that they do not receive recommended vaccinations and that conditions such as growth retardation, chronic ear infections, asthma, and vision problems are not properly diagnosed or treated. Nevertheless, the child's probability of having had a checkup goes down with age, so we control for age dummies in all of our models. We also control for whether a child was covered by private health insurance, Medicaid, or was uninsured in each wave.

Our hypothesis is that the probability a child receives a checkup declines with distance to the nearest hospital. We calculated distance to hospital by determining the latitude and longitude of the respondent's residence in each year and matching to the latitude and longitude of the nearest hospitial. To track respondents from year to year, the National Longitudinal Survey collects street addresses for each respondent. However, prior to 1990, street addresses were recorded only for respondents who moved. Hence, addresses for 1986 and 1988 have to be imputed for those who did not move. To do the imputation, we began in 1979 with the addresses for parents and spouses, because in 1979 respondent addresses were not obtained. By using the household roster to identify respondents who lived with their parents or spouses, we obtained a 1979 address for about 70% of the mothers in the sample. (3) The remaining mothers were assigned an address at the center of their ZIP code, because 1979 ZIP codes were available for a ll mothers.

Until 1984, respondents were asked whether they had moved since the last interview. Thus, in principal, we can start with the 1979 address. If the person moved subsequently, then the new address should be reported. If they report that they did not move and no new address is given, then we use the old address. (4) In 1984 through 1990, we assume that if no new address is reported and the metropolitan statistical area (MSA) or county and state for rural residents remain the same, then the person did not move. Beginning in 1990, we have the street address for all respondents. (5) This database of street addresses was then fed into geographic software that determines the latitude and longitude of each address. (6) A first pass yielded a match rate of approximately 80%. (7)

Only 10% of the potential sample was lost due to an inability to locate mothers. Fourteen percent of the sample was lost due to missing data for family income. Out of 25,233 potential person year observations we obtained a final sample of 16,746. There is no evidence that the sample selection criteria biased the sample. We found that cases that were excluded due to missing data were similar in most respects to those that are included in our sample. For example, the mean of our dependent variable (whether or not a child had a checkup in the past year) did not change when those missing other data were excluded from the sample. A detailed description of the sample construction is available on request.

Table 1 presents sample sizes by location of mothers, children, MSAs, and moves. We have a final sample of 16,746 observations on 3,173 mothers and 6,722 children in 232 different MSAs and rural locations. In our mother fixed-effects model, the coefficient estimate on distance to hospital is identified by mothers who move within each location category. Therefore, we also tabulated the number of within-location-category moves that mothers make and the number of children affected by these moves for each of our subsamples. We define a move as being a change in residential location of at least one-tenth of a mile. In the whole sample we observe mothers moving within each of the location categories on 1,055 occasions, which affected 1,723 children. Of the 3,173 mothers, 896 or 28% ever moved within a location category. Of the 6,722 children, 1,509 or 22% were ever affected by their mothers' moves. Central-city blacks and Hispanics are more likely to move than central city whites. Rural blacks and Hispanics are als o more likely to move than rural whites. Suburban blacks are slightly less likely to move than suburban whites. Suburban Hispanics are the most likely to move of the three suburban groups.

The NLSCM data are matched to data on hospital location from the American Hospital Survey of 1990. This survey is a census of all hospitals. In addition to the exact street address, the survey tells us which hospitals might potentially treat children. Only general and pediatric hospitals were included. We determined hospital latitude and longitude using the methods described. When a match could not be made, we examined maps, which often show hospitals, to determine the location. All but three general and pediatric hospitals were successfully located. Our hospital sample consists of 5,731 general and children's hospitals. We found exact locations for 5,468 of these hospitals and used ZIP code centroids as the location of the remaining 263 hospitals. (8) For each respondent, we calculated distance to the closest hospital. Everyone in our sample lived within 50 miles of a hospital.

In addition to constructing the distance variable, we used the geographic software and address information to refine the "central city" variable that is on the public-use NLSCM tape. The NLSY data set contains a "central city" variable that is defined using ZIP codes. If a ZIP code lies entirely within the city limits of the "main" city of an MSA (population more than 100,000) then the respondent is coded as living in a central city. If the ZIP is entirely in the MSA but not within the city limits then the respondent is coded "MSA-not central city." Finally, if the ZIP laps in and out of the city limits, then the respondent is coded as "don't know central city." As a result, the central city variable is missing for many NLSY respondents. We improved on this measure by overlaying boundary files of central cities on top of the respondents' latitudes and longitudes to eliminate the "don't know central city" category. However the reader should be aware that this "central city" variable is based on city limits. It includes but is not restricted to impoverished inner-city neighborhoods.

Table 2 gives an overview of our data, broken down into four categories: central-city black, central-city white and Hispanic, suburban, and rural. The choice of these four categories reflects the fact that we were unable to find any statistically significant differences in the effects of distance between central-city whites and Hispanics or between racial and ethnic groups outside of the central-city area. Out of our four groups, central-city black children are both most likely to get checkups and more likely to live close to hospitals. They are also the least likely to be uninsured. Although they are less likely than other children to have private health insurance coverage, central-city black children are more likely than other children to have Medicaid coverage.

In some respects, such as family income or not having a father present, black central-city children appear disadvantaged relative to others. However, their mothers are better educated than mothers of other central-city children or of rural children and are about equally likely to work. All of these characteristics are likely to be related to the probability that a child receives a checkup. Thus, we control for them. Models that excluded potentially endogenous characteristics, such as measures of family structure and work status, produced very similar estimates.

The NLSCM also has information about car ownership. Because car ownership may significantly mitigate the effect of distance on utilization of care, we have included this variable in our models, and in some cases have interacted it with the distance variable. Although it may be argued that car ownership is an endogenous choice, many of the omitted factors that are likely to explain ownership will be captured by either city dummies or mother fixed effects. As Table 2 shows, central-city blacks differ significantly from other children in that their families are much less likely to own cars. In the other groups, car ownership rates vary between 79% and 88% compared to only 54% for central-city blacks.

IV. ESTIMATION

Baseline estimates of the effects of distance to hospital on use of preventive care are obtained by estimating linear probability models in which the dependent variable is a zero/one indicator for whether a child had a preventive care visit in the past year. The independent variables include indicators for private and public insurance; whether the child was first born, black, Hispanic, or male; the age of the mother at birth; and indicators for whether the mother was a high school dropout, high school graduate, had some college, or was a college graduate (the omitted category for education is "missing"). In addition, in the results reported we include indicators for whether or not the mother was married, whether she had a car, whether the mother was employed full-time, family size, and the log of family income. Although some may view these variables as endogenous, we include them because they are important determinants of use of preventive care and are arguably predetermined. Our estimates are, however, very similar if these variables are excluded from the model. Estimates of logits or probits rather than linear probability models produced very similar results, which is not surprising given the distribution of our dependent variable.

In addition to these baseline models, we estimate two additional sets of models. One group of models includes city dummy variables. These estimates control for characteristics of cities, such as the quality of public transportation, the population density, and the distribution of medical services, which may mitigate the effects of distance. In these models, the effect of distance to hospital on checkups is identified by comparing children within cities who live at different distances to hospital.

The second set of models includes fixed effects for mothers. These fixed effects control for unobserved time-invariant characteristics of mothers, such as health knowledge, value attached to preventive care, and resourcefulness in finding providers willing to treat her child. In these models, the effect of distance is identified by mothers who move within each of the three location categories.

In principle, one could estimate models that controlled for both mother fixed effects and city dummies, but we found that in practice there was not enough variation in our data to allow us to do this. However, a comparison of the three sets of models sheds light on the importance of omitted city-specific and mother-specific factors that might influence the use of preventive care and allows us to determine whether ordinary least squares (OLS) estimates are biased by failure to control for these factors. We will show that our results are remarkably similar across the three specifications. This finding suggests that our model is capturing the main determinants of use of preventive care and that our results regarding the effects of distance are unlikely to be contaminated by omitted variable bias.

V. RESULTS

OLS regressions of the effects of distance on the probability of a checkup are shown in Table 3 for each group. These linear probability models include all of the variables shown in Table 2. Our main results are robust to the exclusion of marital status, employment status, family size, and family income from the regressions. Logit or probit models produced very similar results, as one would expect, given that our dependent variable is not a rare outcome.

The OLS estimates indicate that among central-city blacks, the probability of a checkup decreases by 4% for each mile of distance from a hospital. Among suburban residents, distance has a small but statistically significant effect. There is no significant effect of distance among nonblack central-city residents or among rural residents.

The next two rows of Table 3 show that insurance increases the probability that children in all four groups have had a checkup in the past year and that the effect is as much as twice as big for Medicaid as for private health insurance coverage. This result is consistent with work by Currie and Thomas (1995) on the effects of insurance coverage using the NSLCM. It is not a surprising result given that Medicaid checkups are free, whereas most private insurance policies have deductibles and copayments and many do not pay for preventive pediatric care at all.

Other notable results are that firstborn children are more likely to have checkups than their siblings. Hispanic central-city children are less likely than other white central-city children to have checkups, although there is no difference between Hispanics and other whites outside the central city. The probability of a checkup increases with maternal education for central-city whites and Hispanics and suburban residents. These results are in keeping with the literature. It is also interesting to note that having a mother who is employed full-time has a significant negative effect on the probability of a checkup among suburban and rural children.

We also find that car ownership has a significant negative effect on the probability of a checkup among white and Hispanic central-city residents. This finding probably reflects the fact that OLS estimates do not control for the fixed unobserved variables at either the city or the maternal level, such as quality of public transportation and maternal resourcefulness, that were discussed in section III.

Table 4 compares the estimated effects of distance from Table 3 (reproduced in panel A) to those obtained from models with city dummies (in panel B). The estimated effects of distance for central-city blacks are robust to this change in specification. The point estimate is from 4% to 3% for each mile of distance but remains strongly statistically significant. In contrast to the OLS estimates that suggested a small negative effect of distance among suburban children, models with city dummies show no significant effect of distance on the probability of checkup for suburban or rural children. City dummies reduce the effects of health insurance and Medicaid by about one-quarter among central-city residents. Only the effect of Medicaid remains statistically significant.

Estimates from models with mother fixed effects are shown in panel C of Table 4. The estimates for central-city blacks are remarkably similar to those obtained from either the OLS models or those with city dummies. This result suggests that we are really measuring an effect of distance to hospital because adding controls for city dummies or mother fixed effects have no impact on our estimates. Moreover, it is striking that the effect of distance to hospital on probability of checkup found for black central-city children does not carry over to nonblack central city children. These findings suggest that location is not the only factor causing black children to rely on hospitals for preventive care. Finally, note that the anomalous negative coefficient on car ownership among nonblack central-city children becomes smaller when city dummies are added to the model and becomes statistically insignificant when the model is estimated with mother fixed effects. (9) Moreover, we now find a significant positive effect of car ownership on checkups among rural children, a finding that makes considerable sense. The estimates in Tables 3 and 4 are consistent with the literature that finds black central-city children to be more reliant on hospitals and their affiliated medical services for their primary care than are other groups.

The literature also suggests that uninsured children and children on Medicaid may be more likely than others to rely on hospitals. This hypothesis is investigated in Table 5, which follows the same three-panel format as Table 4. The OLS estimates indicate that both privately insured and Medicaid-covered black central city children are affected by distance to hospital. Moreover, the effects for these two groups of insured children are roughly equal in magnitude and not significantly different from each other. Once again, the OLS estimates are robust to both the inclusion of city dummies (panel B) and mother fixed effects (panel C).

A surprising finding is that there is no significant interaction between lack of insurance coverage and distance to hospital among central-city blacks (or other groups for that matter) in any specification. The fact that the probability of a checkup is not related to distance to hospital among the uninsured suggests either that they receive nonurgent care only from particular hospitals, so that distance to the nearest hospital is not the relevant concept for this group. As we have seen, black central-city children who are uninsured are less likely than the insured to get regular checkups, which suggests that they often go without primary care rather than being served by nearby hospital clinics that take black central-city children who have some form of insurance. Thus, Table 5 suggests that many central-city black children with insurance rely on hospitals and affiliated medical services for preventive care, whereas uninsured children either do without or find service providers at locations other than the near est hospital.

Among nonblack central-city residents and all suburban residents, our estimates are less consistent across the three specifications. OLS estimates suggest that distance has a significant but very small negative effect on the probability of a checkup among privately insured suburban children. However, this effect becomes insignificant once city dummies are included and actually becomes slightly positive and significant in models with mother fixed effects.

Among white and Hispanic central-city children, distance is not statistically significant except in the mother fixed effects models in which it has a large (3%) positive effect on the probability of a checkup among the privately insured. If these children are more likely to be seen by doctors than hospitals, this result is consistent with work by Fosset and coauthors (1992) who found that the number of doctors in private practice who see these privately insured patients generally increases with distance from central city.

If the estimates truly reflect the effects of distance, then we should expect people with better transportation to be less affected. Hence, we have estimated models interacting distance and car ownership, which are shown in Table 6. Panels A and B of Table 6 show that among central-city blacks, distance has negative effects on the probability of checkups among both car owners and noncar owners. Moreover, the effects for noncar owners are twice as big as for car owners. However, these effects become statistically insignificant once mother fixed effects are added to the model. These mother fixed effects are identified using mothers who changed either distance or car ownership status. A potential problem is that changes in these two factors could be off-setting because someone who moves further away from the center of town and public transportation services is more likely to need a car. We find little effect of distance in any of our specifications among the other three groups.

VI. EXTENSION

We have estimated several other variants of the models discussed. First, we have reestimated models using distance to the nearest hospital that accepts Medicaid patients as the independent variable of interest, rather than distance to the nearest hospital. The hospital survey allows us to calculate the fraction of inpatient days paid for by Medicaid in each hospital. We required the hospital to have at least 1% of patient days paid for by Medicaid to qualify as a hospital serving Medicaid patients. Some hospitals did not report in this section and some hospitals did not serve Medicaid patients. Average distance to a hospital serving Medicaid patients for central-city blacks is 0.25 miles farther away than distance to nearest hospital. The coefficient estimates for black central-city residents become slightly smaller when using distance to nearest hospital serving Medicaid patients. In the simplest OLS specification the coefficient fell from -0.04 to -0.035. Thus, the ballpark estimate of a 3% decline in the probability of a checkup for each mile from a hospital is robust to this change in variable definition.

Second, we estimated models that contained child fixed effects rather than mother fixed effects. The models with mother fixed effects use within-child variation in distance as well as variations in distance between siblings to achieve identification. Models with child fixed effects use only the within-child variation. It is the mother (or parent) who determines whether or not a child will go for a checkup, so her unobserved characteristics are likely to be important. Nevertheless, there may also be characteristics of individual children that would lead them to be more or less likely to receive checkups. Hence, it could be argued that models with child fixed effects are preferred to those presented. We found, however, that there was not enough within-child variation in distance to allow us to precisely identify the effects of distance in these models.

A third issue we have addressed is whether it is appropriate to measure distance to hospital linearly. Nonparametric regressions of distance on the probability of a checkup indicated that the assumption of linearity was reasonable over most of the range of our data.

VII. CONCLUSIONS

Our central finding is that among central-city black children, a longer distance to the nearest hospital reduces the probability of checkups. We cannot reject the hypothesis that the effects are the same for children on Medicaid and children with private health insurance coverage. These results suggest that many central-city black children rely on hospitals and surrounding providers for preventive care, whether or not they have private health insurance coverage.

We find little evidence that white or Hispanic children on Medicaid rely on hospitals for preventive care. It is possible that previous findings that children on Medicaid tend to rely on hospitals for primary preventive care reflect the fact that children on Medicaid are disproportionately likely to be black, rather than an effect of Medicaid per se. Other surprising negative findings are that there is no relationship between distance and checkups among rural residents or among uninsured children. It is important to note that these results do not imply that these groups are receiving adequate preventive care. Our point estimates suggest that these children are significantly less likely to receive checkups than other children but that this poor showing is independent of distance to hospital.

The current system appears to serve black children well in the sense that they are as likely to get checkups as other central-city children (who presumably rely more heavily on other providers). However, there are several reasons for concern. Some of these children are receiving preventive care directly from hospitals or clinics attached to hospitals. Care provided by hospitals is generally more expensive and more likely to lead children to be admitted than care provided in doctor's offices. In addition, quality of care, as measured by such indicators as continuity of care, may be lower in hospital clinics than in doctors' offices. Our results show that trends that threaten the viability of central-city hospitals and ancillary services are likely to have a disproportionate impact on black children. TABLE 1 Sample Sizes and Counts of Mothers, Children, Moves, and MSAs Black White Hispanic Black Central City Central City Central City Suburb # Observations 2,734 1,600 1,558 1,661 # Children 1,316 751 820 812 # Child moves 320 103 230 161 # Children ever move 273 100 201 145 # Mothers 625 477 329 398 # Mother moves 201 74 129 103 # Mothers ever move 163 70 112 89 # MSAs 106 139 70 83 White Hispanic Black White Hispanic Suburb Suburb Rural Rural Rural Total # Observations 4,852 2,289 794 1,670 160 16,746 # Children 2,202 825 403 734 69 6,722 # Child moves 502 193 92 99 23 1,723 # Children ever move 442 167 76 86 19 1,509 # Mothers 1,148 339 111 392 33 3,173 # Mother moves 318 111 45 61 13 1,055 # Mothers ever move 272 90 38 51 11 896 # MSAs 162 79 NA NA NA 232 Notes: The total number of children and mothers is less than the sum of the columns because some children and mothers contribute to more than one sample. For example, if a woman moves from a central city to a suburb, she contributes observations to the central city sample for those years when she lives in the central city and she contributes to the suburban sample for those years when she lives in the suburbs. The reported total number of times a mother or child moves is the sum of the columns and excludes movements across geographic units, such as a move from a central city to a suburb. The total number of MSAs is less than the sum across the columns because the MSAs in each column are not mutually exclusive. TABLE 2 Means by Race and Location Black Nonback Central Central City City Suburban Rural Checkup past year 0.74 0.72 0.69 0.66 (0.44) (0.45) (0.46) (0.47) Distance to hospital 1.51 1.64 4.35 7.31 (1.09) (1.36) (4.36) (6.77) Child characteristics Private health 0.46 0.61 0.72 0.58 insurance (0.50) (0.49) (0.45) (0.49) Medicaid 0.43 0.24 0.15 0.27 (0.49) (0.43) (0.36) (0.45) Uninsured 0.11 0.15 0.13 0.15 Hispanic 0 0.53 0.22 0.07 (0.50) (0.42) (0.25) Black 1 0 0.22 0.33 (0.41) (0.47) Male 0.49 0.53 0.51 0.50 (0.50) (0.50) (0.50) (0.50) First born 0.40 0.48 0.46 0.43 (0.49) (0.50) (0.50) (0.50) Age (years) 6.26 5.63 5.62 6.01 (3.47) (3.39) (3.37) (3.37) Mother characteristics Age @ birth 22.69 23.86 24.06 22.03 (3.99) (4.04) (4.04) (4.01) Education 12.39 12.15 12.52 12.04 (1.67) (2.39) (2.21) (2.17) Car 0.54 0.81 0.88 0.79 (0.50) (0.39) (0.32) (0.41) Married, spouse 0.31 0.66 0.72 0.63 present (0.46) (0.48) (0.45) (0.48) Full-time 0.45 0.46 0.51 0.43 employment (0.50) (0.50) (0.50) (0.49) Family size 5.14 5.22 5.28 5.36 (1.57) (1.44) (1.39) (1.40) Family income 15,912 29,486 32,967 22,992 (24,660) (57,066) (56,351) (47,204) # Observations 2,734 3,158 8,230 2,624 # Cities 103 163 202 - # Mothers 625 806 1,885 567 # Children 1,316 1,571 3,839 1,206 Note: Standard deviations in parentheses. TABLE 3 OLS Regressions of Checkups on Distance Black White/Hispanic Central City Central City Suburban Distance -0.040 -0.004 -0.002 (5.34) (0.76) (2.20) Private health insurance 0.039 0.108 0.038 (1.42) (4.71) (2.51) Medicaid 0.103 0.171 0.091 (3.67) (6.51) (4.62) First born 0.070 0.028 0.032 (3.68) (1.62) (2.85) Black -- -- 0.078 (6.01) Hispanic -- -0.032 -0.005 (1.93) (0.41) Male 0.017 -0.025 -0.004 (1.04) (1.66) (0.38) Age mother @ birth 0.007 -0.003 -0.001 (2.25) (1.29) (0.73) High school dropout 0.025 0.015 -0.051 (0.28) (0.42) (1.79) High school graduate 0.040 0.021 -0.027 (0.44) (0.65) (1.02) Some college 0.058 0.070 -0.001 (0.63) (1.93) (0.046) College graduate 0.068 0.036 0.022 (0.69) (0.86) (0.71) Car -0.027 -0.106 0.003 (1.35) (4.52) (0.15) Married, spouse present -0.013 -0.038 -0.017 (0.60) (1.77) (1.17) Full-time employment -0.016 -0.024 -0.025 (0.75) (1.41) (2.40) Family size -0.001 0.012 -0.023 (0.71) (1.31) (9.60) Log family income -0.012 0.010 0.007 (1.61) (1.45) (1.50) Intercept 0.47 0.46 0.56 (3.60) (4.67) (8.51) [R.sup.2] 0.121 0.153 0.144 # Observations 2734 3158 8230 Rural Distance 0.000 (0.09) Private health insurance 0.065 (2.44) Medicaid 0.150 (4.91) First born 0.077 (3.81) Black 0.021 (0.93) Hispanic 0.043 (1.22) Male 0.020 (1.15) Age mother @ birth 0.002 (1.78) High school dropout -0.034 (0.83) High school graduate -0.016 (0.44) Some college -0.007 (0.15) College graduate 0.014 (0.27) Car -0.009 (0.31) Married, spouse present -0.028 (1.05) Full-time employment -0.087 (4.45) Family size -0.005 (3.21) Log family income 0.006 (0.60) Intercept 0.39 (3.49) [R.sup.2] 0.152 # Observations 2624 Note: t-statistics in parentheses. TABLE 4 Regressions of Checkups on Distance Black Central City White/Hispanic Central City A: Ordinary least squares Distance -0.040 -0.004 (5.34) (0.76) Private health insurance 0.039 0.108 (1.42) (4.71) Medicaid 0.103 0.171 (3.67) (6.51) Car -0.027 -0.106 (1.35) (4.52) [R.sup.2] 0.121 0.153 B: City dummies Distance -0.032 -0.101 (3.52) (0.08) Private health insurance 0.032 0.105 (1.12) (4.30) Medicaid 0.082 0.145 (2.78) (5.35) Car 0.008 -0.058 (0.39) (2.28) [R.sup.2] 0.204 0.244 C: Mother fixed effects Distance -0.034 0.011 (2.98) (0.98) Private health insurance 0.033 0.038 (1.02) (1.28) Medicaid 0.061 0.082 (1.95) (2.68) Car 0.014 -0.047 (0.55) (1.51) [R.sup.2] 0.467 0.482 Suburban Rural A: Ordinary least squares Distance -0.004 0.000 (2.20) (0.09) Private health insurance 0.038 0.065 (2.51) (2.44) Medicaid 0.091 1.50 (4.62) (4.91) Car 0.003 -0.009 (0.15) (0.31) [R.sup.2] 0.144 0.152 B: City dummies Distance -0.001 (0.96) Private health insurance 0.034 (2.20) Medicaid 0.086 (4.31) Car 0.020 (1.07) [R.sup.2] 0.177 C: Mother fixed effects Distance 0.002 0.003 (1.19) (1.03) Private health insurance 0.020 0.049 (1.00) (1.52) Medicaid 0.059 0.141 (2.56) (3.83) Car 0.035 0.084 (1.39) (2.23) [R.sup.2] 0.453 0.475 Notes: t-statistics in parentheses. Aside from the mother and/or city dummies, there regression models are of the same form as those shown in Table 2. TABLE 5 Interactions of Distance and Insurance Status Black Central City A: Ordinary least squares Private health insurance 0.115 (2.34) Private health * distance -0.039 (4.49) Medicaid 0.175 (3.48) Medicaid * distance -0.030 (2.36) Uninsured * distance 0.012 (0.48) [R.sup.2] 0.121 B: City dummies Private health insurance 0.093 (1.87) Private health * distance -0.027 (2.69) Medicaid 0.158 (3.10) Medicaid * distance -0.032 (2.44) Uninsured * distance 0.015 (0.58) [R.sup.2] 0.204 C: Mother fixed effects Private health insurance 0.079 (1.44) Private health * distance -0.027 (2.40) Medicaid 0.127 (2.34) Medicaid * distance -0.037 (2.18) Uninsured * distance 0.005 (0.18) [R.sup.2] 0.468 White/Hispanic Central City Suburban A: Ordinary least squares Private health insurance 0.082 0.066 (2.39) (3.26) Private health * distance 0.002 -0.004 (0.23) (2.54) Medicaid 0.152 0.123 (4.01) (4.81) Medicaid * distance -0.003 -0.004 (0.27) (1.40) Uninsured * distance -0.014 0.002 (1.06) (0.89) [R.sup.2] 0.153 0.144 B: City dummies Private health insurance 0.075 0.054 (2.05) (2.62) Private health * distance 0.007 -0.002 (0.88) (1.34) Medicaid 0.126 0.114 (3.30) (4.39) Medicaid * distance -0.001 -0.003 (0.07) (1.29) Uninsured * distance -0.011 0.002 (0.77) (0.81) [R.sup.2] 0.245 0.177 C: Mother fixed effects Private health insurance -0.027 0.005 (0.62) (0.20) Private health * distance 0.028 0.005 (2.56) (2.24) Medicaid 0.081 0.080 (1.96) (2.62) Medicaid * distance -0.012 -0.005 (0.71) (1.41) Uninsured * distance -0.009 0.001 (0.52) (0.29) [R.sup.2] 0.483 0.453 Rural A: Ordinary least squares Private health insurance 0.031 (0.80) Private health * distance -0.000 (0.036) Medicaid 0.103 (2.36) Medicaid * distance 0.002 (0.81) Uninsured * distance -0.005 (1.27) [R.sup.2] 0.152 B: City dummies Private health insurance Private health * distance Medicaid Medicaid * distance Uninsured * distance [R.sup.2] C: Mother fixed effects Private health insurance 0.011 (0.24) Private health * distance 0.001 (0.44) Medicaid 0.132 (2.54) Medicaid * distance -0.003 (0.79) Uninsured * distance -0.004 (0.84) [R.sup.2] 0.475 Notes: t-statistics in parentheses. Aside from the mother and/or city dummies and interaction terms, these regression models are of the same form as those shown in Table 2. TABLE 6 Interactions of Distance and Car Ownership Black Central City A: Ordinary least squares Car ownership -0.070 (2.35) Car * distance -0.028 (3.01) No car * distance -0.058 (4.83) Private health insurance 0.038 (1.35) Medicaid 0.104 (3.71) [R.sup.2] 0.122 B: City dummies Car ownership -0.027 (0.85) Car * distance -0.022 (2.03) No car * distance -0.046 (3.51) Private health insurance 0.031 (1.09) Medicaid 0.083 (2.81) [R.sup.2] 0.204 C: Mother fixed effects Car ownership 0.06 (0.22) Car * distance -0.005 (0.76) No car * distance -0.014 (1.12) Private health insurance 0.031 (0.93) Medicaid 0.049 (1.52) [R.sup.2] 0.468 White/Hispanic Central City Suburban Rural A: Ordinary least squares Car ownership -0.140 -0.028 0.040 (4.16) (1.26) (1.06) Car * distance -0.001 -0.001 -0.002 (0.21) (1.41) (1.07) No car * distance -0.025 -0.009 0.004 (1.58) (3.08) (0.76) Private health insurance 0.107 0.038 0.061 (4.66) (2.50) (2.31) Medicaid 0.168 0.090 0.144 (6.38) (4.59) (4.71) [R.sup.2] 0.153 0.145 0.153 B: City dummies Car ownership -0.069 -0.005 (1.90) (0.022) Car * distance -0.001 -0.000 (0.8) (0.23) No car * distance -0.07 -0.006 (0.43) (2.16) Private health insurance 0.105 0.034 (4.29) (2.20) Medicaid 0.144 0.086 (5.31) (4.30) [R.sup.2] 0.244 0.177 C: Mother fixed effects Car ownership -0.048 0.022 0.099 (1.39) (0.83) (2.46) Car * distance 0.005 0.002 -0.000 (0.96) (1.82) (0.28) No car * distance 0.004 -0.002 0.004 (0.27) (0.44) (1.25) Private health insurance 0.034 0.020 0.051 (1.14) (1.04) (1.58) Medicaid 0.090 0.063 0.148 (2.82) (2.67) (3.90) [R.sup.2] 0.482 0.459 0.475 Notes: t-statistics in parentheses. Aside from the mother and/or city dummies and interaction terms, these regression models are of the same form as those shown in Table 2.

(1.) Studies by Luft et al. (1990), Burns and Wholey (1992), and Goodman et al. (1997) have focused on the effect of distance to hospital on the choice of hospital conditional on a hospitalization and on delays in seeking hospitalization.

(2.) Hernandez and Charney (1998) provide a useful review of the literature regarding access to health care among Hispanic children. Hispanics at all ages are less likely to usc preventive care than other groups.

(3.) In some cases addresses were recorded for siblings in the same household, and adding this information increased the fraction with addresses by a few percentage points. For respondents who were unmarried and away at school in 1979, we use the parents' address.

(4.) Only ZIP codes are available for 1980. All of the exact street addresses have been lost.

(5.) Exceptions occur when the respondent was in the military and living on a base or ship or was incarcerated. In these cases we have no address. In rural areas addresses are sometimes reported as "the third trailer on the right." In these cases, we assigned addresses at the center of the respondent's ZIP code.

(6.) The geographical software that we used included Arcview, Arc Information, and Geographic Data Technology's Matchmaker/2000. We used Geographic Data Technology's Matchmaker/2000 to geocode complete addresses and assign ZIP code centroids when address information was incomplete. We used maps available in Arcview to check locations of incomplete addresses.

(7.) When a match was not made we searched maps that were also included in the software for a likely location. For example, an address might have been recorded as "221 Morning Glory Circle, Cleveland, Ohio," and the map indicated that only numbers between 1000 and 2000 existed on Morning Glory Circle. If Morning Glory Circle was a reasonably short street, we would assign an address on the midsection of the street. Alternatively, if the software found a "Morning Glory Court," rather than a circle, we would use "221 Morning Glory Court." When we could not come up with a sensible match, we assigned an address at the center of the ZIP code.

(8.) Some hospitals have multiple locations, but only one address is available from the American Hospital survey. Thus, our numbers understate the true diversity of hospital locations.

(9.) In the mother fixed effects models, this coefficient is identified by mothers who change car ownership status. The percentages changing ownership status were relatively large at 25%, 14%, 8%, and 12% for the black central-city, white/Hispanic central city, suburban, and rural groups, respectively.

REFERENCES

Bloom, B. "Health Insurance and Medical Care," in Advance Data from Vital and Health Statistics of the National Center for Health Statistics, no. 188. Washington, DC: Public Health Service, 1990.

Burns, L., and D. R. Wholey. "The Impact of Physician Characteristics in Conditional Choice Models of Hospital Care." Journal of Health Economics, 11(1), 1992, 43-62.

CDC (Centers for Disease Control and Prevention). Access to Health Care Part 1: Children, Series 10, No. 196. Washington, DC: U.S. Department of Health and Human Services, National Center for Health Statistics, 1997.

Children's Defense Fund. 14 Things You Should Know about the New Child Health Program. Washington, DC: Children's Defense Fund, 1997.

Currie, J., and D. Thomas. "Medical Care for Children: Public Insurance, Private Insurance and Racial Differences in Utilization." Journal of Human Resources, 30(1), 1995, 135-62.

Denton, N., and D. Massey. "Hypersegregation in U.S. Metropolitan Areas: Black and Hispanic Segregation Along Five Dimensions." Demography, 26(3), 1989, 373-91.

Fossett, J. W., J. Perloff, P. Kletke, and J. A. Peterson. "Medicaid and Access to Child Health Care in Chicago." Journal of Health Politics, Policy, and Law, 17(2), 1992, 273-98.

Goodman, D. C., E. Fisher, R. A. Stukel, and C. Cheng. "The Distance to Community Medical Care and the Likelihood of Hospitalization: Is Closer Always Better?" American Journal of Public Health, 87(7), 1997, 1144-50.

Halfon, N., P. Newacheck, D. Wood, and R. St. Peter. "Routine Emergency Department Use for Sick Care by Children in the United States." Pediatrics, 98(1), 1998, 28-34.

Hernandez, D., and E. Charney. From Generation to Generation: The Health and Well-Being of Children in Immigrant Families. Washington, DC: National Academy Press, 1998.

Kogan, M., G. R. Alexander, M. A. Teitelbaum, B. W. Jack, M. Kotelchuck, and G. Pappas. "The Effect of Gaps in Health Insurance on Continuity of a Regular Source of Care among Preschool-Aged Children in the United States." Journal of the American Medical Association, 274(18), 1995, 1429-35.

Luft, H., D. Garnick, D. Mark, D. Peltzman, C. S. Phibbs, E. Lichtenberg, and S. J. McPhee. "Does Quality Influence Hospital Choice?" Journal of the American Medical Association, 263(21), 1990, 2899-906.

Rowland, D., and A. Salganicoff. "Commentary: Lessons from Medicaid--Improving Access to Office-Based Physician Care for the Low-Income Population." Public Health Policy Forum, 84(4), 1994, 550-52.

St. Peter, R. F., P. Newacheck, and N. Halfon. "Access to Care for Poor Children: Separate and Unequal?" Journal of the American Medical Association, 267(20), 1992, 2760-64.

U.S. Department of Health and Human Services. States' Assessment of Health Personnel Shortages, Issues and Concerns. Washington, DC: Government Printing Office, HRS-P-OD-90-6, 1990.

Weigers, M. E., R. Weinick, and J. Cohen. MEPS Chart-book #1. Washington, DC: Agency for Health Care Policy and Research, 1996.

Weissman, J., and A. Epstein. Falling through the Safety Net: Insurance Status and Access to Health Care. Baltimore, MD: Johns Hopkins University Press, 1994.

Yudkowsky, B. K., J. Cartland, and S. Flint. "Pediatrician Participation in Medicaid: 1978 to 1989." Pediatrics, 85(4), 1990, 567-77.

RELATED ARTICLE: ABBREVIATIONS

COC: Centers for Disease Control and Prevention

MSA: Metropolitan Statistical Area

NLSCM: National Longitudinal Survey of Youth's Child-Mother File

NLSY: National Longitudinal Survey of Youth

OLS: Ordinary Least Squares

JANET CURRIE and PATRICIA B. REAGAN *

* We are grateful to David Cutler and Mark Duggan for helpful comments. We also thank Randy Olsen for his support of the NLSY79 geocode project. Steve Mulherin, Fernando Bosco, Chris Starrett, Kevin Dippold, and Eric Fischer provided excellent research assistance. Janet Currie thanks the Canadian Institute for Advanced Research and the NICHD for support under grant number R01-HD3101A2. All views expressed are those of the authors and are not necessarily endorsed by any funding agency.

Currie: Professor, Department of Economics, University of California at Los Angeles, 405 Hilgard Ave., Los Angeles, CA 90095. Phone 1-310-206-8380, Fax 1-310-825-9528, E-mail [email protected]

Reagan: Professor, Department of Economics, Ohio State University, Center for Human Resource Research, 921 Chatham Lane, Columbus, OH 43221. Phone 1- 614-487-0667, Fax 1-614-442-7329, E-mail [email protected]