Search tips
Search criteria 


Logo of hsresearchLink to Publisher's site
Health Serv Res. 2009 October; 44(5 Pt 1): 1603–1621.
PMCID: PMC2754550

Measuring Racial/Ethnic Disparities across the Distribution of Health Care Expenditures



To assess whether black–white and Hispanic–white disparities increase or abate in the upper quantiles of total health care expenditure, conditional on covariates.

Data Source

Nationally representative adult population of non-Hispanic whites, African Americans, and Hispanics from the 2001–2005 Medical Expenditure Panel Surveys.

Study Design

We examine unadjusted racial/ethnic differences across the distribution of expenditures. We apply quantile regression to measure disparities at the median, 75th, 90th, and 95th quantiles, testing for differences over the distribution of health care expenditures and across income and education categories. We test the sensitivity of the results to comparisons based only on health status and estimate a two-part model to ensure that results are not driven by an extremely skewed distribution of expenditures with a large zero mass.

Principal Findings

Black–white and Hispanic–white disparities diminish in the upper quantiles of expenditure, but expenditures for blacks and Hispanics remain significantly lower than for whites throughout the distribution. For most education and income categories, disparities exist at the median and decline, but remain significant even with increased education and income.


Blacks and Hispanics receive significantly disparate care at high expenditure levels, suggesting prioritization of improved access to quality care among minorities with critical health issues.

Keywords: Racial disparities, health care expenditures, quantile regression, vigicile

One of the continuing concerns in the health policy literature has been the marked differences in access to and extent of health care utilization across racial and ethnic subgroups. Studies of health care disparities in the early part of this decade found significant racial and ethnic disparities in a number of health care access measures (Collins, Hall, and Neuhaus 1999; Mayberry, Mili, and Ofili 2000; IOM 2002;). Since then, progress has been made in tracking disparities in health care use (Blanco et al. 2007; Cook, Miranda, and McGuire 2007; Gross et al. 2008; Martinez et al. 2008; Cook, McGuire, and Zuvekas 2009b;), understanding the role of geography in disparities (Skinner et al. 2003; Baicker et al. 2004; Baicker, Chandra, and Skinner 2005;), and more rigorously defining a racial/ethnic disparity in the health care context (McGuire et al. 2006; Cook et al. 2009a;).

A number of studies have focused on disparities in medical expenditure (Escarce et al. 1993; McGuire et al. 2006; Cook, Miranda, and McGuire 2007; Cook, McGuire, and Zuvekas 2009b;). This outcome variable improves upon other commonly used indicators of access to care (having any doctor visit in the last year, having a usual source of care) because it also captures the differences in intensity of care, as well as allowing for a more finely grained quantification of disparities among those that are the most medically needy. Cook, McGuire, and Zuvekas (2009b) found that significant disparities existed in total medical expenditures and that between 1996–1997 and 2004–2005 these disparities remained the same for blacks and widened for Hispanics. In another analysis of medical expenditures, Escarce and Kapur (2003) found little or no black–white or Hispanic–white disparities among aged Medicare beneficiaries. The literature on disparities in medical expenditures complements numerous other studies assessing disparities in health care access and utilization summarized in Unequal Treatment (IOM 2002) and elsewhere (Fennell 2005), as well as the National Healthcare Disparities Reports published yearly by the Agency for Healthcare Research and Quality (AHRQ 2007).

Many of these studies of disparities are measured at a population's mean, or at the mean conditional on observable characteristics of the racial and ethnic groups. Although mean differences are essential, they do overlook potential differences in different parts of the distribution, especially the right tail of the distributions of expenditures conditional on covariates that are known to differ across the groups. At lower expenditures, disparities may be driven by differential access for care, possibly reflecting differential access to and preferences for preventive health care. At the higher end of expenditures, the health system is treating individuals who are likely to have more critical health issues, and equity may be of greater concern. With attention turning to caring for particularly vulnerable populations (Mechanic and Tanner 2007), it will be useful for policy makers to assess whether discrepancies are exacerbated or diminished at the high end of expenditures where individuals are likely to have more critical health issues.

Differences in the upper tail of the distribution are likely to be driven by differences in white and minority use of high-cost medical procedures or access to inpatient care. Studies measuring racial disparities in cardiac care treatment and outcomes have compared blacks and whites presenting at cardiac care centers with similar symptoms, finding that blacks have lower rates of cardiac catheterization (Ford, Newman, and Deosaransingh 2000; Kressin and Petersen 2001; Kressin et al. 2004;), other cardiac tests (Ford, Newman, and Deosaransingh 2000), bypass surgery (Kressin and Petersen 2001; Petersen et al. 2002;), coronary angioplasty, and bypass surgery after cardiac catheterization (Peterson et al. 1997; Ford, Newman, and Deosaransingh 2000; Kressin and Petersen 2001; Ibrahim et al. 2003;). Disparities also exist among cancer patients—individuals likely to be in the upper tail of medical expenditure. Black cancer patients were 23 percent less likely than whites to receive radiation and surgery (Martinez et al. 2008), and treatment disparities among patients with breast, colorectal, lung, and prostate cancers did not improve between 1992 and 2002 (Gross et al. 2008). Looking across disease categories, black–white disparities continue to exist in major surgical procedures (Jha et al. 2005) and in the use of a wide range of medical procedures among Medicare fee-for-service beneficiaries (Escarce and McGuire 2004). We add to this literature by presenting a broader look at disparities among black, Hispanic, and white individuals and assessing whether the disparities continue to exist among those that require the largest amount of medical care resources.

The objective of this study is to assess black–white and Hispanic–white disparities in health care services expenditures across the distribution of the dependent variable and covariates. We know that more money is spent on white individuals than minority individuals when they share the same demographic, SES, and health status characteristics, but what about differences between white and minority individuals that share a high risk and incidence of costly health disorders? As one moves toward the upper quantiles of expenditure (conditional on the covariates), do the differences by group increase or abate? To answer these questions, we will use quantile regression, a statistical method previously used to assess differential response across gender, race, and other subgroups on treatments and other covariates (Buchinsky 1994, 1998; Manning, Blumberg, and Moulton 1995; Koenker 2005).



The data are from the 2001–2005 Medical Expenditure Panel Surveys (MEPS), which contain variables related to individuals' health care expenditures, demographic, socioeconomic status, and health status characteristics of a nationally representative sample of the noninstitutionalized civilian population of the United States. We pooled these 5 years of data in order to increase the precision of the estimates. The dependent variable of interest is total medical expenditure, the sum of all direct payments for care provided during the year, including out-of-pocket payments and payments by private insurance, Medicaid, Medicare, and other sources, but not including payments for over-the-counter drugs or nursing home care. Data were taken from responses to the medical provider and household components of the MEPS as reported on the MEPS public use files. Prices were adjusted to 2005 U.S. dollars using the consumer price index.

Covariates used include education level (less than high school, high school graduate, any college, college graduate), income level (below federal poverty level [FPL], near poverty [100–125 percent FPL], low income [125–200 percent FPL], middle income [200–400 percent FPL], and high income [≥400 percent FPL]), region of the country (Northeast, South, Midwest, and West), and insurance coverage (Private, Medicare, Medicaid or other public, and uninsured). Demographic characteristics include gender and age category (18–24, 25–34, 35–44, 45–54, 55–64, 65–74, and 75+), while health status includes any functional limitation, self-assessed health scores (excellent, very good, good, fair, poor), the physical (PCS-12) and mental health (MCS-12) components of the SF-12 (Ware, Kosinski, and Keller 1996) and their squares, and indicators for the chronic diseases and health conditions are as follows: diabetes, asthma, coronary heart disease, angina, myocardial infarction, stroke, emphysema, joint pain, arthritis, and other heart disease. Individuals of any race claiming to be of Latino or Hispanic origin were identified as “Hispanic” in our study. Other respondents were classified as African American or black (hereafter referred to as “black”) or non-Hispanic white (hereafter referred to as “white”) by responses to the Census-based question about race. Asian Americans and Native Americans were excluded due to small sample sizes.

As with other nationally representative surveys, the MEPS contain a large number of missing values on certain variables. The sample size of the 5 years of MEPS data was initially 175,309 but was trimmed to 93,320 given our exclusion criteria and to account for missing data. Individuals were included in the sample if they were 18 years of age or older, non-Hispanic white, black, or Hispanic, and had existing data on health status and health conditions, education, marriage, and income (individuals with some but not all income measures missing were included by using imputed income values available in the public use dataset). To account for differential missingness by race/ethnicity, and to maintain generalizability of the data to the adult (age≥18) white, black, and Hispanic population, we tracked exclusions due to missingness. We reweighted the included individuals to represent their propensity to be like individuals with missing values. This was done by estimating a logit regression of the probability of being missing on race/ethnicity, education, and sex, and interactions between race and education and race and sex, generating a predicted probability of being missing for each individual. Weights that account for missingness were calculated by multiplying the final sampling survey weight by the inverse of 1 minus the predicted probability of being missing. This is predicated on the assumption that the item nonresponse occurs at random, conditional on a core set of characteristics.

Statistical Analysis

To examine racial/ethnic disparities over the distribution of expenditures, we first make unadjusted comparisons by race across the distribution of expenditures. Second, we use quantile regression to estimate disparities at the median, the upper quartile, upper decile, and upper 5th centile of expenditures, conditional on covariates. We test the sensitivity of the results to comparisons based only on health status and to a two-part generalized linear model (GLM) that accounts for the extreme skewness and large zero mass of the data (Blough, Madden, and Hornbrook 1999; Buntin and Zaslavsky 2004;).

Unadjusted Comparisons of Racial/Ethnic Expenditure Distributions

To assess the cumulative distribution function (CDF) of medical expenditures for each racial/ethnic group, we first compared the CDF of medical expenditures for each group. The intercept of the CDF with the vertical axis shows the difference in the likelihood of no medical care during the year. To provide more detail, at the 25th, 50th, 75th, 90th, and 95th percentiles, we calculated and compared expenditures for each of the racial/ethnic groups. To assess how disparities in health care expenditures change as need for health care increases, we present unadjusted expenditures by racial/ethnic group at each of the deciles of the physical health composite score of the SF-12 (Ware, Kosinski, and Keller 1996).

Assessing Differential Response at Upper and Lower Quantiles Using Quantile Regression

We measure differences in disparities at the median, 75th, 90th, and 95th using quantile regression (Koenker and Hallock 2001; Koenker 2005;), assessing the differential response by race and ethnic group and the differential impact of education and income on these racial/ethnic differences. Quantile regression is not a regression estimated on a quantile, or subsample of data as the name may suggest. Instead, the method employs differential weighting of positive and negative absolute residuals across the distribution of data. For example, for a 75th quantile regression, one can think of quantile regression as passing a regression plane so that 75 percent of the observations are below the regression plane and 25 percent are above. Alternatively, this can be viewed as a specific form of iteratively reweighted regression, weighting negative residuals by 0.5 and positive residuals by 1.50 to ensure that minimization occurs when 75 percent of the residuals are negative. In ordinary least squares (OLS) regression, slopes are treated as constants and the error term as additive. In OLS, shifting the error up or down moves one through the distribution of the outcome, conditional on the covariates, but assumes that the shift is parallel on the scale of estimation across quantiles of the error. In quantile regression, the intercept and the slopes are both allowed to vary across the quantiles. The model still assumes the presence of a linear index function—the response is linear in the coefficients—but the values of the slope coefficients are not constrained to be the same across the distribution.

We employed quantile regression on the natural log of expenditures. Because our focus is on the higher levels of expenditure, we look at log(U.S.$+a constant). Because quantiles are order statistics, a monotonic transformation does not change the results, other than to look at the response as proportional rather than additive. Another concern is that retransformation of regression results to dollar terms may create biased results (Duan 1983). However, we avoid this problem by using regression models to measure disparities at upper quantiles of expenditures rather than at the mean. Retransformation of a quantile regression does not raise the same bias issue of retransformation at the mean using other regression estimators based on transformed dependent variables. As Parzen (2004) points out, there is no retransformation issue per se when retransforming at a quantile if a nondecreasing monotonic transformation function is used (our transformation satisfies that requirement). That is, if g is the transformation, and QY(u) is the uth quantile of Y, then “Qg(Y)(u)=g(QY(u))” (Parzen 2004, p. 655).

Model covariates were chosen to assess racial/ethnic differences in medical expenditures while adjusting for the subjects' differences in health status, demographic, regional, health insurance status, and SES characteristics. Interactions between SES and race–ethnicity variables were included to allow for the differential return on these variables by race that has been noted in previous work on the MEPS data (Cook, McGuire, and Zuvekas 2009b). Variables used in the interaction terms were centered by subtracting their mean so that main effects results are readily interpretable (Kraemer and Blasey 2004). The model is as follows:

equation image

where Yi is log(US$ medical expenditures+constant), and Racei, Healthi, Demographicsi, Regioni, Insurancei, and SESi represent groups of vectors of covariates as described in the data section.

Quantile regression operates similarly to OLS in that it estimates group differences in the dependent variable after adjustment for all variables. Race/ethnicity coefficients for each quantile regression can thus be interpreted as the difference between the minority and the non-Hispanic white group in expenditures at quantile q, after adjustment for all observed health status, demographic, regional, and SES variables. Using standard tests for linearity, adapted to quantile regression, we checked that the model as estimated was appropriate. We used Pregibon's Link Test (Pregibon 1980) to assess the fit for the regression models, and the modified Hosmer–Lemeshow test (Hosmer and Lemeshow 1989; Archer and Lemeshow 2006;) to assess systematic misfit overall in terms of predicted expenditures, as well as the model misfit for major covariates.

We examined how the minority–white differentials either expand or contract as we move from the median to the outer quartile, outer decile, and 95th percentile. By adding a strictly positive constant to expenditures, we were able to maintain the percentiles for the distribution, allowing for the ties at zero. Thus, the 50th percentile of log(US$+c) corresponds to the 50th percentile Q(50) of US$, by retransforming the result, exp(Q(50)) minus the constant. The significance of these disparities was determined by tests of the significance of the race coefficient (black or Hispanic) at each of these quantiles relative to the white contrast group. Tests for overall differences over the distribution of health care expenditures were conducted using simultaneous quantile regression with bootstrapped standard errors that corrected for design effects from both differential sampling and the clustered research design. In addition, F-tests between linear combinations of education and income main effects and interaction coefficients were compared for each of the quantile regression results to assess the differential impact of education and income across quantiles, and to assess the differences between adjacent categories of education and income within quantiles.

Sensitivity Tests

To assess the extent to which disparities results are driven by disparities among the elderly population, we separately measured expenditure disparities at deciles of the PCS-12 levels for individuals above and below 65 years.

As a further sensitivity check, we used a GLM two-part model to examine the robustness of the results to an alternative method designed to account for a large mass of zeros and extremely skewed expenditure data. Specifically, we estimated a multivariate two-part model based on GLMs (Blough, Madden, and Hornbrook 1999) using the same covariate specification as the quantile regressions to determine predicted mean total medical expenditures as a function of observable characteristics, and to extend the covariates to estimate predicted expenditures at the 75th, 90th, and 95th quantiles. This model separately analyzes the likelihood of receiving care by logistic regression by MLE and the level of expenditure conditional on care was estimated by quasi-likelihoods (McCullagh and Nelder 1989). Using diagnostics in Manning and Mullahy (2001) and Buntin and Zaslavsky (2004), we identified the optimal GLM to have a log link, and a gamma distribution to correspond to the conditional variance (power function, λ=2). As above, we assessed fit of the model using Pregibon's Link Test (Pregibon 1980) and the modified Hosmer–Lemeshow test (Archer and Lemeshow 2006).

We also conducted descriptive analyses of individuals predicted to be in the upper quantiles of expenditures in order to speculate on the underlying drivers of disparities among those with critical health care needs. We assessed differences in access to type of service (outpatient, inpatient, ER, and prescription drug use), number of visits and expenditures for these services, and type of payer (private, Medicare, Medicaid, and self/family out-of-pocket). We used the GLM two-part model predictions to identify individuals in the 15th, 18th, and 19th vigicile of predicted expenditures within their racial/ethnic group (these vigiciles are 5 percent ranges of individuals with predicted expenditures that correspond to the 75th, 90th, and 95th quantiles) and compared racial/ethnic group means using chi-square and t-test statistics.


The comparison of the racial/ethnic groups' medical expenditure distributions shows that fewer non-Hispanic whites have zero expenditures (9.8 percent) than blacks (20.6 percent) and Hispanics (32 percent), and that black and Hispanic distributions of expenditures do not catch up with the white distribution of expenditures until the very highest level of expenditures (greater than e10, or ~US$22,000) (Figure 1). More details of the distributions of medical expenditures can be seen in Table 1. Median medical care expenditures for blacks were less than half that of the median for whites (US$666 versus US$1410) but converged to be not significantly different from those of whites at the 99th percentile. The differences in proportional terms are shrinking as well. For Hispanics, the median expenditures were approximately one-fifth that of the median for whites (US$299 versus US$1410). In contrast to the experience of blacks, these disparities persist through to the 99th percentile of expenditures (US$25,902 versus US$43,730). In proportional terms, the differences decreased from Hispanics being 79 percent lower at the median to being a smaller but still statistically significant and appreciable 41 percent lower at the 99th percentile. We next calculated disparities after stratification by decile of the physical health composite score of the SF-12 (Table 2). Black–white disparities exist among all except those with the most critical need (the lowest decile of PCS-12 scores) and Hispanic–white disparities, in general, are reduced as health worsens, but remain statistically significant.

Table 2
Expenditure Disparities by Decile of SF-12 Physical Health Composite Score (PCS) (Ware, Kosinski, and Keller 1996)
Table 1
Weighted Unadjusted Health Expenditure (US$FY 2005) for Blacks, Hispanics, and Whites Age ≥18
Figure 1
Cumulative Distribution Function (CDF) of Total Medical Expenditure (Log Scale)

According to race/ethnicity indicator coefficients from quantile regressions, disparities persist throughout the distribution in fully adjusted models for Hispanics and blacks, conditional on the observed covariates that we know differ across the three racial/ethnic groups (Table 3). Comparing results from the 25th and the 75th quantiles and the 50th and 95th quantiles, we found that disparities do diminish significantly, though not completely, in the upper quantiles. The MEPS data indicate that the response to race and ethnicity is more complex than a simple model of positive expenditures with a constant proportional response or one with interactions of the covariates with indicators for race and ethnicity. The race–education and race–income terms, as well as the main effects for race/ethnicity at the outer quantiles, are significantly different (p<.001) from those at the median.

Table 3
Racial and Ethnic Differences in Total Medical Expenditure as Measured by Quantile Regression Coefficients

There was a differential effect of education and income for both blacks and Hispanics compared with whites across the distribution of expenditures, conditional on the full set of covariates (Table 4). Black–white disparities in medical expenditure exist for all education levels at the 50th and 75th percentile. Disparities persist for all except those who attended or graduated from college at the 90th and 95th percentile. Black–white disparities in medical expenditure exist for nearly all poverty level categories at the 50th, 75th, 90th, and 95th percentiles with the exception of the near poverty group in the 95th percentile. Hispanic–white disparities follow a similar pattern with disparities existing in nearly all education and income categories at all quantiles of medical expenditure. The main exceptions are that Hispanic–white disparities are small, but insignificant, for those who attended college in the 90th percentile of expenditure, and disparities completely disappear for those who attended college in the 95th percentile conditional on the covariates. Across nearly all education and income categories, disparities are reduced as analysis moved from the median out to the upper tails of expenditures, and across nearly all percentiles of expenditures, disparities are reduced as individuals move into higher educational and income categories.

Table 4
Quantile Regression Coefficients Testing Differential Response by Race × Income and Race × Education

To assess whether disparity patterns are similar for the elderly and nonelderly, we applied the analysis of racial/ethnic disparities in health care expenditures by PCS-12 decile after further stratification by age (<65 and ≥65). We found that the same patterns largely held for the nonelderly and the Hispanic elderly, but not for the black elderly (small sample sizes in some deciles limits the precision of these estimates). Statistically significant black–white disparities exist in the second healthiest decile but in no other decile.

The results of the quantile regression are robust to the use of the two-part model, which assumes a constant response across the distribution in the dependent variable and covariates (data available upon request). Racial/ethnic disparities in two-part model-predicted expenditures estimated using white coefficients and minority covariate distributions decreased (but never disappeared) in upper quantiles.

Comparing racial/ethnic groups on different types of medical care at upper quantiles of total expenditure predicted by the two-part model, we found that outpatient expenditures were lower for blacks than whites at the 50th and 75th quantile, and lower for Hispanics than whites at the 50th, 75th, 90th, and 95th quantiles (data available upon request). Compared with whites, prescription drug expenditures were lower for blacks at the 50th and 75th quantile and lower for Hispanics at the 50th, 75th, 90th, and 95th quantiles. We found no differences in ER or inpatient expenditures among blacks and whites, though blacks were significantly more likely than whites to use the ER at the 50th, 75th, and 90th quantiles. Compared with whites, inpatient service use was lower for Hispanics across quantiles, and inpatient expenditures were lower at the 95th quantile. For all quantiles of predicted expenditure, private insurance, Medicare, and out-of-pocket expenditures were higher for whites than for blacks and Hispanics, and Medicaid expenditures were higher for blacks and Hispanics than for whites.


Numerous studies have documented black–white and Hispanic–white disparities in access to medical care. This study assesses the magnitude of these disparities across the distribution of health care expenditures to determine whether the effect is uniform within these minorities compared with non-Hispanic whites or whether the disparity shrinks as one moves to the upper end of the distribution. While disparities were reduced in higher quantiles compared with the median and lower quantiles, these high expenditure minority groups were still receiving significantly less health care. These results from unadjusted analyses persisted after adjustment for health status alone, and after adjustment for demographic, socioeconomic, and health status differences among the groups. Having higher education and income diminishes these disparities across the distribution of expenditures, but for higher education categories, disparities continue to exist at the median and 75th percentiles. Even more remarkably, for higher income categories, disparities continue to exist from the median through to the 95th percentile of expenditures after covariate adjustment.

We recognize three potential limitations to the validity of these findings. First, there may be unobserved health status variables omitted from our quantile regression models. While the MEPS has a rich set of health status variables compared with other nationally representative health care datasets, there may be other variables that are especially influential among the most severely ill cases, potentially biasing our upper quantile regression estimates. Second, it is possible that non-Hispanic whites are over-treated while the two minority groups are appropriately treated. Unfortunately, we do not have the data to refute this hypothesis. A third potential problem is that measurement error may be differential by racial/ethnic group (see Medical Care, special issue, 44(11) Suppl. 3). Future work in this area is needed.

Our preliminary analyses of the underlying causes of upper quantile disparities show black–white expenditure disparities among the nonelderly but not the elderly population, suggesting that Medicare may be a leveling force. We also found blacks' lower use of office-based visits (75th and 95th quantiles) and lower spending on prescription drugs (75th quantile) to be likely drivers of the upper quantile black–white disparities. Hispanics used less outpatient, inpatient, and prescription drug care across the distribution of expenditures, suggesting initiatives to improve access to care for Hispanics with critical health care needs. The greater private insurance expenditures for whites and greater Medicaid expenditures for blacks and Hispanics across the upper quantiles leads us to speculate that the combination of higher minority Medicaid enrollment rates and private insurance companies' more generous coverage of expensive medical procedures may be driving disparities in the upper quantiles. More rigorous analysis of the underlying mechanisms of upper quantile disparities was outside of the scope of this paper and should be investigated in future work.

Our findings from the MEPS indicate that the differences of these two minority groups compared with non-Hispanic whites is much more complex than existing analyses indicate. Of particular note is that disparities in health care expenditure appear to persist even among minorities with the most critical health issues. This finding is bolstered by previous studies finding disparities in cancer treatment (Gross et al. 2008; Martinez et al. 2008;), cardiac care (e.g., Kressin et al. 2004), and major surgical procedures (Escarce and McGuire 2004; Jha et al. 2005;). To the extent that medical care is more vital to the health of the critically ill compared with healthier individuals, it may be desirable to disproportionately target resources and tracking efforts toward reducing health care disparities among sicker individuals.

Previous studies have taken a number of approaches to adjusting for patient characteristics, including adjustment for all covariates, adjustment for health status variables in accordance with the IOM definition of racial/ethnic health care disparities, and no adjustment. In this study, we implement all three approaches and find consistent results. Our preliminary estimations of expenditure disparities—without adjustment and stratified by decile of PCS-12 scores—allow differences due to SES to enter into disparity predictions and find that black–white and Hispanic–white disparities exist at high levels of need. Our main quantile regression results find similar patterns of disparities, even after adjustment for a fuller set of SES and health status covariates.

We recommend that future research on medical expenditure disparities use methods that measure disparities throughout the distribution, especially if their concern is primarily about the sickest end of the spectrum.


Joint Acknowledgment/Disclosure Statement: We would like to acknowledge funding from the National Institute of Mental Health (R03 MH 082312 and P50 MHO 73469). We would like to thank John Ayanian, Philip Cook, Bianca Frogner, Tom McGuire, and audience members at the 2008 American Society of Health Economics conference for their comments and suggestions.

Disclosures: None.

Disclaimers: The opinions expressed here are not those of the Harvard Medical School, the Cambridge Health Alliance, the University of Chicago, or the National Institute of Mental Health. Any remaining errors are those of the authors.

Supporting Information

Additional supporting information may be found in the online version of this article:

Appendix SA1: Author Matrix.

Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.


  • AHRQ. National Healthcare Disparities Report, 2007. Rockville, MD: Agency for Healthcare Research and Quality; 2007.
  • Archer KJ, Lemeshow S. Goodness-of-Fit Test for a Logistic Regression Model Fitted Using Survey Sample Data. Stata Journal. 2006;6(1):97–105.
  • Baicker K, Chandra A, Skinner JS. Geographic Variation in Health Care and the Problem of Measuring Racial Disparities. Perspectives in Biology and Medicine. 2005;48(Suppl 1):S42–53. [PubMed]
  • Baicker K, Chandra A, Skinner JS, Wennberg JE. Who You Are and Where You Live: How Race and Geography Affect the Treatment of Medicare Beneficiaries. Health Affairs (Millwood) 2004;(Suppl Web Exclusives):VAR33–44. [PubMed]
  • Blanco C, Patel SR, Liu L, Jiang H, Lewis-Fernández R, Schmidt AB, Liebowitz MR, Olfson M. National Trends in Ethnic Disparities in Mental Health Care. Medical Care. 2007;45(11):1012–9. [PubMed]
  • Blough DK, Madden CW, Hornbrook MC. Modeling Risk Using Generalized Linear Models. Journal of Health Economics. 1999;18(2):153–71. [PubMed]
  • Buchinsky M. Changes in the U.S. Wage Structure 1963–1987: Application of Quantile Regression. Econometrica. 1994;62(2):405–58.
  • Buchinsky M. Recent Advances in Quantile Regression Models: A Practical Guideline for Empirical Research. Journal of Human Resources. 1998;33(1):88–126.
  • Buntin MB, Zaslavsky AM. Too Much Ado about Two-Part Models and Transformation? Comparing Methods of Modeling Medicare Expenditures. Journal of Health Economics. 2004;23(3):525–42. [PubMed]
  • Collins KS, Hall A, Neuhaus C. U.S. Minority Health: A Chartbook. New York: The Commonwealth Fund; 1999. 161pp.
  • Cook BL, McGuire TG, Zaslavsky AM, Meara E. Adjusting for Health Status in Non-Linear Models of Health Care Disparities. Health Services Outcomes and Research Methodology. 2009a;9(1):1–21. [PMC free article] [PubMed]
  • Cook BL, McGuire TG, Zuvekas SH. Measuring Trends in Racial/Ethnic Health Care Disparities. Medical Care Research and Review. 2009b;66(1):23–48. [PMC free article] [PubMed]
  • Cook BL, Miranda J, McGuire TG. Measuring Trends in Mental Health Care Disparities, 2000–2004. Psychiatric Services. 2007;58(12):1533–40. [PubMed]
  • Duan N. Smearing Estimate: A Nonparametric Retransformation Method. Journal of the American Statistical Association. 1983;78(383):605–10.
  • Escarce JJ, Epstein KR, Colby DC, Schwartz JS. Racial Differences in the Elderly's Use of Medical Procedures and Diagnostic Tests. American Journal of Public Health. 1993;83(7):948–54. [PubMed]
  • Escarce JJ, Kapur K. Racial and Ethnic Differences in Public and Private Medical Care Expenditures among Aged Medicare Beneficiaries. Milbank Quarterly. 2003;81(2):249–75. [PubMed]
  • Escarce JJ, McGuire TG. Changes in Racial Differences in Use of Medical Procedures and Diagnostic Tests among Elderly Persons: 1986–1997. American Journal of Public Health. 2004;94(10):1795–9. [PubMed]
  • Fennell ML. Racial Disparities in Care: Looking Beyond the Clinical Encounter. Health Services Research. 2005;40(6, part 1):1713–21. [PMC free article] [PubMed]
  • Ford E, Newman J, Deosaransingh K. Racial and Ethnic Differences in the Use of Cardiovascular Procedures: Findings from the California Cooperative Cardiovascular Project. American Journal of Public Health. 2000;90(7):1128–34. [PubMed]
  • Gross CP, Smith BD, Wolf E, Andersen M. Racial Disparities in Cancer Therapy: Did the Gap Narrow between 1992 and 2002? Cancer. 2008;112(4):900–8. [PMC free article] [PubMed]
  • Hosmer DW, Lemeshow S. Applied Logistic Regression Analysis. New York: John Wiley & Sons; 1989.
  • Ibrahim SA, Whittle J, Bean-Mayberry B, Kelley ME, Good C, Conigliaro J. Racial/Ethnic Variations in Physician Recommendations for Cardiac Revascularization. American Journal of Public Health. 2003;93(10):1689–93. [PubMed]
  • IOM. Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care. Washington, DC: National Academies Press; 2002. [PMC free article] [PubMed]
  • Jha AK, Fisher ES, Li Z, Orav EJ, Epstein AM. Racial Trends in the Use of Major Procedures among the Elderly. New England Journal of Medicine. 2005;353(7):683–91. [PubMed]
  • Koenker R. Quantile Regression. Cambridge University Press; 2005. Econometric Society Monograph Series. Cambridge, England.
  • Koenker R, Hallock KF. Quantile Regression. Journal of Economic Perspectives. 2001;15:143–56.
  • Kraemer HC, Blasey CM. Centring in Regression Analyses: A Strategy to Prevent Errors in Statistical Inference. International Journal of Methods in Psychiatric Research. 2004;13(3):141–51. [PubMed]
  • Kressin NR, Chang BH, Whittle J, Peterson ED, Clark JA, Rosen AK, Orner M, Collins TC, Alley LG, Petersen LA. Racial Differences in Cardiac Catheterization as a Function of Patients' Beliefs. American Journal of Public Health. 2004;94(12):2091–7. [PubMed]
  • Kressin NR, Petersen LA. Racial Differences in the Use of Invasive Cardiovascular Procedures: Review of the Literature and Prescription for Future Research. Annals of Internal Medicine. 2001;135(5):352–66. [PubMed]
  • Manning W, Mullahy J. Estimating Log Models: To Transform or Not to Transform? Journal of Health Economics. 2001;20(4):461–94. [PubMed]
  • Manning WG, Blumberg L, Moulton LH. The Demand for Alcohol: The Differential Response to Price. Journal of Health Economics. 1995;14(2):123–48. [PubMed]
  • Martinez SR, Robbins AS, Meyers FJ, Bold RJ, Khatri VP, Goodnight JE., Jr. Racial and Ethnic Differences in Treatment and Survival among Adults with Primary Extremity Soft-Tissue Sarcoma. Cancer. 2008;112(5):1162–8. [PubMed]
  • Mayberry RM, Mili F, Ofili E. Racial and Ethnic Differences in Access to Medical Care. Medical Care Research and Review. 2000;57(Suppl 1):108–45. [PubMed]
  • McCullagh P, Nelder JA. Generalized Linear Models. London: Chapman & Hall; 1989.
  • McGuire TG, Alegria M, Cook BL, Wells KB, Zaslavsky AM. Implementing the Institute of Medicine Definition of Disparities: An Application to Mental Health Care. Health Services Research. 2006;41(5):1979–2005. [PMC free article] [PubMed]
  • Mechanic D, Tanner J. Vulnerable People, Groups, and Populations: Societal View. Health Affairs. 2007;26(5):1220–30. [PubMed]
  • Parzen E. Quantile Probability and Statistical Data Modeling. Statistical Science. 2004;19(4):652–62.
  • Petersen LA, Wright SM, Peterson ED, Daley J. Impact of Race on Cardiac Care and Outcomes in Veterans with Acute Myocardial Infarction. Medical Care. 2002;40(Suppl 1):I86–96. [PubMed]
  • Peterson ED, Shaw LK, DeLong ER, Pryor DB, Califf RM, Mark DB. Racial Variation in the Use of Coronary-Revascularization Procedures. Are the Differences Real? Do They Matter? New England Journal of Medicine. 1997;336(7):480–6. [PubMed]
  • Pregibon D. Goodness of Link Tests for Generalized Linear Models. Applied Statistics. 1980;29:15–24.
  • Skinner J, Weinstein JN, Sporer SM, Wennberg JE. Racial, Ethnic, and Geographic Disparities in Rates of Knee Arthroplasty among Medicare Patients. New England Journal of Medicine. 2003;349(14):1350–9. [PubMed]
  • Ware JE, Kosinski M, Keller SD. A 12-Item Short-Form Health Survey: Construction of Scales and Preliminary Tests of Reliability and Validity. Medical Care. 1996;34(3):220–33. [PubMed]

Articles from Health Services Research are provided here courtesy of Health Research & Educational Trust