Notes : OLS estimates of equations (2) and (3) . Panel (A) presents outcomes on utilization. Utilization data are from the harmonized version of the National Health Interview Survey (NHIS) available from IPUMS and merged with restricted identifiers for use in the Restricted Data Center (RDC) and cover the period 1969–1977. Post is an indicator variable equal to 1 in the years following 1972. The unit of observation is the individual, and the sample includes non-veteran black and white men and women ages 45–74. In column (1) the outcome is the number of outpatient physician interactions in the past 12 months. In column (2) the outcome is an indicator variable for any outpatient physician visit in the last 12 months. In columns (3) and (4) the outcome variables are any hospital admission and the number of nights in a hospital, respectively. Panel (B) presents outcomes on mortality. Mortality data are from the compressed mortality files from the CDC and cover the period 1968–1987. The unit of observation is a demographic group within a state economic area (SEA) and the sample includes black and white men and women ages 45–74 who died in the United States. Rates are constructed biennially, and post is an indicator equal to 1 in the years following 1972/1973. In columns (5) and (6) the outcomes are the log and level of age-adjusted mortality from all causes, respectively. In columns (7) and (8) the outcomes are the log and level of age-adjusted chronic mortality, respectively. In addition to the listed fixed effects, utilization regressions control for age, education, marital status, urbanization, whether the respondent has a telephone and income. Mortality regressions include controls for the log of total health (e.g. Medicaid and Medicare) expenditures, the log of Social Security expenditures, the density of hospitals, hospital beds and physicians and the presence of community health centers. Regressions using NHIS data are weighted using provided survey weights. Standard errors are clustered at the state level or SEA level.
Estimated coefficients in column (1) of Table I indicate that black men experienced sharp declines in the probability of visiting a doctor in the years following 1972 as a function of their proximity to the Tuskegee Study’s location. These estimates indicate that a standard deviation increase in proximity to the study’s home county was associated with reduced outpatient interactions of 0.9 visits per year — approximately 22% of the pre-disclosure black male sample mean. From column (2), a one standard deviation increase in proximity to Macon County, Alabama, was associated with a 2.5 percentage point decline in the probability of having an outpatient doctor’s visit, approximately 4% of the sample mean.
For hospitalization outcomes, estimates for β 1 in column (3) of Table I indicate a reduction in the probability of hospital admission for black men after 1972 of 1 percentage point per standard deviation of geographic distance. These results, although not statistically significant, indicate a post-Tuskegee Study disclosure reduction in acute medical care of 7% of the pre-disclosure mean. Although hospital admission rates went down, column (4) of Table I indicates black men appear to have had more advanced illness on presentation as reflected in longer hospital stays: black men experienced an increase in duration of stay of 0.5 nights per standard deviation. These estimates are large, representing 22% of the pre-1972 sample mean.
In Appendix Table A.2 , we estimate this baseline equation on another healthcare utilization outcome less likely to have been affected by the Tuskegee disclosure: dental visits over the last 12 months. This outcome is available for a slightly truncated period of time (1969–1975), and shows black men in closer proximity to Macon County, Alabama, are slightly more likely to visit the dentist after 1972, although the estimated coefficient is not statistically significant. This result provides suggestive evidence that racial gaps in other forms of care stabilized after 1972 and that relative avoidance was specific to institutions most reminiscent of the Tuskegee Study.
In columns (5) and (7) of Table I , we estimate that a one-standard deviation increase in proximity to Tuskegee was associated with a post-1972 increase in both all-cause and chronic-cause age-adjusted mortality for black men of 3.7 and 4.5 log points, respectively. When we estimate the model in levels, we find a one standard deviation increase is associated with 1.1 more deaths from any cause per 1000 population (column (6)) and 0.8 more deaths from chronic causes per 1000 population (column (8)). Based on our estimates, the elasticity of all-cause age-adjusted mortality with respect to physician visits ranges from 0.14 to 0.18; in other words, a 1% decrease in physician visits would increase mortality by approximately 0.14 – 0.18%. 31 We view these estimates as an upper bound since men might have not only stopped interacting with physicians, they might have also delayed going to the hospital or heeding public health messaging around smoking, for instance (see Appendix Table A.2 column (2)). However, this value is consistent with similar estimates from Medicare and community health center expansion. 32 Notably, we do not observe an effect of the Tuskegee disclosure on mortality rates from acute causes of death associated with violence, accidents, or other external causes (see Appendix Table A.6 ). In unreported results, the largest impacts appear to come from deaths due to cardiac conditions, diabetes, and respiratory diseases.
In almost all cases, estimates of β 2 and β 3 are insignificantly different from zero, implying no systematic post-1972 change in the gender gap in mortality for whites or the racial gap in mortality for women as a function of proximity to Macon County, Alabama. The estimated value of β 2 is positive and significant for the log mortality outcomes, but the significance of the coefficient belies two important observations. First, there is no change in the trajectory of the β 2 coefficient in 1972 (see Figure IV , Panel E). Rather, there is a low, positive rate of growth in the geographic mortality gradient for white men compared to white women with a slope an order of magnitude smaller than β 1 . Second, unlike the β 1 effect, the β 2 effect is not centered on Macon County, Alabama, as demonstrated in Appendix Figure A.5 .
In Table II , we explore heterogeneous effects by splitting the sample in various ways and reporting estimates of β 1 , β 2 , and β 3 for these subsamples. First, we explore the roles of income and education by dividing the sample at the median of black male household income (columns (1) and (2)) and at the median of black male education (columns (3) and (4)). Because the entire sample is divided at the black male median, above- and below-median samples are not equally sized. In both cases, we conjecture that black men lower on the socioeconomic ladder would respond more strongly to the Tuskegee Study disclosure if the channel for utilization effects is empathy or salience. Indeed, although the estimates of β 1 are negative and significant for both income groups, the point estimates for the effect of the Tuskegee disclosure on the health seeking behavior of poorer, less educated black men are significantly distinguishable from higher income, better educated black men at the 1% level. Black men in the lowest 50 th percentile of the income distribution reduce their utilization by approximately 3 times more than black men in the top half of the income distribution. At the same time, black men with below-median education reduced their utilization by 1.9 visits per standard deviation of proximity to Macon County, Alabama, while the upper 50 th percentile of the education distribution exhibits a point estimate of approximately zero.
Heterogeneous Effects, Utilization
(1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | |
---|---|---|---|---|---|---|---|---|
Dependent Variable: Number Outpatient Visits | ||||||||
By Income Level | By Educational Status | By Prevalence Black Doctors | By Marital Status | |||||
| | | | | | |||
P *post *black *male | −0.546 (0.548) | −1.725 (0.705) | −0.061 (0.409) | −2.801 (0.839) | −1.359 (0.373) | −2.052 (1.460) | −1.398 (0.326) | −1.665 (1.061) |
P *post *male | 0.060 (0.084) | −0.049 (0.184) | −0.013 (0.092) | −0.130 (0.294) | 0.042 (0.082) | −0.110 (0.238) | 0.029 (0.106) | −0.188 (0.240) |
P *post *black | −0.150 (0.221) | −0.211 (0.108) | −0.465 (0.215) | 0.468 (0.617) | −0.019 (0.115) | −0.566 (0.726) | 0.419 (0.204) | −0.511 (0.191) |
Fixed Effects | State-Year, Race-Gender-Year, Race-Gender-State | |||||||
Observations | 143,554 | 77,400 | 178,756 | 42,198 | 176,032 | 44,922 | 160,335 | 60,619 |
No. Clusters | 49 | 49 | 49 | 49 | 25 | 24 | 49 | 49 |
Adj R-squared | 0.013 | 0.030 | 0.014 | 0.036 | 0.016 | 0.017 | 0.014 | 0.030 |
Notes : OLS estimates of equation (2) assessing heterogeneous effects by income level, education level, black doctor prevalence, and marital status. Specifically, in the first two columns we divide the sample by median black male income. In the following two columns we divide the sample by median black male education. In the next two columns, we calculate black physicians as a percentage of all physicians using occupational data from the 1970 Census and bifurcate states as above or below median on this dimension. In the last two columns, we divide the sample by whether the survey respondent was married. Utilization data are from the harmonized version of the National Health Interview Survey (NHIS) available from IPUMS and merged with restricted identifiers for use in the Restricted Data Center (RDC) and cover the period 1969–1977. Post is an indicator variable equal to 1 in the years following 1972. The unit of observation is the individual, and the sample includes non-veteran black and white men and women ages 45–74. The outcome variable across all panels and columns is the number of physician interactions in the last 12 months. In addition to the listed fixed effects, individual-level controls in every specification for utilization include indicator variables for educational status, income, age, marital status, telephone ownership, and rural/urban status. Regressions are weighted using provided survey weights. Standard errors are clustered at the state level.
Next, we examine the moderating effect of black physicians on our baseline results in columns (5) and (6) of Table III . We hypothesize that the availability of a black physician would have reduced the rate at which black men downgraded their expectation of encountering a "good" doctor. For these results, we utilized data on the number of black and white physicians in each U.S. state from the 1970 U.S. Census of Population ( U.S. Census Bureau 1970 ). Importantly, these counts are measured before, and are therefore not endogenous to, the 1972 disclosure. When we split the sample by places above and below the median number of black doctors (as a percentage of all doctors), we find suggestive, but weak, evidence for a moderating effect of black doctors. The coefficient for locations above the median is smaller than that for locations below this median, although these differences are not statistically significant. 33
Alternative Measures of Proximity, Robustness Checks
(1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | |
---|---|---|---|---|---|---|---|---|
Dependent variable: | ||||||||
| ||||||||
Migrant Treatment | South Only | Kids (Placebo) | Migrant Treatment | South Only | Kids (Placebo) | |||
P *post *black *male | −1.794 (1.027) | −0.152 (0.151) | 0.066 (0.033) | 0.171 (0.548) | ||||
P *post *male | 0.417 (0.423) | 0.016 (0.030) | 0.003 (0.012) | −0.019 (0.010) | ||||
P *post *black | 0.794 (0.434) | 0.148 (0.112) | −0.040 (0.027) | −0.386 (0.125) | ||||
Migrant *post *black *male | −10.18 (3.307) | −8.356 (3.351) | 0.220 (0.072) | 0.140 (0.069) | ||||
Migrant *post *male | 0.838 (1.306) | 0.099 (1.715) | 0.016 (0.018) | 0.015 (0.023) | ||||
Migrant *post *black | 2.775 (1.755) | 3.071 (1.981) | −0.090 (0.069) | −0.092 (0.048) | ||||
Fixed Effects | State-Year, Race-Gender-Year, Race-Gender-State | SEA-Year, Race-Gender-Year, Race-Gender-SEA | ||||||
Observations | 216,984 | 65,495 | 69,465 | 299,688 | 17,103 | 6,973 | 7,413 | 18,600 |
No. Clusters | 48 | 16 | 17 | 49 | 451 | 175 | 186 | 465 |
Adj R-squared | 0.017 | 0.016 | 0.016 | 0.044 | 0.799 | 0.920 | 0.923 | −0.027 |
Notes : OLS estimates of equations (2) and (3) . Panel (A) presents outcomes for outpatient visits. Utilization data are from the harmonized version of the National Health Interview Survey (NHIS) available from IPUMS and merged with restricted identifiers for use in the Restricted Data Center (RDC) and cover the period 1969–1977. Post is an indicator variable equal to 1 in the years following 1972. In columns (1) through (4) the outcome is the number of outpatient physician interactions in the past 12 months. Columns (1) and (2) use an alternative measure of proximity – the fraction of black migrants from Alabama – instead of geographic proximity to Tuskegee. Alabama is excluded from these regressions. Column (2) is identical to column (1), but restricts the sample to the South. Columns (3) and (4) both use geographic proximity to Tuskegee as an instrument (as in Table I ). Column 3 restricts the sample to the South, and column (4) uses the placebo outcome of children’s utilization. Panel (B) presents outcomes for mortality. Mortality data are from the compressed mortality files from the CDC and cover the period 1968 to 1987. The unit of observation is a demographic group within a state economic area (SEA) and the sample includes black and white men and women ages 45–74 who died in the United States. Rates are constructed biennially, and post is an indicator equal to 1 in the years following 1972/1973. In columns (5) and (6) the outcome is log age-adjusted chronic mortality, and fraction of black migrants from Alabama is used as the measure of proximity. Column (6) repeats the specification in (5), restricting the sample to the South. Column (7) restricts the sample to the South and uses geographic distance to Tuskegee. In column (8), the outcome variable is the level of age-adjusted child mortality. Regressions using NHIS data are weighted using provided survey weights. Standard errors are clustered at the state level or SEA level.
Last, we divide the sample by marital status. In the absence of individual birth histories in the NHIS, we use marital status as a proxy for whether or not a woman had given birth. By 1960, 97% of all births were in hospitals ( Feldhusen 2000 ), and unmarried women had fewer children compared to married women. 34 Thus, if marriage is a proxy for more childbirth-related experience with the healthcare sector, β 2 for unmarried black women should be more similar to β 1 , the coefficient for black men. These results are tested in columns (7) and (8) of Table II . In estimates for β 1 , we find that unmarried men have a more negative response to the news of the Tuskegee Study than their married counterparts, though the difference is relatively small and not statistically significant. But unmarried black women, represented by estimated coefficients on P s · I t post · I r black in column (8) of Table II , respond differently from married black women in column (7), both in sign and magnitude. Mirroring the behavior of black men, unmarried black women reduce health provider interactions in proportion to their proximity to Macon County, Alabama, whereas married black women continue to exhibit convergence in the South. For both married and unmarried women, the size of the coefficient is about a third of that for black men, albeit with opposing signs for married women.
We perform several tests to bolster a causal interpretation for our results. First, we test the robustness of our estimates by incorporating migration networks as an alternative measure of proximity and demonstrating similarly signed and statistically significant estimates. Second, we use placebo locations to show that the main results are specific to gradients of proximity to Macon County, Alabama. Third, we use placebo populations to demonstrate that the main results for both utilization measures and mortality rates are not observed when we estimate the baseline equations on younger population samples. We also show that our results are robust to a South-only sample; to dropping all control variables; to limiting to non-South observations; and to parametric estimating equations that allow for trends in unemployment and incarceration in the mortality specifications. We discuss the findings of the main robustness checks below, and the remainder are described in the Appendix .
The reduction in healthcare utilization and uptick in mortality for black men as a function of geographic proximity to Macon County, Alabama, is also apparent using an alternative measure of proximity. Table III , Panel A reports these checks for the utilization results regarding the number of physician interactions and Panel B provides analogous results for log age-adjusted chronic mortality. In both cases, we replace the baseline proximity measure, geographic proximity to Macon County, Alabama, with the percentage of 1935–1940 black migrants to a particular state or SEA who originated in Alabama. (All Alabama SEAs and the state of Alabama are excluded from this analysis.) We observe statistically significant differences in the post-1972 utilization of primary healthcare and in the post-1972 mortality rates for black men as a function of this variable as well. Specifically, we estimate that a 10% increase in the share of black migrants from Alabama reduces utilization by 1 interaction per year and increases mortality by approximately 2%.
To test whether the geographic gradients documented in our baseline interacted DDD results are specific to distance from Macon County, Alabama, we run placebo regressions, replacing the baseline proximity measure (proximity to Macon County, Alabama) with proximity to the geographic centroid of every other state or SEA and re-estimating the model. These regressions serve as placebo tests, evaluating whether we find the same (or stronger) utilization effects as a function of the gradient to other locations in the U.S. Figure V presents the distribution, in histogram form, of the estimated values of β 1 in each of these tests when the outcome is the intensive margin of primary care utilization (Panel A) and log age-adjusted chronic mortality (Panel B). The vertical line indicates the estimated coefficient from Table I when the proximity measure is the baseline proximity to Macon County, Alabama. In both cases, the value of the Macon County estimate is greater (in absolute value) than 96% of placebo estimates; other outcomes exhibit similar patterns.
Notes : Frequency of the “false” β 1 coefficient estimated using distance from every other state (exclusive of Alabama) or SEA (exclusive of the one containing Macon County) in the sample and estimating equation (2) in Panel (A) or equation (3) in Panel (B). The vertical line denotes β 1 from baseline estimates using the true treatment distance (to Macon County, Alabama) as reported in Table I .
For the mortality results, we can also display heat maps of the coefficient estimates of β 1 as well as β 2 and β 3․ . These results, contained in Appendix Figure A.5 , Panel B, demonstrate that the handful of locations where the estimates for β 1 are greater than those using proximity to Macon County, Alabama, are also close geographically and thus highly correlated with the treatment. The coefficients for β 2 and β 3․ using these alternative proximity measures exhibit a far smaller range of estimate values. For β 2 , these values are largest in the North Central states, while β 3․ values are actually smallest in the U.S. South.
To evaluate whether the paper’s measured impacts simply reflect a general southern effect, columns (2), (3), (6), and (7) of Table III restrict the analysis sample to southern residents only. This reduces the scope for identification threats to things correlated with geographic proximity to Macon County, Alabama, in particular, and not to the South in general. The estimated value of β 1 is relatively unchanged in these specifications compared to our baseline estimates, particularly when using geographic proximity as the treatment variable. In Appendix Table A.3 , we include additional robustness checks for the utilization and mortality outcomes, including dropping the South entirely and adding interactions between region fixed effects and post, black, and male indicators. Results are similar to those presented in Tables I and III .
We next limit the sample to children under 14 years of age for both mortality and utilization results. The post-1972 difference in outpatient care for black and white male children exhibits no geographic gradient, nor does the difference in utilization for black male and black female children as evidenced by coefficients statistically indistinguishable from zero in column (4) of Table III . For mortality outcomes, there is no differential change in the mortality rate of black male children after 1972 as a function of proximity to Macon County, Alabama, although male children and black children, in general, experienced lower death rates along this geographic gradient. This result is based on a mortality outcome in levels (instead of logs) due to the large number of zeroes, but caution should still be used in interpreting these results given the small death counts compared to older age groups. The non-result for black male children is consistent with both a lower salience of the study’s abuses for younger children and with results showing that black married women were seemingly unaffected by the study’s disclosure. Although the perceptions of their fathers could feasibly have affected the demand for children’s healthcare utilization, the non-result here may indicate that decision-making for children at this time was mostly driven by maternal preferences. Given these results, the scope of identification threats is narrowed substantially to those factors correlated with the primary care utilization and mortality rates of older black men centered on Macon County, Alabama, after 1972 which did not affect child mortality or children’s healthcare utilization.
Our measure of how much trust one has in their doctor comes from the GSS, a repeated cross section extending from 1972 to the present. The earliest year questions were asked about doctors was 1998, when several questions were included. In particular, participants were asked about whether “doctor’s judgment trusted" and whether "doctors deny me the treatment needed". As a placebo outcome, we also examine respondents’ views on whether people, in general, can be trusted. 35 Although we do not have data for both before and after the disclosure, we can test whether individuals living closer to Macon County, Alabama, exhibit greater post-disclosure mistrust. For this analysis, we ask whether there is a racial (gender) gap in medical trust that varies across genders (races) as a function of proximity to Macon County, Alabama. The estimating equation for survey responses for individual i residing in state s of race r and gender g is given by:
where P s is proximity to Macon County, τ represents a current state of residence fixed effect, and θ denotes race-by-gender fixed effects. State fixed effects ensure that β coefficients capture the geographic gradient in mistrust for each demographic group net of state-specific attitudes common across all groups. The sample includes individuals at least 10 years of age at the time of the disclosure. 36 X contains age, education, and marital status fixed effects. Standard errors are clustered at the level of treatment (state of residence). The results, contained in Table IV , demonstrate that black men exhibit a strong, statistically significant geographic gradient in mistrust of doctors and similar (though not statistically significant) concern regarding treatment denial. A one standard deviation increase in proximity to Macon County, Alabama, is associated with a 14 percentage point increase in medical mistrust and a similar increase in treatment denial suspicion among black men as compared to white men, net of such racial gaps for women. Note that β 2 (the geographic gradient in mistrust for white men as compared to white women) and β 3 (the gradient for black women compared to white women) are often oppositely signed and not significant. It is not the case that black men closer to Tuskegee are simply less trusting; β 1 in column (3) where the outcome is general mistrust is opposite signed and not statistically significant.
Effect of Tuskegee on Beliefs About Medical Care
(1) | (2) | (3) | |
---|---|---|---|
| | ||
P *black *male | 0.176 (0.071) | 0.157 (0.127) | −0.073 (0.197) |
P *male | −0.016 (0.030) | −0.002 (0.039) | −0.005 (0.048) |
P *black | −0.051 (0.047) | −0.024 (0.115) | −0.052 (0.055) |
Fixed Effects | State , Race*Gender | ||
Observations | 801 | 801 | 801 |
Adj R-squared | 0.024 | 0.054 | 0.103 |
No. Clusters | 36 | 36 | 36 |
Notes : OLS estimates of equation (4) . The data are from the General Social Survey for the year 1998. The sample contains black and white males and females at least 10 years old in 1972. The outcome variable for column (1) is whether the respondent disagrees with the statement that doctors can be trusted. The outcome variable in column (2) is whether the respondent believes they will be denied needed treatment by the medical profession. The outcome variable in column (3) is general mistrust (whether people, in general, can be trusted). In addition to the controls listed above, every specification includes indicator variables for age categories, marital status, state of current residence fixed effects, and indicators for being black, male, and the interaction. Standard errors are clustered at the state of residence level.
Finally, we have posited the impact of Tuskegee would be more poignant for individuals who had limited experience with the healthcare sector prior to the disclosure. To isolate the role of experience as a mitigating factor on belief formation, we examine the post-1972 differences in healthcare utilization between men who have served in the military versus those who have not, again as a function of proximity to Macon County, Alabama. To do so, we drop women from the utilization sample, add back veteran men, and replace I g male in equation (2) with an indicator variable for whether an individual was a non-veteran, i.e., never drafted or entered into the military.
Estimated coefficients from this modified specification, contained in Table V , indicate a marked geographic gradient in the veteran/non-veteran gap in primary care utilization in the years following the Tuskegee disclosure. This finding is restricted to black men (as evidenced by the null or oppositely signed β 2 coefficient). Although black men’s utilization declined overall (see β 3 in columns (1) and (2)), this effect was far more pronounced for non-veterans. For example, veteran black males reduced the number of outpatient visits per year by 0.59 per thousand kilometer proximity to Macon County, Alabama, while nonveteran black males reduced their utilization by about one additional visit per year. 37 There are several possible explanations for this observation, including that veteran black men may have expected doctors to treat them as "veteran patients" rather than as "black male patients" and were therefore less worried about facing discrimination. Nevertheless, the results for veterans are consistent with predictions from the multi-period model wherein past experience with the medical profession dampens the response to the news of Tuskegee.
All Male Sample, Veterans vs. Non-Veterans
(1) | (2) | (3) | (4) | |
---|---|---|---|---|
| | | | |
P *post *black *nonvet | −1.161 (0.347) | −0.040 (0.009) | −0.027 (0.007) | −0.417 (0.312) |
P *post *nonvet | 0.146 (0.072) | 0.004 (0.005) | 0.002 (0.003) | 0.051 (0.056) |
P *post *black | −0.592 (0.179) | −0.024 (0.011) | −0.012 (0.010) | 0.280 (0.247) |
Fixed Effects | State-Year, Race-Non-Veteran-Year, Race-Non-Veteran-State | |||
Observations | 135,635 | 135,635 | 135,635 | 135,635 |
No. Clusters | 49 | 49 | 49 | 49 |
R-squared | 0.020 | 0.023 | 0.016 | 0.014 |
Notes : OLS estimates of equation (2) testing differences between “experienced” (veteran) and “less experienced” (non-veteran) males in response to the disclosure of the Tuskegee study. Utilization data are from the harmonized version of the National Health Interview Survey (NHIS) available from IPUMS and merged with restricted identifiers for use in the Restricted Data Center (RDC) and cover the period 1969–1977. Post is an indicator variable equal to 1 in the years following 1972. The unit of observation is the individual, and the sample includes non-veteran and veteran black and white men ages 45–74. The outcome varies across columns and is given by the column heading. In addition to the listed fixed effects and controls in the table, individual-level controls in every specification for utilization include indicator variables for educational status, income, age, marital status, telephone ownership, and rural/urban status. Standard errors are clustered at the state level.
The paper’s baseline results for the mortality of older adults, aged 45–74, can be converted into predicted reductions in life expectancy. To do so, we use levels of all-cause mortality as the outcome in equation (3) . Predictions based on the full complement of terms, including the interaction of P a · I t post · I r black · I g male multiplied by β 1 ^ , reflect mortality with the Tuskegee Study revelation. Predictions based on equation (3) , but without this product, reflect counterfactual mortality in the absence of the revelation. The average value of these predictions are weighted so that they reflect the mortality patterns of black men given their population distribution circa 1970. These predicted mortality rates allow us to construct abridged, current life tables, in the presence and absence of the Tuskegee revelation, as described in the London Health Observatory Technical Supplement (2010) .
For black men in our sample, the observed life expectancy, conditional on reaching age 45, is 23.7 years. Counterfactual life expectancy in the absence of the calculated treatment effect is 25.2, an increase of 1.5 years. This estimated decrease in the life expectancy of black men attributable to the Tuskegee revelation represents approximately 35% of the racial gap in male life expectancy in 1980 and 25% of the gender gap in black life expectancy at the same juncture ( National Center for Health Statistics 1985 ). As a benchmark for these results, Black et al. (2015) find that migration to the North in the first waves of the Great Migration (1916–1932) resulted in decreased life expectancy at age 65 of at least 1.5 years. Because their most robust analysis is based on Medicare data, these authors do not provide a similar estimate for life expectancy at age 45. Another useful benchmark is the hypothetical removal of smoking, which has been projected to increase life expectancy of an 18-year old by 1.7 years ( Stewart, Cutler, and Rosen 2009 ) .
The Tuskegee Study was one of the most egregious examples of medical exploitation in U.S. history. Our estimates indicate that the years immediately following the study’s disclosure brought significantly lower utilization of both outpatient and inpatient medical services by older black men in closer geographic proximity to the study’s subjects. The effects are particularly heightened among less-educated and lower income men, a socioeconomic profile shared by the men targeted by PHS investigators. Moreover, the reductions in healthcare utilization we document paralleled a significant increase in the probability that older black men died before the age of 75. The data indicate no corresponding effects for younger black males or for white males or black women. Our results are robust to accounting for a wide range of policies, economic forces, and individual characteristics thought to shape health behaviors. These findings underscore the importance of trust for economic relationships involving imperfect information, including in the provision of medical care.
Although our study has focused on how Tuskegee generated mistrust and shaped demand for healthcare services by blacks, the Tuskegee example also revealed racial inequities inherent in the provision of healthcare. A modern literature indicates these inequities persist. For example, Hoffman et al. (2016) document false beliefs among medical students and residents regarding race-based biological differences in pain tolerance that resulted in racial differences in treatment. As long as biased beliefs, policies, and practices are still prevalent in the U.S. healthcare system, mistrust is a rational response that may continue to contribute to health disparities.
* We thank the editor, Lawrence Katz, and four anonymous reviewers for constructive comments that improved the paper. For detailed feedback at an early stage of our work, we thank Nathan Nunn, Arun Chandrasekhar, Martha Bailey, Pascaline Dupas and William Collins. We are also grateful to John Parman, Achyuta Adhvaryu, Rebecca Diamond, Claudia Goldin, Melanie Morten, Mark Duggan, Mark Cullen, Melissa Dell, Nancy Qian, Ran Abramitzky, Rema Hanna, Grant Miller and seminar participants at NBER DAE, NBER Cohort Studies, University of Tennessee, Vanderbilt Health Policy, Carnegie Mellon and University of Pittsburgh Joint Seminar, University of Copenhagen, University of Pennsylvania Health Policy, ASSA 2016, PACDEV 2016, Berkeley Population Center, University of Chicago Harris School of Public Policy, Stanford Health Policy, University of California-Davis, University of Maryland Population Center, Stanford Social Science and History Workshop, University of South Carolina, Florida State University, University of Richmond, Highland Hospital of Oakland, Dartmouth College, Harvard Medical School, University of Michigan, University of California Berkeley, Simon Fraser University and CIREQ Montreal for constructive comments. We thank the CDC for providing access and the administrators at the Atlanta and Stanford Census Research Data Centers for their help in navigating the restricted data. We thank Michael Sinkinson, Martha Bailey, Andrew Goodman-Bacon and Walker Hanlon for sharing data and methods. Mario Javier Carrillo, Anlu Xing and Afia Khan provided excellent research assistance.
1 For a comprehensive review of racial inequalities in U.S. medical care, see Institute of Medicine (2003) .
2 In our model of medical mistrust, we consider Bayesian updating as the benchmark, although behavioral models could deliver similar results (e.g. Becker and Rubinstein 2011 ).
3 Syphilis can also be transmitted mother to child and cause severe congenital problems, including stillbirth.
4 This lack of minority representation in clinical trials may have important spillover effects, including on the speed and direction of innovation for ailments that heavily afflict their communities. See Hamilton et al. (2016) for a structural estimation of these losses.
5 Vann Newkirk II (2016) highlights the cultural role of Tuskegee, even in the absence of specific details, saying: "Like me, several other black men that I interviewed throughout the rural South were either inculcated from birth or from experience living in 1972 with the idea that the American health-care system is not for them. Young boys and old men felt it alike, and even if the Tuskegee Study was not known by name, it was a definite part of a vivid shared cultural memory. References to injection of ‘bad blood’, government research, or conspiracies about HIV were clearly influenced by details of Tuskegee, even if the details weren’t always quite right."
6 For example, 75% of black men believed that "African Americans are more likely to be treated poorly in health research studies" compared to 58% of black women and 37% of whites. Sample sizes here are quite small, however, so caution is warranted.
7 Additional beliefs were that "HIV was deliberately created in a laboratory in order to infect black people, that AZT is a plot to poison them, that condom distribution campaigns are a scheme to reduce the number of black babies, and that needle distribution programs are a plot to encourage drug abuse," ( Quinn 1997 ). Similar Tuskegee-attributed mistrust related to the origins of the HIV virus emerged in a 1996 ABC News 20/20 forum and in a 1997 Atlanta-based forum audience.
8 The life expectancy gap between black and white men can be substantially reduced by controlling for stage at diagnosis. See American Cancer Society (2008) and Silber et al. (2014) . For evidence on these racial gaps, see Boulware et al. (2003) ; Wiltshire, Person, and Allison (2011) ; and Hood et al. (2012) .
9 In a 2016 study of low adherence to antiretroviral treatment among HIV-positive black men, researchers documented that 63% of the study’s subjects held a "conspiracy belief", for example that the federal government was responsible for HIV’s introduction into the black population. These beliefs, in turn, were associated with a reduced likelihood of adhering to a physician-prescribed treatment regimen ( Bogart et al. 2016 ). See also Halbert et al. (2009) , Kayaniyil et al. (2009) ; and Laveist, Nickerson, and Bowie (2000) .
10 Blacks were more likely than whites to utilize public and community-based healthcare resources, including charity clinics, public health departments, community health centers, and city and county hospitals and less likely to make use of private and non-profit hospitals. Some of this difference was the result of rapid suburbanization of the white population ( Byrd and Clayton 2002 , p. 391).
11 The level of utilization for black (and white) men is lower than for women in every period. Note that we do not have a continuous series of outpatient utilization data in the pre-1969 period.
12 The Appendix also contains figures for Southerners alone ( Appendix Figure A.1 ). In Appendix Figure A.2 , younger black men exhibit an increase in relative mortality beginning in the 1980s. This is likely due to the evolving HIV/AIDS crisis in this community. Appendix Section V.A . contains more detail.
13 The question we use regarding outpatient interactions is unchanged between 1969 to 1981. In 1978 the race categorization changed, with mutiple categories included, making 1977 a natural stopping point. But we have also extended our analysis to 1981 (the year the question on doctor visits changes and harmonization is no longer possible) and obtain similar (though not disclosed) results. Inpatient (hospital) utilization data is available prior to 1969 and the broad patterns are discussed in Section II.B. Our results are robust to the inclusion of these earlier data. IHIS is now known as IPUMS Health Surveys: NHIS.
14 We accessed the IHIS harmonized data, with location-based treatment variables (geographic proximity and migration shares) attached, inside Census Research Data Centers (RDCs) at Atlanta and Stanford. In later years, the NHIS data have been linked to mortality files, but these linked data are not available for our study period.
15 The NHIS data do not contain consistent measures of individual health insurance coverage over our time period; however, in all mortality regressions we control for local, time-varying Medicare and Medicaid expenditures, interacted with race and gender. Because phone calls to medical providers were counted as physician interactions, all NHIS regressions control for whether the household has a telephone.
16 Recent examples include Almond, Chay, and Greenstone (2006) ; Alsan and Goldin (2015) ; Bhalotra and Venkataramani (2011) ; Bailey and Goodman-Bacon (2015) ; and Goodman-Bacon (2015) .
17 State economic areas are groups of counties reflecting relatively homogenous areas within states. Results using county-level and annual mortality rates are in Appendix Table A.5 .
18 Some examples include blood pressure and glucose control for the management and prevention of cardiovascular and diabetic complications such as heart attack, stroke, and kidney failure, as well as counseling on prevention, early detection, and treatment of cancers. See Appendix Section IV for a discussion of medical and public health innovations over this time period.
19 Briefly, to construct age-adjusted rates, age-specific mortality rates, calculated as deaths per 1000 relevant at risk population, are weighted by a reference population. We follow the demographic literature and use the standard 1940 population for weighting. See the Appendix for further details.
20 After 1987, the paper’s conclusions are likely to be compromised by the evolving HIV/AIDS epidemic. But in the 1981–1987 period, HIV/AIDS was a nascent health threat, particularly among older black men. In this period, there were roughly 9,000 AIDS-infected men older than 45, of which perhaps 25% were black ( CDC 2001 ). Young people were at higher risk of acquiring HIV, and by 1994 AIDS had become the leading cause of death among black males aged 25–44. This increased mortality among younger black males in the late 1980s is apparent in Appendix Figure A.2 .
21 The Appendix contains a fully-specified model of how prior beliefs, prior experience with the medical profession, shared characteristics with the Tuskegee study subjects, and proximity to the study interact to determine health-seeking behavior
22 Data from the 1979 Survey of Black Americans support the notion that black men from the South identified more with the men from the study. When asked "how close does the respondent feel to black people who are poor", 78% of black men born in the South answered they felt "very close" compared to 65% of men born elsewhere ( Jackson and Gurin, 1999 ). We thank Trevon Logan for pointing out this source. As another example, although the Rodney King beating, which took place in March 1991, was widely publicized in many media markets, opinions on the police force shifted most markedly for black men. Though we lack geographic identifiers in the survey data, comparing 1989 and 1992 polling data, the percentage of people who disagree with the statement: "These days police in most cities treat Blacks as fairly as they treat Whites" jumped 18% among black men, 7% among black women, and declined among whites ( ABC 1992 ; NBC 1989 ).
23 We exclude the SEAs containing Los Angeles and San Francisco from the analytic sample. These two cities hold most of the black population in the West region, and they are more connected to Alabama (via migration) than our distance-based proxy for study salience would suggest (see Appendix Figure A.7 ). Our migration-based results are robust to the inclusion of these two cities. See Appendix Table A.5 , Column (1).
24 Using later census years would preclude us from using the detailed geographic information publicly available for 1940. We normalize by black migrants for two reasons. First, doing so gives a measure of what percentage of new information is coming from Alabama. (If the denominator was the black population, for example, the new information measure would be diluted.) Second, and more importantly, we only observe the migration variables in the five year period of 1935 to 1940. But the migration of this generation of men extended from 1918 up through 1960. The patterns of migration were persistent over time, so even though the 1935–1940 measure is not an accurate measure of the absolute number of migrants, the relative measure (those from Alabama divided by all migrants) is a good proxy for how connected the stock of black residents in 1972 would have been to Alabama, given several decades of migration preceding 1972.
25 Age-adjusted mortality rates for older adults (ages 45–74) are measured at the level of each SEA (denoted a ) for each race ( r ) and gender ( g ) combination: black males, black females, white males, white females. Bins in the middle of the country with few older black men were paired.
26 The estimating equation is Y gat k = α + β k ( I g male · I t post ) + γ k ( I t post ) + ϕ a , male + ε rat for k ∈ [1, K ], where g reflects gender.
27 Specifically, we use max(distance)–(distance s ) so that locations farthest from Tuskegee receive a proximity value of 0. Distance is measured at the state level in the utilization data and at the SEA or county level for the mortality data. In the mortality data, distances are measured from the geographic centroid of each SEA to the centroid of Macon County, Alabama. For the NHIS outcomes, which are available at the state level, we use county-level black adult population statistics and geographic county centroids to create black population-weighted state centroids which are used to construct proximity as above. These centroids represent the average latitude and longitude of black individuals in each state based on the black population of counties ( Haines 2010 ).
28 In the Appendix , we show that our results are robust to controls for local measures of incarceration and unemployment as well. See Appendix Table A.5 columns (3) and (6).
29 We find no evidence that expenditure levels for these social programs exhibited a temporal or geographic pattern mirroring our main empirical results. In Appendix Table A.7 , both hospital beds per capita and Medicare expenditures show a continuous increase over this time period, and more so in locations closer to Tuskegee. But there is no trend break in 1972 that would have induced utilization and mortality changes in that year. (See Appendix Figure A.8 .)
30 Due to data limitations, we cannot acurately generate a longer pre-period for the mortality or utilization results. NCHS mortality data are available for earlier years, but the 1960 U.S. Census of Population, from which population counts would need to be derived for such an exercise, only reports local population counts by age and gender for the nonwhite population. In an attempt to construct a denominator for black mortality rates for years prior to 1968, we limit the sample to locations where at least 60% of the 1960 non-white population was black. (Our results are robust to other choices of this cutoff.) Extended event study results (in Appendix Figure A.3 ) generated under these limitations also demonstrate insignificant and relatively flat pre-period estimates of the black male mortality penalty as a function of proximity to Macon County, Alabama.
31 The elasticity was obtained by dividing a one standard deviation percent change in all-cause mortality by a one-standard deviation change percent change in outpatient physician interactions. The range is due to the inclusion or exclusion of veteran males from the utilization estimates; the higher elasticity is from a sample including veterans (whose utilization response was more muted).
32 Bailey and Goodman-Bacon (2015) show that exposure to a community health center in one’s county of residence between 1965 and 1974 reduced adult all-cause mortality by as much as 2%; implying an elasticity of approximately −0.18. Lichtenberg (2002) reports an elasticity of −0.095.
33 We do not find the same pattern if we use black physicians per black capita, suggesting that black men responded to the "whiteness" of the local medical profession, rather than to the availability of a black physician.
34 In 1965, approximately 75% of black infants were born to married mothers ( Akerlof, Yellen, and Katz 1996 ). In related work, Thomasson and Treber (2004) find that the hospitalization of childbirth in the 1930s and, in particular, the introduction of sulfa drugs may have disproportionately reduced maternal mortality among black women, despite the segregation of hospitals and otherwise racially inequitable access. Because older women in our sample would have given birth between 1925 and 1960 (roughly), a substantial share of them would have benefited from this "positive" healthcare experience.
35 Individuals who replied "no" or "don’t know" to the question of whether people can be trusted were coded as 1 and those who replied "yes" were coded as 0. A similar coding strategy was applied to the outcome variables for medical mistrust (measured on a four point Likert scale) so that the coefficients can be easily interpreted as marginal effects.
36 This implies that age in the sample ranges from 36 to 89, with an average in the mid-fifties. Note that if we restrict to age<10 in 1972, those aged 18 to 35 in our sample, our coefficients on medical mistrust and deny treatment generally have the "wrong sign".
37 Appendix Figure A.9 contains event study coefficients for the within-male comparision.
Marcella Alsan, Stanford University, NBER and BREAD.
Marianne Wanamaker, University of Tennessee, NBER and IZA.
As you were browsing something about your browser made us think you were a bot. There are a few reasons this might happen:
To regain access, please make sure that cookies and JavaScript are enabled before reloading the page.
Intended for healthcare professionals
In 1965 Peter Buxtun had just started working as a contact tracer with the Public Health Service (PHS) in San Francisco when he overheard colleagues talking about a syphilis study. Wanting to learn more, he called the PHS office in Atlanta, now part of the Centers for Disease Control and Prevention (CDC). A helpful clerk sent him a large manila envelope crammed with documents. The more Buxtun read, the more convinced he became that the study was racist, unethical, and immoral.
The PHS had launched the Tuskegee Study of Untreated Syphilis in the Negro Male 1 in 1932 in the Tuskegee area of Alabama, where the incidence of syphilis was high. It enrolled 399 infected men and 201 negative control participants, most of whom were illiterate farm workers. The study’s purpose was to observe the natural course of the disease at a time when treatment options were limited.
The PHS told the men that the study was about “bad blood,” not the fact that they were infected with a disease that, by the 1940s, was treatable with penicillin. They were promised medical exams, meals, and burial expenses but …
BMA Member Log In
If you have a subscription to The BMJ, log in:
Subscribe from £184 *.
Subscribe and get access to all BMJ articles, and much more.
* For online subscription
Access this article for 1 day for: £50 / $60/ €56 ( excludes VAT )
You can download a PDF version for your personal record.
Buy this article
Official websites use .gov
A .gov website belongs to an official government organization in the United States.
Secure .gov websites use HTTPS
A lock ( ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.
After the U.S Public Health Service’s (USPHS) Untreated Syphilis Study at Tuskegee, the government changed its research practices.
In 1974, the National Research Act was signed into law, creating the National Commission for the Protection of Human Subjects of Biomedical and Behavioral Research . The group identified basic principles of research conduct and suggested ways to ensure those principles were followed.
In addition to the Commission’s recommendations, regulations were passed in 1974 that required researchers to get voluntary informed consent from all persons taking part in studies done or funded by the Department of Health, Education, and Welfare (DHEW). They also required that all DHEW-supported studies using human subjects be reviewed by Institutional Review Boards, which decide whether research protocols meet ethical standards.
The rules and policies for human subjects research have been reviewed and revised many times since they were first approved and efforts to promote the highest ethical standards in research are ongoing.
An Ethics Advisory Board was formed in the late-1970s to review ethical issues of biomedical research. As a result of their work, the 1979 publication commonly known as The Belmont Report summarized the three ethical principles that should guide human research: respect for persons ; beneficence ; justice. From 1980-1983, the President’s Commission for the Study of Ethical Problems in Medicine and Biomedical and Behavioral Research reported “every two years on the adequacy and uniformity of the Federal rules and policies, and their implementation, for the protection of human subjects in biomedical and behavioral research.” In 1991, federal departments and agencies (16 total) adopted the Federal Policy for the Protection of Human Subjects .
In October 1995, President Bill Clinton created a National Bioethics Advisory Commission , funded and led by the Department of Health and Human Services. The commission’s task was to review current regulations, policies, and procedures to ensure all possible safeguards are in place to protect research volunteers. It was succeeded by the President’s Council on Bioethics , which was established in 2001, and then the Presidential Commission for the Study of Bioethical Issues established in 2009.
IMAGES
VIDEO
COMMENTS
Tuskegee Experiment: The Infamous Syphilis Study
The Tuskegee Study of Untreated Syphilis in the Negro Male [1] (informally referred to as the Tuskegee Experiment or Tuskegee Syphilis Study) was a study conducted between 1932 and 1972 by the United States Public Health Service (PHS) and the Centers for Disease Control and Prevention (CDC) on a group of nearly 400 African American men with syphilis. [2] [3] The purpose of the study was to ...
Tuskegee syphilis study | US Government Experiment ...
Starting in 1932, 600 African American men from Macon County, Alabama were enlisted to partake in a scientific experiment on syphilis. The "Tuskegee Study of Untreated Syphilis in the Negro Male," was conducted by the United States Public Health Service (USPHS) and involved blood tests, x-rays, spinal taps and autopsies of the subjects. The goal was to "observe the natural history of ...
Tuskegee Syphilis Study Timeline. By 1943, penicillin was the treatment of choice for syphilis and becoming widely available, but the participants in the study were not offered treatment.. In 1972, an Associated Press story about the study was published. As a result, the Assistant Secretary for Health and Scientific Affairs appointed an Ad Hoc Advisory Panel to review the study.
The official title was "The Tuskegee Study of Untreated Syphilis in the Negro Male.". It is commonly called the Infamous Tuskegee Syphilis Experiment. Beginning in 1932 and continuing to 1972 the United States Public Health Services lured over 600 Black men, mostly sharecroppers in Tuskegee, Alabama, into this diabolical medical experiment ...
In 1932, 399 African American men in Tuskegee and Macon County, Alabama were enrolled in a Public Health Service study on the long-term effects of untreated syphilis.At that time, there was no cure for syphilis, though many ineffective and often harmful treatments, such as arsenic, were used.
Surgeon General Thomas Parran boasted that in Macon County, Ala., where Tuskegee is located, the syphilis rate among the African-American population had been nearly 40% in 1929 but had shrunk to ...
The U.S. Public Health Service (USPHS) Untreated Syphilis Study at Tuskegee was conducted between 1932 and 1972 to observe the natural history of untreated syphilis. As part of the study, researchers did not collect informed consent from participants and they did not offer treatment, even after it was widely available. The study ended in 1972 ...
Summary. Between 1932 and 1972, the US Public Health Service (PHS) ran the Tuskegee Study of Untreated Syphilis in the Male Negro in Macon County, Alabama, to learn more about the effects of untreated syphilis on African Americans, and to see if the standard heavy metal treatments advocated at the time were efficacious in the disease's late latent stage.
The PHS began working with Tuskegee Institute in 1932 to study hundreds of black men with syphilis from Macon County, Alabama. Compensation for Participants. As part of the class-action suit settlement, the U.S. government promised to provide a range of free services to the survivors of the study, their wives, widows, and children.
The experiment, called the Tuskegee Study began in 1932 with about 600 black men mostly poor and uneducated, from Tuskegee, Ala., an area that had the highest syphilis rate in the nation at the time. One-third of the group was free of syphilis; two-thirds showed evidence of the disease. In the syphilitic group, half were given the best ...
The Tuskegee syphilis study, as the experiment is often called today, began in 1932 with the recruitment of 600 Black men, 399 with syphilis and 201 without, to serve as the control group ...
The now-infamous 1932 Tuskegee Syphilis Study was conducted by the U.S. Public Health Service in Macon County, Alabama. During the experiment 600 impoverished black men were studied, 399 of whom had syphilis that went untreated although the health care practitioners knew of their illness. The men were never educated about syphilis, were never ...
'You've got bad blood': The horror of the Tuskegee syphilis ...
Fiftieth Anniversary of Uncovering the Tuskegee Syphilis ...
JEL Codes: I14, O15 For forty years, the Tuskegee Study of Untreated Syphilis in the Negro Male passively monitored hundreds of adult black males with syphilis despite the availability of effective treatment. The study's methods have become synonymous with exploitation and mistreatment by the medical profession. To identify the study's effects on the behavior and health of older black men ...
THE EXPERIMENT AND HEW'S ETHICAL REVIEW Racism and Research: The Case of the Tuskegee Syphilis Study by ALLAN M. BRANDT In 1932 the U.S. Public Health Service (USPHS) initiated an experiment in Macon County, Alabama, to determine the natural course of untreated, latent syphilis in black males. The test comprised 400 syphilitic men, as well as ...
Tuskegee community members were aware of the study but understood it to be a special government health care program. 1. According to the Assistant Secretary for Health and Scientific Affairs' Ad Hoc Advisory Panel's published report, "…the Macon County Health Department and Tuskegee Institute were cognizant of the study."
July 26, 2022 at 6:00 a.m. EDT. In the fall of 1932, the fliers began appearing around Macon County, Ala., promising "colored people" special treatment for "bad blood.". "Free Blood Test ...
Tuskegee Syphilis Experiment What is the Tuskegee Syphilis Experiment? The Tuskegee Syphilis Experiment (TSE) was done during 1932-1972 in Macon County, Alabama. The study consisted of 600 black men, 400 infected with syphilis which was a sexually transmitted disease and 200 not infected black men who were seen as controls. During the experiment a medication called Penicillin was created to ...
The Tuskegee Syphilis Experiment is a notorious example of unethical investigative practices on vulnerable minorities. Conducted between 1932 and 72, the study aimed to observe the natural history of untreated syphilis (Kim, Oliver J.; Magner, Lois N. (2018). ... The Belmont Report summarized the basic ethical principles and corresponding ...
The PHS had launched the Tuskegee Study of Untreated Syphilis in the Negro Male1 in 1932 in the Tuskegee area of Alabama, where the incidence of syphilis was high. It enrolled 399 infected men and 201 negative control participants, most of whom were illiterate farm workers. The study's purpose was to observe the natural course of the disease ...
After the U.S Public Health Service's (USPHS) Untreated Syphilis Study at Tuskegee, the government changed its research practices. In 1974, the National Research Act was signed into law, creating the National Commission for the Protection of Human Subjects of Biomedical and Behavioral Research.The group identified basic principles of research conduct and suggested ways to ensure those ...