Abstract
Objectives. We tested the data artifact hypothesis regarding the Hispanic mortality advantage by investigating whether and to what degree this advantage is explained by Hispanic origin misclassification on US death certificates.
Methods. We used the National Longitudinal Mortality Study, which links Current Population Survey records to death certificates for 1979 through 1998, to estimate the sensitivity, specificity, and net ascertainment of Hispanic ethnicity on death certificates compared with survey classifications. Using national vital statistics mortality data, we estimated Hispanic age-specific and age-adjusted death rates, which were uncorrected and corrected for death certificate misclassification, and produced death rate ratios comparing the Hispanic with the non-Hispanic White population.
Results. Hispanic origin reporting on death certificates in the United States is reasonably good. The net ascertainment of Hispanic origin is just 5% higher on survey records than on death certificates. Corrected age-adjusted death rates for Hispanics are lower than those for the non-Hispanic White population by close to 20%.
Conclusions. The Hispanic mortality paradox is not explained by an incongruence between ethnic classification in vital registration and population data systems.
Mortality rates reported for Hispanics in the United States are lower than those for non-Hispanic Whites. Because the average socioeconomic status of Hispanics is lower than that of non-Hispanic Whites, the relatively low mortality of Hispanics has been called an epidemiologic paradox.1 The explanation of lower Hispanic mortality remains a matter of investigation.2–5 Palloni and Arias3 identified 4 groups of explanations: (1) poor data quality (the data artifact hypothesis), (2) selective immigration of healthier persons (the healthy migrant hypothesis), (3) selective return migration of less healthy immigrants (the “salmon-bias” hypothesis), and (4) positive cultural influences on individual health and lifestyle behaviors, family structure, and social networks.
Information about Hispanic mortality comes from 2 types of studies. In the first type of study, death counts from the US Vital Statistics System (NVSS) are used as numerators and population counts from the US Census are used as denominators for death rate calculations. In the second type of study, sometimes called passive follow-up studies, records from large survey samples or administrative databases are linked to death certificates. Matched subjects are coded as deceased on the date reported on the death certificate, and unmatched cases are presumed alive. This information is used to calculate mortality rates for the subjects in the sample or administrative database. Both types of studies report lower age- and gender-adjusted mortality rates for Hispanics in the United States than for non-Hispanic Whites.5–7
Each type of study is subject to data problems, which can be grouped into 3 types: ethnic classification errors, record linkage errors, and age misstatement.3 Ethnic classification errors are inherent in data systems that rely on 2 distinct data sources—NVSS and census—for the estimation of vital rates. Mortality follow-up studies are vulnerable to failures of the algorithm linking the survey or administrative records to death records if the linkage algorithms fail at a differential rate for the 2 groups being compared. Mortality rates calculated in both types of studies are vulnerable to age misstatement.
The debate concerning the role of these data problems on the Hispanic paradox has not been resolved. In particular, the question remains whether the lower mortality of Hispanics observed in US vital statistics data results from discrepancies between the assignment of Hispanic origin on the death certificate that supplies the numerator for vital rates and on the census forms that provide the denominator. All evidence confirms that some degree of correction of Hispanic rates for these errors is warranted, although no consensus exists on the size of the errors and the resulting impact on vital rates.8–12 The ethnic identity of the decedent is recorded on the death certificate by the funeral director, who is responsible for completing the demographic portion of the death certificate and registering it with the state vital statistics office. Whether the funeral director consults with the family of the deceased is not ascertained. Ethnic identity is most typically reported on the census form by a respondent living in the household. These identities may not agree because the funeral director will likely have less information about the ethnic ancestry and personal identity of decedents.
Before the 1980 US Census, populations of Hispanic origin were identified indirectly by matching surnames of census respondents and decedents to a list of common Spanish surnames in several Southwestern states.8,13,14 The 1980 US Census introduced a Hispanic origin item for the first time on all census forms. By 1980, 21 states had introduced a Hispanic origin item to their death certificates. However, Florida, an important Hispanic origin state, did not include a Hispanic origin item on the death certificate until 1989. California was missing responses on the Hispanic origin item for more than 50% of death certificates through 1982.
The leading source of data about discrepancies in ethnic classification on the death certificate and population surveys or censuses is the National Longitudinal Mortality Study (NLMS). The NLMS is a database consisting of US Census Bureau data from Current Population Surveys (CPSs) and the 1980 US Census combined with death certificate information to identify mortality status and cause of death. The NLMS has been used in several passive follow-up studies of Hispanic mortality.6,7 The NLMS can also be used to investigate the validity of death certificate ethnic classification by comparing ascertainment of Hispanic ethnicity in survey responses and on the death certificate for linked records.
Studies using data from an early release of the NLMS reported a net ascertainment ratio of 1.07 for survey compared with death certificate assignment of Hispanic origin.9,10 This finding suggested a small correction to death rates for Hispanics, which remain lower than those of non-Hispanic Whites after correction.5 These studies had important limitations, however. The NLMS at this time contained a sample of only 600 linked death certificate records that were classified as Hispanic in survey responses. Estimated ascertainment ratios were imprecise, and subpopulation analysis was not possible. The NLMS at this time included deaths that took place from 1979 through 1985, a period during which only 21 states included a Hispanic origin item on their death certificates.9,10
We tested the data artifact hypothesis as it pertains to ethnic classification across 2 distinct data systems: vital registration and census data. The data artifact hypothesis, as it pertains to ethnic classification, posits that incongruence between ethnic classification in the numerators and denominators of death rates leads to artificially lower death rates for Hispanics than for non-Hispanics, and therefore the appearance of a mortality advantage. We used the latest release of the NLMS to carry out our analysis for deaths that took place in 1979 through 1998. We used the sample of survey records that were matched to death certificate information in the NLMS to assess the sensitivity, specificity, and net ascertainment of Hispanic ethnicity on the death certificate compared with the identification of Hispanic ethnicity by a household respondent. We also investigated differential rates of net ascertainment in relation to sociodemographic and area characteristics, including age, gender, immigrant status, census region, urban or rural status, and residence in a county with a high concentration of Hispanics. We calculated death rates and death rate ratios for the United States by comparing Hispanics and non-Hispanic Whites with correction for death certificate misclassification of Hispanic identity.
METHODS
To evaluate misclassification of Hispanic origin on death certificates, we used the NLMS to compare classification of Hispanic origin on census survey records with that on death certificates for the sample of survey records that were linked to death certificates. The NLMS consists of data from the CPS, Annual Social and Economic Supplements, and a sample of the 1980 US Census combined with death certificate information from the NVSS to identify mortality status and cause of death.15 The CPS is a multistage, stratified probability sample of the US noninstitutionalized civilian population, with a response rate of approximately 95%.16 The NVSS consists of a voluntary contractual agreement between the National Center for Health Statistics (NCHS) and individual registration areas to collect the nation's birth and death information. NVSS coverage includes more than 99% of deaths that occur in all 50 states, the District of Columbia, and US territories.17
Data
To date, the NLMS includes 26 files, March 1973 CPS, February 1978 CPS, March 1979 CPS, April 1980 CPS, August 1980 CPS, December 1980 CPS, 1980 Decennial Census E Sample, March 1981 CPS through March 1998 CPS, and September 1985 CPS, which add up to 2.3 million records. Through linkage to the NCHS National Death Index for 1979 through 1998, 252 627 deaths have been identified. We used records in the files listed above that were matched to death certificates. We excluded the 1980 Decennial Census E Sample and the September 1985 CPS files because their records do not contain sampling weights. A more detailed description of the use of the NLMS sample to investigate racial and ethnic classification on US death certificates is reported elsewhere.18
A Hispanic origin item was introduced on death certificates at different times in different states between 1978 and 1997. In 1979, only 20 states included the item. By 1990, only Louisiana, New Hampshire, and Oklahoma still did not have a Hispanic origin item on their death certificates. Beginning in 1997, coverage was complete. We excluded records from states for years that the state did not include a Hispanic origin item on the death certificate. In the remaining states, fewer than 1% of cases were missing a Hispanic origin classification: 0.29% and 0.59% in the CPS and death certificates, respectively. The CPS began imputing unknown Hispanic origin in the mid-1980s. The nonresponse rate for this item remained around 5% throughout the years covered in this study.15,16 The NVSS imputes unknown Hispanic origin subgroups on the death certificate for reporting states by use of birthplace. If birthplace is Mexico, Puerto Rico, or Cuba, then the Hispanic origin is accordingly assigned. If birthplace is elsewhere, then the code is “Other and Unknown Hispanic” or “Unknown.”19 We did not make any adjustments to these imputations, but excluded the small fraction of records with unknown or unclassified Hispanic origin.
The final sample we used for our analysis consisted of 114 869 matched records. Of these, 6656 records were coded as Hispanic on the CPS, and 6320 were coded as Hispanic on death certificates.
We have noted concerns that there may be ethnic differentials in the rate at which the linkage algorithms in linked data sets such as the NLMS correctly match records from 2 distinct data systems. These differentials are critical to studies in which record linkage is used to identify vital status, because unlinked records are presumed alive. Because we used matched cases only to assess the validity of death certificate classification, linkage errors would only bias reported results to the degree that incorrectly unmatched cases compared with matched cases differed in the rate of agreement between the CPS and death certificate ethnic classifications. However, there is no documented evidence of the existence or degree of differential matching rates by race or ethnicity in the NLMS.
To assess the effects of Hispanic origin misclassification on the death certificate on mortality measures, we pooled the number of deaths for 1999 through 2001 from the NVSS and used the NCHS bridged April 1, 2000, population estimates to estimate age-specific and age-adjusted death rates that were unadjusted and adjusted for Hispanic origin misclassification on the death certificate. The bridged file used a bridging algorithm developed by the NCHS to reassign multiple-race persons to single-race categories.20
Analyses
We adopted Hispanic origin as reported on the CPS as the standard for comparison with the death certificate classification. Survey reporting has its own limitations resulting from question wording effects and item nonresponse.21 Public policy embodied in Office of Management and Budget Statistical Directive as revised in 1997 emphasizes respondent self-identification as the standard for collection of racial and ethnic identifiers, and we adopted that assumption.22 We first analyzed classification as a Hispanic. We also analyzed subgroup-specific classification, including Mexican, Puerto Rican, Cuban, Central and South American, or other Hispanic. The “other” group included individuals not falling into the previous Hispanic categories, such as Dominicans and Spaniards, as well as Hispanic individuals for whom specific country of origin was not reported. Persons were coded as Hispanic without regard to their racial classification.
We reported ratios of CPS Hispanic origin counts to death certificate counts for the sample of identified NLMS decedents described above (hereinafter classification ratios). These ratios were the net difference in assignment of Hispanic origin between the 2 data systems. We also reported 2 measures of record-level agreement between the CPS and death certificates. Sensitivity was the percentage of respondents in a CPS-identified ethnic group who were correctly identified on the death certificate. Predictive value positive was the percentage of decedents identified by the death certificate in a specific ethnic group who were self-identified in the same group on the CPS.5,11,12 All measures were weighted by CPS sample weights. We produced these measures for 2 time periods, 1979 to 1989 and 1990 to 1998, to assess whether the quality of ethnic classification changed over time.
We reported classification ratios by decedents' age (0–24, 25–44, 45–54, 55–64, 65–74, and 75 years or older), gender, nativity (US versus foreign born), urban or rural status, census region of residence, and degree of co-ethnic geographic concentration for the 1990 to 1998 period. We restricted these analyses to this period because national coverage of Hispanic mortality dramatically improved during the 1990s. We also preferred more recent estimates of death certificate misclassification for use as death rate adjustment factors. These measures were derived from the death certificate and pertained to values at time of death. Co-ethnic concentration is a dichotomous variable contrasting decedents who died in a county that falls within the first 50th percentile of ethnic-specific ranked number of deaths by county with decedents who died outside these counties during the 1990s. A list of counties with high Hispanic (and Hispanic subgroup) concentrations is available from the authors.
We estimated uncorrected age-specific and age-adjusted death rates for non-Hispanic Whites and persons of Hispanic origin as follows:
 |
and
 |
where Diyr is the number of deaths in specific age group i and specific year yr, Pi2000 is the population in specific age group i and year 2000, and Wi is the age-specific weight based on the US Standard Population.23
We corrected the observed age-specific death rates with the age-specific classification ratios derived from the NLMS and then re-estimated the age-adjusted death rates. We limited our correction to age because of the sample size limitations. Adjustment was done as follows:
 |
where CRi is the age-specific classification ratio derived from the NLMS.
RESULTS
Correspondence between death certificate and survey identification of all Hispanics was high in both the 1979 to 1989 and 1990 to 1998 periods and declined slightly between the 2 periods (Table 1). Sensitivity and predictive value positive were always near or above 90% in both periods. The classification ratio was 1.04 in the 1979 to 1989 period, which meant that the survey responses identified an additional 4% of Hispanics compared with the death certificates. In the 1990s, the classification ratio increased to 1.05. This difference was not statistically significant.
TABLE 1.
Sensitivity, Predictive Value Positive, and Classification Ratio for US Deaths Occurring in 1979–1989 and 1990–1998
| Sensitivity, % |
Predictive Value Positive, % |
Classification Ratioa (SE) |
No. of Records (1979–1989) |
No. of Records (1990–1998) |
||||||
| Hispanic Origin | 1979–1989 | 1990–1998 | 1979–1989 | 1990–1998 | 1979–1989 | 1990–1998 | CPS | Death Certificate | CPS | Death Certificate |
| Non-Hispanic White | 99.6 | 99.6 | 99.1 | 99.2 | 1.00 (0.001) | 1.00 (0.001) | 12 786 | 12 860 | 81 797 | 82 146 |
| All Hispanics | 92.8 | 88.1 | 95.9 | 92.5 | 1.04 (0.010) | 1.05 (0.008) | 1516 | 1465 | 5140 | 4855 |
| Mexican | 79.2 | 86.0 | 93.2 | 91.4 | 1.17 (0.021) | 1.06b (0.011) | 1018 | 864 | 2887 | 2622 |
| Puerto Rican | 85.8 | 79.7 | 88.8 | 85.0 | 1.06 (0.047) | 1.07 (0.036) | 150 | 144 | 527 | 496 |
| Cuban | 82.1 | 86.6 | 87.0 | 90.5 | 1.05 (0.069) | 1.04 (0.026) | 84 | 80 | 532 | 505 |
| Central and South American | 32.3 | 64.3 | 75.9 | 67.1 | 2.35 (0.432) | 1.05b (0.063) | 67 | 28 | 250 | 235 |
| Other Hispanic | 46.7 | 39.2 | 22.3 | 38.7 | 0.50 (0.042) | 0.99b (0.045) | 197 | 349 | 944 | 997 |
Note. CPS = Current Population Survey.
Ratios were based on weighted data.
Difference in ratios across the 2 time periods was statistically significant at the 1% level.
Agreement between death certificate and survey identification improved significantly between the 2 periods for Mexicans, Central and South Americans, and “other” Hispanics as measured by the classification ratios and, for the former 2, by the measure of sensitivity. The proportion of Mexicans and Central and South Americans falling in the “other” Hispanic group on the death certificate because of missing information on country of origin declined between 1979 to 1989 and 1990 to 1998 from 17% and 26%, respectively, to 7% and 8%, respectively.
Two characteristics emerged as important correlates of ethnic misclassification: nativity status and ethnic concentration. For foreign-born decedents, a classification ratio of 1.02 indicated strong agreement in counts on death certificates and in survey responses. The classification ratio for US-born decedents was 1.07 (Table 2). This relation held for subgroup reporting for Mexicans, Puerto Ricans, Cubans, and Central and South Americans, although the US-born ratios were unreliable for the last 2 groups because of the very small number of US-born decedents of these nationalities (Table 3).
TABLE 2.
Classification Ratios for the Current Population Survey (CPS) Compared With US Death Certificates for All Hispanics, by Various Characteristics: 1990–1998
| Characteristic | Classification Ratioa (SE) | CPS, No. | Death Certificate, No. |
| Total | 1.05 (0.008) | 5140 | 4855 |
| Nativity | |||
| US-born | 1.07b (0.013) | 2902 | 2698 |
| Immigrant | 1.02 (0.008) | 2221 | 2141 |
| Age, y | |||
| 0–24 | 0.96 (0.053) | 125 | 125 |
| 25–44 | 1.06 (0.026) | 527 | 496 |
| 45–54 | 1.05 (0.025) | 474 | 451 |
| 55–64 | 1.03 (0.020) | 728 | 700 |
| 65–74 | 1.07 (0.017) | 1151 | 1081 |
| ≥ 75 | 1.05 (0.012) | 2135 | 2002 |
| Gender | |||
| Men | 1.04 (0.011) | 2892 | 2750 |
| Women | 1.06 (0.012) | 2248 | 2105 |
| Region | |||
| Northeast | 1.08 (0.028) | 617 | 593 |
| Midwest | 1.02 (0.040) | 293 | 278 |
| South | 1.05 (0.012) | 1911 | 1793 |
| West | 1.04 (0.011) | 2315 | 2217 |
| Locale | |||
| Urban | 1.05 (0.008) | 4406 | 4161 |
| Rural | 1.05 (0.026) | 721 | 691 |
| Co-ethnic concentration | |||
| Yes | 1.02b (0.008) | 2507 | 2459 |
| No | 1.08 (0.014) | 2633 | 2396 |
Ratios were based on weighted data.
Difference in ratios across the 2 dimensions of the variable was statistically significant at the 1% level.
TABLE 3.
Classification Ratios for the Current Population Survey (CPS) Compared With US Death Certificates for Selected Hispanic Subgroups, by Co-ethnic Concentration and Nativity: 1990–1998
| Mexican |
Puerto Rican |
Cuban |
Central/South America |
|||||
| Concentration and Nativity | Classification Ratio (SE) | No. of CPS Records/No. of Death Certificate Records | Classification Ratio (SE) | No. of CPS Records/No. of Death Certificate Records | Classification Ratio (SE) | No. of CPS Records/No. of Death Certificate Records | Classification Ratio (SE) | No. of CPS Records/No. of Death Certificate Records |
| Co-ethnic concentration | ||||||||
| Yes | 1.02a (0.011) | 1348/1321 | 1.04 (0.042) | 256/246 | 1.02 (0.025) | 343/334 | 0.99 (0.078) | 391/386 |
| No | 1.11 (0.019) | 1539/1301 | 1.09 (0.058) | 271/250 | 1.10 (0.059) | 189/171 | 1.12 (0.101) | 131/114 |
| Nativity | ||||||||
| US Born | 1.09a (0.016) | 1948/1704 | 1.14(0.133) | 122/112 | 1.92b (0.558) | 27/15 | 1.30b (0.381) | 22/17 |
| Foreign born | 1.01 (0.011) | 928/912 | 1.04 (0.028) | 402/383 | 1.02 (0.023) | 502/490 | 1.04 (0.063) | 228/217 |
Note. Classification ratios are based on weighted data.
Difference in ratios across the 2 dimensions of the variable was statistically significant at the 1% level.
Ratio was unreliable because either the unweighted number of Current Population Survey deaths or the unweighted number of death certificate deaths or both were based on fewer than 20 deaths.
Agreement between death certificate and survey self-identification was stronger in areas of higher ethnic concentration. For the total Hispanic population, the co-ethnic concentration classification ratio was 1.02, compared with a ratio of 1.08 outside these areas (Table 2). Net ascertainment ratios for the Mexican population were 1.02 and 1.11 in areas with high Mexican concentration versus outside these areas, respectively. The ratios were 1.02 and 1.10 for the Cuban population, 0.99 and 1.12 for the Central and South American population, and 1.04 and 1.09 for the Puerto Rican population (Table 3). Thus, as expected, ethnic classification on the death certificate varied significantly according to whether a death occurred in an area with a significant number of co-ethnic residents. We found no statistically significant effects of age, gender, census geographic region, or urban or rural place of residence on the classification ratios.
Correction for death certificate misclassification did not have a large effect on death rates for the Hispanic population. Relative to the non-Hispanic White population, the total Hispanic population maintained its mortality advantage after correction for death certificate misclassification. The Hispanic age-adjusted death rate increased from 79% to 83% of the non-Hispanic White age-adjusted death rate (Table 4). The age-specific mortality pattern after correction suggested that the Hispanic mortality advantage was somewhat more pronounced in the older age groups. For example, for age groups between 0 and 54 years of age, the corrected rate ratios compared with non-Hispanic Whites ranged between 0.87 and 0.99, whereas for older ages they were lower and flat, varying only between 0.80 and 0.83. The lower mortality in the oldest age groups could be a reflection of the salmon-bias effect or age misreporting, but because of the data we were working with and the focus of this study, this interpretation is speculative.
TABLE 4.
Hispanic Age-Specific, Age-Adjusted Death Rates Uncorrected and Corrected for US Death Certificate Misclassification and Hispanic to Non-Hispanic White Death Rate Ratios, 1999–2001
| No. of Deaths | Population | Uncorrected Ratea (SE) | Corrected Rate (SE) | Reported Rate Ratio | Corrected Rate Ratio | |
| Age, y | ||||||
| 0 | 13 698 | 2 313 159 | 592.2 (5.1) | 568.5 (5.0) | 1.00 | 0.96* |
| 1–4 | 2 673 | 8 840 763 | 30.2 (0.6) | 29.0 (0.6) | 1.02 | 0.98 |
| 5–14 | 3 117 | 20 361 276 | 15.3 (0.3) | 14.7 (0.3) | 0.91* | 0.87* |
| 15–24 | 14 356 | 19 743 219 | 72.7 (0.6) | 69.8 (0.6) | 0.99 | 0.95* |
| 25–34 | 16 703 | 19 530 705 | 85.5 (0.7) | 90.7 (0.7) | 0.93* | 0.99 |
| 35–44 | 24 324 | 15 387 930 | 158.1 (1.0) | 167.7 (1.0) | 0.87* | 0.93* |
| 45–54 | 31 594 | 9 408 309 | 335.8 (1.9) | 352.8 (1.9) | 0.86* | 0.91* |
| 55–64 | 38 389 | 5 131 320 | 748.1 (3.8) | 771.0 (3.9) | 0.79* | 0.82* |
| 65–74 | 58 793 | 3 229 857 | 1 820.3 (7.5) | 1 948.8 (7.8) | 0.77* | 0.83* |
| 75–84 | 66 500 | 1 518 792 | 4 378.5 (17.0) | 4 600.1 (17.4) | 0.77* | 0.81* |
| ≥ 85 | 54 083 | 452 124 | 11 962.0 (51.4) | 12 567.3 (52.7) | 0.76* | 0.80* |
| Age adjusted | 672.1 (1.3) | 705.5 (1.3) | 0.79* | 0.83* | ||
*Rate ratio was statistically significant at the 5% level.
Number of deaths per 100 000 population.
DISCUSSION
Several important findings emerged from our study. First, Hispanic origin reporting on the death certificate was reasonably good. Net ascertainment of Hispanic deaths was just 5% higher for survey classification compared with death certificate classification for deaths occurring in the 1990s. Sensitivity and specificity were both at or above 90%. By 1997, national coverage of the Hispanic population was complete, and all 50 states and the District of Columbia included a Hispanic origin item on the death certificates. By the early 1990s, the overwhelming majority of states had missing rates of less than 1% on the Hispanic origin item. The combination of these findings led us to conclude that, with minor adjustment, Hispanic mortality estimates based on US vital statistics are reasonably reliable.
Significant problems remained with the mortality data for specific Hispanic subgroups, however, particularly populations other than the 3 main groups (Mexicans, Puerto Ricans, and Cubans). Although the quality of death certificate reporting for Hispanic subgroups improved markedly in the 1990s compared with the 1980s, the quality of census reporting significantly deteriorated. Sixteen percent of Hispanics did not provide a country of origin when responding to the Hispanic question in the 2000 US Census. Research suggests that this was a result of question design and wording.21,24
Characteristics of the decedent and the place of residence of the decedent affected ethnic classification on the death certificate. Foreign-born Hispanics were more likely to be correctly classified than were US-born Hispanics. This finding is understandable because a funeral director must also inquire about birthplace when filling out the death certificate and can use that information to assign Hispanic ethnicity. Hispanics who died in counties where co-ethnic residents formed a large share of the county's population were more likely to be correctly identified than those dying in counties with low representation of co-ethnic residents. These results suggest that states and substate areas with relatively small Hispanic populations that are predominantly US-born may experience substantial underascertainment of Hispanic deaths.
Finally, adjustment for Hispanic misclassification on the death certificate required only a modest upward restatement of age-adjusted mortality rates. This correction left intact the finding of relatively low mortality for the national population of Hispanics. Overall corrected age-adjusted mortality rates for Hispanics were lower than those of the non-Hispanic White population by close to 20%. Our findings reaffirm those reported in previous evaluation studies that used data from the NLMS for deaths occurring between 1979 and 1985.9,10 Although not entirely independent of reports based on these earlier releases of NLMS data, those findings are put on substantially firmer footing, because of the much larger number of matched CPS and death certificate records in the NLMS sample used in this study, the inclusion of data from death certificates for virtually all states, and the use of more recent data.
Our findings contribute to a growing body of evidence that the lower mortality of the Hispanic population compared with non-Hispanic Whites is not an artifact of poor data quality. Elo et al.5 and Turra and Elo25 used mortality follow-up for Medicare and Social Security enrollees and reported consistently lower mortality for Hispanic than for non-Hispanic White subjects, with mortality follow-up based on administrative follow-up of vital status of beneficiaries of the Social Security Administration. Hummer et al.2 examined mortality in linked infant birth–death files and found lower Hispanic mortality in seconds and minutes after birth, discounting record linkage errors, return migration, and ethnic misclassification as explanations of low Hispanic infant mortality. Palloni and Arias3 conducted a theoretically rigorous analysis of data from a passive follow-up study using the National Health Interview Survey–Multiple Cause of Death data set and reported patterns of lower mortality for some Hispanic foreign-born populations that were not easily explained as a result of data linkage errors but were consistent with the salmon-bias effect. In this study, we used the best available data to investigate the validity of death certificate reports of Hispanic ethnicity and found that errors were too small to nullify the consistent finding of lower Hispanic mortality compared with non-Hispanic Whites in US vital statistics.
Our study had some limitations. We did not address explanations of Hispanic mortality patterns apart from the data artifact of ethnic misclassification. We did not investigate either the salmon-bias hypothesis or the healthy migrant hypothesis, because NVSS data lack information about duration of immigration. We did not investigate effects of age misstatement in census or death certificates on Hispanic mortality.5,26 CPS data pertain to the noninstitutionalized population of the United States only. Finally, we did not adjust for the impact of differential census undercount of Hispanics compared with non-Hispanic Whites on calculated rate ratios, noting that that differential was not statistically significant in the 2000 US Census.27
Our findings reaffirm the validity of mortality data in the NVSS for the investigation of ethnic differences in mortality between the Hispanic population of the United States and other groups. This finding is qualified by caution about the use of data for Hispanic national-origin subgroups, for comparisons between native and foreign-born Hispanics, and when using data for areas of low Hispanic concentration.
Acknowledgments
Karl Eschbach was supported by the University of Texas Medical Branch Center for Population Health and Health Disparity, P50CA105631-05.
We are grateful to Robert N. Anderson, Julia Holmes, Virginia Cain, and Jennifer Madans of the National Center for Health Statistics (NCHS) and the 3 anonymous reviewers for their insightful and helpful reviews. We also acknowledge the National Longitudinal Mortality Study (NLMS) team, US Census Bureau, for their excellent work in producing and maintaining the NLMS database. We also thank Betzaida Tejada-Vera of NCHS for her assistance with table formatting.
Human Participant Protection
No protocol was needed for this study because the data were obtained from secondary sources.
References
- 1.Markides KS, Coreil J. The health of Hispanics in the southwestern United States: an epidemiologic paradox. Public Health Rep. 1986;101(3):253–265 [PMC free article] [PubMed] [Google Scholar]
- 2.Hummer RA, Powers DA, Pullum SG, Gossman GL, Frisbie WP. Paradox found (again): infant mortality among the Mexican-origin population in the United States. Demography. 2007;44(3):441–457 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Palloni A, Arias E. Paradox lost: explaining the Hispanic adult mortality advantage. Demography. 2004;41(3):385–415 [DOI] [PubMed] [Google Scholar]
- 4.Franzini L, Ribble JC, Keddie AM. Understanding the Hispanic paradox. Ethn Dis. 2001;11(3):496–518 [PubMed] [Google Scholar]
- 5.Elo IT, Turra CM, Kestenbaum B, Fergusson BR. Mortality among elderly Hispanics in the United States: evidence and new results. Demography. 2004;41(1):109–128 [DOI] [PubMed] [Google Scholar]
- 6.Sorlie PD, Backlund E, Johnson NJ, Rogot E. Mortality by Hispanic status in the United States. JAMA. 1993;270(20):2464–2468 [PubMed] [Google Scholar]
- 7.Singh GK, Siahpush M. Ethnic-immigrant differentials in health behaviors, morbidity and cause-specific mortality in the United States: an analysis of two national data bases. Hum Biol. 2002;74(1):83–109 [DOI] [PubMed] [Google Scholar]
- 8.Smith DP, Bradshaw BS. Rethinking the Hispanic paradox: death rates and life expectancy for US non-Hispanic White and Hispanic populations. Am J Public Health. 2006;96(9):1686–1692 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Sorlie PD, Rogot E, Johnson NJ. Validity of demographic characteristics on the death certificate. Epidemiology. 1992;3(2):181–184 [DOI] [PubMed] [Google Scholar]
- 10.Rosenberg HM, Maurer JD, Sorlie PD, et al. Quality of death rates by race and Hispanic origin: a summary of current research, 1999. Vital Health Stat 2. 1999;Sep(128):1–13 [PubMed] [Google Scholar]
- 11.Swallen K, Guend A. Data quality and adjusted Hispanic mortality in the United States, 1989-1991. Ethn Dis. 2003;13(1):126–133 [PubMed] [Google Scholar]
- 12.Eschbach K, Kuo YF, Goodwin JS. Ascertainment of Hispanic ethnicity on the California death certificate: implications for the explanation of the Hispanic mortality advantage. Am J Public Health. 2006;96(12):2209–2215 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Buechley RW. A reproducible method of counting persons of Spanish surname. J Am Stat Assoc. 1961;56(293):88–97 [Google Scholar]
- 14.Sullivan TA, Gillespie FP, Hout M, Rogers RG. Alternative estimates of Mexican-American mortality in Texas, 1980. Soc Sci Q. 1984;65(2):609–617 [Google Scholar]
- 15.National Longitudinal Mortality Study. Reference manual; version 2. 2007. Available at: http://www.census.gov/nlms. Accessed April 28, 2009.
- 16.Current Population Survey Design and methodology. Technical paper 63RV. US Census Bureau. 2002. Available at: http://www.census.gov/prod/2002pubs/tp63rv.pdf. Accessed April 28, 2009
- 17.National Center for Health Statistics Mortality technical appendix. Vital statistics system. Available at: http://www.cdc.gov/nchs/datawh/statab/pubd/ta.htm. Accessed April 28, 2009
- 18.Arias E, Schauman WS, Eschbach K, Sorlie PD, Backlund E. The validity of race and Hispanic origin reporting on death certificates in the United States. Vital Health Stat 2. 2008;Oct(148):1–24 [PubMed] [Google Scholar]
- 19.National Center for Health Statistics Instruction Manual, Part 4: Demographic Classification and Coding Instructions for Death Records. Hyattsville, MD: National Center for Health Statistics; 1993 [Google Scholar]
- 20.Ingram DD, Parker JD, Schenker N, et al. United States Census 2000 population with bridged race categories. Vital Health Stat 2. 2003;Sep(135):1–55 [PubMed] [Google Scholar]
- 21.Cresce AR, Schmidley AD, Ramirez RR. Identification of Hispanic Ethnicity in Census 2000: Analysis of Data Quality for the Question on Hispanic Origin. Population Division Working Paper; no. 75 Suitland, MD: US Census Bureau; 2004 [Google Scholar]
- 22.Office of Management and Budget Revisions to the standards for the classification of federal data on race and ethnicity. Fed Regist 1997;62FR58782-58790 [Google Scholar]
- 23.Anderson RN, Rosenberg HM. Age standardization of death rates: implementation of the year 2000 standard. Natl Vital Stat Rep. 1998;47(3):1–16 [PubMed] [Google Scholar]
- 24.Cresce AR, Ramirez RR. Analysis of General Hispanic Responses in Census 2000. Population Division Working Paper; no. 72 Suitland, MD: US Census Bureau; 2003 [Google Scholar]
- 25.Turra CM, Elo IT. The impact of salmon bias on the Hispanic mortality advantage: new evidence from social security data. Popul Res Policy Rev. 2008;27(5):515–530 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Preston SH, Elo IT, Stewart Q. Effects of age misreporting on mortality estimates at older ages. Popul Stud (NY). 1999;53(2):165–177 [Google Scholar]
- 27.Mulry M. Summary of Accuracy and Coverage Evaluation for Census 2000. Research Report Series, Statistics; no. 2006-3 Suitland, MD: US Census Bureau; 2006 [Google Scholar]