Rethinking the Hispanic Paradox: Death Rates and Life Expectancy for US Non-Hispanic White and Hispanic Populations

Abstract

Objectives. We examined the “Hispanic paradox,” whereby persons of Hispanic origin seemed to experience lower mortality than the non-Hispanic White population. This paradox coincided with a change from the classification of deaths and population by Spanish surname to the use of Hispanic-origin questions in the census and vital statistics.

Methods. To estimate US Hispanic and non-Hispanic White mortality, we applied a familiar relation between death rates for population subgroups to Hispanic and non-Hispanic White population death rates. We calculated age-specific death rates for the Hispanic population and the non-Hispanic White population and computed life tables for each.

Result. For Texas between 1980 (surname) and 1990 (origin), the change in Hispanic deaths in persons aged 65 years or older was only half as great as the change in population size, implying a relative omission of 15% to 20% of deaths. By a different approach, the life tables for the US Hispanic and non-Hispanic White populations pointed to a similar omission.

Conclusions. There is no “Hispanic paradox.” The Hispanic paradox described in past research derives from inconsistencies in counts of Hispanic-origin deaths and populations.

Since at least 1930, the US Census Bureau has tried different methods for identifying the Hispanic population in the United States. Some of these methods have been reasonably successful, although only 1 (classification by Spanish or non-Spanish surname) has so far been capable of replication in classifying vital records and persons enumerated in the census. The problems of consistency between census data and vital statistics for the Hispanic population of the southwestern states under different ascertainment conventions have been reviewed rather extensively.1^–3

After the 1980 census, as had been done after each census beginning in 1950, the Census Bureau coded the surnames of respondents in 5 southwestern states (Arizona, California, Colorado, New Mexico, Texas) according to whether the names were of Spanish or non-Spanish origin, using a surname list developed over a number of years.4^,5 Several of the states, including Texas, used the list, or a linguistically based computer program that gave comparable results, to classify surnames of infants and decedents. Vital statistics for these states were thus defined consistently with census data for the Hispanic and non-Hispanic populations. Several researchers6^–8 used these data to create life tables for the Spanish-surname and White non-Spanish-surname populations between 1950 and 1980. The Texas Department of Health also published Spanish-surname life tables in some years, and a number of life tables have been prepared for local populations, such as San Antonio9^,10 and Houston.11

Starting in 1980, life tables for both Hispanic- and non-Hispanic-origin populations have been calculated from self- or informant-identified Hispanic origin in censuses, and from the origin questions on death certificates. The latter, however, use different wording from that of censuses and are typically filled in by medical or funeral home personnel. These individuals may have no close ties to the families of the deceased and no means of confirming that their designation of ethnicity on the death certificate agrees with the designation likely to have been given for the decedent in the previous census. It is also relevant that death certificates are legal documents that are used in ways the census is not, placing explicit responsibility on physicians and funeral directors to use informed judgment. The census places no such burden on respondents, who answer the origin question according to their self-perception or preference.

In an investigation of the consistency of death data and census data, Sullivan et al.8 found 1980 life expectancies of 69.6 years at birth for White Spanish-surnamed males and 77.1 years for White Spanish-surnamed females, but 71.0 years for Spanish-origin males and 78.7 years for Spanish-origin females when origin questions in death certificates and the census were used. The former are less than, and the latter greater than, the life expectancies they derived for White non–Spanish-surnamed males (70.2 years) and females (78.1 years). Their results imply a remarkably high life expectancy for the population that is of Spanish origin but not Spanish surnamed, or else a serious discrepancy between the 2 sets of identifiers as applied to populations and to deaths.

The higher life expectancies and lower Hispanic death rates that are observed when data from Hispanic-origin questions are used to classify people have given rise to the claim of a “Hispanic paradox.” That is, despite lower average incomes and educational levels, the Hispanic population appears to have higher life expectancies than the more advantaged non-Hispanic White population. In a review of the literature, Franzini et al.12 found the balance of the evidence for the paradox convincing, but such evidence is harder to find in studies with complete death records13^,14 than in studies in which it is less evident that all deaths have been identified.1^,15^,16 Problems also arise because the term Hispanic is not itself clearly or consistently defined, or definable,17 requiring at least some substitution of faith for evidence in its usage.

We developed new estimates of US Hispanic death rates and life expectancies for 1990 and 2000 that are adjusted for the evident omission of deaths relative to the population as suggested by Sullivan et al.8 Our results are inconsistent with the existence of a Hispanic paradox. We do not assert that our own estimates are accurate, but we believe them to be better than measures based on presently available Hispanic-origin death and population data.

METHODS

Our initial task was simply to compare rates of change in the Hispanic and non-Hispanic population and counts of deaths from 1980 to 2000.

To develop estimates of US Hispanic and non-Hispanic White mortality from what may be flawed data, we exploited a familiar relation among rates. If we are given 2 population subgroups, for example, Hispanic and non-Hispanic, whose death rates differ, but the death rate of each subgroup is similar across regions, we should expect the death rates for the total population in 2 regions to differ strictly as a consequence of the different proportions of the 2 groups in each place. A familiar example is infant mortality rates: rates among states vary in a consistent fashion with the proportion of state births that are Black, owing to the higher mortality seen among Black infants throughout the United States.

Applying the same logic to Hispanic and non-Hispanic White population death rates, we created 2 subpopulations (we used Texas and the United States excluding Texas) whose Hispanic proportions differed, and defined the following:

P_{US N} =Non-Hispanic White population of the United States excluding Texas

P_{US H} =Hispanic population of the United States excluding Texas

P_{TX N} =Non-Hispanic White population of Texas

P_{TX H} =Hispanic population of Texas

D_US =Deaths for the combined populations (United States excluding Texas) P_{US N} and P_{US H}

D_TX =Deaths for the combined populations (Texas) P_{TX N} and P_{TX H}

M_US = D_US /(P_{US N} + P_{US H} ) = death rate of non-Hispanic White population and Hispanic population of the United States excluding Texas

M_TX = D_TX /(P_{TX N} + P_{TX H} ) = death rate of non-Hispanic White population and Hispanic population of Texas

If death rates for the Hispanic populations of the 2 regions we were using were essentially the same, and if those for the non-Hispanic White population were essentially the same, then for the United States as a whole we could find death rates for each ethnicity where

M̂_N = death rate to be estimated for the non-Hispanic White population of both areas

M̂_H =death rate to be estimated for the Hispanic population of both areas,

by using the following equalities:

(1)

(2)

Rearranging terms in these 2 expressions and solving for M̂ _N and M̂_H gives

(3)

(4)

Solving the expressions for each age interval gave us sets of age-specific death rates for the 2 ethnicities, which could be compared directly or through life tables. We could also find expected deaths by age for each population, which would correctly add to total deaths but would differ at each age from the number of deaths actually reported for each group if some deaths had been misassigned in the source data. Because the reported deaths were not used in the calculations, the difference between reported and expected deaths was a measure of the extent of misclassification of deaths relative to population for the 2 areas and 2 ethnicities.

For the results of the calculations to be meaningful, we required that death rates be essentially the same in the populations P_{US N} and P_{TX N} , and in the populations P_{US H} and P_{TX H} , so that differences between M_US and M_TX would be wholly because of differences in the proportions of the 2 subgroups in each area. We also required that the subgroups be enumerated by the same criteria, because errors of misclassification were not addressable by our formulas, and that deaths not be misclassified as occurring to a population other than those we were working with. These assumptions were satisfied by our use of Hispanic and non-Hispanic White populations but did not extend to finer distinctions, because national origin was not reported for a substantial part of the Hispanic population in the data sources we used (US Censuses, and national and Texas vital statistics).18^–22

In constructing death rates and mortality tables for non-Hispanic White and Hispanic populations, we opted to work around 2 difficulties in the source data. First, in the group aged 65 years or older, the Hispanic-origin population remained younger than the non-Hispanic population, and there may have been greater age exaggeration in the reported data.23^,24 Both those would produce lower age-specific death rates and better life expectancies for the Hispanic population than they actually had. To minimize distortion in the life tables from these differences, we used Keyfitz-Frauenthal estimates of the mortality probabilities _nq_x , which correct for age composition effects in the tables.25^–27 At the terminal age (75 years or older) we substituted National Center for Health Statistics (NCHS) White male and White female life expectancies28^,29 for all population groups. The NCHS values were 12.0 years for women and 9.4 years for men aged 75 years in 1990, and 12.3 years for women and 10.1 years for men aged 75 years in 2000.

Second, although we included age-specific death rates for age 0 years in the tables for completeness, we caution readers that the rates at age 0 are inaccurate owing to under-enumeration of infants in the censuses, which provided denominators. When we contrast the 2000 census population with 2000 registered births adjusted for mortality,30 the census omission was about 4% for non-Hispanic White infants and 5% for Hispanic infants if the mother’s reported ethnicity was consistent in the 2 sources, which we could not test. With this correction, or with the use of linked birth and death files,31 age-specific death rates for 2000 for the 2 ethnicities would be approximately 0.00570 and 0.00560, respectively. These values are markedly more similar than those we display in our tables and are reasonable given that neither population had a high incidence of low or very low birth weights. We did not have information comparable to 2000 for 1980 and 1990, or appropriate correction factors at the state level for 2000. Consequently, although we report death rates for this age interval, we have little confidence in them. Their use distorts life expectancies, but only trivially.

RESULTS

When we examined population and mortality changes for Texas over the 1980–1990 decade, we found an increase in the Hispanic population aged 65 years and older from 1980 (when surnames were used to classify) to 1990 (when origin questions were used to classify) of approximately 75% (Table 1 ▶). In sharp contrast, when the same origin identifiers were used, the increase in deaths among Hispanics at these ages was 43%, or roughly half as great (Table 1 ▶). In Texas, from 1990 to 2000, when reported Hispanic origin was used to classify at both points, the increase in both the Hispanic population aged 65 years or older and deaths in Hispanic persons aged 65 years or older was about 50%.18^,19 In effect, there is no repeat of the previous difference in percentage change in population and deaths. When one combines the 2 periods from 1980 (classification by surname) to 2000 (classification by origin), the increase in the Hispanic population aged 65 years or older was 164%, and the increase in deaths in Hispanic persons aged 65 years or older was 117%. Corresponding to the change in ascertainment of Hispanic origin, the Hispanic death rate for those aged 65 years or older fell from 0.0485 in 1980 to 0.0396 in 1990 and 0.0399 in 2000.

TABLE 1—

Populations, Deaths, and Death Rates Among Non-Hispanic White and Hispanic Texans Aged 65 Years or Older, by Source and Method of Ascertainment, and Adjusted Estimates Assuming No Change in Hispanic Death Rates After 1980: 1980, 1990, and 2000

	Non-Hispanic White			Hispanic
	Men	Women	Total	Men	Women	Total
Texas 1980 (surname)
Population	434 544	655 522	1 090 066	59 403	71 900	131 303
Deaths	26 592	27 299	53 891	3 378	2 984	6 362
Death rate	0.06120	0.04164	0.04944	0.05687	0.04150	0.04845
Texas 1990 (origin)
Population	535 980	794 284	1 330 264	97 830	131 530	229 360
Deaths	31 247	35 086	66 333	4 556	4 516	9 072
Death rate	0.05830	0.04417	0.04986	0.04657	0.03433	0.03955
Texas 1980–1990 changes
Population	1.233	1.212	1.220	1.647	1.829	1.747
Deaths	1.175	1.285	1.231	1.349	1.513	1.426
Texas 1990 (resetting 1990 Hispanic death rate to 1980 death rate)
Population	535 980	794 284	1 330 264	97 830	131 530	229 360
Deaths	30 239	34 144	64 383	5 564	5 458	11 022
Death rate	0.05642	0.04557	0.04840	0.05687	0.04150	0.04806
Texas 1980–1990 changes (resetting 1990 Hispanic death rate to 1980 death rate)
Population	1.233	1.212	1.220	1.647	1.829	1.747
Deaths	1.137	1.251	1.195	1.647	1.829	1.747
Texas 2000 (origin)
Population	648 050	902 693	1 550 743	146 420	200 216	346 636
Deaths	36 002	45 721	81 723	6 563	7 255	13 818
Death rate	0.05555	0.05065	0.05270	0.04482	0.03624	0.03986
Texas 1990–2000 changes
Population	1.209	1.136	1.166	1.497	1.522	1.511
Deaths	1.152	1.303	1.232	1.441	1.607	1.523
Texas 1980–2000 changes
Population	1.491	1.377	1.423	2.465	2.785	2.640
Deaths	1.354	1.674	1.516	1.943	2.431	2.172
Texas 2000 (resetting 2000 Hispanic death rate to 1980 death rate)
Population	648 050	902 693	1 550 743	146 420	200 216	346 636
Deaths	34 238	44 667	78 905	8 327	8 309	16 636
Death rate	0.05283	0.04948	0.05088	0.05687	0.04150	0.04799
Texas 1980–2000 changes (resetting 2000 Hispanic death rate to 1980 death rate)
Population	1.491	1.377	1.423	2.465	2.785	2.640
Deaths	1.288	1.624	1.464	2.465	2.785	2.615

Nothing similar occurred in the larger non-Hispanic White population in Texas. The 1980 (White non-Spanish surname) to 1990 (White non-Hispanic origin) changes in population and deaths in persons aged 65 and older in Texas were 22% and 23%, respectively. These figures would change little, becoming 22% and 19%, respectively, if we were to transfer some 1990 deaths from the non-Hispanic to the Hispanic population to equalize 1980 and 1990 Hispanic death rates in those aged 65 years or older (to 0.05687 for males and 0.04150 for females in both years), in essence treating the improvement in the Hispanic death rates as an artifact of misassigning deaths. From 1990 (classified by origin) to 2000 (classified by origin), the increases for the non-Hispanic White population were 17% in population and 23% in deaths.

When the periods were combined, from 1980 (classified by surname) to 2000 (classified by origin), the increases in the Texas non-Hispanic White population and deaths in those aged 65 years or older were 42% and 52%, respectively. If we again transferred deaths from the non-Hispanic White to the Hispanic population to yield 2000 Hispanic death rates of 0.05687 for males and 0.04150 for females aged 65 years or older, equal to the 1980 death rates, the increases in the non-Hispanic White population aged 65 years or older and deaths in those aged 65 years or older became 42% and 46%, respectively. The adjustment fixed the increase in the Hispanic-origin population to 164% and the increase in deaths to 161%, compared with 164% for population and 117% for deaths in the unadjusted series. The shortfall in deaths relative to population implied in the adjustment was 18% in 1990 and 17% in 2000 in the Hispanic-origin population; the corresponding overcounts in the non-Hispanic White population were 3% in both years.

In using the crude 10- and 20-year changes, we introduced some distortion in comparisons between death rates, as the age composition of the population aged 65 years or older is not constant, and mortality has improved, but these errors are of small magnitude and do not alter the fundamental impression the numbers give: the change from the use of surnames to self-designation or informant designation has not been benign.

Table 2 ▶ summarizes 1980 life expectancies at birth for infants born in Texas who have Spanish surnames, and for those born in Texas and the United States excluding Texas for 1990 and 2000, based on Hispanic origin. Reestimated life expectancies are also shown for 1990 and 2000. Detailed results are presented in Tables 3 ▶ and 4 ▶. From Table 2 ▶, one can see that for girls the effect of reestimation was to narrow the apparent Hispanic advantage in life expectancy in 1990 for the United States excluding Texas from about 1.3 years to approximately 0.2 years, essentially the same as the Texas estimate. Neither the 1980 nor the reestimated 2000 tables showed Hispanic-origin female life expectancy to be above non-Hispanic White life expectancy. For boys, the reestimated tables for 1990 show a Hispanic mortality disadvantage of about 3 years, which requires comment (see later this column), compared with a 0.3–0.5 year Hispanic advantage in Texas and the United States excluding Texas tables derived from Hispanic-origin data. For 2000, the Hispanic mortality advantage in the source rates was similar to that of 1990, whereas in the reestimated series, life expectancy was lower for both Hispanic males and Hispanic females than for the non-Hispanic White population, by approximately 0.5 year for females and 1.1 years for males. Better coverage in the 2000 census may be a factor in the differences from 1990, but the small gain in Hispanic female life expectancy during the decade in the reestimated results, and the exceptional gain in Hispanic male life expectancy, point to the likelihood of more complex errors in the data or in our assumptions.

TABLE 2—

Life Expectancy at Birth for Non-Hispanic Whites and the Hispanic-Origin Population for Selected Years, From Reported Population and Death Distributions, and Reestimations From Differences in Mortality

	Girls		Boys
Year and Source	Non-Hispanic White	Hispanic	Non-Hispanic White	Hispanic
1980
1980 Texas (surname)	78.5	77.6	70.6	69.8
1990
1990 United States excluding Texas (origin)	79.1	80.4	72.5	73.0
1990 Texas (origin)	79.1	79.3	71.9	72.2
1990 Reestimated	79.2	79.4	72.8	69.6
2000
2000 United States excluding Texas (origin)	79.9	81.4	74.6	75.9
2000 Texas (origin)	79.7	80.4	74.4	74.8
2000 Reestimated	80.1	79.6	74.8	73.7

TABLE 3—

Age-Specific Death Rates for Non-Hispanic White and Hispanic Populations, by Gender: United States Excluding Texas and Texas, 1990

	United States		United States Excluding Texas		Texas		Reestimated for United States
	Non-Hispanic White	Hispanic	Non-Hispanic White	Hispanic	Non-Hispanic White	Hispanic	Non-Hispanic White	Hispanic
Women
Age, y
0.0	0.00741	0.00879	0.00743	0.00866	0.00710	0.00941	0.00737	0.00900
1.0	0.00035	0.00044	0.00035	0.00046	0.00034	0.00036	0.00038	0.00031
5.0	0.00018	0.00017	0.00018	0.00017	0.00019	0.00019	0.00017	0.00023
15.0	0.00047	0.00043	0.00047	0.00044	0.00052	0.00037	0.00047	0.00046
25.0	0.00061	0.00066	0.00061	0.00069	0.00064	0.00055	0.00062	0.00061
35.0	0.00117	0.00113	0.00117	0.00113	0.00116	0.00112	0.00118	0.00108
45.0	0.00310	0.00252	0.00310	0.00246	0.00298	0.00277	0.00312	0.00228
55.0	0.00838	0.00622	0.00838	0.00587	0.00835	0.00768	0.00825	0.00808
65.0	0.01940	0.01429	0.01941	0.01337	0.01915	0.01812	0.01924	0.01766
75.0	0.07363	0.04810	0.07367	0.04553	0.07292	0.05881	0.07334	0.05584
Ratioa	1.005	0.879	1.006	0.844	0.996	1.026	…	…
Men
Age, y
0.0	0.00973	0.01064	0.00973	0.01053	0.00967	0.01119	0.00958	0.01133
1.0	0.00044	0.00052	0.00044	0.00050	0.00047	0.00063	0.00039	0.00076
5.0	0.00027	0.00026	0.00026	0.00026	0.00030	0.00026	0.00025	0.00033
15.0	0.00122	0.00163	0.00121	0.00162	0.00139	0.00168	0.00113	0.00213
25.0	0.00167	0.00238	0.00165	0.00241	0.00199	0.00224	0.00153	0.00332
35.0	0.00257	0.00353	0.00256	0.00367	0.00277	0.00290	0.00258	0.00350
45.0	0.00544	0.00541	0.00542	0.00535	0.00575	0.00572	0.00525	0.00763
55.0	0.01484	0.01140	0.01481	0.01096	0.01533	0.01331	0.01444	0.01724
65.0	0.03425	0.02398	0.03420	0.02272	0.03510	0.02904	0.03362	0.03707
75.0	0.09732	0.06740	0.09722	0.06446	0.09917	0.07852	0.09612	0.09560
Ratioa	1.021	0.729	1.019	0.715	1.059	0.789	…	…

^aRatio of reported to adjusted deaths.

TABLE 4—

Age-Specific Death Rates for Non-Hispanic White and Hispanic Populations, by Gender: United States Excluding Texas and Texas, 2000

	United States		United States Excluding Texas		Texas		Reestimated Rates for United States
	Non-Hispanic White	Hispanic	Non-Hispanic White	Hispanic	Non-Hispanic White	Hispanic	Non-Hispanic White	Hispanic
Women
Age, y
0.0	0.00511	0.00524	0.00518	0.00535	0.00416	0.00482	0.00578	0.00332
1.0	0.00024	0.00028	0.00024	0.00026	0.00037	0.00034	0.00015	0.00055
5.0	0.00014	0.00013	0.00014	0.00013	0.00015	0.00015	0.00013	0.00017
15.0	0.00043	0.00032	0.00042	0.00031	0.00052	0.00036	0.00037	0.00055
25.0	0.00057	0.00043	0.00056	0.00042	0.00066	0.00049	0.00050	0.00073
35.0	0.00128	0.00100	0.00128	0.00101	0.00137	0.00098	0.00124	0.00125
45.0	0.00285	0.00224	0.00285	0.00216	0.00293	0.00256	0.00276	0.00303
55.0	0.00742	0.00548	0.00742	0.00517	0.00739	0.00682	0.00726	0.00724
65.0	0.01891	0.01423	0.01890	0.01354	0.01901	0.01701	0.01859	0.01856
75.0	0.07769	0.05521	0.07749	0.05280	0.08152	0.06491	0.07581	0.09737
Ratioa	1.025	0.658	1.023	0.632	1.064	0.763	. . .	. . .
Men
Age, y
0.0	0.00636	0.00607	0.00644	0.00615	0.00510	0.00574	0.00712	0.00385
1.0	0.00032	0.00032	0.00032	0.00031	0.00037	0.00033	0.00030	0.00040
5.0	0.00020	0.00018	0.00020	0.00017	0.00022	0.00020	0.00018	0.00024
15.0	0.00103	0.00108	0.00103	0.00105	0.00115	0.00120	0.00092	0.00147
25.0	0.00123	0.00120	0.00123	0.00117	0.00126	0.00135	0.00115	0.00150
35.0	0.00234	0.00211	0.00234	0.00208	0.00236	0.00225	0.00229	0.00238
45.0	0.00498	0.00439	0.00496	0.00433	0.00524	0.00462	0.00482	0.00579
55.0	0.01171	0.00966	0.01169	0.00921	0.01207	0.01153	0.01131	0.01423
65.0	0.02931	0.02288	0.02930	0.02213	0.02944	0.02591	0.02896	0.02786
75.0	0.09162	0.06904	0.09154	0.06650	0.09314	0.07873	0.09052	0.09203
Ratioa	1.018	0.784	1.017	0.761	1.037	0.876	. . .	. . .

^aRatio of reported to adjusted deaths.

What is distinct in the 1990 Hispanic and non-Hispanic female tables is that age-specific death rates and life expectancies were virtually identical for Texas and for the United States excluding Texas (Tables 2 ▶ and 3 ▶), despite the much higher proportion of the Texas population that was of Hispanic origin (30% in Texas vs 6% in the rest of the United States). Those for 2000 (Tables 2 ▶ and 4 ▶) differed only marginally, but in the direction we expected, life expectancy being slightly lower for Texas. Taking both years together, we could not be confident of the direction of differences between Hispanic and Non-Hispanic females, but the error is more likely to belong with the 1990 estimates. Using the life table approximation e_t = P_t /D_t =1/M_t between the death rate for the terminal age interval M_t and the life expectancy e_t, we calculated for age 75 years in 1990 the life expectancies 1/0.04810 = 21 years, 1/0.04553 = 22 years, and 1/0.05584 = 18 years for the Hispanic female population in Table 3 ▶, far above the NCHS estimate of 12 years, and above reason. The comparable estimates for 2000 (Table 4 ▶) were 1/0.05280 = 19 years, 1/0.06491 = 15 years, and 1/0.09737 = 10 years, closer to the 12.3 years in NCHS 2000 life tables. We reached the tentative conclusion that Hispanic female life expectancy was probably marginally below non-Hispanic White female life expectancy, in line with the reestimated series of Table 4 ▶. For males, all series pointed to lower Hispanic than non-Hispanic White life expectancy. The 2000 differences (Tables 2 ▶ and 4 ▶) may be approximately correct.

What underlay all these results was a consistent pattern of higher age-specific death rates in Texas than in the rest of the United States for the combined Hispanic and non-Hispanic White populations, holding for both sexes and across most ages, despite the mortality advantage Texas’s substantial Hispanic population should have given it.

We have tested our results for sensitivity to our source data, first by excluding California from the analysis, a modification that was possible with 2000 data (tables not shown). With the exclusion of California, which had the effect of testing our findings under more relaxed assumptions about the closeness of Texas and US mortality rates, the ethnic differences narrowed but persisted. The differences also persisted when we substituted the US Mexican and other Hispanic populations (i.e., Hispanic omitting Cuban and Puerto Rican populations) in place of all Hispanics, to more closely match the US Hispanic and Texas Hispanic populations.

DISCUSSION

Apart from Hispanic-origin females in 1990, the reestimated death rates in Tables 3 ▶ and 4 ▶ required a substantial transfer of deaths from the non-Hispanic White population to the Hispanic population, or a substantial transfer of population from Hispanic origin to non-Hispanic White. Except for 1990 females, the transfers were on the order of 15% to 25% for the Texas series and occurred at all ages, in line with the magnitude of the transfers in those aged 65 years or older suggested by Table 1 ▶. That consistency served as a check of the methodology underlying Tables 3 ▶ and 4 ▶, as we derived Table 1 ▶ independently. The transfers conveyed only small adjustments to the estimated non-Hispanic White population or deaths—no more than a 2% shift for the United States excluding Texas and 5% for Texas in 2000, as the Hispanic population aged 65 years or older, in which mortality is concentrated, was relatively small. The large errors we found in Hispanic mortality estimates arose through very minor discrepancies between the census and vital statistics data for the non-Hispanic White population.

Also, except for the problems in infancy and old age we have identified, the reestimated death rates were unremarkable. Indeed, if they had been generated from tables using either surname or origin data it is unlikely they would have been challenged.

By our methodologies, no Hispanic paradox exists. Life expectancies were apparently close for non-Hispanic White and Hispanic females, but less close for non-Hispanic White and Hispanic males. Our results could be made more robust with adjustment for the obvious errors at age 0 years and better insight into survival at ages 75 years or older, in which group reported ages in the Hispanic population are suspect.

What we need more, however, is a better starting point. The apparent discrepancy between Hispanic-origin information given in the census and on death certificates would not arise in census and mortality tables based on surnames, as was last done after the 1980 census, and urged by Rosenwaike and Bradshaw.2 A surname listing exists, and the cost of running it against the 2000 census and 2000 (or 1999–2001) deaths should not be great. Without it, we will not derive reasonable Hispanic survival estimates from what should be our 2 best sources. At the very least, the questions on Hispanic origin on vital records should be made entirely consistent with those in the decennial census, and appropriate efforts should be made to collect origin information. The Hispanic population will continue to increase in importance in the United States, and it is imperative that we be able to assess as accurately as possible its levels and trends in mortality.