As noted, we have estimated the coverage of young Hispanics by the direct application of the demographic "accounting" equation to: (1) recorded Hispanic births, (2) estimated Hispanic deaths, and (3) recorded and estimated immigration data. We used three approaches, based on assumptions about these data, to generate three successive estimates regarding the 1990 census coverage of young Hispanics. In the first approach, we applied the demographic "accounting" equation (i.e. births minus deaths plus net immigration) to the given data. In the second approach, we increased the number of estimated intercensal Hispanic births used. And in the third approach, we used the increased number of Hispanic births, but slightly lowered the 1990 census allocation counts of young Hispanics by using long- form questionnaire (17 percent sample census) allocation rates instead of the short- form (100 percent census) allocation rates. For each of the three different approaches, the immigration figures were maintained the same. Later, our third set of coverage results were modified to account for the effect of internal migration (see Appendix and tables).
First Set of Coverage Results.
The survived intercensal Hispanic births (classified as Hispanic by origin of mother) plus the calculated net intercensal immigration of Hispanics age 0 to 9 years during the 1980-1990 decade yielded the cohorts age 0 to 4 years and 5 to 9 years at the 1990 census date. Comparing these cohorts with the actual 1990 census counts of Hispanics in those age groups showed that, in most of the 22 States of our study, the estimated young Hispanic groups were substantially smaller than the corresponding census counts of those groups, indicating that the 1990 census had substantially overcounted Hispanics age 0 to 9 years in most of the States !!! (see Table 1)
Although some States did show census undercounts of young Hispanics, these undercounts occurred mostly in typically "Hispanic" States (i.e. California: 1-percent undercount; New Jersey, 12-percent; and Texas, 2-percent). But most of the States - in this first round of results - showed a census overcount of young Hispanics. For example, Arkansas showed a 45 percent "overcount" of young Hispanics; Georgia, 74 percent; Indiana, 72 percent; Mississippi, 95 percent; and Nevada, 50 percent. Overall, the census counts showed an average of 4 percent overcount of Hispanics age 0 to 9 years in the U.S.
Intuitively we believe it improbable that the 1990 census could have overcounted young Hispanics - especially to the degree shown for the non-Hispanic States noted above. In fact, the Post Enumeration Survey (PES) showed an undercount of Hispanics in all States! Specifically, the PES showed that - on average - about 5 percent of Hispanic origin persons in the U.S. were missed in the census, and that for the 22 States of our study, there was an undercount of Hispanics which ranged between 3 to 7 percent for those States. Interestingly, and contrary to our first set of results, the PES showed that the 1990 census Hispanic undercount levels were particularly severe in the "Non-Hispanic" States of our study.
This disparity between our first set of coverage results and the PES strongly suggested that something was amiss in our coverage estimation scheme of young Hispanics. Could the 1990 census truly have overcounted young Hispanics in most of the 22 States of our study? Or are the overcounts simply aberrant results produced by underestimating Hispanic births in the implementation of our residual method? Are the census counts of Hispanic origin persons inflated by excessive 100 percent Hispanic-question allocation rates? Are there other factors operating here? The second and third round of results, shown below, will help answer some of these questions.
Some comments on the first round of results. - Since results show significant improbable overcounts of Hispanics in most "Non-Hispanic" States, and overcounts and some small undercounts in the "Hispanic" States, it is very likely we are underestimating our State intercensal Hispanic births and/or immigrants. If so, we should question our sources and our methods when using these items.
Second Set of Coverage Results
In our first set of results, the census "overcounts" of young Hispanics underscores the incomparability that sometimes occurs when two different statistical-collecting systems are used to collect data and produce results from a single conventional population. In this case: vital registration and the decennial census undoubtedly differently classify the same item: Hispanic origin. For instance, in the vital registration system of States births are classified as Hispanic - or other origin - according to the origin of the mother only; in the 1990 census, however, respondents are asked to report whether or not they are of Hispanic origin, and since the question elicits a subjective response, some children were probably reported as Hispanic solely because of the Hispanic origin of the father. Therefore, the classification of a person as Hispanic origin in the vital registration system is presently not equivalent to identification as Hispanic in our decennial census, and we must tread cautiously when interpreting the analytical results derived from combining data on the same characteristic (in this case: Hispanic origin) but identified in different statistical collecting-systems.
Theoretically, the resulting Hispanic cohort created from surviving intercensal Hispanic births and adding immigrants may not be comparable to the equivalent Hispanic cohort enumerated in the census, for example, if the intercensal births - and to a lesser extent: immigrants - are understated during the decade, the estimated "expected" Hispanic population age 0 to 9 years at census date would be expectedly smaller than the census-enumerated Hispanic population of that age group. Therefore, to provide for better comparability, we increased the registered Hispanic births for each individual State by 13.9 percent - which is an estimate of the proportion of Hispanic births by origin of father when the origin of the mother is non-Hispanic.6/ ( Table 2).
The significant aspect, then, of our second set of results was that we classified Hispanic births according to Hispanic origin of mother plus Hispanic origin of father (when the mother's origin was non-Hispanic).7/ The outcome was that the increased Hispanic births substantially reduced the overcount level in most of the "Non-Hispanic" States of our study and concurrently increased the undercount rates in the so-called "Hispanic" States. The overall U.S. net census coverage of young Hispanics in the 1990 census was now estimated as an undercount of about 8 percent instead of as an overcount of 4 percent estimated by our first approach (see Table 3), and the individual States poignantly reflected this overall transformation. For example: Arizona, instead of an overcount of Hispanics 0 to 9 years, now indicated an undercount of this group of about 2 percent. Colorado now showed a 1 percent undercount; Illinois, 11 percent; New Mexico, 6 percent. Evidently, the increase in the number of Hispanic births (i.e. now based on origin of mother or father) produced indicators of net 1990 census coverage of young Hispanics which were much more reasonable than those derived in the first set of results in which Hispanic births were based solely on origin of mother.
In summary, our second set of coverage results are more plausible than the first set (Table 6). Increasing the Hispanic births made our estimated young Hispanic population base more comparable to that of the census, in which children may be identified as Hispanic according to origin of mother or father. This second set of results underscores the importance of using comparable data in different statistical collecting systems.
Third Set of Coverage Results
Although increasing the number of intercensal Hispanic births gave more realistic census coverage rates of young Hispanics, for a few States the rates still appear unreasonable. For example, in the second set of results some "Non- Hispanic" States still showed large overcounts and some "Hispanic" States still trivially small undercounts. Therefore, either we were still understating the processes of intercensal change (e.g. births and immigration) for Hispanics age 0 to 9 years during the 1980-1990 intercensal period, or the 1990 census counts of that age group were in some way artificially overinflated, possibly through some aspect of census-questionnaire processing.
During the 1990 decennial census, questions that were left incomplete, had inconsistent or unreasonable responses, or simply were left entirely unanswered, had responses "imputed" by computer. These "imputed" responses were called imputations or allocations. The imputation rate (i.e. allocation rate) for the 1990 census Hispanic origin question of persons age 0 to 9 years was 7.2 percent. Generally, this was much higher than for other short-form census questions. In the short-form questionnaire, over three-quarters of a million persons age 0 to 9 years were allocated as "Hispanic origin"; but in the 1990 census long-form (i.e. sample census) questionnaires, about half that number of persons were allocated as Hispanic.
We suspect that the number of response-imputations of "Hispanic origin" inserted in the short-form questionnaires during 1990 census-processing was excessive, and generally we believe that the greater number of questions and, therefore responses available for editing the census long-from questionnaire offers a more correct selection of response-allocations than the short-form questionnaire. Hence, the number of allocations of "Hispanic" origin inserted in the long-form questionnaires should be more credible than the number from the short-form questionnaires. Since in our study we compared the residual-method estimates of young-Hispanics with the 1990 census 100 percent counts of these persons, the noticeable census excess (i.e. overcount) for certain States could have been caused - at least partly - by an over-allocation of young-Hispanic origin persons in the census. (Table 4)
Therefore, in an attempt to get more accurate and realistic estimates of young Hispanic census coverage, we derived yet a third set of results by adjusting the 1990 census young Hispanic counts downward by using the long-form allocation rates of "Hispanic origin" responses instead of the "inflated" short-form allocation rates used to produce the 1990 census (100-percent) Hispanic counts. (Table 5). The rationale for our action was to try again to legitimately reduce the inexplicably high "overcounts" of Hispanics in the NonHispanic States of our study. At this point, we should note that in these States, where the concept of Hispanic origin is not as well recognized as in other areas, we would expect more non-responses in the Hispanic origin question and thus more erroneous allocations ("false positives) of "Hispanic origin" than in other States. In any case, we believe that in most States the imputation rates of "Hispanic origin", based on long-form questionnaire responses were probably more correct than the rates based on short form questionnaires responses.
For our third and final set of results, therefore, we used: (1) the same intercensal immigration factor as in the other two sets of results (Table 5); (2) Hispanic births by origin of father (mother non-Hispanic) in addition to Hispanic births by origin of mother; and (3) census long-form allocation rates instead of short- form allocation rates. With these modifications, this final set of results generated a more plausible (higher) indication of undercount of young Hispanics in the "Hispanic" States and substantially lower overcounts in the "Non-Hispanic" States. For the entire U.S., the net 1990 census undercount rate of Hispanics age 0 to 9 years then became 11.2 percent, or over half-a-million persons (Table 5).
The census coverage errors for young Hispanics now seemed more plausible because they showed normally expected undercounts in most of the heavily "Hispanic" States and somewhat reduced overcounts in the "NonHispanic" States. Prior to simple adjustment due to internal migration (shown in the Appendix), in this third set of results "Hispanic" States such as New York and New Jersey showed more than a 25-percent undercount of young Hispanic children;8/ Texas, Illinois, and California about 15 percent undercount each; Hawaii, over 10 percent; New Mexico, 8 percent; Colorado, 4 percent; Arizona, 3 percent; and Florida, 1 percent.9/ Overall, the implied young Hispanic net census undercount for the combined "Hispanic" States of our study was 15.5 percent.
Most of the "Non-Hispanic" States, however, continued to show overcounts of young Hispanics in the census, and for these States combined the net overcount rate reached 18.7 percent. Particularly high levels were noted for Mississippi, which showed an overcount of over 55 percent followed by Georgia, over 45 percent; Maine, over 40 percent; Nevada and Kansas over 25 percent, etc. (Figure 1). These "overcounts" are still rationally unexplainable. Could they be caused, at least in part, by mixing two different identification modes for Hispanics in our estimating-scheme, namely: administrative data (such as: the vital registration of Hispanic births, etc.) and the decennial census responses to the subjective question used to identify persons as: "Hispanic" origin?
As noted previously, another factor to consider in determining the magnitude of the coverage rates is the internal migration effect. However, for the "NonHispanic" States, this effect does not reduce the overcounts significantly.
The original goal of this study was to analytically determine the net census coverage error of Hispanics age 0 to 9 years enumerated in the 1990 census. To do this, we survived the 1980 to 1990 occurring Hispanic births and calculated the net Hispanic immigration during that period. To these cohorts we applied the "demographic accounting equation" to calculate the residual between the resulting survived intercensal Hispanic cohort age 0 to 9 years and the enumerated 1990 census population of that age. This residual ostensibly represented an estimate of net 1990 census coverage error for Hispanics age 0 to 9 years. At first, the unreasonable estimates of overcoverage from our first set of results persuaded us to use: (1) Hispanic births also based on Hispanic origin of father, and (2) 1990 census sample-questionnaire allocations of "Hispanic origin" instead of the 100 percent allocations. These refinements narrowed the unreasonable gap between each of the estimates and the census counts and improved our final net census coverage rates. In general, the "high" undercount rates shown for some "Hispanic" States in our third set of results (see Table 6) may not be unreasonable. But the overcounts in the "Non-Hispanic" States, although reduced somewhat, are still highly improbable. In any case, we can still make some very positive inferences from this study. (Figure 2).
Firstly, we must stress that the undercount levels shown for the "Hispanic" States are simply plausible "indicators" of net census coverage error of young Hispanics rather than direct "estimates" of coverage. Still, these "indicators" can add to our understanding of the 1990 census coverage estimates of the overall U.S. Hispanic population and lead the way to developing analytic measures of coverage in the 2000 census.
Secondly, it appears that the intercensal Hispanic births tabulated by the National Center for Health Statistics and used by the Census Bureau in its estimates and projections programs are understated relative to their use with the reporting of Hispanic origin in the census. We believe it imperative that, to improve the comparability of Hispanic origin data collected by these two agencies, Hispanic births be classified according to origin of father as well as by origin of mother. Our second set of results alone shows the enormous impact that this extension would have in estimating overall census Hispanic population coverage.
Thirdly, in an effort to reduce the unexpected high census overcounts of young Hispanics in most of the "NonHispanic" States, we used the sample-census (long-form) questionnaire allocation rates of "Hispanic" origin. As noted above, the short-form rates were about double the long-form rates and our contention is that the former probably overinflated the Hispanic origin counts in some States; /10 although this overinflation factor was probably more than compensated for, at least in the "Hispanic" States, by the anticipated census undercount of Hispanic persons.
In general, then, for selected States (particularly those with "large" Hispanic populations) we believe we have derived reasonable indicators of 1990 census undercount levels of Hispanic origin persons age 0 to 9 years. Furthermore, a by- product from this study - which emerges conspicuously - is the realization that we should examine closely the appropriateness of the annual Hispanic fertility and migration rates used in the Census Bureau's estimates and projections programs, and in the analysis and estimation of the net census coverage error of our decennial censuses./11