Revised estimating procedure

This report presents estimates of the Hispanic population based on a revised estimating procedure. The Bureau of the Census adjusted the Hispanic population totals from both the 1986 and 1987 CPS to conform with independently derived estimates of the Hispanic population. This general procedure was used on an experimental basis for the first time in the March 1982 CPS.1 The Census Bureau subsequently revised the methodology and used it to develop post-census estimates of Hispanics for 1983 through 1985.2 The procedure will be refined further as new data on births, deaths, emigration, and immigration become available.

Beginning with population estimates and CPS controls for January 1986, the Census Bureau made two major modifications in the methods used to produce national estimates for the population by age, sex, race, and Hispanic origin. The first change was an allowance for net undocumented immigration into the United States that had occurred since the 1980 census. This change added 200,000 persons per year to the estimate for the total population. The second change was an increase in the estimate of migration out of the United States by legal residents from 36,000 per year to 160,000. The net effect of these two changes was to add 76,000 persons per year to the estimate for the total population.3

Some undocumented immigrants from Spanish culture countries (approximately 1.4 million) were counted in the 1980 census.4 These undocumented immigrants were, therefore, reflected in the post-census independent estimates for Hispanics that were used for 1982 to 1985. These previous post-census estimates, however, included no allowance for net undocumented immigration that occurred after 1980 because there were no empirically based estimates available. More recent research has suggested

'See appendixes A and B, Current Population Reports, Series P-20, No. 396, Persons of Spanish Origin in the United States: March 1982. 2See U.S. Bureau of the Census report, Current Population Reports, Series P-20, No. 422, The Hispanic Population in the United States: March 1985, for a detailed explanation of the methodology used in 1983 through 1985.

3 Jeffrey S. Passel, "Changes in the Estimation Procedure in the Current Population Survey Beginning in January 1986," Employment and Earnings 33 (2 February 1986), pp. 7-10.

*Jeffrey S. Passel and Karen A. Woodrow, "Geographic Distribution of Undocumented Immigrants: Estimates of Undocumented Aliens Counted in the 1980 Census by State," International Migration Review 18 (Fall 1984), pp. 642-671.

that the overall undocumented population has grown annually by between 100,000 and 300,000 since 1980.5 About 70 percent of the undocumented population is estimated to be Hispanic. As a result of the inclusion of this component in the estimation procedure (instituted in January 1986), about 141,000 persons were added to the current independent estimates of the Hispanic population for each year since 1980.

Research over the last decade suggests that emigration of legal foreign-born residents from the United States was much higher than the figures being used. In order to avoid understating net immigration, these higher estimates of legal emigration were not incorporated into the international migration component of the post-census population estimates until an allowance for net undocumented immigration could be incorporated. The effect of the new figures for legal emigration is a decrease of about 31,000 per year in the estimated Hispanic population for years since 1980.

The net effect on the Hispanic population of the new figures for legal emigration and net undocumented immigration is an increase of about 110,000 per year. This increase amounts to about 18 percent of the 4.3-millionperson increase in the Hispanic population between the 1980 census and the March 1987 CPS.

Fluctuations in the data

The data in this report are based on sample surveys and are, therefore, subject to two types of errors: sampling and nonsampling. Sampling error arises from variations that occur by chance because a sample rather than a complete census is used to survey the population. Caution should be exercised in the use of survey data for small groups of the population (e.g., smaller Hispanic subgroups) because they are more likely to be affected by sampling variability. (For a more detailed explanation of sampling errors and variability, see the "Reliability of Estimates" section in appendix C.)

5 Jeffrey S. Passel and Karen A. Woodrow, "Change in Undocumented Alien Population in the United States, 1979-1983," International Migration Review 21 (Winter 1987), pp. 1304-1334, and Karen A. Woodrow, Jeffrey S. Passel, and Robert Warren, Preliminary Estimates of Undocumented Immigration to the United States, 1980-1986: Analysis of the June 1986 Current Population Survey. Paper presented at the 1987 annual meeting of the American Statistical Association, San Francisco, California, August 1987.

Random annual fluctuations in the CPS estimates for the total Hispanic population were noted in the past but have been reduced through the use of current independently derived estimates of the Hispanic population as controls. (See "Revised Estimating Procedure" section.) Estimates for the Hispanic subgroup populations, however, are not controlled to independent estimates. The estimates of the subgroup populations in the CPS are obtained through ratio estimation using the proportional distribution of the Hispanic subgroups obtained from the CPS supplement for the particular date. Consequently, annual fluctuations in the subgroups can be the result of sampling and nonsampling errors as well as real changes in the subgroups.

Fluctuations in the estimates of the Puerto Rican and Cuban populations have caused concem. Although increases in the Puerto Rican and Cuban populations were evident between the March 1982 CPS and the March 1987 CPS, fluctuations have been noted in the annual CPS estimates of these groups during this time period. For instance, the Puerto Rican population estimate obtained from the March 1985 CPS was 2,562,000 (± 109,000), compared with the March 1986 CPS estimate of 2,340,000 (± 105,000) and the 1987 CPS estimate of 2,284,000 (± 104,000). The Cuban population experienced a similar apparent change; the March 1986 and 1987 CPS estimates of Cubans are 1,003,000 (72,000) and 1,017,000 (± 72,000), respectively; for March 1985, it was 1,036,000 (± 73,000). (See table A-1.)

The apparent change in the Cuban population could be attributed to sampling variability since the 1985, 1986, and 1987 estimates of the Cuban population do not differ statistically (at the 90-percent level of confidence). The change in the Puerto Rican population, however, cannot be attributed entirely to sampling variability since the differences between the 1985, and 1986 and 1987 estimates are statistically significant at the same level. (The 1986 and 1987 estimates of Puerto Ricans are not statistically different at the 90-percent level of confidence.)

Several possible causes of the apparent decrease in the Puerto Rican population over the 1985-1987 period relating to CPS methodology have been investigated. In addition, a component population estimate of Puerto Ricans in the United States was developed and compared to the CPS estimate of Puerto Ricans.

The comparison of the component population estimate to the CPS estimate of Puerto Ricans suggests that the 1985 CPS estimate was probably too high and that the 1986 CPS estimate was probably more accurate. Research into CPS methodology suggests that a larger than expected number of Puerto Ricans classified as "movers" between November 1984 and March 1985 may have possibly resulted in an estimate that was too large (see section below entitled "Movers Between November 1984 and March 1985"). Also, as a result of the CPS redesign during this period, fewer Hispanic households were in sample, and, as expected, the average Hispanic weight increased.

However, even though the average weight for Puerto Rican households did not change, fewer of these households were in sample. Consequently, the Puerto Rican estimate for 1986 was significantly lower than the 1985 estimate.

Growth Rate

The Census Bureau's "component" population estimate was derived by adding births and net migrants, and subtracting deaths from an estimate of the 1980 census Puerto Rican civilian noninstitutional population. This analysis suggests that the 1985 CPS estimate of 2,562,000 Puerto Ricans was probably too high and that the 1986 estimate of 2,340,000 was probably more accurate. The implied growth rate if we accept a model growth rate which intersects the 1982 and 1986 CPS estimates, is 3.3 percent. (See table A-2.)

If, on the other hand, the 1985 CPS estimate is accepted as correct, then the Puerto Rican mainland population must have grown at about 7.4 percent per year since 1982. This rate of growth appears to fit the 1983, 1984, and 1985 CPS estimates well, and could be possible for a short period of time. However, a growth rate of 7.4 percent per year would be impossible to maintain over a long period of time in the absence of large sustained net migration from the island of Puerto Rico.

According to the National Center for Health Statistics, the Puerto Rican birth rate in 1985 was no more than 20 per 1,000 people, and the death rate was no less than 4 per 1,000, which would produce about a 41,000-person increase in that year. Based on the 1985 CPS estimate, the balance of the growth would have to be migrants from Puerto Rico. Between 1985 and 1986, 156,000 people would have had to migrate to the mainland to sustain the 7.4 percent growth rate between 1985 and 1986. Yet during the present decade according to passenger traffic data, no more than 46,000 Puerto Ricans migrated here in any one year. (See table A-3.)

If the 7.4 percent growth rate were maintained, the Puerto Rican mainland population would reach 3.7 million by 1990, but 1.1 million new migrants would be required to reach this total. This would leave only 2.6 million people on the island, compared to 3.3 million residing there in 1986. By the year 2000, the Puerto Rican population on the mainland would reach 7.8 million, but there would be no one left on the island.

6A 3.3-percent growth is a very rapid rate of growth, which, if sustained, would cause the population to double in size approximately every 20 years.

"Estimates of the net movement of persons between Puerto Rico and the United States mainland were based on the movement of passengers as reported by the Puerto Rico Planning Board.

In fact, there would be a shortfall of about a half million immigrants, so that the 7.8 million would not be attainable by the year 2000.

These figures suggest that the 7.4 percent growth rate is not realistic, and that it is more likely that the Puerto Rican mainland population is growing at about 3 percent per year. Even this rate of growth is not possible to sustain without continued migration from the island because about one-half of current growth is due to migration. Thus, it appears from this analysis that the 1985 estimate is too high, and that the 1986 estimate is probably in the correct range.

Movers Between November 1984 and March 1985

The March CPS supplement from which we derive our Hispanic estimates is composed of a double sample of Hispanic households. In addition to the Hispanic households selected to be interviewed in March, Hispanic households from the previous November CPS are reinterviewed. These Hispanic samples are referred to as the March and November samples, respectively. This effectively doubles the number of Hispanic households in sample in March and, therefore, increases the reliability of the Hispanic data. To inflate the sample to represent the Hispanic population of the United States, every household in the March and November samples, except those classified as "movers," receives half a weight. Persons classified as "movers" in the March sample receive a full weight, and those in the November sample receive a zero weight. Weights are factors by which the number of persons interviewed are inflated in order to represent their actual number in the population. For example, in the 1986 CPS, 2,201 Puerto Ricans were interviewed. Through the use of weights, this sample was inflated to 2.34 million, which represents the estimate of all Puerto Ricans on the U.S. mainland in that year.

All Hispanic households interviewed in March are asked if they lived at their current address in November of the previous year. If the answer is no, then persons in that particular household are classified as "movers." Some Hispanics interviewed in November are no longer at the same address and cannot be followed up for interview easily without great time or expense. Instead, "movers" from the March sample are given full weights so that they represent themselves, as well as the November movers. The underlying assumption is that the number of Hispanic movers from the March and November samples are the

same.

Between November 1984 and March 1985, the percentage of Puerto Rican "movers" was unusually high compared to other survey years and other Hispanic subgroups. It is suspected that most of the movers were from the March sample, thus, more Puerto Rican households received full weights than received zero weights resulting in an inflated estimate for the Puerto Rican population in March 1985.

A similar occurrence was noted for Central and South Americans in March 1986. It is possible that the majority of

Central and South American movers came from the March sample resulting in an unusually large estimate of Central and South Americans in March 1986.

The number of "movers" interviewed in the March sample may be greater or smaller than the actual number of November movers. In addition, the "movers" interviewed in March may not be of the same Hispanic-origin group as those movers from the November sample.

CPS Sample Redesign

Since the inception of the CPS in 1940, the sample has been redesigned several times. The most recent redesign began in April 1984. It was designed to improve the efficiency and reliability of the employment data, particularly at the State level. Prior to this date, the design was based solely on 1970 census data. By July 1985, the 1970 sample design was completely phased out by a new design based on 1980 census data.

As indicated in the report The Hispanic Population in the United States: March 1986 and 1987 (Advance Report), Series P-20, No. 416, preliminary research suggested that the new design had no effect on the fluctuations of the Puerto Rican estimates. Subsequent research has shown that the new design did not alter the proportion of Puerto Ricans among Hispanic respondents. It appears, however, that the redesign may have caused a decrease in the average weight of Puerto Rican respondents as a proportion of the total Hispanic population weights. This resulted in an estimate that was more comparable to the 1986 component estimate of Puerto Ricans in the United States using a 3-percent growth rate and, thus, appears to be a more realistic estimate.

In the new design, a smaller sample was selected from the Southwestern States, where the majority of the Hispanics are located. As expected, fewer Hispanic households were in sample, and the average weight per Hispanic household increased. In New York and New Jersey, where most Puerto Ricans live, the sample size effectively stayed the same. The average weight per Hispanic household did not change substantially, but the number of Puerto Rican households in sample unexpectedly decreased. As a result, the 1986 Puerto Rican estimate was lower than that of 1985.

Summary

First, a comparison of CPS estimates with independent estimates of the Puerto Rican population on the mainland suggests the 1985 CPS estimate of Puerto Ricans was too high and that the 1986 estimate is more realistic.

Second, it is possible that the unusually high percentage of Puerto Rican movers falling into the March sample might have caused an artificially high estimate of Puerto Ricans in the March 1985 CPS.