The alternative methods for dealing with nonresponses still need to be evaluated. (See the description of Project F above.)

As already indicated, moderate NA rates, of themselves, are not in general a satisfactory indicator of quality of the census measurements for characteristics. The NA rate for the monthly Current Population Survey, which is regarded as of high quality relative to the census, is approximately 4 or 5 percent for most items, and considerably higher for some. In part, this NA rate results from the fact that interviewing must be completed within a very short time.

On the other hand, where NAs run to 25 or 50 percent or more for a particular area, it can be interpreted as providing poor data for that area. This, for example, was the situation in the city of Chicago where, among census tracts having 1,000 inhabitants or more, there were three tracts, each containing one or more ED's, in which allocation rates were 50 percent or more for a majority of the sample items.

**Enrollment data available only for persons 5 to 29 years of age in 1950.

The increase in nonresponse for age, which was a 100-percent item, presumably resulted primarily from the collection of information in 1960 by date of birth instead of age in years as of the last birthday.

Analysis of Age Data.--In each census in which data on single years of age have been collected, there have been overstatements of ages ending in certain digits and understatements for other digits. In each census since 1880, this "age heaping" has declined, and another substantial reduction was achieved in the 1960 census.

The percentage distribution of ages by final digit for the population aged 23 to 99 years as reported in the 1960 census is the following:

1

An index of age heaping can be constructed by taking one-half the sum of the deviations from 10.0 percent, each taken without regard to sign. For the 1960 census, this index is 0.8. Comparisons with other censuses back to 1880 are shown below:

In 1960 and in 1900, "date of birth" rather than "age at last birthday" was the inquiry, and the improved question together with the opportunity for self-enumeration in 1960 may be largely responsible for the decrease in age heaping.

The higher nonresponse rate in 1960 may also have contributed to the reduction in this index. The reductions in 1960 compared to 1950 occurred for both males and females in both the white and nonwhite groups:

Comparison of Census Statistics and Independent Estimates.--Labor force data from the 1950 and 1960 censuses were compared with data from the April 1950 and April 1960 monthly Current Population Survey (CPS). The results are shown in table 6.

Without exception, for all population groups shown, there is evidence, according to CPS results, of census undercount of persons in the labor force. However, except for persons employed in agriculture, the differences relative to CPS results are smaller for 1960 than for 1950. Differences for employment in agriculture may be especially affected by the variations in time reference.

For the census figures on enrollment in the public schools, comparisons were made with two independent estimates. The first was an estimate of total enrollment from kindergarten through the 12th grade, based on a questionnaire sent to all offices of State school systems in the fall of 1959 requesting enrollment figures as of Oct. 1, 1959, or as close to that date as enrollment stabilization appeared to have occurred. The census data referred to school enrollment between February 1, 1960, and the census date. The second independent estimate was from a school enrollment question in the October 1959 CPS, and is subject to the usual CPS sampling error. The figures on public school enrollment, kindergarten through 12th grade, from the three sources are as follows:

1 Myers, Robert J., "Accuracy of Age Reporting in the 1950 United States Census, Journal of the American Statistical Association, Vol. 49, No. 268, December 1954, pp. 826-831.

College enrollment data from the 1960 census can also be compared with the October 1959 CPS and with the information obtained from college and university officials in the 1959-60 biennial Office of Education survey:

Although the CPS and Office of Education figures are in close agreement, the census differs from both to a marked degree. A special inquiry concerning fall and spring enrollment, sent to the largest universities in the six States where the census figures fell appreciably below the Office of Education figures, showed that the attrition rate was very close to the difference between the two sets of fig

ures.

Data on income for both the 1960 and 1950 censuses were compared with CPS data and with estimates of the National Income Division of the Office of Business Economics. All figures relate to aggregate income for the United States in the years 1959 and 1949:

2

3

Persons 14 years of age and over, excluding inmates and members of Armed Forces living on base. Total population, all ages.

4 Estimates based on preliminary sample tabulations because final data do not contain distribution of income by type.

A comparison of the census and the Office of Business Economics (OBE) estimates by type of income shows that for both 1949 and 1959 there was close agreement for wages and salaries and that there was evidence of substantial underreporting of income other than earnings in the censuses. In each case the census was in closer agreement with the OBE estimate than were the CPS figures. There was a substantial reduction in underreporting in the 1960 census compared to the 1950 census.

Response variance.--The index of inconsistency has been discussed in the Introduction (page 127) to this chapter. The indexes of inconsistency for the labor force classifications in the 1960 and 1950 censuses as compared to CPS data for the identical populations 14 years old and over are shown below:

*Minus sign indicates greater unreliability in 1960 census than in 1950 census; plus sign indicates greater unreliability in 1950 census.

The intensive reinterview survey described earlier in this chapter (Project E) produced the estimates of bias and of indexes of inconsistency shown in table 7 for the specified characteristics in the 1960 census.

Where an equivalent comparison of cases was available from the 1950 census and the 1950 postenumeration survey, the differences between the 1950 and 1960 biases and indexes of inconsistency are shown.

The measurements of bias shown in table 7 result from comparisons of responses for persons enumerated in both the census and the reinterview survey. A larger bias for 1960 than for 1950 could result from an improved reinterview study in 1960, a reduced accuracy in the 1960 census, or both. Conversely, reductions in the indexes of inconsistency in 1960 compared with 1950 might have occurred as a result of improvement in the 1960 census or the 1960 reinterview survey, or both. The results are different to interpret. Generally, they are consistent with the view that either the 1960 census had less response error than the 1950 census, or the 1960 evaluation study had less response error than the 1950 evaluation study, or both.