to say anything significant about that group. Consequently, the sample was stratified into six groups: three racial groups whites, blacks, and nonblack minorities each divided according

to male and female.

The number of employees included in each of the six stratified groups was determined on the basis of a priori assumptions concerning the need for information about the particular groups. We attached equal importance to information concerning blacks and whites, and males and females. Because nonblack minorities comprise only 3 per cent of the Department's workforce it was felt that a smaller sample size could be used for this group compared to that for whites and blacks. These a priori assumptions yield the sample distribution shown below.

Since the overall study included only GS employees, the

sample contains only these workers.

Sample selection

The sample was selected from a computer printout of all DOL employees listed separately according to the six stratified groups. Within each group, the employees were listed on the printout by Social Security number in ascending order. From each of the six listings a random sample of employees was chosen corresponding to the number of employees allotted to represent that group. Because of the geographic and age bunching that appears in an ordered listing by Social Security number, the technique which is described below seemed best

· suited to choosing a random sample from each of the six groups. This technique was applied to each of the groups as follows:

1.

2.

3.

The number of employees in the group is divided by the number of employees allotted in the sample for that group to yield a "ratio of employees sampled." For example, since the sample includes 214 white males out of the total 4,964 (4964 23) the "ratio" 214

=

of white males sampled to all white males would be 1 out of 23.

A random number smaller than the "ratio of employees
sampled" is chosen to select the first employee included
in the sample. In the example above this number would
be between one and 23. The first employee chosen is
the one designated by the random number counting down
the listing of employees. For example, if the random
number is five, then the fifth employee is the first one
selected for the sample.

After choosing the first employee, the remainder
are selected according to the "ratio of employees
sampled." In the example used here this would
mean that starting with the first employee selected,
every 23rd would be chosen consecutively over the
remainder of the list.

Sample collection

The most efficient means of collecting the information in the sample appeared to be by retrieving the educational data from the employees' personnel records. We realized that the information contained in the files might not be completely up-to-date. However, the alternative of gathering the data from each of the employees in the sample would have consumed much more time. Trading off the factors of accuracy against time it seemed that collecting the data quickly was the overriding consideration. We felt that gathering the information from the files would provide a sufficient degree of accuracy, and would satisfy a severe time constraint.

Questionnaires were sent out to the seven central personnel offices and the eight RAOS. The forms listed the employees in the sample and requested the following information:

1. Years of education

2. Highest degree earned

3. Major field of study

A copy of the questionnaire is attached.

Responses were received for all of the 1,000 employees for whom information was requested. One aspect of the response which should be noted is that 8 per cent of the personnel folders were not located where the computer master file indicated. Almost none of this could be accounted for

by the time lag between the date of the computer run and
when the questionnaires were received. Nearly all the errors
occurred because the employees had been transferred or had
resigned months before our computer run. We were told that
some of the employees had never worked for the administration
where they were listed as employed. In locating the personnel
folders the master file was more accurate for the Field
6 per cent error than for the National office 10 per cent
error. In the National office the largest rate of error was
for WSA and MA. It should be noted that the error discussed
here applies only to individuals not being located where the
master file placed the, and does not refer to the factual
validity of the information.

-

Attached is a table which provides a breakdown of the survey response by bureau and region.

Testing the validity of sample

There are three factors which need to be considered concerning the validity of using our sample to generalize about the educational characteristics of DOL employees. First, it is important to determine whether the characteristics of the educational data itself permit the population to be represented by a sample of the general size of the one used here. It is possible that the distribution of years of education among DOL employees would be such that the sample required to predict the population's characteristics would need to be almost as large as the population itself. Preliminary testing indicates that this is not a problem. 2/

The second factor to be considered is the actual validity of the educational data collected by the sample. As noted before, the data was gathered from the sometimes out-of-date personnel folders which raises a question concerning the accuracy of the data in the sample. To evaluate this, 40 employees in the sample were called to determine if the information collected about them was, in fact, correct. The results of the test indicated that the years of education attained were understated for a significant number of employees. However, the magnitude of the understatement was in most cases small. As a result, the degree actually attained was different from that reported in the sample for only two of the 40 employees. It is clear that the data in the sample is not entirely correct, and, consequently, will not give a completely accurate representation of the educational attainment of DOL employees. Nonetheless, the sample would appear to provide better than a rough indication of the level of education for various groups of employees and this should be adequate for our purposes.

2/ This test was accomplished by calculating the estimated variance of the distribution of the sample means.

The final factor to be considered in evaluating the validity of the sample consists of determining whether the sample is representative of the DOL workforce. Since the sample was stratified into six groups, the appropriate criterion is for the sample of a particular group to represent the characteristics of that group. The basis for this comparison can be demographic characteristics which are available for all DOL employees average salary, average length of service, and average age. For example, the average salary of black male employees in the sample can be matched against the average salary of all black male employees. Using measures of the characteristics named above, for both the sample and the population, there exist statistical techniques to determine precisely how closely the sample represents the population. test of the hypothesis that the sample me an equals the population mean for salary, service, and age discloses that we cannot reject this hypothesis. 3/ The values for the t-test are shown in the tables following this page. On this basis it can be concluded that the sample is representative of professionals in the DOL workforce.

A t

3/ Except for nonblack minority females.

Values of a t-Test on the demographic characteristics of employees in the educational sample