The employment estimates appearing in the establishment data section of this issue have been adjusted to reflect complete employment counts as of March 1967. These adjustments, which affect most of the published series, generally mean that the employment series have been revised back to March 1966 (the previous complete count) and forward to the current month's estimate. Estimates of hours and earnings and labor turnover, which are weighted by employment estimates, may also be revised as a result of the changes in employment levels.

Each year, universe employment counts called benchmarks, are established for each of the most detailed industrial classifications for which estimates are available. Revisions are introduced at these levels and then are aggregated through successively inclusive series to total nonagricultural employment. Benchmarks are an integral part of the Bureau of Labor Statistics' establishment payroll survey program. They provide accurate measures of the employment level, and the monthly reports submitted by a sample of employers provide a measure of the monthto-month changes in that level. New benchmark levels are determined for March of each year, and the corresponding current estimates are adjusted to the new levels.

The March 1967 total benchmark count of 64.8 million workers on establishment payrolls was higher than the original estimate by 26,000--a difference of only 0.04 percent. The benchmarks for each of the eight industry divisions ranged from 1.6 percent below the estimate for contract construction to 0.5 percent above the estimate for manufacturing. Estimates for the government division were not revised because new benchmarks are available only at 5-year intervals. The March

Of the Division of Industry Employment Statistics

Monthly estimates of employment, hours, and earnings are prepared and published in considerable industrial detail--the threedigit SIC level for most nonmanufacturing industries and the four-digit SIC level for many manufacturing industries. Of the 196 three-digit categories published, the employment estimates for only 17 were revised by 5 percent or more. The distribution of the 196 industry groups, by employment-size class and by percentage adjustment required,illustrates the tendency for the amount of relative change to vary inversely with the size of employment. (See table 2.)

30 percent of nonagricultural workers are employed in manufacturing industries. The importance of this division in size and influence requires that estimates be published in 4-digit SIC detail. The degree of accuracy that can be assigned to the estimates over time is of primary interest. The March 1967 estimates and benchmark levels for the major industry groups in manufacturing

Industry and Percent Difference

Between BLS Estimates and March 1967 Benchmarks

One reason for differences between estimates and benchmarks lies in the limitation of any sample in representing a universe; that is, a certain amount of error is to be expected from sample-derived estimates. A complete monthly count of employment would reflect all changes in the level from month to month, but complete coverage involving several million reports each month would be prohibitively expensive and time consuming. The BLS establishment series, derived from a sample of approximately 150,000 reports, provides estimates at moderate cost within a month following the reference week.

A second reason for differences between estimates and benchmarks is the procedure used in keeping the industrial classification of establishments up to date. Establishments are classified by industry according to their major activities which are determined by the principal product produced or handled, or services rendered. Many establishments engage in more than one activity. When the composition of their output or services changes so that what was once a secondary product or activity becomes a primary one, the classification of the establishment is changed to the industry of its new major activity. This change is not introduced into the employment estimates at the time it occurs but at the time of the annual benchmark

ported annually. Thus, differences between estimates and benchmarks for an industry may result because the estimates are tied to the former benchmark levels and do not reflect intervening classification changes for individual establishments.

At the more detailed industry levels, particularly within manufacturing, changes in classification are the major cause of benchmark adjustment. The larger revisions in the ordnance and accessories and the petroleum and coal products groups are to a large extent due to classification changes. (See table 3.)

A third reason for benchmark differences arises from the use of bias adjustment factors based on past experience. The benchmark link-relative employment estimating technique, which is a form of ratio estimation, results in biased estimates, because the employment of new firms entering the economy during the current month cannot be measured with this technique. The size of the bias (the influence of the employment of new firms)is considered negligible for most purposes; however, in the BLS establishment survey, for which the previous month's estimate is used as a base in computing the current month's estimate, such bias would cumulate if steps were not taken to counteract the situation. Accordingly, small bias correction factors are applied to the employment estimates each month. Since the size of these factors must be determined on the basis of past experience, small errors may arise if the rate at which new firms enter an industry increases or decreases. Between 1966 and 1967, the rate of entry of new firms into manufacturing industries was slightly greater than anticipated, whereas the rate of entry of firms in retail and wholesale trade was somewhat smaller than expected.

A fourth, generally infrequent, cause of benchmark differences arises from improvements in the quality of the benchmark data. The most recent example of this type of revision was the more complete data on religious and charitable institutions introduced into the employment levels with the 1965 benchmark revision.

In most instances of recent benchmark revisions, the estimates for major industry divisions have varied from benchmarks by less than 1 percent. A comparison of the size of the revisions made since 1963 is presented in table 4.

The difference between estimates and benchmarks is assumed to have accumulated in constant increments over the previous 12 months. Most series, therefore, are adjusted by wedging or tapering out the difference over the period from the new benchmark to the preceding one, 12 months earlier. Estimates subsequent to the new benchmark are revised by projecting the new level forward to the current month using the sample trend. The latter part of the revision is then subject to change when the March 1968 benchmarks are established.

Benchmark Source Material

The most important sources of benchmark information are the tabulations by industry and employment-size group of reporting units, compiled annually by the Bureau of Employment Security, for the first quarter of the calendar year. These tabulations provide monthly employment counts of establishments covered under State unemployment insurance laws. Each calendar quarter, covered employers file a report with their respective State employment security agencies. This report includes total employment for the week including the 12th for each month of the quarter. State tabulations of these data, summarized according to industry, are provided to the Bureau of Labor Statistics for benchmark purposes. Unemployment insurance (UI) data account for

A number of industries such as railroads, private schools, hospitals, and nonprofit organizations, are exempt from mandatory UI coverage. In addition, the UI laws in 31 States (as of January 1967) provide that employers of fewer than a specified number (usually 4) of workers be exempt from coverage. Benchmark data for the latter as well as for certain nonprofit institutions are obtained from the tabulations of employment and taxable wages of employees covered under social security laws as published in County Business Patterns by the Bureau of the Census.

For the remaining industries, benchmark data are obtained from several public and private agencies including the Interstate Commerce Commission (interstate railroads), the American Hospital Association (private nonprofit hospitals), the U.S. Office of Education and the National Catholic Welfare Conference (private schools, colleges, and universities), the U.S. Civil Service Commission (Federal Government), and the Governments Division of the Bureau of the Census (State and local Government).

The Bureau's reporting sample is also an important source of benchmark information. Since sample reports are current and are reviewed monthly, reporting errors are disclosed that otherwise may remain undetected. The industry classification of each sample establishment is reviewed annually on the

ployer. Changes in industry classification of sample reports often precede such changes in other sources of information. Insofar as sample reports are known to differ from the corresponding employer's reports included in other benchmark source material, the data in the other source are modified accordingly.

Relation of Benchmarks to Other Series

Benchmarks are not available for the hours and earnings and labor turnover series. The levels shown are derived from the BLS reporting sample only. For primary estimating cells, i.e., region and/or size strata within the most detailed industry classifications, the series are computed directly from reported figures. Series for more inclusive categories, however, require a weighting mechanism to yield meaningful averages. The employment benchmarks are used as weights in computing the hours and earnings averages and labor turnover rates for broader industry groupings.

Adjustment of the estimates to new benchmarks may result in a reallocation of weights which, in turn, may change the averages. To influence the average of a broad group, changes in employment have to be relatively large and must affect industries which have substantially higher or lower averages than the other industries in their group. Generally speaking, the introduction of new benchmarks does not change hours and earnings and labor turnover series for broader groupings by more than 0.1 hour, 1 cent, or 0.1 per 100 rate respectively. The changes caused by the 1967 benchmark revision are summarized in table 5.

Corrections of Hours and Earnings Levels

As a result of improvements in sample data or newer stratification patterns, it may be necessary to correct published historical hours and earnings levels of selected industries. Such corrections also are introduced with the annual benchmark adjustment. Corrections introduced with the 1967 benchmark adjustment affected the Communications in