CONSTRUCTION OF THE 1951-77 AVERAGE WAGE SERIES

The amended Act requires the use of an average wage for indexing described in various sections of the law as "the average of the total wages (as defined in regulations of the Secretary. . .)." Such general language leaves a wide range of possibilities for a definition of such a wage series. In order to provide cost estimates to the Congress for its deliberations on the decoupling provisions in the 1977 Amendments, the Office of the Actuary developed a tentative set of average wages for 1951-77 which we refer to in the following text as the preliminary series. However, following the enactment of those amendments, further consideration was given to the development of the best possible series of wages for indexing that was consistent with the general intent of the law. These deliberations resulted in the series of average wages for the period 1951-77 that was published in the Federal Register on December 29, 1978. In the following sections, we examine the data sources and the construction of the preliminary series and its successor series which is the final set of average wage figures currently used by the Social Security Administration (SSA) for indexing earnings histories.

1. Historical precedents for wage indexing and average wage

The precise intent of the general language "average of the total wages" can best be illuminated by studying both the historical precedents for the use of wage-indexing procedures under the Social Security Act as well as the legislative intent of Congress as expressed in the congressional committee reports accompanying the legislation that eventually became the 1977 Amendments.

As mentioned in the introduction, the idea of using an indexing procedure based on some measure of increases in average earnings in order to update certain program amounts was introduced into the Act in 1965. The 1965 Amendments contained a provision for offsetting benefits paid to disabled workers and their families if benefits plus workmen's compensation payments exceeded 80 percent of a certain measure of average current earnings. This measure of average current earnings was to be updated periodically according to increases in wage levels in the economy. At that time it was decided to measure the increase in wage levels from internal social security data that provided an average wage per wage item for wages reported to SSA for the first quarter of each year. (A wage item represents a report from one employer for one employee, so that a single employee may be represented by more than one wage item.) For all years before 1978, wages above the earnings base amount were not reported to SSA. Therefore, the average annual amount of reported taxable wages was lower than the average total wages, and the year-to-year increases in the annual average wage were distorted by the dampening effect of the earnings base. However, since very few workers earn more than the annual maximum taxable amount in the first quarter, the average first-quarter wage is not significantly affected by the earnings base.

The use of first-quarter wage data was continued in the 1972 Amendments which instituted the indexing of the contribution and benefit base and the retirement text exempt amount. At that time, the law prescribed that the indexing be based on an average wage per employee where the wage data were to be obtained from the processing of employer's wage reports following the posting of earnings for the first quarter of a given year to the Summary Earnings Record (SER). In order to provide a consistent measure from year-to-year and, at the same time, provide the data soon enough to meet the publication requirements of the law, it was decided to use the data extracted from the SER following the September posting cycle of the same year for which the first-quarter data were being extracted. The use of the data obtained in this manner was specified in a general way in the law, and in a very explicit way in regulations promulgated by the Secretary.

In the case of the 1972 Amendments, as well as in the earlier implementation of the workmen's compensation offset provisions, a deliberate decision was made to include only wages in the indexing measure. Self-employment income (SEI) was excluded from the measure for a variety of reasons. The most important reason at that time was that data on SEI are available only on an annual basis from Schedules SE that are attached to income tax returns. Thus, the SEI data could not be combined with the first-quarter wage data.

All of these factors were considered during the congressional committee deliberations that preceded the enactment of the 1977 Amendments. It was the intent of the congressional committees that the wage indexing series to be used in the new decoupled benefit formula be consistent with such indexing procedures already in the law. The average wage figures for the first quarter of each year 1973-77 had been published in the announcements of the automatic determinations of the contribution and benefit base and retirement test exempt amounts for each year 1975-79. Therefore, they had been subjected to public scrutiny and were apparently acceptable to the public and to Congress. The use of the first-quarter average wage figures was also suggested by the "Report of the Actuarial and Economic Consultants to the 1975 Advisory Council Subcommittee on Financing." For the purpose of using average wages to index earnings under their decoupling proposal, these consultants said "... average wages for a calendar year are determined by the same method as is currently used for adjustment of the taxable earnings base and the exempt amounts under the retirement test." (page 129, Appendix A of the 1975 Advisory Council Report). Thus, in constructing the indexing series for years 1951-77, the first quarter wage data for years 1973-77 that already had the sanction of law and regulations were used as the basis of the new series. Unfortunately, data on amounts of wages and numbers of employees comparable to the 1973-77 data, on a 100-percent basis, are not available for other years; nor could such data be obtained now, since the data must be obtained through posting operations. Therefore, various other sources of data were investigated with the aim of constructing a series of average wages for years 1951-72 to complement the 1973-77 series.

2. The preliminary series of average wages

The pre-1973 segment of the preliminary average wage series (Table 3) was constructed from tabulated data on 100 percent of the wages and wage items reported to SSA for the first quarter of each year 1951-72. The data on amounts of wages and numbers of wage items were tabulated from first-quarter employer report cards processed as of September of the same year for which the tabulation was being made. The information on the total amounts of wages reported for the first quarter obtained during this employer report card operation is essentially consistent with the data on amounts of wages used for each year 1973-77 which were obtained during the operation of posting the earnings to individual earnings records. However, because of possible employment of one employee by more than one employer, the employer report card operation did not provide sufficient information for compiling a tabulation of the numbers of employees corresponding to these amounts of wages. Therefore, from this source of information it is only possible to compute a series of averages per wage item rather than averages per wage earner. Nevertheless, the 100-percent nature of the data, and the essential comparability of the wage totals from this data with the wage totals in the post-1972 data suggested the use of this wage-item data in constructing a series of averages for years prior to 1973. This series was constructed by calculating an average per wage item from the employer report card data for each year 1951-73, measuring the annual percentage increases in average wages from the series of averages per wage item, and then using these annual percentage increases to extend the 1973-77 series of average wages per employee backward in time to 1951. This procedure assumes that the ratio of the number of wage items to the number of employees, in the first quarter of each year, remained constant throughout the period 1951-72 at a level equal to the level observed in the first quarter of 1973.

In addition to the linking procedure used to correct the conceptual difference between wage items and employees, further adjustments to the wage-item data were made to take into account changes in patterns of coverage and irregularities in reporting and posting practices. One of the major modifications to the wage-item data was the total exclusion of all military wages and counts of wage items. This exclusion was required to eliminate the inconsistent effect of the military wages on the increases in average wages that resulted from the irregularities in the timeliness of reporting from the various uniformed services.

A second modification to the data was made for changes in coverage. During the period 1951-58, the number of State and local government employees covered under social security increased rapidly. The increases in average taxable wages were affected by this rapid growth in State and local coverage. Furthermore, prior to 1957, the data on State and local wages were not processed and tabulated in as timely a manner as they were after 1956. Therefore, wages and wage items reported for State and local government employees were not included in the data for years prior to 1958 in the derivation of the preliminary set of average wages.

Finally, the wage-item data for 1951 contained information on the number of wage items for the first quarter of that year, but not a tabulation of the corresponding amount of wages. Therefore, to complete this average wage series, it was necessary to estimate the amount of wages based on the total amount of wages for the first quarter of 1951 that was certified by SSA to the Department of the Treasury.

Thus, although the wage-item data for years prior to 1973 are on a 100-percent basis and the wage counts from these data were somewhat comparable to the post-1972 data already in use, a number of modifications and simplifying assumptions were required to produce a series of average wages conceptually consistent both internally and with the 1973-77 series. This created uncertainty regarding the extent to which the preliminary series properly reflected percentage increases in average wages. In particular, the assumption of a constant ratio of wage items to wage earners throughout a period of varying economic conditions seemed questionable. Furthermore, the exclusion of the effects of military wages on the increases in average wages for the period 1957-72 made these data inconsistent with the data in use for the period 1973-77 which do include military wages, and the data from W-2 Forms which will be in use for future years and which will also include data on military wages. Consequently, an attempt was made to construct a series for the period 1951-72 from data which would permit the direct computation of an average wage per employee, and which would properly allow the inclusion of effects of military wages on the increases in average wages.

3. Limitations on other historical sources of data available from SSA

Although no suitable sources of data on wages and employees tabulated on a 100-percent basis are available for years prior to 1973, other sources of data on wages and employees are available from SSA tabulations of statistical samples of workers covered under social security. The largest such sample collected in a consistent manner almost since the inception of the social security program is the 1-percent sample of all social security numbers. Such 1-percent sample data have been collected annually since 1940 and provide the source data for several basic record files which make up the Continuous Work History Sample (CWHS). Two of these data files which contain detailed information on wages and wage earners under the social security program are the 1-percent 1937-to-date CWHS file and the 1-percent 1957-to-date Longitudinal Employee-Employer Data (LEED) file. The CWHS file is compiled from data in the SER after the posting of earnings to individual records with a cutoff date of September of the year following the latest year represented in the file. This file contains information from the 1-percent sample population on amounts of wages and numbers of wage earners for each year in the desired period 1951-72; however, it does not contain a quarterly breakdown of those wages. Therefore, the information in this file is not suitable for comparison with the established set of first-quarter data to be used in 1973-77.

The LEED file is assembled from the 1-percent sample of annual employee-employer records which are prepared yearly during the operation of posting earnings to the SER with a cutoff date of the September following the year for which the annual records are being compiled. It is estimated that this cutoff date provides for the inclusion of over 99 percent of the wage items reported for the first quarter for workers in the 1-percent sample. In the annual files, one record is created for each employee-employer combination during the year. In the longitudinal file, the original records from the various annual files have been resequenced and merged so that all records associated with an employee over the time span of the file appear together. The detailed nature of the information compiled in the LEED file provides not only data on quarterly amounts of wages and numbers of wage earners in the sample population, but also data on numbers of employer-employee combinations exhibited by the sample population and a breakdown of wage data by type of employer. The consistent September cutoff date provides a year-to-year comparability throughout the entire time span of the file and makes it possible to use the data for measurement of variation of average wages through time. Also, this cutoff date is one year later than that used in the 100-percent wage-item data and therefore includes more complete information on certain segments of the population (e.g., the military) which were inconsistently represented in the 100-percent wage-item data. However, although the LEED file is an exceptionally rich source of data, it has been compiled only since 1957 and therefore provides no information for the period 1951-56.

A final source of detailed data on wages and wage earners is the smaller 0.1-percent sample, a subset of the 1-percent sample of social security numbers. The 0.1-percent sample provides information on quarterly wages since 1937. It has been the basis for quarterly estimates of average taxable wages per employee which are produced regularly by the Office of Research and Statistics (ORS) for publication in the Social Security Bulletin. The originally published estimates are based on raw data from the 0.1-percent sample adjusted to the level established by the 1-percent sample. Furthermore, the estimates for a given year are revised over a period of time following the original publication to reflect more complete information derived from the 1-percent sample which is collected at a later point in time. More detailed information regarding these sample sources of data may be found in the papers: "The Continuous Work History Sample (CWHS): Description and Content," by Warren Buckler and Creston Smith, ORS-SSA, and "Sampling Variability in the 1-percent Continuous Work History Sample," by Robert H. Finch, Jr., ORS-SSA.

4. The construction of the final average wage series for 1951-77

In making a final decision on a series of average wages to be used for indexing under the 1977 Amendments, three main considerations played important roles. First, the average wage figures should be based on the most complete and accurate source of data available for the given time period subject to the restrictions imposed by the second and third considerations. Second, the series of averages should represent averages per employee as envisioned by legislative history of the 1977 Amendments and legislative precedent provided by the Act and its associated regulations as in effect prior to the 1977 Amendments. Third, since the purpose of developing this historical series is for use in indexing of earnings according to increases in average wages, the major concern is that the final series represents a consistent and accurate measure through time of the annual percentage increases in average wages per employee.

These three considerations suggested the following choices of data sources for construction of the final series of average wages. For 1973 through 1977, increases in average wages should be measured by the averages per employee calculated from the 100-percent first-quarter tabulations as provided by the law in effect prior to the 1977 Amendments. For 1957 through 1973, increases in average wages should be measured by the averages per employee calculated from first-quarter taxable wages recorded in the 1-percent sample LEED file.

Finally, for 1951 through 1957, the increases in average wages should be measured by the published estimates of first-quarter taxable wages based on the 0.1-percent sample.

Furthermore, for purposes of indexing in benefit computations, the only requirement of a series of average wages is the measurement of year-to-year percentage increases in average wages, rather than the measurement of absolute dollar levels of averages in given years. Therefore, it is possible, using the year-to-year percentage increases, to link these three somewhat disparate sources into a series of averages that provides a consistent and accurate measure of increases in average wages throughout the entire time period 1951-77. This linking to produce the final series presented in Table 1 was accomplished as follows. First, the first-quarter averages per employee for each year 1973-77 were retained as they had originally been computed. Next, an average wage per employee was calculated for the first quarter of each year 1957-73 from the data tabulated from the 1-percent sample LEED file, and the annual percentage increases measured by these averages were used to extend the 1973-77 series backward in time to establish a set of average wages for the entire period 1957-77.

Finally, the revised estimates based on the 0.1-percent sample of average taxable wages per employee for the first quarter of each year 1951-57 were compiled from various issues of the Social Security Bulletin and the annual percentage increases measured by these averages were used to extend the 1957-77 series backward in time to establish a set of average wages for the entire period 1951-77. The basic averages used for constructing the final series are presented in Table 4, along with the adjusted first-quarter series. The series of first-quarter averages produced in this way was then multiplied by four to produce the final series to be used for indexing the 1977 average wage levels.

5. The final series—indications of reliability

To test the relative accuracy of the final indexing series in measuring the pattern of changes in average wages since 1951, another series of average wages was derived from estimates of amounts of wages and numbers of wage earners for all industries combined as prepared by the Department of Commerce. These estimates, prepared regularly for publication in the "Survey of Current Business," are based largely on data on the population covered by unemployment insurance, compiled by the Bureau of Labor Statistics (BLS). The amounts for 1951-72 were derived from estimates appearing in the "The National Income and Product Accounts of the United States 1929-74 - Statistical Tables - A Supplement to the Survey of Current Business." The amounts for 1973-77 were derived from estimates appearing in the July 1977 and July 1978 issues of "Survey of Current Business." The Department of Commerce total wage series includes some industries not in the SSA wage universe, e.g., railroad and most Federal civilian workers, and excludes overseas possessions, such as Puerto Rico, which are included in SSA's wage data. Furthermore, the employment series is defined as an annual average of employment, rather than the SSA definition of employment at any time during the year. For purposes of comparison, Table 5 shows the annual percentage increases in the final indexing series, the preliminary series and the series derived from the Department of Commerce estimates.

The final series is closer in its pattern of annual increases to the Department of Commerce series than to the preliminary series based on wage-item data. This can largely be explained by the fact that the final series and the Department of Commerce series are both true per-employee series, whereas the preliminary series is an artificial series based on the assumption of a constant ratio through time of wage items to wage earners. Data from the 1-percent sample LEED file, collected since the construction of the preliminary series, tend to confirm that this assumption is not an accurate one when attempting to measure increases in average wages per employee. Table 6 presents data from the 1-percent sample, for each year 1957-73, on the numbers of employees and their corresponding numbers of employer-employee combinations. The number of employer-employee combinations here corresponds to the number of wage items that would have been recorded for these employees. Therefore, the ratio of number of employer-employee combinations to number of employees is a measure of the number of wage items per employee. The sample data indicate that this ratio in reality varies significantly through time both above and below the 1973 level, but was assumed constant in the construction of the preliminary series. The consequences of this discrepancy can be observed, for example, in the percentage increase in first-quarter average wages from 1957 to 1958. As expected during an economic slowdown, this increase was small. But at the same time, the number of wage items per employee was shrinking. Adjusting the wage-item data to reflect this shrinkage produced a much more accurate final wage series than if the constant ratio assumption discussed earlier had been used.

Some discrepancies can be observed between the increases resulting from the final series and the increases from the series compiled from the Department of Commerce estimates. For example, in the final series there is a relatively low percentage increase in average wages from 1964 to 1965 and an unusually large increase from 1971 to 1972. These effects, though not present in the Department of Commerce series, can reasonably be explained as variations present in first-quarter data which are smoothed out in annual averages. This can be substantiated by looking at unemployment insurance data for the first quarter, compiled by the BLS, which are part of the annual data on which the Department of Commerce estimates are largely based (Table7). For example, the percentage increase in average wages measured by the final series from first quarter 1971 to first quarter 1972 is large compared to increases for years just before and after this period. This same large change from 1971 to 1972 appears in the first-quarter BLS data, although it is not reflected by the Department of Commerce estimates. Other differences, for example the recent faster growth in average wages in the Department of Commerce series, may be explained in part by the differences in the covered population being represented by the two series.

An overall measure of the difference or similarities of various average wage series can be obtained by observing the effects in using these wage series to compute the Average Indexed Monthly Earnings (AIME) for workers with a selected earnings history. Table 8 shows projected AIME's for maximum wage earners retiring at age 62 in each year 1979-90. In these projections, each of the three series was extended after 1977 by the same series of assumed increases in average wages. The indexing series based on Department of Commerce data produces the highest AIME's, while the preliminary wage-item series produces the lowest AIME's. The final average wage series produces AIME's intermediate to those of the other two sets. However, because of the smoothing effect of the averaging process used in calculating the AIME, the results using the preliminary series and the series based on Department of Commerce estimates vary at most 2 percent on either side of the results using the final series. Calculations of AIME's for workers with average and low earnings records show similar patterns of difference.

The higher AIME's resulting from the use of the final average wage series, instead of the older preliminary series, could have been anticipated by observing the pattern of higher indexing factors produced by the proposed series (Table 5). While this difference in indexing factors is partly attributable to the difference in the basic nature of the two series (per-employee averages as opposed to per-wage-item averages), specific causes for the level of differences can be identified by recalling some details of the construction of the preliminary series. For example, in that construction, military wage-item data were excluded from the computation of averages for years prior to 1973 to eliminate the effects of erratic reporting practices of the uniformed services in those years. In computing the averages for the final series, the wage records of military employees in the 1-percent sample were properly included in the computation as was done in the calculation of averages based on the 100-percent data for years after 1972. Examination of detailed information on military wages and employees from the 1-percent sample shows that excluding military employees from the computation of averages would have lowered somewhat the indexing factors in the final series, but would not have lowered them to the level of the preliminary series indexing factors.

Notes published in the 1980s