Following is the portion of a recent report by the President’s Pay Agent explaining how the government determines the pay gap between federal and non-federal jobs, a method that shows federal employees are substantially behind on average, in contrast to some other studies that show them substantially ahead on average.
How Local Pay Disparities Are Measured
Locality-based comparability payments are a function of local disparities between Federal and non-Federal pay. Pay disparities are measured for each locality pay area by comparing the base GS pay rates of workers paid under the General Schedule pay plan in an area to the annual rates generally paid to non-Federal workers for the same levels of work in the same area. Under the NCS program, BLS surveys or models salaries for all non-Federal jobs deemed to match GS positions, as shown in the crosswalk in Appendix VII to the 2002 Pay Agent’s report. BLS can also produce equivalent estimates using its new model and OES data.
Non-Federal rates are estimated on a sample basis by BLS area surveys. The rate for each non-Federal job is an estimate of the mean straight-time earnings of full-time non-Federal workers in the job, based on the BLS survey sample. GS rates are determined from Federal personnel records for the relevant populations of GS workers. Each GS rate is the mean scheduled annual rate of pay of all full-time, permanent, year-round GS workers in the relevant group.
The reference dates of the BLS surveys vary over the cycle of non-Federal salary surveys conducted for the GS locality pay program. To ensure that local pay disparities are measured as of one common date, it is necessary to “age” the BLS survey data to a common reference date before comparing it to GS pay data of the same date. March 2010 is the common reference and comparison date used in this report. The Employment Cost Index (ECI) based on wages and salaries for civilian workers was used to age the BLS NCS data.1 BLS aged data for the OES model prior to sending it to OPM.
Because 5 U.S.C. 5302(6) requires that each local pay disparity be expressed as a single percentage, the comparison of GS and non-Federal rates of pay in a locality requires that the two sets of rates be reduced to one pair of rates, a GS average and a non-Federal average. An important principle in averaging each set of rates is that the rates of individual survey jobs and job categories are weighted by Federal GS employment in equivalent classifications. Weighting by Federal employment ensures that the influence of each non-Federal survey job on the overall non-Federal average is proportionate to the frequency of that job in the Federal sector.
We use a three-stage weighted average in the pay disparity calculations. In the first stage, job rates (based on survey results or modeled data) are averaged within PATCO2 category by grade level. Both the NCS and OES programs cover virtually all GS jobs since only jobs that were not randomly selected in any BLS survey area cannot be modeled. For averaging within PATCO category, each job rate is weighted by the full-time permanent year-round employment3 in GS positions that match the job. The reason for national weighting in the first stage is explained below.
When the first stage averages are complete, each grade is represented by up to five PATCO category rates in lieu of its original job rates. Under the NCS or OES program, all PATCO/grade categories with Federal incumbents are represented, except for any where BLS had no data at all and could not model results.
In the second stage, the PATCO category rates are averaged by grade level to one grade level rate for each grade represented. Thus, at grade 5, which has Federal jobs in all five PATCO categories, the five PATCO category rates are averaged to one GS-5 rate. For averaging by grade, each PATCO category rate is weighted by the local full-time permanent year-round GS employment in the category at the grade.
In the third stage, the grade averages are weighted by the corresponding local full-time permanent year-round GS grade level employment and averaged to a single overall non-Federal rate for the locality. This overall non-Federal average salary is the non-Federal rate to which the overall average GS rate is compared. Under the NCS or OES programs, all 15 GS grades can be represented.
Since GS rates by grade are not based on a sample, but rather on a census of the relevant GS populations, the first two stages of the above process are omitted in deriving the GS average rate. For each grade level represented by a non-Federal average derived in stage two, we average the scheduled rates of all full-time permanent year-round GS employees at the grade in the area. The overall GS average rate is the weighted average of these GS grade level rates, using the same weights as those used to average the non-Federal grade level rates.
The pay disparity, finally, is the percentage by which the overall average non-Federal rate exceeds the overall average GS rate.4 See Appendix V for more detail on pay gaps using NCS data and Appendices VII and VIII using the OES model.
As indicated above, at the first stage of averaging the non-Federal data, the weights represent national GS employment, while local GS employment is used to weight the second and third stage averages. GS employment weights are meant to ensure that the effect of each non-Federal pay rate on the overall non-Federal average reflects the relative frequency of Federal employment in matching Federal job classifications.
The methodology employed by the Pay Agent to measure local pay disparities does not use local weights in the first (job level) stage of averaging because this would have an undesirable effect. A survey job whose Federal counterpart has no local GS incumbents will “drop out” in stage one and have no effect on the overall average. For this reason, national weights are used in the first stage of averaging data. National weights are used only where retention of each survey observation is most important—at the job level or stage one. Local weights are used at all other stages.5
Publishability and Substitute Data
Since the beginning of the locality pay program in 1994, BLS was never able to publish data for all survey jobs in every survey area. The fact that the set of available jobs varies from area to area was a concern because the disparity between Federal and non-Federal pay varies by job as well as by area. If area pay disparities are not based on the same set of jobs in each area, the differences between those disparities are caused not only by differences in the pay of Federal and non-Federal workers for the same jobs (as intended), but also by differences in the set of jobs for which pay data are available.
Since 1995, the Council and the Pay Agent have used estimates of non-Federal pay produced by a multiple regression model to estimate salaries for jobs not available in individual BLS surveys. OPM staff developed the original model to estimate local non-Federal pay rates for the survey jobs with Occupational Compensation Survey Program (OCSP) data. The OCSP model produced estimates of the pay of unpublished jobs based on multiple regression analysis of the pay of published jobs. The model assumed that pay varies with three factors—geographic area, occupation, and work level. A technical report on the original OPM model was provided in Appendix II to the 1994 Report, and a summary of subsequent years’ models appeared in Appendices II or III of later reports.
BLS staff developed and implemented a similar model using NCS data to produce pay estimates for missing non-Federal jobs in NCS. Both the NCS and the OCSP models predict pay as a function of location, occupation, and grade level.6 The NCS model accounts for about 83 percent of the variations in pay, which is very good for models of this type. BLS also developed a model for use of OES data in locality pay. The OES model is described in Appendix VI.
Use of modeling is a generally accepted practice, and we have used modeled data for most of the history of the locality pay program. The models used in both the original OCSP surveys and the NCS program are similar in concept and form. They are also similar to the curve-fitting process used in the pay comparability system prior to FEPCA. All jobs included on the crosswalk shown in Appendix VII to the 2002 Pay Agent’s report were included in developing the NCS and OES models, with the exception of a handful of jobs for which BLS had no data.
While the use of modeled data is a standard technique, the Federal Salary Council noted that there is a relatively large amount of data modeled under the NCS program. Based on GS employment weights used to combine the data at the job level, an average of about 74 percent of the NCS data are modeled in this year’s surveys. This varies by area from a high of 89 percent modeled in Buffalo to a low of 34 percent modeled in the Rest of U.S. locality pay area. The amount of modeled data also varies considerably by grade level and ranges from an average of 38 percent modeled at GS-4 to an average of 98 percent modeled at GS-15. All data used in the pay comparisons under OES are modeled. BLS research concluded that the new method produces pay gap estimates that are comparable to the existing method.