Federal Personnel: Federal/Private Sector Pay Comparisons (Chapter
Report, 12/14/94, GAO/OCE-95-1).

Although official estimates consistently find that federal pay is low
compared with private sector compensation, academic studies have found
the opposite.  Individuals in both the media and academia have
criticized official estimates of the pay gap, arguing that the
methodology used to compare pay is defective. They claim that data from
sources other than the official surveys, when analyzed using different
methodologies, lead to a different conclusion: that federal pay levels
are higher than those for employees in the private sector with
comparable characteristics, such as education and work experience.  GAO
examined this contradiction and identified factors that may account for
the differing conclusions.

--------------------------- Indexing Terms -----------------------------

 REPORTNUM:  OCE-95-1
     TITLE:  Federal Personnel: Federal/Private Sector Pay Comparisons
      DATE:  12/14/94
   SUBJECT:  Wage surveys
             Compensation
             Federal employees
             Demographic data
             Labor statistics
             Evaluation methods
             Statistical methods
             Non-government enterprises
             Salary increases
             Comparative analysis
IDENTIFIER:  Professional, Administrative, Technical, and Clerical Pay 
             Survey
             Census Bureau Current Population Survey
             
**************************************************************************
* This file contains an ASCII representation of the text of a GAO        *
* report.  Delineations within the text indicating chapter titles,       *
* headings, and bullets are preserved.  Major divisions and subdivisions *
* of the text, such as Chapters, Sections, and Appendixes, are           *
* identified by double and single lines.  The numbers on the right end   *
* of these lines indicate the position of each of the subsections in the *
* document outline.  These numbers do NOT correspond with the page       *
* numbers of the printed product.                                        *
*                                                                        *
* No attempt has been made to display graphic images, although figure    *
* captions are reproduced. Tables are included, but may not resemble     *
* those in the printed version.                                          *
*                                                                        *
* A printed copy of this report may be obtained from the GAO Document    *
* Distribution Facility by calling (202) 512-6000, by faxing your        *
* request to (301) 258-4066, or by writing to P.O. Box 6015,             *
* Gaithersburg, MD 20884-6015. We are unable to accept electronic orders *
* for printed documents at this time.                                    *
**************************************************************************


Cover
================================================================ COVER


Report to Congressional Committees

December 1994

FEDERAL PERSONNEL -
FEDERAL/PRIVATE SECTOR PAY
COMPARISONS

GAO/OCE-95-1

Federal Personnel


Abbreviations
=============================================================== ABBREV

  ADF - Annual Demographic File
  BLS - Bureau of Labor Statistics
  CBO - Congressional Budget Office
  CPS - Current Population Survey
  CRS - Congressional Research Service
  CSRS - Civil Service Retirement System
  ECI - Employment Cost Index
  EEO - equal employment opportunity
  FEPCA - Federal Employees Pay Comparability Act of 1990
  GS - general schedule
  NIST - National Institute of Standards and Technology
  OMB - Office of Management and Budget
  OPM - Office of Personnel Management
  PATC - National Survey of Professional, Administrative, Technical,
     and Clerical Pay
  TCC - total compensation comparability

Letter
=============================================================== LETTER


B-256047

December 14, 1994

The Honorable John Glenn
Chairman
The Honorable William V.  Roth, Jr.
Ranking Minority Member
Committee on Governmental Affairs
United States Senate

The Honorable William L.  Clay
Chairman
The Honorable John T.  Myers
Ranking Minority Member
Committee on Post Office and Civil Service
House of Representatives

This report presents the results of our review of the methodologies
underlying various estimates of the differences between federal
white-collar pay and the salaries of similar workers in private
industry.  This report was prepared pursuant to our basic statutory
authority.  We are sending it to you because of your committee's
responsibility for federal personnel matters. 

We are sending copies of this report to other interested Members of
Congress, as well as the President's Pay Agent (the Secretary of
Labor and the Directors of the Office of Personnel Management and of
the Office of Management and Budget).  Copies will also be made
available to other interested parties upon request. 

Appendix VIII lists the major contributors to this report.  Please
contact me at (202) 512-6209 if you have any questions concerning
this report. 

James R.  White
Acting Chief Economist


EXECUTIVE SUMMARY
============================================================ Chapter 0


   PURPOSE
---------------------------------------------------------- Chapter 0:1

The federal government's official surveys of the pay--wages and
salaries--of federal and private sector employees have indicated that
federal pay has lagged behind prevailing levels for comparable jobs
in private enterprise and that the pay gap has grown over the last 2
decades. 

However, these official estimates of the pay gap have been subjected
to criticism in both academic studies and media accounts.  Critics
argue that the official methodology for performing pay comparisons is
defective.  They claim that data from sources other than the official
surveys, when analyzed using different methodologies, lead to a
different conclusion:  that federal pay levels are higher than
prevailing levels for employees with comparable characteristics, such
as education and work experience, in private enterprises. 

In view of these opposing conclusions, GAO identified and analyzed
possible explanations for the discrepancy between official estimates
of the federal private pay gap and those of the critics.  Because
federal personnel management policy includes factors that are beyond
the scope of this report, such as the level of fringe benefits and
judgments concerning the desired quality of the federal workforce,
GAO did not reach conclusions about the appropriateness of
comparability estimates or the level of federal pay. 


   BACKGROUND
---------------------------------------------------------- Chapter 0:2

GAO analyzed data from 1978 through 1987, the most recent period for
which suitable data were available.  For that period, applicable
federal laws required that federal pay be comparable with pay of
private enterprise for the same level of work.  The National Survey
of Professional, Administrative, Technical, and Clerical Pay (PATC),
which was conducted by the Bureau of Labor Statistics (BLS),
collected data on annual pay for jobs in private enterprise, which
were then compared with pay data for corresponding jobs in each grade
of the general schedule in the federal civil service (a position
comparison approach).  PATC data formed the basis for official
estimates of the pay gap.  Over the years, pay gap estimates based on
PATC data have consistently shown that employees in the federal
government are paid less than those in the private sector. 

In distinct contrast to the PATC-based results, a set of academic
studies based on a human capital approach (which compares the
earnings of individuals with similar personal characteristics, such
as years of education, rather than similar occupations) has
consistently shown that federal employees are paid more than their
private sector counterparts. 


   RESULTS IN BRIEF
---------------------------------------------------------- Chapter 0:3

GAO's review of academic studies identified two factors that might
explain differences between the human capital and official estimates
of the pay gap. 

First, official pay comparisons compare the pay of federal employees
to that of employees of predominantly large employers in similar
occupations, while the human capital studies compare the pay of
federal employees to that of employees of nonfederal employers of all
sizes.  Because employees of small private employers tend to be paid
less than their counterparts in large firms, the choice of private
sector comparison group can affect estimates of the pay gap.  When
GAO adjusted human capital estimates of the federal private pay gap
for the effect of employer size on pay, the discrepancy between human
capital and official estimates of the pay gap was decreased. 

Second, official estimates compare pay for employees in the same
occupation and work level, ignoring such personal characteristics as
race and gender.  By contrast, academic studies implicitly compare
federal employees to private sector employees of the same age,
education, race, and gender, while largely ignoring occupation. 
Because privately employed women and minorities tend to be paid less
than their counterparts in the federal sector, after adjusting for
education and work experience, the choice of comparison group can
affect estimates of the pay gap.  When GAO adjusted human capital
estimates of the pay gap so that all federal employees were compared
to private sector white males, rather than to all private sector
employees, the discrepancy between the two estimates also decreased. 

The combined effect of these two adjustments produced human capital
estimates that are similar to the official estimates.  GAO did not
reach conclusions about the appropriateness of the adjustments. 
Because federal personnel management policy includes important
factors that are beyond the scope of this report, and which may be
influenced by the level of pay, our analysis cannot by itself be used
to judge the appropriateness of comparability estimates or the level
of federal pay. 

GAO's analysis shows the importance of considering the effects of
employer size and race and gender on both official and human capital
estimates of the gap between federal and private pay.  The official
position comparison estimates and human capital estimates are based
on different methodologies, both of which have limitations--neither
method is clearly superior. 


   GAO'S ANALYSIS
---------------------------------------------------------- Chapter 0:4


      HUMAN CAPITAL MODEL
      ESTIMATES OF PAY GAP DIFFER
      FROM OFFICIAL ESTIMATES
-------------------------------------------------------- Chapter 0:4.1

GAO analyzed data from the Current Population Survey (CPS), the
principal data source used by analysts who have produced human
capital pay comparisons.  GAO's analysis of the CPS data, using
standard econometric techniques for comparisons based on the human
capital method (including standard specifications for education, work
experience, race, and gender), showed that federal employees were
consistently paid more than their private sector counterparts with
similar personal characteristics.  The estimated size of the pay
premium ranged from 7 to 15 percent over this period.  This finding
is broadly consistent with the findings of the numerous human capital
analysts whose research GAO reviewed for this report.  The official
pay gap estimates of the President's Pay Agent, based on data from
PATC for this period, yielded the opposite conclusion, namely, that
federal employees were paid less than their private sector
counterparts with similar jobs.  (See fig.  1.)

   Figure 1:  The Pay Gap as a
   Percentage of Private Sector
   Pay

   (See figure in printed
   edition.)

Source:  GAO analysis of CPS data and Pay Agent's reports for various
years. 


      THE EFFECT OF EMPLOYER SIZE
      ON PAY GAP ESTIMATES
-------------------------------------------------------- Chapter 0:4.2

CPS provided information on employer size at three points in the
period of GAO's analysis:  1978, 1982, and 1987.  For these 3 years,
GAO adjusted the standard human capital model to account for the
relatively high pay of employees of large private employers. 

Even after allowing for employee characteristics, numerous studies
have found that larger employers pay higher wages and salaries than
smaller employers.  Further, surveys that compare pay on a
position-by-position basis, such as PATC, have a higher percentage of
large-firm employees than do CPS data, which are commonly used for
human capital estimates of the pay gap.  The specialization and
distinct level of responsibility associated with many federal jobs
mean that BLS is more likely to find matches in larger organizations. 
These factors could cause human capital estimates of the pay gap to
differ from official estimates. 

When GAO compared federal white-collar pay to that of employees of
large private employers, it was able to demonstrate the effect of
employer size on human capital estimates of the pay gap.  The effect
of relatively high pay at large private employers is substantial. 
However, the effect of this factor alone is less than the difference
between estimates of the pay gap.  (See fig.  2.)

   Figure 2:  The Pay Gap Adjusted
   for Employer Size

   (See figure in printed
   edition.)

Source:  GAO analysis of CPS data and Pay Agent's reports for various
years. 


      THE EFFECT OF FEDERAL PAY OF
      WOMEN AND MINORITIES ON PAY
      GAP ESTIMATES
-------------------------------------------------------- Chapter 0:4.3

In addition, GAO adjusted the standard human capital model to reflect
the federal pay of women and minorities.  Federally employed women
and minorities earn relatively more than privately employed women and
minorities, at least in part because of occupational differences. 
One way human capital estimates of the pay gap can be computed
compares federal workers to all private sector workers, controlling
for education, work experience, race, and gender in a standard
fashion.  GAO followed this procedure in estimating the standard
human capital model. 

GAO also analyzed an alternative method of computing the pay gap in
which the pay of federal white males, women, and minorities is
compared to the pay of private sector white males.  The choice of a
private sector comparison group involves implicit assumptions about
the reasons for race- and gender-based differences in pay within the
private sector.  By comparing federal employees to all private sector
employees the analyst allows the lower pay of private sector women
and minorities, relative to private sector white males, to influence
the size of the pay gap.  Such a comparison implicitly assumes that
private sector pay differences by race and gender are caused by
unobserved productivity differences that are not necessarily related
to education and work experience.  By comparing all federal employees
to private sector white males the analyst does not allow the lower
pay of private sector women and minorities, relative to private
sector white males, to influence the size of the pay gap.  Such a
comparison implicitly assumes that private sector pay differences by
race and gender are caused by labor market discrimination. 

To the extent that both productivity differences and labor market
discrimination influence private sector race- and gender-specific pay
differences, this alternative represents an upper limit on the effect
of private sector labor market discrimination on human capital
estimates of the pay gap.  Discrimination can take other forms.  For
example, productivity may be influenced by previous discrimination in
education.  GAO did not reach any conclusions about the
appropriateness of either method of adjusting for race and gender;
the analysis shows the significance of the choice. 

The effect of using private sector white males as the comparison
group, as shown in figure 3, is substantial.  Again, the effect of
this factor alone is less than the difference between estimates of
the pay gap. 

   Figure 3:  The Pay Gap Adjusted
   for the Federal Pay of Women
   and Minorities

   (See figure in printed
   edition.)

Source:  GAO analysis of CPS data and Pay Agent's reports for various
years. 


      THE EFFECT OF EMPLOYER SIZE
      AND FEDERAL PAY OF WOMEN AND
      MINORITIES COMBINED
-------------------------------------------------------- Chapter 0:4.4

Finally, GAO adjusted the human capital model to reflect both
employer size and the federal pay of women and minorities.  The
result is shown in figure 4.  The combined effects of these two
adjustments produce human capital estimates of the federal private
pay gap that are similar to the official estimates. 

   Figure 4:  The Pay Gap Adjusted
   for Employer Size and the
   Federal Pay of Women and
   Minorities

   (See figure in printed
   edition.)

Source:  GAO analysis of CPS data and Pay Agent's reports for various
years. 

Although the literature GAO reviewed suggested the two factors that
were selected for analysis, there may be other factors that also
affect estimates of the pay gap. 


      IMPLICATIONS OF GAO'S
      ANALYSIS
-------------------------------------------------------- Chapter 0:4.5

GAO's analysis demonstrates the importance of considering the effects
of employer size and race and gender on private sector pay when
evaluating the two approaches for measuring pay comparability.  For
example, human capital estimates of the pay gap may be sensitive to
the choice of comparison groups. 

In addition, GAO's analysis should be interpreted within the broader
framework of federal personnel management policy.  Federal personnel
management policy includes important factors, such as workforce
quality, recruitment and retention, affirmative action, and employee
benefits, which may be influenced by the level of pay.  Because these
factors are beyond the scope of this report, GAO's analysis cannot by
itself be used to judge the appropriateness of comparability
estimates or the level of federal pay. 

GAO's analysis found that both the position comparison and the human
capital estimates have limitations.  For example, neither method can
account for how qualified employees are for the jobs they do or for
the level at which they perform. 

There is no easy answer to the question of the appropriateness of
federal general schedule compensation--pay and benefits.  Any
limitations of pay comparisons do not necessarily invalidate the use
of such comparisons in determining appropriate levels of
compensation.  Even if there were no disagreement as to the size of
the pay gap, paysetters and lawmakers would need to carefully weigh
all aspects of the compensation question to determine the appropriate
level of federal compensation. 


   RECOMMENDATIONS
---------------------------------------------------------- Chapter 0:5

This report contains no recommendations. 


   AGENCY COMMENTS
---------------------------------------------------------- Chapter 0:6

GAO received written comments on a draft of this report from BLS and
the Office of Personnel Management (OPM).  BLS offered a number of
clarifications and technical corrections, which have been
incorporated into the report as appropriate.  OPM said the report was
useful.  The Office of Management and Budget (OMB) was also provided
a draft of this report but declined to comment on it. 


INTRODUCTION
============================================================ Chapter 1

The federal government's official surveys of the pay--wages and
salaries--of federal and private sector employees have indicated that
federal pay has lagged behind prevailing levels for comparable jobs
in private enterprise and that the pay gap has grown over the last 2
decades.  However, these official estimates of the pay gap have been
subjected to criticism in both academic circles and in the media. 
Critics argue that the official methodology for performing pay
comparisons is defective and that data from sources other than the
official surveys, when analyzed using a different methodology, lead
to a different conclusion-- that federal pay levels are higher than
prevailing levels for employees with comparable characteristics, such
as education and work experience, in private enterprise. 

In view of these opposing conclusions, we identified two possible
explanations for the discrepancy between official estimates of the
federal private pay gap and those of the critics.  We then performed
a statistical analysis to determine the empirical significance of
these explanations for estimates of the pay gap.  Our analysis does
not address whether and to what extent federal employees are under-
or overpaid. 


   BACKGROUND
---------------------------------------------------------- Chapter 1:1

Personnel management policy encompasses such issues as pay and
benefits, recruitment, promotion, retention, and in certain
circumstances, the management of reductions in force.  Personnel
management also encompasses such issues as ethics policies (e.g.,
restrictions on postfederal employment) and equal employment
opportunity (EEO) policies. 

A number of tools are available to employers, including federal
agencies, as they attempt to achieve their personnel management
goals.  These include such tools as allocating resources to
recruitment and providing on-the-job training for employees. 


      COMPENSATION POLICY IS AN
      IMPORTANT COMPONENT OF
      PERSONNEL MANAGEMENT POLICY
-------------------------------------------------------- Chapter 1:1.1

The level of compensation--pay and benefits--that government agencies
offer their employees can have a substantial impact on the success of
these agencies in recruiting and retaining qualified employees.  As
such, the process by which compensation is determined is an important
component of overall personnel management policy. 


      FEDERAL PAY REFORM ADOPTED
      LOCALITY-BASED PAY
-------------------------------------------------------- Chapter 1:1.2

The Federal Employees Pay Comparability Act of 1990 (FEPCA) is the
most recent comprehensive reform of the federal paysetting process. 
Under FEPCA, federal pay is compared to that of nonfederal (including
employees of state and local governments) employees rather than only
those in private enterprise.  The annual governmentwide adjustment
has been separated into two components--national and local.  The
national component when granted would prevent federal salaries from
falling substantially further behind nonfederal sector salaries. 
This result is accomplished by linking the annual governmentwide
increase to the percentage increase in the Employment Cost Index
(ECI). 

Under FEPCA, the paysetting process relies on position comparison
information to measure the local pay gaps.  Federal employees may
receive an additional increase in pay that is designed to reduce the
local pay gaps.  Partial adjustments (based on a formula specified in
FEPCA) are accorded eligible employees until the pay gap for their
area becomes sufficiently small. 

This paysetting process relies directly upon a position comparison
method for gathering and interpreting the data needed to determine
local pay comparability.  Before FEPCA, a similar method was used to
determine national comparability.  (See ch.  2.)


      EXPERTS HAVE SUGGESTED
      FURTHER CHANGES IN FEDERAL
      PAY POLICY
-------------------------------------------------------- Chapter 1:1.3

Critics of current federal pay policies claim that even with recent
reforms under FEPCA, the policies and their mechanisms are deficient
in several specific respects.  For instance, some have raised the
issue of whether the general schedule (GS) is sufficiently flexible
to permit federal agencies to compete effectively in the variety of
circumstances in which they must recruit and retain employees.\1 In
particular, the GS prescribes uniform pay rates that do not
necessarily take into account differences in prevailing rates of pay
in particular occupations.\2


--------------------
\1 The GS is a pay table that governs the salaries of most federal
employees in professional, administrative, clerical, and technical
occupations.  Federal employees covered by the GS comprise more than
50 percent of the federal civilian workforce.  There are several
salary schedules governing other groups of federal employees.  Among
these are the executive schedule, the senior executive service
schedule, the postal schedule, and the judicial salaries schedule. 

\2 However, in some instances federal agencies can obtain authority
to offer higher pay for selected occupations, if they can demonstrate
difficulties in recruiting and retaining employees in those
occupations. 


      PAY COMPARISONS AND
      PAYSETTING ARE TWO DIFFERENT
      CONCEPTS
-------------------------------------------------------- Chapter 1:1.4

The principle of comparability between federal and private (and under
FEPCA, nonfederal) sector pay--and the use of comparability
surveys--has played an important role in the paysetting process for
more than 30 years.  However, it does not necessarily follow that
future decisions concerning the level of compensation should be
completely determined by the findings of comparability surveys. 
Policymakers may want to provide for other factors to be taken into
account, such as

  possible differences in fringe benefits and other nonpay
     characteristics (e.g., job security) of federal and nonfederal
     employment;

  evidence of difficulty in recruiting and retaining federal
     employees, either in general or in specific occupations and
     localities;

  budgetary pressures faced by the federal government at any given
     point in time (e.g., a perceived need to control deficit
     spending); and

  judgments concerning the desired quality of the federal workforce. 


      THE VALIDITY OF OFFICIAL
      ESTIMATES OF THE PAY GAP HAS
      BEEN QUESTIONED
-------------------------------------------------------- Chapter 1:1.5

The government's official comparisons of federal and private sector
pay have indicated that federal pay has lagged behind prevailing
levels for comparable jobs in private industry and that the pay gap
is growing.  On the basis of evidence such as this, some analysts of
federal pay policy have said that pay is low and has led to personnel
management problems.  For instance, the National Commission on the
Public Service (commonly known as the Volcker Commission) has said
that a "quiet crisis"--due in part to low pay--in federal employment
threatens the quality of the government's day-to-day performance. 

Other analysts of federal pay policy dispute the contention that
federal employees are underpaid.  To some extent, their position
reflects disagreements concerning such policy issues as whether
federal agencies should attempt to attract and retain the "best and
brightest" talent.  However, on a more technical level, some of these
analysts--mostly academic labor economists--have questioned the
validity of official estimates of the pay gap.  They point to
evidence on the relative pay of federal and nonfederal employees from
data sources other than those used to determine the official pay gap. 
According to these analysts, this evidence would seem to suggest that
federal employees are, if anything, overpaid when compared with their
nonfederal counterparts.  In other words, federal employees are said
to receive a premium.  Further, these analysts say that evidence on
the number of applicants for federal employment and the rate at which
federal agencies retain employees does not support the notion that
federal agencies generally have problems in recruiting and retaining
employees.  Although these analysts usually have not identified what
they consider to be specific defects in the methodology underlying
the official pay gap estimates, they claim their evidence shows that
something must be wrong with it. 


   OBJECTIVES, SCOPE, AND
   METHODOLOGY
---------------------------------------------------------- Chapter 1:2

In light of the ongoing controversy concerning the existence and size
of the official pay gap, we reviewed the methodologies that have been
used to estimate the federal private pay gap.  Our objective was to
evaluate possible explanations for the apparent disagreement about
the existence and the size of the pay gap. 

We limited our review to technical issues related to the collection
and interpretation of data on pay comparability.  We did not address
broader issues in compensation policy, such as desired employee
quality; nor did we attempt to determine whether federal white-collar
employees are over- or underpaid. 

We limited the scope of our review to the pay gap as it applies to
the general schedule, which governs the salaries of most federal
white-collar employees.  We did not review issues relating to the
accuracy and quality of the data sources used to estimate the pay
gap.\3 Finally, we limited the scope of our empirical work to
comparisons of pay.  We did not analyze data on nonwage job
characteristics, success in recruitment and retention, or other
potentially relevant variables.\4

We reviewed the relevant literature, including academic research
studies, reports by government agencies, and studies prepared under
contract to government agencies.  We also interviewed analysts at the
Bureau of Labor Statistics, the Office of Personnel Management, the
National Institute of Standards and Technology (NIST), the
Congressional Budget Office (CBO), and other organizations.  On the
basis of this information, we identified possible explanations that
might reconcile the different estimates of the pay gap produced by
the different methodologies. 

We developed and estimated an econometric model using data from the
Current Population Survey, a major survey of the U.S.  workforce. 
The purpose of this analysis was to test hypotheses of why different
methodologies lead to different estimates of the pay gap.  In
particular, we analyzed relatively unexploited data on employer size
collected as part of CPS for the years 1978, 1982, and 1987, the most
recent years for which these data were available.  We did our work in
Washington, D.C., in accordance with generally accepted government
auditing standards. 

We received comments from several academic labor economists at
various stages of our work.  We received written comments on a draft
of this report from BLS (see app.  VI) and OPM (see app.  VII).  BLS
offered several clarifications and technical corrections, which we
have incorporated into the report.  OPM said the report was useful. 
The Office of Management and Budget was also offered an opportunity
to review this report but declined. 


--------------------
\3 We reported on the quality of the data used for official estimates
in Additional Improvements Needed in the National Survey of
Professional, Administrative, Technical, and Clerical Pay
(GAO/FPCD-82-32, Apr.  5, 1982), and Federal Pay:  Changes to the
Methods of Comparing Federal and Private Sector Salaries
(GAO/GGD-87-8, May 14, 1987). 

\4 See appendix I for a review of previous analyses of federal
nonfederal differences in nonpay conditions of employment.  On the
basis of this review, we determined that it was not feasible to
incorporate data on these conditions in our analysis.  Therefore, we
limited the scope of our analysis to comparisons of the pay of
federal and private sector employees, consistent with the
methodologies that are used by both the Pay Agent and the academic
studies discussed in chapters 2 and 3 of this report.  However, we
note that the evidence from the studies we reviewed suggests that if
we were able to construct broader measures of compensation for
federal and private employees, the resulting pay gap estimates
defined in terms of such measures would most likely not differ
substantially from those reported here. 


ACADEMIC STUDIES APPEAR TO
CONTRADICT OFFICIAL PAY
COMPARISONS
============================================================ Chapter 2

The method of calculating the government's official estimates of the
pay gap has been an ongoing source of controversy.  Over the years,
the official estimates have been contradicted by academic studies on
the pay gap.  The official estimates have consistently shown that
federal employees are paid relatively less than comparable private
sector employees.  However, the academic studies we reviewed
generally concluded that federal employees are paid relatively more. 

In this chapter, we discuss the different methods employed by the
government and the academic researchers.  We summarize the findings
of a number of academic studies and contrast them with official pay
gap estimates for the same period.  We then identify possible
explanations for why these analyses produce such opposing
conclusions. 


   OFFICIAL ESTIMATES FIND FEDERAL
   PAY LOW
---------------------------------------------------------- Chapter 2:1

For the period covered by our review, applicable federal law required
that federal pay rates be "comparable" to those of private sector
employees for the same level of work.  Different mechanisms exist to
establish pay levels for various segments of the federal workforce. 
Until 1989, pay gaps for the largest of these segments--white collar,
nonpostal employees covered under the general schedule--were usually
computed annually on the basis of the National Survey of
Professional, Administrative, Technical, and Clerical Pay (PATC).\1
PATC, which was conducted by BLS, provided nationwide salary
information on selected white-collar occupations in the private
sector.  The Pay Agent (the Secretary of Labor and the Directors of
OMB and OPM) was charged with selecting PATC occupations, and
ensuring that they appropriately represented a broad range of federal
white-collar occupations.  On the basis of PATC, the Pay Agent
determined and reported annually to the President the pay adjustments
necessary to maintain pay comparability.  The President had the
option of submitting an alternative proposal for pay increases to the
Congress. 

PATC consistently showed that federal employees were paid less than
their private sector counterparts.  Until the mid-1970s, federal
salaries were raised most years by an amount that, according to the
Pay Agent, would achieve pay comparability.  From 1977 until 1989,
however, the President has recommended increases that were lower than
those needed to achieve pay comparability as determined by the Pay
Agent.  As reported by the Pay Agent, the result of these successive
recommendations for lower-than-comparable pay increases and
subsequent congressional action has been to sharply reduce the
relative pay of federal employees in all GS levels.  As table 2.1
shows, the official federal pay gap increased from 10 percent in 1979
to 26 percent in 1989. 



                          Table 2.1
           
             General Schedule Pay Adjustments for
                           1979-89

                                  Pay gap as         Size of
                             reported by Pay        increase
Date                                   Agent        provided
-------------------------  -----------------  --------------
October 1979                          10.41%           7.00%
October 1980                           13.46            9.10
October 1981                           15.10            4.80
October 1982                           18.47            4.00
January 1984                           21.51            4.00
January 1985                           18.28            3.50
January 1986                           19.15            0.00
January 1987                           23.79            3.00
January 1988                           23.74            2.00
January 1989                           26.28            4.10
------------------------------------------------------------
Source:  Office of Personnel Management, and Congressional Research
Service. 


--------------------
\1 We discuss PATC in more detail, and the paysetting process in
general, in appendix II.  As we noted in chapter 1, there have been
recent changes to the process, which are also discussed in appendix
II.  Despite the changes, such as locality pay, the paysetting
process continues to rely on position-based pay comparisons similar
to those used in PATC. 


      PAY COMPARISONS BASED ON THE
      HUMAN CAPITAL APPROACH FIND
      FEDERAL PAY HIGH
-------------------------------------------------------- Chapter 2:1.1

In marked contrast to PATC, academic studies have consistently
concluded that federal employees are paid more than their private
sector counterparts.  Those studies generally employed a human
capital approach.  Rather than comparing the pay of similar jobs, as
did the official pay comparisons, the human capital method compares
the pay of individuals with similar personal characteristics, such as
education and work experience. 

Under human capital theory, employees are seen as embodying a set of
skills that can be "rented" out to firms through employment.  The
more valuable the knowledge and skills an employee possesses, the
higher the rent (i.e., the employee's pay).  An individual can
acquire more valuable knowledge and skills through education and work
experience.  Each of these activities generally requires that the
individual incur some initial costs, either in the form of
out-of-pocket expenses (e.g., tuition) or opportunities forgone
(e.g., rejecting a better paying but "dead-end" job in favor of one
with more opportunity for advancement).  When an individual decides
to incur some initial cost to acquire knowledge and skills that will
lead to higher pay, such a decision is analogous to a business
deciding to buy a new machine in order to obtain returns from its
services in the future.  These examples show how the knowledge and
skills of an employee can be viewed as productive "human capital,"
analogous to the physical capital that business plant and equipment
represent, and the initial costs to acquire knowledge or skills can
be viewed as "investments" in human capital.  The human capital
approach assumes that, to the extent that education and training are
valued only because they enhance pay, individuals will not invest in
such human capital unless the return in the form of enhanced earnings
over the employee's life at least covers all of their costs,
including interest. 

Therefore, with the human capital approach, it appears that
differences in earnings among individuals and groups can partly be
explained by observable differences in investments in human capital. 
Labor economists have used this method to study the effect education
and work experience have on the level, time pattern, and distribution
of earnings.  Statistical methods have been employed to develop
empirical formulas that implement this approach; such formulas are
called earnings functions. 

The human capital approach has also been applied to study whether
employees in one group are paid the same as those in other groups
with comparable investments in human capital.  Examples of such
applications are studies of pay differences between men and women,
minorities and whites, and union and nonunion members.  By employing
statistical methods that take into account the effect of education
and work experience, researchers have estimated the percentage of pay
differences that are attributable to gender, race, and union status. 

A number of academic studies have employed the human capital approach
to estimate the federal private sector pay gap.  The source for the
data that are most commonly used in these studies is the CPS, which
we discuss in appendix III.  One early study based on a 1978 CPS
sample indicated that federal male employees were overpaid by 11
percent and federal female employees by 21 percent.\2 The official
pay gap estimate that was based on the 1978 PATC survey indicated
that the federal pay was lower than private sector pay by about 8
percent.\3

The study did not take into account differences in employee
characteristics other than years of education and work experience. 
It also did not capture the effects of differences in nonwage job
attributes, such as work environment and fringe benefits.  To account
for the effects of some of these factors, later studies, each
employing elaborate and sophisticated econometric techniques, have
made a variety of modifications to the standard human capital
model.\4 Nevertheless, their findings, as shown in table 2.2, are
more or less similar.  The studies all indicated that federal pay was
higher than private sector pay. 



                                    Table 2.2
                     
                       Human Capital Studies on the Federal
                                 Private Pay Gap


                                  Studie   Over-                          Federa
Author                     Publ.       d     all    Male  Female   Total       l
------------------------  ------  ------  ------  ------  ------  ------  ------
Smith                       1981    1978      \a      11      21  13,148      \a
Venti                       1987    1982      \a       4      22  10,625     318
Gyourko-Tracy               1988    1977    17.6      \a      \a  13,907     431
Krueger                     1988   1984,    11.0      \a      \a   3,844      59
                                    1986
--------------------------------------------------------------------------------
Note:  For this table, the pay gap represents the percentage by which
federal salaries exceed private salaries. 

\a Not reported. 

Source:  See footnote 4. 


--------------------
\2 Sharon Smith, "Public/Private Wage Differentials in Metropolitan
Areas," Public Sector Labor Markets, eds.  Peter Mieszkowski and
George E.  Peterson (Washington, D.C.:  Urban Institute, 1981). 

\3 The numbers that we present in this section and in the remainder
of the report differ slightly from those reported by the Pay Agent,
which we present in table 2.1.  We report the pay gap as a percentage
of private sector pay, whereas the Pay Agent reported the pay gap as
a percentage of federal pay. 

\4 Steven F.  Venti, "Wages in the Federal and Private Sectors,"
Public Sector Payrolls, ed.  David Wise (Chicago:  The University of
Chicago Press, 1987); Joseph Gyourko and Joseph Tracy, "An Analysis
of Public- and Private-Sector Wages Allowing for Endogenous Choices
of Both Government and Union Status," Journal of Labor Economics,
Vol.  6 (1988), pp.  229-53; Alan B.  Krueger, "Are Public Sector
Workers Paid More Than Their Alternative Wage?  Evidence from
Longitudinal Data and Job Queues," When Public Sector Workers
Unionize, eds.  Richard B.  Freeman and Casey Ichniowski (Chicago: 
The University of Chicago Press, 1988), pp.  217-240; Brent R. 
Moulton, "A Reexamination of the Federal-Private Wage Differential in
the United States," Journal of Labor Economics, Vol.  38, No.  2
(1990), pp.  270-293. 


   POSSIBLE EXPLANATIONS FOR THE
   DIFFERENT FINDINGS
---------------------------------------------------------- Chapter 2:2

On the basis of our literature review and discussions with experts in
this area, we identified two possible explanations for the
discrepancy between the Pay Agent's estimates and those reported in
the studies by academic researchers.\5 One such explanation for the
discrepancy is that a pay comparison that uses data from a survey
like PATC compares the pay of federal employees to that of employees
of predominantly large companies in similar occupations, while the
academic studies compare the pay of federal employees to that of
employees of nonfederal employers of all sizes, regardless of the
employee's occupation.  Employees of small private employers with
given investments in human capital tend to be paid less than their
counterparts in large private firms.  As we discuss below, human
capital pay gap estimates may reflect the lower pay of employees of
small employers. 

The other explanation for the discrepancy is that position
comparisons compare pay for employees in the same occupation and work
level, ignoring the personal characteristics of the employees
compared.  By contrast, human capital methods implicitly compare
employees of the same age, education, race, and gender, largely
ignoring occupation and responsibilities.  Privately employed women
and minorities with given investments in human capital tend to be
paid less than their counterparts in the federal sector.  As we
discuss below, this fact may have different implications for position
comparison and human capital pay gap estimates. 

The two factors that we have identified were suggested by our
literature review and discussions with experts, and they lend
themselves to further analysis with the data that are available to
us.  However, there may be other factors that have contributed to the
discrepancy.\6


--------------------
\5 A discussion of these explanations can be found in Richard B. 
Freeman, "How Do Public Sector Wages and Employment Respond to
Economic Conditions," Public Sector Payrolls, ed.  David A.  Wise
(Chicago:  The University of Chicago Press, 1987), especially pp. 
189-193. 

\6 For instance, one expert on federal personnel management policy
has advanced the hypothesis that employees with given investments in
human capital tend to have different levels of responsibility in the
federal and private sectors.  See Robert W.  Hartman, Federal Pay and
Pensions (Washington:  Brookings Institution, 1983), pp.  40-45.  In
addition, an OMB official suggested to us the possibility that the
process of identifying position matches may be imperfect. 


      EMPLOYER SIZE
-------------------------------------------------------- Chapter 2:2.1

The empirical evidence of a positive relationship between pay and
employer size is overwhelming.  Even after allowing for employee
characteristics, numerous studies have found that larger employers
pay more.\7 The same relationship also appears to apply outside of
the United States.\8 Moreover, one study finds that this relationship
prevails when analysts compare the pay of employees of organizations
of various sizes within the public sector as well.\9

Position comparison surveys like PATC tend to reflect the
compensation levels of larger employers.  The specialization and
distinct level of responsibility associated with many federal
occupations mean that matches are more likely to be found in larger
nonfederal organizations.  Once a position match is found, there are
likely to be more employees employed in any such position when the
match is found in a large organization than for a smaller one.\10

By contrast, human capital estimates of the pay gap generally have
been based on data from sources such as CPS, which cover employees of
employers of all sizes.  Thus, a PATC-based approach compares federal
employees to nonfederal employees of predominantly large companies
while most human capital estimates compare federal employees to
nonfederal employees of companies of all sizes. 

Because large employers pay more than small ones, employer size could
affect estimates of the federal private pay gap.  To date, few
academic studies of the federal private sector pay gap have attempted
to isolate the effect of employer size on the difference in pay
between the two sectors.\11


--------------------
\7 Stanley H.  Masters, "Wages and Plant Size:  An Interindustry
Analysis," Review of Economics and Statistics, Vol.  51 (1960), pp. 
341-345; Sherwin Rosen, "Unionism and the Occupational Wage Structure
in the United States," International Economic Review, Vol.  11
(1970), pp.  269-286; Charles T.  Haworth and Carol Jean Reuther,
"Industrial Concentration and Interindustry Wage Determination,"
Review of Economics and Statistics, Vol.  60 (1978), pp.  85-95;
Wesley Mellow, "Employer Size and Wages," Review of Economics and
Statistics, Vol.  64, No.  3 (1982), pp.  495-501; John E.  Garen,
"Worker Heterogeneity, Job Screening, and Firm Size," Journal of
Political Economy, Vol.  93, No.  4 (1985), pp.  715-739; Charles
Brown and James Medoff, "The Employer Size- Wage Effect," Journal of
Political Economy, Vol.  97, No.  5 (1989), pp.  1027-1059. 

\8 Robert Evans, for example, finds that in Japan larger employers
pay substantially more than smaller ones in "Pay Differentials:  The
Case of Japan," Monthly Labor Review, Vol.  107, No.  10 (1984), pp. 
24-29. 

\9 Charles C.  Brown and James L.  Medoff, "Employer Size, Pay, and
the Ability to Pay in the Public Sector," When Public Sector Workers
Unionize, eds.  Richard B.  Freeman and Casey Ichniowski (Chicago: 
The University of Chicago Press, 1988), pp.  217-240. 

\10 BLS has undertaken initiatives to include more small employers in
its surveys.  These efforts to increase smaller employers'
representation have been costly and have not appreciably affected
official comparability estimates because few position matches were
found in the smaller private firms surveyed by BLS. 

\11 One such study is Dale Belman and John S.  Heywood, "The Effect
of Establishment and Firm Size on Public Wage Differentials," Public
Finance Quarterly, Vol.  18, No.  2 (1990), pp.  221-235.  They found
that when employer size is taken into account in human capital models
it is unclear that federal pay is higher than private sector pay;
this is contrary to the findings of most other human capital studies. 


      THE FEDERAL PAY OF WOMEN AND
      MINORITIES
-------------------------------------------------------- Chapter 2:2.2

Federal personnel management policy implements the government's
commitment to prohibit all types of illegal discrimination and takes
affirmative action to ensure equal employment opportunity (EEO). 
Although similar legal requirements apply to private sector
employers, several human capital studies show that on average private
employers are likely to pay lower wages than federal employers to
women and minorities with comparable investments in human capital.\12
Thus, there is a strong possibility that differences in the levels of
pay for women and minorities between the two sectors may affect
estimates of the pay gap. 

Further, there is evidence that to the extent that the pay of women
and minorities tends to be lower in the private sector, it largely
takes the form of a higher concentration of women and minorities in
lower-paying occupations, as opposed to unequal pay within narrowly
defined occupations.  To this extent, a pay comparison that is based
on position comparisons within categories that are defined in terms
of both occupation and work level, such as PATC, is likely to be less
affected by race and gender effects.  By contrast, human capital
methods, which compare pay across occupations, are likely to be
affected.  Hence, human capital estimates may be sensitive to the
specific assumptions that analysts make regarding race and gender
effects. 

One decision regarding race and gender effects that analysts
implicitly make when estimating federal private sector pay gaps
concerns the choice of private sector comparison group.  The possible
options involve implicit assumptions about the reasons for race- and
gender-based pay differences within the private sector. 

One way in which pay gaps can be computed in human capital models is
to compare federal employees to all private sector employees.  Such
comparisons can produce a single estimate of the pay gap, assumed to
be the same for all race-gender groups, or they can produce separate
estimates of the pay gap for each race-gender group.  The studies
that we cite in table 2.2 use all private sector employees as the
comparison group.  Comparing federal employees to all private sector
employees means the lower pay of private sector women and minorities,
relative to private sector white males, will influence the size of
the pay gap.  Such a comparison implicitly assumes that private
sector pay differences by race and gender are caused by unobserved
productivity differences that are not necessarily related to
education and work experience.\13

However, several researchers have argued that it is possible that a
federal pay premium could result from federal white males being paid
the same as private sector white males while federal women and
minorities are paid more than their private sector counterparts,
after controlling for education and work experience.\14 Based on this
argument, an alternative method of computing the pay gap would be to
compare the pay of federal white males, women, and minorities to the
pay of private sector white males.  Comparing all federal employees
to private sector white males means the lower pay of private sector
women and minorities, relative to private sector white males, will
not influence the size of the pay gap.  Such a comparison implicitly
assumes that private sector pay differences by race and gender are
caused by discrimination. 

Using private sector white males as the benchmark for comparison
could be described as a method that measures the upper limit of the
contribution of private sector labor market discrimination to an
explanation of the discrepancy between estimates of the pay gap.  It
is an upper limit if labor market discrimination and unobservable
productivity differences share responsibility for private sector pay
differences by race and gender.\15


--------------------
\12 Martin Asher and Joel Popkin, "The Effect of Gender and Race
Differentials on Public-Private Wage Comparisons:  A Study of Postal
Workers," Industrial and Labor Relations Review, Vol.  38, No.  2
(1984), pp.  16-25.  See also Sharon Smith, "Pay Differential between
Federal Government and Private Sector Workers," Industrial and Labor
Relations Review, Vol.  29 (1976), pp.  179-197, and Equal Pay in the
Public Sector:  Fact or Fantasy (Princeton, N.J.:  Princeton
University Press, 1977). 

\13 Human capital estimates are sensitive to how well education and
work experience are measured.  See chapter 4 for further discussion. 

\14 Asher and Popkin, 1984; Jeffrey M.  Perloff and Michael L. 
Wachter, "Wage Comparability in the U.S.  Postal Service," Industrial
and Labor Relations Review, Vol.  38, No.  2 (1984), pp.  26-35. 

\15 We recognize the possibility that productivity differences may
themselves be partly caused by discrimination in such factors as
education or past employment. 


   CONCLUSION
---------------------------------------------------------- Chapter 2:3

On the basis of our review of the relevant studies, we have
identified two factors for further analysis that may affect estimates
of the federal private pay gap.  One factor is the effect of employer
size on estimates of the pay gap.  The other factor is the pay of
federal women and minorities, relative to private sector white males. 
By identifying these two factors for further analysis, we do not mean
to rule out the possibility that there are other contributing
factors. 


THE EFFECTS OF EMPLOYER SIZE AND
EMPLOYEE RACE AND GENDER ARE
SIGNIFICANT FACTORS IN PAY
COMPARISONS
============================================================ Chapter 3

In light of the opposing conclusions we discussed in chapter 2, we
sought to determine why the estimates of the federal private pay gap
that were reported by the Pay Agent differed from those derived from
a human capital earnings function.  Our review of academic studies
pointed to private employer size and the federal pay of women and
minorities as possible explanations.  Our objective was to measure
the effects of these possible explanations on human capital estimates
of the pay gap. 

This chapter presents the results of our human capital analysis of
CPS data.\1 This analysis consisted of two parts.  In the first part,
the analysis of data on the earnings of full-time federal and
nonfederal employees for each year from 1978 through 1987 used the
standard human capital model.  This part of the analysis served two
purposes--first, to document trends in the size of the pay gap over
this period, as measured using both the Pay Agent's and human capital
methods; and second, to determine the extent to which these human
capital estimates were consistent with those found by the academic
researchers. 

In the second part, we analyzed supplemental CPS data on earnings for
the years 1978, 1982, and 1987\2 to determine the extent to which the
opposing conclusions of the Pay Agent and the human capital analyses
of CPS data could be accounted for by variations in employer size and
by the earnings of federal women and minorities, respectively.  We
selected these years for analysis because CPS provided more detailed
information on the characteristics of the respondents' employers,
e.g.,(firm and establishment size) in these years, thereby enabling
us to examine the possible explanations mentioned above.\3


--------------------
\1 The March CPS Annual Demographic File contains information on
earnings and demographics that is commonly used by academics to
estimate pay gaps using the human capital method.  See appendix III
for more information on this survey. 

\2 We used May CPS supplements on pension and employee benefits that
contained matching earnings and demographic information from the
preceding March CPS.  See appendix III for a more detailed discussion
of this survey. 

\3 See appendix IV for a more detailed discussion of the methodology
used. 


   HUMAN CAPITAL ESTIMATES OF PAY
   GAP DIFFER FROM OFFICIAL
   ESTIMATES
---------------------------------------------------------- Chapter 3:1

GAO analyzed CPS data on full-time employees, ages 18 to 65, for the
years 1978 through 1987 by estimating standard human capital earnings
functions.  The resulting estimated pay gaps and the corresponding
pay gaps reported by the Pay Agent are shown in figure 3.1.\4

   Figure 3.1:  The Pay Gap as a
   Percentage of Private Sector
   Pay

   (See figure in printed
   edition.)

Source:  GAO analysis of CPS data and Pay Agent's reports for various
years. 

Estimates based on the standard human capital analysis of CPS data
are strikingly different from those reported by the Pay Agent.  The
estimates that are based on standard human capital analysis of CPS
data consistently show that federal employees are paid relatively
more than their private sector counterparts, while official estimates
of the pay gap based on PATC show the opposite. 

These opposing conclusions mirror the findings of the human capital
studies of the federal private pay gap we discussed in chapter 2. 
However, both the Pay Agent's and the human capital's estimates agree
about the decrease in the relative pay of federal employees.  The pay
of federal employees, relative to the private sector, decreased by 6
to 14 percentage points over the period from 1978 to 1987. 


--------------------
\4 We remind the reader that we have computed the pay gaps that we
report here and elsewhere in this chapter in the manner that we
describe in chapter 2.  Also, we present the sample statistics and
complete regression results that underlie this and other figures in
appendix V. 


      CPS PENSION SUPPLEMENT DATA
      CONFIRM PAY GAP DISCREPANCY
-------------------------------------------------------- Chapter 3:1.1

We estimated the pay gap for the years 1978, 1982, and 1987 by
applying the standard human capital method to CPS pension supplement
data.  We narrowed the CPS sample to federal and private sector
white-collar employees to better match those included in PATC.  While
the resulting pay gap estimates were lower than those shown in figure
3.1, both series of CPS-based estimates differed substantially from
those based on PATC data.  Pay gap estimates from the CPS-based human
capital comparisons indicate that federal employees are paid
relatively more than their private sector counterparts.  Our
estimates based on CPS pension supplement data and the corresponding
Pay Agent's numbers are shown in figure 3.2.  These human capital
estimates based on the CPS pension supplement show that federal pay
declined over the 10-year period by a little more than 12 percentage
points. 

   Figure 3.2:  May CPS Pay Gap
   Estimate

   (See figure in printed
   edition.)

Source:  GAO analysis of CPS data and Pay Agent's reports for various
years. 


   ANALYSIS OF DIFFERENCES BETWEEN
   THE PAY AGENT'S REPORTS AND THE
   HUMAN CAPITAL ANALYSIS
---------------------------------------------------------- Chapter 3:2

Chapter 2 described two possible explanations for a discrepancy
between estimates in the Pay Agent's reports and those published in
academic studies.  The first explanation concerns the relative pay of
employees of large and small employers.  The second explanation
concerns the relative pay of women and minorities in private and
federal employment.  This section explains how we analyzed CPS data
for selected years to determine the empirical importance of each of
these possible explanations. 


      THE EFFECT OF EMPLOYER SIZE
      ON THE PAY GAP ESTIMATES
-------------------------------------------------------- Chapter 3:2.1

Although annual CPS data did not regularly contain information on
employer size, at approximately 5-year intervals a supplemental CPS
survey on pensions and employee benefits collected the needed
information on employer size.  We expected that by using this
employer size data when we produced human capital estimates of the
pay gap, we could measure the effect that employer size has on
estimates of the federal private pay gap.  To measure the effect of
employer size on the pay gap, we again estimated the human capital
earnings functions.  In doing this estimate, we allowed for the
effects of employer size, so that we compared the earnings of federal
employees to the earnings of large private sector employers.\5

The results of our adjustment are shown in figure 3.3.  The figure
shows that the discrepancy in the estimated pay gap was smaller after
we adjusted for private sector employer size.  This result strongly
suggests that the greater proportion of employees from small
employers in CPS compared with PATC contributes to the finding of a
positive pay premium for federal employment in the CPS-based
estimates.\6

   Figure 3.3:  The Pay Gap
   Adjusted for Employer Size

   (See figure in printed
   edition.)

Source:  GAO analysis of CPS data and Pay Agent's reports for various
years. 


--------------------
\5 As explained in appendix IV, our analysis provides a comparison of
the average federal employee to the average employee in a private
sector establishment with over 1,000 employees, after adjusting for
other characteristics, such as education and experience.  This
represents an approximation to the effect that employer size could
have on the discrepancy in pay gap estimates, because not all private
establishments surveyed in PATC have over 1,000 employees. 

\6 This finding is broadly consistent with that of the Belman-Heywood
study cited in chapter 2. 


      THE EFFECT OF THE FEDERAL
      PAY OF WOMEN AND MINORITIES
      ON PAY GAP ESTIMATES
-------------------------------------------------------- Chapter 3:2.2

We also reestimated human capital earnings functions in a way that
allowed us to compare the pay of federal employees in all race and
gender groups to the pay of private sector white males, after
controlling for education and work experience.  We then computed pay
gap estimates as a weighted average of the race gender-specific
federal private pay gaps.  (See app.  IV). 

Figure 3.4 shows that the discrepancy in the estimated pay gap is
smaller after this adjustment for the higher federal pay of women and
minorities.  This result shows that the manner in which the analyst
accounts for the higher federal pay of women and minorities can
affect estimates of the pay gap.  The gap is smaller when federal
white males, women, and minorities are compared to private sector
white males rather than to private sector white males, women, and
minorities, respectively. 

   Figure 3.4:  The Pay Gap
   Adjusted for the Federal Pay of
   Women and Minorities

   (See figure in printed
   edition.)

Source:  GAO analysis of CPS data and Pay Agent's reports for various
years. 


      THE COMBINED EFFECT OF
      EMPLOYER SIZE AND THE
      FEDERAL PAY OF WOMEN AND
      MINORITIES
-------------------------------------------------------- Chapter 3:2.3

Our adjustments for the effects of employer size and the federal pay
of women and minorities each account for a substantial amount of the
difference between human capital and official estimates of the pay
gap.  If we were to add these two potential effects together, the
total would exceed the difference between the Pay Agent's estimate of
the federal private pay gap and the alternative measure from the
simple human capital method. 

This suggests that the effects of employer size and employee race and
gender together potentially could account for the full discrepancy in
measuring the pay gap.  However, these factors may be interrelated in
a statistical sense.  In this case, both adjustments may be measuring
roughly the same thing.  The addition of the separately estimated
effects would then be misleading. 

To determine whether the effects of employer size and employee race
and gender were interrelated in the human capital method, both sets
of factors need to be adjusted simultaneously.  By including controls
for the effects of employer size and employee race and gender, we
measured the joint effect of these factors on estimates of the pay
gap.  Our results are shown in figure 3.5. 

   Figure 3.5:  The Pay Gap
   Adjusted for Employer Size and
   the Federal Pay of Women and
   Minorities

   (See figure in printed
   edition.)

Source:  GAO analysis of CPS data and Pay Agent's reports for various
years. 

The combined effect of the two possible causes of the discrepancy is
roughly equal to the difference between the two pay gap measures. 
Our analysis explained the discrepancy by adjusting for the effects
of employer size and sector-specific pay gaps related to race and
gender.  The effects of employer size and employee race and gender
appear to be substantially independent.  Although we analyzed the two
effects identified in our review of previous analyses of the pay gap,
there may be other factors that contribute to the opposing
conclusions. 


   CONCLUSION
---------------------------------------------------------- Chapter 3:3

Our analysis of CPS data for 1978 to 1987 has shown that the human
capital method, as applied in a manner similar to that of other
analyses, has consistently yielded estimates of the pay gap that
differ substantially from those produced by the Pay Agent for the
same period.  This finding is consistent with the findings of the
other academic researchers whose work we reviewed in the previous
chapter.  Our analysis also shows that the position comparison and
human capital methodologies agree that federal pay compared to
private sector pay has declined over the same period. 

Our analysis of CPS data for the years 1978, 1982, and 1987 shows the
significance of the two factors we identified.  We found substantial
narrowing of the differences between the position comparison and
human capital estimates of the pay gap after adjusting for the effect
of employer size on earnings.  Further, we found a substantially
smaller discrepancy in measured pay gaps after adjusting the human
capital estimates so that all federal employees were compared to
private sector white males.  The combined effects of these two
adjustments produce human capital estimates of the pay gap that are
similar to the official estimates. 


IMPLICATIONS OF GAO'S ANALYSIS
============================================================ Chapter 4

Our analysis of employer size and employee race and gender as
potential explanations of the differences between the results of the
position comparison and human capital approaches must be understood
within the broader framework of federal personnel management policy. 
Federal personnel management policy includes such important factors
as workforce quality, recruitment and retention, affirmative action,
and employee benefits, which may be influenced by the level of pay. 
Because these factors are beyond the scope of this report, our
analysis cannot by itself be used to judge the appropriateness of
comparability estimates or the level of federal pay. 

Our human capital analysis shows the importance of considering the
effect of employer size and employee race and gender on private
sector pay when evaluating the two approaches for measuring pay
comparability.  Further, both the position comparison and the human
capital method have limitations in estimating pay gaps. 


   POSITION COMPARISONS AND HUMAN
   CAPITAL ESTIMATES ARE LIMITED
   IN MEASURING PAY COMPARABILITY
---------------------------------------------------------- Chapter 4:1

Position comparisons and human capital estimates are different
methods for comparing federal and nonfederal pay.\1 Each method has
strengths, but each also has weaknesses; neither method is clearly
superior.  Although annual comparability adjustments are no longer
linked to the PATC survey, the Federal Employees Pay Comparability
Act calls for locality pay adjustments to the general schedule based
on position comparisons.  Our analysis contributes to discussions on
the strengths and weaknesses of using a position comparison method to
compare federal and nonfederal pay. 


--------------------
\1 There is a difference between comparing salaries and setting
salaries.  Although both position comparisons and human capital
methods are used to compare salaries, we know of no cases where the
human capital method is used by employers to set or adjust pay. 


      POSITION COMPARISONS
-------------------------------------------------------- Chapter 4:1.1

Position comparisons are based on the specific characteristics of a
job and the pay associated with such a job rather than on the
individual characteristics of the employee in the job.  Position
comparisons address what other employers pay staff in a specific job. 
They are used to measure the pay associated with a particular job. 
To do such a comparison, job descriptions from different employers
are matched and the accompanying levels of pay are compared. 

Position comparisons are an accepted way for employers to learn what
other employers are currently paying employees to perform specific
jobs.  Many nonfederal employers purchase such information from
compensation consultants and use it for such purposes as setting
starting pay, adjusting pay, and determining the competitiveness of
compensation.  The occupational detail and the number of matched jobs
and employees is typically smaller in most applications of the
position comparison method than was the case with PATC.  To average
position comparison data, the Pay Agent blends nonfederal salaries
using the federal occupation and pay distribution and arrives at a
pay gap for each GS grade level.  The official pay gap estimate that
has been criticized by some academics is a weighted average of the
pay gaps for each GS grade level. 

Pay data produced by position comparisons will reflect the nonfederal
pay for federal occupations regardless of an employee's race or
gender.\2 A position comparison survey like PATC does not distinguish
whether a job is being staffed by men, women, whites, or minorities. 
Thus, the use of a position comparison method will neutralize the
effect of race and gender in comparing the salaries of federally
employed women and minorities in occupations that are commonly
staffed by white males in the private sector. 

Most of the pay differences by race and gender within an organization
are attributable to the narrowly defined job categories in which
individuals are employed.  Evidence exists that women and minorities
in the private sector are concentrated in lower paying jobs.  It is
certainly possible that this concentration reflects, at least partly,
discrimination.  Using position comparison data for such occupations
on a job-by-job basis would extend the lower pay for these positions
to the federal sector.\3

Position comparison surveys like PATC tend to reflect the pay level
of large employers.  Because of the specialization and distinct level
of responsibility associated with many federal jobs, position matches
for such jobs are more likely to be found in large nonfederal
organizations.  Once a position match is found, there are likely to
be more matching employees employed in any such job when the match is
found in a large organization.  BLS is now including more small
employers in its surveys.  These initiatives to increase the
representation of small employers have been costly.  Also, they have
not appreciably affected official pay gap estimates because few
position matches were found with the smaller employers surveyed by
BLS.\4

Because position matches are more likely to be found with large
employers, using the position comparison method makes it inevitable
that the collected pay data will tend to reflect the pay of large
nonfederal employers and therefore will be higher than the average
nationwide pay.  Thus, basing federal pay on position comparisons
could make the federal workforce appear to be higher paid in
comparison to the nation as a whole. 

Position comparisons are only as good as the quality of the job
matches and the position descriptions.  If the matches are poor or
the comparison group is poorly chosen, the survey data on the pay of
nonfederal jobs could be misleading.  It is equally important to
ensure that the position descriptions accurately reflect the duties,
responsibilities, and qualifications of the federal employees. 
Otherwise the comparison by job description will not be valid. 

We examined the quality of position matching from past PATC data and
reported that the result was accurate.\5 Although there is no
guarantee that such accuracy has been maintained, our report
indicates that pay surveys based on position comparisons have been
conducted effectively by the federal government. 


--------------------
\2 Although average nonfederal pay varies systematically by race and
gender, these differences within the narrowly defined occupations of
an employer are relatively small. 

\3 Some analysts have argued that this is a moot point because the
applicable law requires that federal pay be compared to prevailing
private sector pay, as opposed to the pay of white males or any other
subset of the private workforce, or hypothetical prevailing pay
levels in the absence of discrimination.  See Perloff and Wachter,
op.  cit. 

\4 See Changes to the Methods of Comparing Federal and Private Sector
Salaries (GAO/GGD-87-8, May 14, 1987). 

\5 See Changes to the Methods of Comparing Federal and Private Sector
Salaries (GAO/GGD-87-8, May 14, 1987). 


      HUMAN CAPITAL ESTIMATES
-------------------------------------------------------- Chapter 4:1.2

Human capital estimates link differences in individual employees' pay
to common measurable characteristics, such as race, gender, union
membership, and federal employment.  These estimates also account for
individual differences attributable to education and accumulated work
experience.  This approach is commonly used by many labor economists
for studying these kinds of pay differences. 

An attractive feature of the human capital approach is that the
analyst easily obtains a pay gap estimate without resorting to costly
position comparisons.  The data that are used to compute human
capital estimates have usually been collected for other purposes. 
Therefore, these data are both widely and inexpensively available. 
However, little work has been done to answer the question of how
representative these data are of nonfederal comparison groups. 

By using the human capital approach, the process of estimating the
pay gap can be simplified.  However, there are data and specification
issues that raise concerns as to the applicability of such estimates
of the pay gap.  Our work in chapter 3 suggests the analysts should
exercise caution when they use the empirical results of the human
capital estimates to compare pay.  We demonstrated that the effect of
race and gender on the human capital estimates of the pay gap is
sensitive to the choice of the private sector comparison group.  The
reason for this sensitivity is because privately employed women and
minorities tend to be concentrated in lower-paying occupations.  In
addition, we found that employer size also affects the pay gap. 

Human capital estimates of differences in pay between groups, such as
between male and female employees or federal and private employees,
reflect the average pay for groups of employees that share common
measurable characteristics.  The appropriateness of using such pay
gap estimates for federal paysetting depends on how well these
characteristics are measured and on the importance of any unmeasured
characteristics.  The human capital earnings functions that were
estimated for this report attributed approximately 40 percent of the
differences in the pay being compared to the following factors: 
years of education, age, race, gender, employer size, and sector of
employment.\6

Although this amount is quite good by academic standards, such
results still leave a majority--approximately 60 percent--of the
differences in pay unexplained.  Much of this difference is
attributable to factors, such as ability, intelligence, leadership,
and motivation, that analysts are unable to observe directly. 

The human capital estimates that we present in this report also
reflect the assumption that measured characteristics are equivalent
for all groups.  An example of such a measured characteristic is the
years of schooling.  Each year of formal education is counted as a
year of schooling and each year is assumed to be equivalent.  Pay
differences that are due to choice of college major, type of graduate
degree, quality of instruction, completion of studies, and academic
honors earned are typically ignored.  When groups that are being
compared in a human capital earnings function differ in these
characteristics, the estimated differences in pay could reflect these
differences in the quality of education.  For example, if federal and
nonfederal employees have advanced degrees from universities and
colleges of different quality, then pay differences that are
attributed to federal employment may in fact be due to the
differences in quality of education. 

Also, years of potential work experience are assumed to be
equivalent,\7 and the human capital model assumes that average pay
grows over time in a similar way for all employees.  However, if
advancement opportunities are better for some employees, then the
relative pay of this group will increase directly with their average
age.  The pay difference appears largest when we compare individuals
at the height of their careers.  Any estimated pay difference could
then reflect differences that are due to choice of career and labor
force participation decisions. 

We know of no studies that answer the question of the appropriateness
of the nonfederal employees surveyed in CPS--or in other similar data
sources--as a comparison group.  Only to the extent that the
nonfederal group is appropriately comparable will the results of a
human capital comparison be useful in determining pay comparability. 
Ideally, such a comparison group would consist of individuals that
the federal government would be willing to hire, drawn from an
occupational mix that is comparable to the federal government's.  The
use of the human capital approach can result in the comparison of
nonfederal computer programmers to federal secretaries or federal
lawyers to nonfederal librarians.  Such a comparison may yield an
estimate of relative pay that is partially attributable to
differences in the occupational distribution, rather than providing
useful information on the comparability of federal and nonfederal
pay.  As we discussed earlier, other critical characteristics to
consider include the type of job, type and quality of education,
on-the-job training, career paths, and advancement opportunities. 

The position comparison and human capital methods are different
methods for comparing federal to nonfederal salaries.  The position
comparison approach goes to great lengths to ensure the comparability
of occupations and then arrives at an overall average that obscures
much of the detailed information gathered.\8 The human capital
approach arrives at an average without the need for the occupational
detail but may be sensitive to the choice of nonfederal comparison
groups. 


--------------------
\6 A commonly used measure of how well an econometric model accounts
for variations in the data being analyzed is the adjusted R-squared. 
A typical adjusted R-squared for human capital models is in the
neighborhood of 0.4, which implies that approximately 40 percent of
the variation in salaries across individuals is explained by the
estimated human capital earnings function. 

\7 Potential work experience is defined as age in years minus years
of education minus the 6 years before the individual started first
grade.  An additional problem arises when the link between actual and
potential work experience varies for groups studied.  Potential work
experience is greater than actual experience when individuals are
both out of work and out of school for long periods of time.  Actual
experience may exceed the potential for individuals who worked full
time while attending college or graduate school. 

\8 Although the law intends that pay comparability increases be
determined separately for each GS level, historically, the practice
has been to grant uniform comparability increases for all GS levels. 


      COMMON LIMITATIONS OF PAY
      COMPARISONS
-------------------------------------------------------- Chapter 4:1.3

Any method that is used to arrive at a single comparability number is
bound to be open to criticism.  An appropriate level of compensation
cannot be arrived at without considering the consequences for
personnel management.  In addition, nonpay aspects of compensation,
such as fringe benefits, job security, working conditions,
advancement opportunities, and on-the-job training, substitute to
some degree for purely monetary rewards.\9

Any method that is used to arrive at a single comparability number
cannot be expected to apply with precision to every individual being
compared.  Any single number is likely to be an average of many
differently paid individuals with different skills and
responsibilities.  The inevitable result of averaging is that
individuals on each side of the comparison are paid higher and lower
than any one comparability number might suggest. 

Both of the methods that we discuss in this report are typically
unable to take into account the quantity or quality of employees'
work, because no quality or quantity indicators in the compensation
data are currently available. 


--------------------
\9 We discuss attempts to include these nonpay aspects of
compensation to arrive at a measure of total compensation
comparability in appendix I. 


   COMPENSATION LEVELS ARE NOT
   INDEPENDENT OF PERSONNEL
   MANAGEMENT PRIORITIES
---------------------------------------------------------- Chapter 4:2

The appropriate level of compensation for a job does not exist in a
vacuum.  Pay and benefits provide not only compensation for services
rendered but also incentives for improvements in employee
performance.  Compensation can be used to attract and retain
employees.  In addition, pay and benefits serve to some degree as
substitutes for each other. 

There is no easy answer to the question of the appropriateness of
federal general schedule compensation.  Any shortcomings of pay gap
estimates do not necessarily invalidate the use of such estimates in
determining appropriate levels of compensation.  Whether there are
doubts as to the accuracy of such comparisons, the federal government
must consider pay competitiveness in the broader context of federal
personnel management policy.  Federal personnel management priorities
such as turnover, retention, recruiting, workforce quality, labor
market competition, and the achievement of EEO/affirmative action
goals are all considerations that are influenced by compensation and
have their own implications for the appropriate level of
compensation.  Critics of the federal pay system sometimes cite one
or more of those priorities as evidence that federal compensation is
high or low, while ignoring other personnel management priorities. 
Paysetters and lawmakers need to carefully weigh all aspects of the
compensation question when determining the appropriate level of
federal compensation.  Compensation is considered high or low only in
relation to the personnel management goals being considered. 


ANALYSES OF TOTAL COMPENSATION
COMPARABILITY
=========================================================== Appendix I

Many factors, in addition to pay, may affect the success of employers
in recruiting and retaining employees.  Some of these factors are

  pension benefits;

  health insurance;

  the risk of on-the-job injuries;

  the risk of being laid off;

  vacations, sick leave, and holidays;

  working conditions;

  the inherent ("psychic") satisfaction of the job. 

A number of experts in labor market analysis have suggested that
federal private compensation comparisons that focus exclusively on
pay may be misleading.  They have said that systematic differences
tend to exist between federal and private sector employment
concerning the nonpay conditions of employment. 

As an alternative to the principle of pay comparability as it is
currently defined and implemented,\1 these experts have suggested
that the principle of total compensation comparability (TCC) be
implemented.  Under the TCC approach, a monetary value for
employer-subsidized fringe benefits is imputed for federal and
comparable private sector jobs.  While these experts recognize that
not all differences in job characteristics between the two sectors
can be quantified, they think that those elements that can be
quantified can capture a substantial portion of the differences
between jobs. 

We identified several previous attempts to create total compensation
measures for the comparison of federal and nonfederal compensation. 
In this appendix we review these studies, along with evaluations of
their analyses. 


--------------------
\1 See appendix II. 


      OFFICE OF PERSONNEL
      MANAGEMENT STUDY
------------------------------------------------------- Appendix I:0.1

In the mid-1970s, OPM began a program of research to assess the
feasibility of TCC as a basis for setting federal pay.  In embarking
on this effort, OPM relied heavily on analyses undertaken by
actuarial specialists under contract.  These analyses were based on
data on private sector employee benefit plans gathered by the Bureau
of Labor Statistics.  A former official of OPM who was heavily
involved with this effort told us that it was extremely difficult to
perform meaningful comparisons, and that this effort was
discontinued. 

Despite the difficulty in comparing benefits between the federal and
private sectors, in 1981 the Administration recommended a change in
the pay-setting process that was based on the TCC concept. 
Specifically, the Administration recommended that increases in the
general schedule be limited in order to attain a level of pay that is
94 percent of the level of comparability with the private sector, as
determined by the results of the PATC survey, rather than the
100-percent target prescribed by the Federal Pay Comparability Act of
1970.  In large part this recommendation was based on intangible
aspects of federal employment, including the presumed greater
employment stability associated with federal employment compared with
the private sector, as well as the greater attractiveness of federal
nonwage benefits.  Further, the 94-percent target was admittedly
judgmental.  In reviewing this recommendation, GAO found that the
justification for this proposal was inadequate.\2


--------------------
\2 Proposal to Lower the Federal Compensation Comparability Standard
Has Not Been Substantiated (FPCD-82-4, Jan.  26, 1982). 


      CONGRESSIONAL RESEARCH
      SERVICE STUDY
------------------------------------------------------- Appendix I:0.2

In 1984 the House Post Office and Civil Service Committee published a
report on options for the design of a new retirement system for
federal civilian employees.\3 This report was largely based on
analyses conducted by the Congressional Research Service (CRS), as
well as by actuarial consultants under contract to CRS.  One segment
of this report compared the federal retirement benefits (including
survivor and disability benefits) with typical plans in the private
sector.  The analysts found that federal employees received
retirement benefits that were one-third more generous than the most
generous private plans.\4

GAO notes that these computations were based on the package of
employee benefits offered to federal employees at that time under the
Civil Service Retirement System (CSRS).  Although the purpose of the
CRS report was to estimate the cost of several alternatives to CSRS,
none of the alternatives that were analyzed exactly corresponded to
the replacement retirement plan that was finally adopted. 


--------------------
\3 U.S.  Congress, House Committee on Post Office and Civil Service,
Designing a Retirement System for Federal Employees Covered by Social
Security, December 1984. 

\4 This finding is based on a comparison of employer cost associated
with the various plans. 


      GAO BENEFIT COMPARISONS
------------------------------------------------------- Appendix I:0.3

GAO reported a comparison between federal and private employment with
respect to benefits in 1985.  GAO presented comparisons between
federal and private sector pay and benefits that took into account
health insurance, life insurance, and annual and sick leave and
holidays, as well as retirement benefits.\5 GAO found that private
employers tended to offer benefits other than retirement benefits
that were at least as good as those of the federal government, and,
in the case of health and life insurance, were significantly more
generous. 


--------------------
\5 Comparison of Federal and Private Sector Pay and Benefits
(GAO/GGD-85-72, Sept.  4, 1985). 


      NATIONAL INSTITUTE OF
      STANDARDS AND TECHNOLOGY
      DEMONSTRATION PROJECT
------------------------------------------------------- Appendix I:0.4

The National Institute of Standards and Technology, formerly the
National Bureau of Standards, is currently conducting a demonstration
project testing alternative compensation schemes designed to enhance
NIST's flexibility in meeting competition from the private sector for
scientists, engineers, and other staff.\6 Among other things, the
NIST project had been assessing the feasibility of basing pay on a
total compensation comparability principle.  However, this aspect of
the project was discontinued for budgetary reasons.\7


--------------------
\6 GAO reported on this project in Federal Workforce:  Information on
the National Bureau of Standards Personnel Demonstration Project
(GAO/GGD-88-59FS, Apr.  5, 1988). 

\7 Federal Personnel:  Special Authorities Under the Demonstration
Project at Commerce (GAO/GGD-92-124BR, July 13, 1992). 


FEDERAL PAY COMPARABILITY PROCESS
========================================================== Appendix II

Congress directed BLS to conduct an annual survey of private sector
salaries to provide the Pay Agent with data to make annual pay
comparability recommendations.  For the period covered by our report,
BLS responded to this mandate by conducting the National Survey of
Professional, Administrative, Technical, and Clerical Pay. 

In this appendix, we discuss the process by which pay comparability
recommendations are developed, focusing on those features of PATC
that are most relevant to this report.\1 We also include information
on changes to the paysetting process that have been enacted since the
period covered by our review.\2


--------------------
\1 In 1989, PATC was renamed the White-Collar Pay Survey. 

\2 BLS has discontinued the White-Collar Pay Survey.  The locality
pay data gathering effort combines pay information for private
employees with that for state and local government employees.  This
information is used on a locality basis to measure locality pay gaps. 


   PATC SURVEY OF PRIVATE SECTOR
   SALARIES
-------------------------------------------------------- Appendix II:1

The initial stage in the pay comparability process involved the
collection of private sector pay data.  Several steps were
involved.\3


--------------------
\3 Unless otherwise noted, the information in this section is taken
from U.S.  Department of Labor, Bureau of Labor Statistics, National
Survey of Professional, Administrative, Technical, and Clerical Pay: 
Private Workservice Industries, March 1988, Bulletin 2317 (November
1988). 


      PRIVATE OCCUPATIONS
------------------------------------------------------ Appendix II:1.1

In administering the pay survey, BLS and OPM developed narrowly
defined work levels for selected occupations in the private sector to
reflect the same level of work performed in GS grades 1 through 15. 
PATC survey occupations and work levels were selected on the basis of
three criteria.  First, an occupation had to be surveyable in private
enterprise establishments.  Second, it had to be representative of
occupational groups that are numerically important in both the
federal and private sectors.  Finally, a job had to be of essentially
the same nature in both sectors.  The occupational coverage of PATC
was continually revised and expanded over the years.  The most recent
survey covered 30 occupations and 133 work levels.  (See table II.1.)



                          Table II.1
           
           Occupations and Work Levels Surveyed by
                         PATC in 1988

                                                      Number
                                                          of
                                                        work
Occupational category                                 levels
--------------------------------------------------  --------
Professional
Accountants                                                6
Attorneys                                                  6
Auditors                                                   4
Chemists                                                   8
Chief accountants                                          5
Engineers                                                  8
Job analysts                                               4
Registered nurses                                          4
Administrative
Buyers                                                     4
Computer programmers                                       5
Computer systems analysts                                  5
Computer systems analyst
supervisors/managers                                       4
Directors of personnel                                     5
Technical
Civil engineering technicians                              5
Computer operators                                         6
Drafters                                                   5
Engineering technicians                                    5
Licensed practical nurses                                  3
Nursing assistants                                         4
Photographers                                              5
Clerical
Accounting clerks                                          4
File clerks                                                3
General clerks                                             4
Key entry operators                                        2
Messengers                                                 1
Personnel clerks/assistants                                5
Purchasing clerks/assistants                               4
Secretaries                                                5
Stenographers                                              2
Typists                                                    2
------------------------------------------------------------
Source:  Bureau of Labor Statistics. 


      GS-EQUIVALENT LEVELS
------------------------------------------------------ Appendix II:1.2

PATC was designed to provide salary data for the occupational work
levels defined jointly by BLS and OPM.  OPM provided the translation
into GS-equivalent grades.  These definitions were designed to
reflect duties and responsibilities of employees in private
enterprise that were translatable into the specific GS grades.  Table
II.2 shows examples of occupational work levels translated into
GS-equivalent grades. 



                          Table II.2
           
            Selected GS-Equivalent Grades of PATC
                     Work Levels in 1988

GS-equivalent
grades              PATC work levels
------------------  ----------------------------------------
GS-7                Accountants II

                    Auditors II

                    Buyers II

                    Chemists II

                    Civil engineering technicians IV

                    Computer programmers II

                    Drafters V

                    Engineers II

                    Engineering technicians IV

                    Medical machine operating technicians IV

                    Public accountants

                    Personnel clerks/assistants V

                    Personnel specialists II

                    Photographers III

                    Registered nurses I

                    Secretaries IV

GS-12               Accountants V

                    Attorneys III

                    Chemists V

                    Chief accountants II

                    Computer programmers V

                    Computer systems analysts III

                    Computer systems analysts supervisors/
                    managers I

                    Directors of personnel II

                    Engineers V

                    Personnel specialists V

                    Personnel supervisors/managers II

                    Public accountants IV

                    Registered nurses IV

GS-15               Attorneys VI

                    Chemists VIII

                    Chief accountants V

                    Computer systems analysts
                    supervisors/managers IV

                    Directors of personnel V

                    Engineers VIII

                    Personnel supervisors/managers V
------------------------------------------------------------
Source:  Bureau of Labor Statistics. 


      DATA COLLECTION
------------------------------------------------------ Appendix II:1.3

Each year, field economists from BLS who were specially trained in
job matching either personally visited or interviewed by telephone
approximately 3,500 to 4,000 sample establishments.  To match actual
jobs in the sample establishments to the survey's occupational work
level definitions, the BLS field economists used the employers'
organization charts, position descriptions, and other personnel
records.  For each job match, pay rates were collected for each
individual in that position.  The collected pay rates were those that
were paid to full-time employees for a standard work schedule. 


      PAY COMPARABILITY
      RECOMMENDATIONS
------------------------------------------------------ Appendix II:1.4

After the fieldwork was completed, the Pay Agent took several steps
to develop a pay comparability recommendation for the president. 


         GS-EQUIVALENT AVERAGES
---------------------------------------------------- Appendix II:1.4.1

The Pay Agent used a set of statistical techniques to arrive at the
pay comparability recommendation.  The average pay for each
GS-equivalent grade was calculated using the median private pay rate
for each surveyed work level.  There are 14 GS-equivalent grades. 
These grades range from GS-1 through GS-15, omitting GS-10.  To make
the calculation, weighting procedures were used to ensure that jobs
that are more common within the federal government were given greater
weight in the pay comparability process. 


         FEDERAL COMPARABILITY
         PAYLINE
---------------------------------------------------- Appendix II:1.4.2

A curve, called a payline, was then fitted to the 14 data points that
resulted from the calculation described above to produce a smooth
pattern of pay rates across GS-equivalent grades in the private
sector.  A payline for the federal sector was similarly fitted to
federal median salaries at each grade.  Each median GS salary in the
federal sector was determined using the actual federal salary
distribution.  The Pay Agent then calculated the percentage
difference between the two paylines at each grade.  These percentages
reflected the amounts that federal salaries for each grade needed to
be adjusted to be comparable with the private sector.  In 1989, these
calculations resulted in pay increase recommendations that ranged
from 20.04 percent at GS-1 to 36.69 at GS-15. 


         THE PRESIDENT'S OPTIONS
---------------------------------------------------- Appendix II:1.4.3

The Pay Agent annually sent a report summarizing the federal private
comparability findings to the President.  The President had the
following two options:  proposing a pay adjustment that agreed with
the Pay Agent's recommendations or proposing an alternative plan. 
The President could propose an alternative plan to the Congress if he
considered a full comparability pay adjustment inappropriate because
of "national emergency or economic conditions affecting the general
welfare." The President's alternative plan would become effective
unless a majority of either house of the Congress adopted a
disapproving resolution within 30 days of the submission of the
President's plan.  Each year from 1978 until FEPCA went into effect,
the President proposed and the Congress agreed on an alternate pay
adjustment that granted increases that were less than those that
would have been required for full comparability, as determined by the
Pay Agent. 


   RECENT CHANGES IN THE PROCESS
-------------------------------------------------------- Appendix II:2

The Federal Employees Pay Comparability Act of 1990 changed the
paysetting process.  The annual governmentwide comparability
adjustment is now broken into two parts:  national and local
comparability.  All federal general schedule employees are to receive
an annual comparability increase based on the percentage increase in
the Employment Cost Index rather than on the presidential
recommendation.  While the PATC survey no longer has the central pay
comparability role, the paysetting process still relies on position
comparison information to measure locality pay gaps. 

The local portion of the annual pay adjustment varies by geographic
area.  Eligible federal employees receive a locality pay adjustment. 

Under this paysetting process, governmentwide pay increases are now
based on the ECI, an index of nonfederal sector labor costs.  This
ensures that governmentwide pay increases closely follow increases in
nonfederal payrolls.  Such a process seeks to maintain current gaps
rather than to redress past comparability differences. 

The locality component of the new paysetting process is designed to
address federal nonfederal pay discrepancies.  Eligible federal
employees receive an additional increase in pay designed to reduce
the local pay gap.  Locality wage gaps are measured by a position
comparison method to determine the amount of any locality adjustment. 
Partial adjustments (based on a formula specified in the legislation)
are accorded eligible employees until the pay gap for their locality
becomes sufficiently small. 


CURRENT POPULATION SURVEY
========================================================= Appendix III

The Bureau of the Census' Current Population Survey is the principal
source of official government statistics on employment and
unemployment.  In addition to monthly labor force data, CPS provides
a large amount of detailed and supplementary data.  For the monthly
survey, households are scientifically selected on the basis of area
of residence to represent the nation as a whole as well as individual
states and other specified areas.  The monthly CPS sample consists of
approximately 58,000 households that together contain about 122,000
individuals age 14 and older.  The universe is the civilian
noninstitutional population of the United States.  A probability
sample is used in selecting housing units.  Each household is
interviewed once a month for 4 consecutive months and again for the
corresponding period 1 year later.  In March of each year,
supplemental data are collected for men in the Armed Forces who
reside with their families in civilian housing units or on a military
base.  The March CPS, which is known as the Annual Demographic File
(ADF), is also supplemented with a sample of Spanish-speaking
households that were identified the previous November.  These
additions result in the addition of about 2,500 households in the
March CPS. 

Although the main purpose of CPS is to provide information on
employment, an important secondary purpose is to collect demographic
information, such as age, race, gender, and level of educational
attainment.  In addition, questions on income, employer size, and
other subjects are included from time to time.  ADF contains the
basic monthly demographic and labor force data as well as
supplemental data on work experience, income, noncash benefits, and
migration. 

The Survey of Employee Benefits is a May supplement to CPS.  At the
time we performed our analysis, it had been conducted most recently
in May 1988.\1 That supplement provided information on pension and
retirement plan coverage, employer size, and other questions asked of
all persons employed for pay who had participated in the prior ADF. 
The supplemental information was matched to ADF to pick up detailed
income and demographic information. 


--------------------
\1 The May 1979 supplement was referred to as Pension Plan Data.  The
May 1983 supplement was named Pension and Retirement Plan Coverage. 
Although different names have been used for these May supplements,
the information collected is similar enough for the purposes of this
report. 


GAO'S ECONOMETRIC ANALYSIS: 
DETAILED DESCRIPTION AND
METHODOLOGICAL CONSIDERATIONS
========================================================== Appendix IV

In chapter 3, we presented estimates of the pay gap based on an
econometric analysis of CPS data.  We analyzed CPS data to determine
the potential effects of employer size and employee race and gender
on the differences between the federal pay gap estimates that have
been reported by the Pay Agent and those derived from a human capital
earnings model.  Econometric analyses necessarily involve elements of
professional judgment.  To do our analysis, we had to make a number
of methodological decisions concerning such issues as model
specification. 

In this appendix we review these issues, explain our decisions, and
discuss the extent to which our findings are sensitive to the
specifications that we adopted.  First, we present a detailed
description of the human capital earnings function used by labor
economists to measure pay gaps.  Then, we show how this model is used
to calculate the pay gap.  Finally, we discuss some of the
statistical and methodological problems we encountered. 


      HUMAN CAPITAL EARNINGS MODEL
------------------------------------------------------ Appendix IV:0.1

The human capital approach to earnings implies that annual earnings
are mathematically related to an employee's years of formal education
and work experience.  Stated mathematically, this relationship takes
the form of

(1) ln Y = ln Y0 + b1 S + b2 E + b3 E\2 ,

where Y is annual earnings, Y0 is the initial earning power of an
individual without any work experience or any formal education, S is
years of education, E is years of work experience, and the bis are
coefficients reflecting the returns to acquire additional education
or work experience. 

Equation (1) is generally assumed to hold true for a relatively
homogeneous group of individuals.  In other cases, certain factors
may raise or lower the level of annual earnings.  These factors can
be allowed for by inserting a dummy variable and a coefficient into
the earnings equation as in

(2) ln Y = ln Y0 + b1 S + b2 E + b3 E\2 + b4 D,

where D is a dummy variable that equals 1 to indicate the presence of
some individual characteristic and equals 0 otherwise, and where b4
is the approximate percentage difference of annual earnings between
otherwise identical individuals with the characteristic as opposed to
those without the characteristic. 

Often, more than one dummy variable is included in equation (2) to
account for the many factors other than education and work experience
that are associated with differences in earnings.  We examined
specific characteristics in this report, such as employer size,
employee gender and race, and federal employment. 

An alternate approach to measuring differences in group earnings
using the human capital model is to allow the coefficients associated
with work experience and education to differ between groups and to
include the dummy variable.  A specification such as

(3) ln Y = ln Y0 + a1 D + b1 S + b2 E + b3 E\2 + b1d S*D + b2d E*D

   + b3d E\2 *D

is equivalent to calculating equation (1) separately for the two
demographic groups.  This equation can be rewritten as

(3a) ln Y = ln Y0 + b1 S + b2 E + b3 E\2

for the group without the characteristic represented by the dummy
variable and

(3b) ln Y = ln Y0 + a1 +(b1+b1d)S +(b2+b2d)E + (b3+b3d)E\2

for the group with the characteristic.  Equations (3a) and (3b) could
be used to calculate the estimated mean earnings of the groups.  For
example, one could calculate (3a) for private sector employees.  Then
one could use the results to estimate the average earnings for
federal employees if they were employed in the private sector.  By
comparing this calculation to the actual average federal earnings,
one can obtain an estimate of the pay gap that is attributable to
federal employment. 

In labor economics research, both methods are frequently used and
generally result in similar estimates of any pay gap.\1


--------------------
\1 For a further discussion, see Robert Willis, "Wage Determinants: 
A Survey and Reinterpretation of Human Capital Earnings Functions,"
Handbook of Labor Economics, Volume I, eds.  Orley Ashenfelter and
Richard Layard (Amsterdam:  North-Holland, 1986). 


      ESTIMATING THE FEDERAL/
      PRIVATE PAY GAP
------------------------------------------------------ Appendix IV:0.2

We decided to use a simple specification of the human capital
earnings equation to focus attention on the investigation of the
possible explanations:  employer size and employee race and gender. 
Specifically, we elected to use the dummy variable method associated
with equation (2) above as the method of estimating the pay gap.  Our
basic specification of the earnings function was

(4) ln Y = ln Y0 + b1 S + b2 E + b3 E\2 + b4 Db + b5 Dw + b6 Df +

   b7 Dw Db

where Y is annual earnings, S is years of formal education, E is
years of potential work experience, and the Dis are dummy variables
that take a value of 1 for black employees, female employees, and
federal employees, respectively.\2


--------------------
\2 In exploratory regressions, we also included dummy variables for
geographic region, urban residence, and broad occupational groups. 
The addition of these variables had a small and inconsistent effect
on the federal coefficient.  We decided to drop these variables from
the analysis to focus attention on the factors of interest, employer
size, and employee race and gender. 


      PAY GAP ESTIMATES IMPLIED BY
      THE DUMMY VARIABLE METHODS
------------------------------------------------------ Appendix IV:0.3

In equation (4), the regression coefficient for federal employment is
an estimate of the pay gap after making standard adjustments for
education, work experience, race, and gender.  The pay gap, which we
express as a percentage, is assumed to be the same for white males,
women, and minorities.  The pay gap calculated from the standard
version of equation (4) provides a comparison of the average federal
employee to the average private sector employee, adjusting for other
characteristics. 

Because the dependent variable is the natural log of earnings, the
maximum likelihood estimate of the proportional pay gap for otherwise
identical individuals with any one characteristic in common equals
the antilog of the corresponding regression coefficient minus 1.  For
example, the coefficient on the federal dummy variable in the basic
earnings regression for May 1983 is 0.07016124.\3 This implies a
federal earnings advantage of exp{0.07016124} - 1 = 0.07268, or a
7.3-percent federal earnings advantage.\4 This procedure was used to
generate the estimates of the pay gap in figures 3.1 to 3.3.  A
complete summary of the pay gaps is provided in table IV.1. 



                          Table IV.1
           
           Human Capital Estimates of the Federal/
              Private Pay Gap As Measured by the
                    Federal Dummy Variable

                                                 Figure 3.3:
                     Figure 3.1:                     Pay gap
                    Pay gap as a                adjusted for
                      percentage   Figure 3.2:     potential
                      of private   May CPS pay      employer
Year                         pay  gap estimate   size effect
------------------  ------------  ------------  ------------
1978                    15.34%\a      14.30%\a        -3.12%
1979                     10.95\a
1980                     12.82\a
1981                     13.17\a
1982                     12.11\a        7.27\a       -4.72\a
1983                      9.66\a
1984                     12.55\a
1985                     11.49\a
1986                      7.73\a
1987                      9.26\a          1.98       -9.04\a
------------------------------------------------------------
\a The underlying regression coefficient is significant at the
5-percent level. 

Source:  GAO analysis of CPS data. 

In order to document the persistent discrepancy between traditional
human capital measures of the federal private pay gap and the annual
Pay Agent pay comparability measure, we estimated standard human
capital earnings functions using CPS cross-sectional data from March
1979 to March 1988.  Our primary sample included all full-time
employees between the ages of 18 and 65. 

The resulting regression estimates were consistent with published
academic estimates.  We found an earnings premium associated with
federal employment that was statistically significantly greater than
zero (at the 5-percent significance level) for every year.  The size
of this premium declined during this time period. 

We modified equation (4) to provide the basis for pay gap estimates
that are adjusted for employer size and the federal pay of women and
minorities.  The modifications included adding dummy variables for
employer size and sector-specific race and gender dummy variables. 
We report the exact specifications that we used in appendix V. 

For those regressions that adjust for the effect of employer size,
the federal dummy variable provides a comparison of the average
federal employee with the average employee in a private sector
establishment with over 1,000 employees, after adjusting for other
characteristics, such as education and work experience.\5

For those regressions with sector-specific race- or gender-specific
dummy variables, the calculation of the percentage pay gap estimate
is more involved.  The federal dummy variable in these regressions
provides a comparison of the average white male federal employee to
the average white male private sector employee, after adjusting for
other characteristics such as education and work experience.  For
other race gender groups, the pay gap relative to private sector
white males must be calculated by combining dummy variables, as we
describe in the next section.  The pay gap is allowed to differ by
race and gender and the overall pay gap is a weighted average of the
individual gaps.\6


--------------------
\3 See table V.6. 

\4 This is the maximum likelihood estimate of the federal/private pay
gap.  Under the usual statistical assumptions that underlie multiple
regression analysis, estimates of the regression coefficients have a
normal distribution.  Taking the antilog of a normal random variable
results in a lognormal random variable.  Because of this
transformation, the expected value of this estimate of the
federal-private pay gap is biased upward by a small amount.  To
correct this bias, one would need to divide this estimate by the
antilog of one-half the variance of the regression coefficient.  In
practice, estimates of this variance are usually small. 

\5 Not all private establishments that were surveyed in PATC have
over 1,000 employees.  Given the data that are available in CPS and
the lack of information about the exact distribution of employer
sizes in PATC, we chose to represent the effect of employer size by
comparing federal employees to private sector employees in
establishments with over 1,000 employees. 

\6 See chapter 2, pp.  24-26, for a discussion of the implications of
the choice of private comparison group. 


      RELATIVE FEDERAL EARNINGS
      USING PRIVATE SECTOR WHITE
      MALES AS THE BENCHMARK FOR
      COMPARISON
------------------------------------------------------ Appendix IV:0.4

The pay gap for federal sector white males is calculated using the
regression coefficient for federal employees.  For the other specific
race gender groups, the pay gap is calculated by adding the
coefficient for federal employees to the coefficient for the specific
race gender group of federal employees.  To determine the pay gap for
federal sector black males, one would add the coefficient of the
dummy variable for federal employment to the coefficient of the dummy
variable for black federal sector males.  This calculation would give
the logarithm of the estimated earnings difference for federal sector
black males relative to otherwise identical private sector white
males.  The overall federal pay gap is then calculated as a weighted
average of the federal sector pay gaps where the weights are the
percentages of the federal sample made up by each specific race
gender group.  As example, table IV.2 shows how by using May 1988
data, we calculate a 14.4 percent federal earnings disadvantage.\7



                          Table IV.2
           
           An Example of the Calculation of the Pay
                  Gap as a Weighted Average

                    Federal   Black     Black     White
          Co-       white     federal   federal   federal
          efficien  male -    male -    female -  female -
Step 1:   ts        0.064     0.046     0.119     0.272
--------  --------  --------  --------  --------  ----------
Step 2:   Relative  White     Black     Black     White
          log       male -    male      female    female
          earnings  0.064     -0.046    -0.119    -0.272
                              -0.064    -0.064    -0.064
                    -0.064    -0.110    -0.183    -0.336

Step 3:   Percenta  White     Black     Black     White
          ge        male -    male -    female -  female -
          earnings  0.062     0.104     0.167     0.285
          gap

x         Share in
          federal   x 0.515   x 0.083   x 0.092   x 0.309
          workforc  -0.032    -0.009    -0.015    -0.088
          e

Step 4:   Add up    (-        (-        (-        (-0.088)
          weighted  0.032)+   0.009)+   0.015)+
          gaps
                    =-0.144
------------------------------------------------------------
This procedure was used to generate figures 3.4 and 3.5.  The
underlying race gender-specific pay gaps and sample proportions for
federal employees are listed in table IV.3. 



                          Table IV.3
           
              Data For Federal Private Pay Gaps
               Calculated as a Weighted Average

------------------------  ----------  ----------  ----------
Distribution of federal employees by race and gender
------------------------------------------------------------
Group                       May 1979    May 1983    May 1988
White men                     60.67%      53.08%      51.52%
White women                    22.73       31.92       30.89
Black men                       9.20        8.16        8.35
Black women                     7.40        6.84        9.24

Pay gap for specific race gender groups without adjusting
for employer size (data used
for Fig. 3.4
------------------------------------------------------------
Group                       May 1979    May 1983    May 1988
White men                   11.33%\b       0.29%    -6.21%\b
White women                 -22.53\b    -25.61\b    -28.53\b
Black men                    -0.32\b    -23.05\b      -10.44
Black women                 -28.04\b    -25.80\b      -16.71
Overall                        -0.35      -11.66      -14.43

Pay gap for specific race gender groups after adjusting for
employer size (data used for
Fig. 3.5)\
------------------------------------------------------------
Group                       May 1979    May 1983    May 1988
White men                    -5.51\b    -10.81\b   -16.49%\b
White women                 -34.65\b    -34.18\b    -36.71\b
Black men                   -15.77\b    -31.75\b      -20.93
Black women                 -39.26\b    -34.38\b      -26.54
Overall                       -15.57      -21.59      -24.04
------------------------------------------------------------
\a Differences between groups or over time for one group may not be
statistically significant. 

\b The underlying race gender-specific regression coefficient is
significantly different from zero at the 5-percent level. 

Source:  GAO analysis of CPS data. 


--------------------
\7 See tables V.12 and V.14. 


      DATA LIMITATIONS AND
      ADJUSTMENTS MADE
------------------------------------------------------ Appendix IV:0.5

Analysts must make many decisions when they conduct statistical
analyses of survey data.  In this section we discuss a number of
decisions that we implemented in carrying out our analysis. 


         COMPARABILITY OF TIME
         PERIODS FOR ANNUAL
         EARNINGS
---------------------------------------------------- Appendix IV:0.5.1

The annual Pay Agent's reports were issued late in the calendar year. 
For example, the September 1979 Pay Agent's report, which was used
for the fiscal year 1980 pay adjustment, was based on private and
federal salaries from 1978 and 1979.  The earnings data from the
March and May supplements to CPS correspond to earnings in the
previous calendar year.  Continuing our example, we decided to
compare the human capital pay gap estimates using 1979 March or May
CPS data with Pay Agent estimates reported later in 1979.  Since CPS
earnings information reflects annual salaries from 1978, we have
labeled this information as 1978 data in our figures. 


         WORK EXPERIENCE
---------------------------------------------------- Appendix IV:0.5.2

Because CPS does not directly measure years of work experience, we
used a proxy for years of work experience to estimate the human
capital earnings functions.  We chose a frequently used proxy: 
substituting potential years of work experience.  Potential years of
work experience is defined as years of age minus years of schooling
minus the 6 years before grade school.  While this procedure is
widely used\8 it is thought to be a better indicator of actual
experience for white males than for women and blacks. 


--------------------
\8 See Willis, op.  cit. 


         ANNUAL WAGE AND SALARY
         INFORMATION
---------------------------------------------------- Appendix IV:0.5.3

PATC measures pay as the annual salary for a position, including
vacation, holidays, and some overtime but excluding some bonuses and
other pay premiums. 

The greatest problem that we faced with annual earnings data was that
the CPS censored the reported income beyond certain values.  While
few federal employees would have salary income beyond the cutoff, a
consequential fraction of private sector employees did have salaries
beyond this cutoff ($100,000 in 1988 for example).  Rather than
statistically imputing a value to these censored salaries, we chose
to understate them by considering their value equal to the cutoff
point.  In this regard, we may be understating any estimated federal
earnings disadvantage.\9

On the other end of the salary spectrum, the reported salary
information for some of the CPS respondents was substantially below
that to be expected of someone working 40 hours per week, 50 weeks
per year at the minimum wage.  This seemed unreasonable to us, and we
chose to omit these respondents from the sample rather than to impute
an income for them. 


--------------------
\9 We experimented with other methods of adjusting for the censoring
of annual earnings data in the CPS.  We used a tobit estimation
technique to predict the value of earnings for those whose earnings
were censored.  Because there was not enough variation in individual
characteristics for individuals censored on income as opposed to
those not censored, this technique did not materially affect the
regression results.

Additionally, we used the Pareto distribution to impute a mean value
for the censored earnings amounts.  This resulted in greater
estimates of the federal earnings disadvantage. 


         WEIGHTED LEAST SQUARES
---------------------------------------------------- Appendix IV:0.5.4

We estimated the earnings regressions using weighted least squares
for two reasons.  First, the CPS is a stratified random sample of the
United States, and the sampling weights differ across geographic
regions.  In cases like this, weighted least squares will lead to
consistent estimates.  Second, the parameter of interest is the gap
in earnings between the two sectors.  In calculating the gap, sample
proportions for black and white men and women in the federal sector
were used to form a weighted average of the race gender-specific pay
gaps.  Since we chose to use these sampling weights to arrive at the
group proportions, we also used these sampling weights to calculate
the regression estimates. 


MAY CPS SAMPLE STATISTICS AND
REGRESSION RESULTS
=========================================================== Appendix V

In this appendix we provide additional documentation for the
econometric analysis that we present in chapter 3 and appendix IV. 
First, we define the variables that we used to estimate the earnings
functions.  Second, we present the results of several regressions
that we estimated using CPS data collected in 1979, 1983, and 1988. 
Finally, we present sample statistics for the variables that are used
in the several regression equations. 



                          Table V.1
           
               Variable Names and Descriptions

Variable name       Description
------------------  ----------------------------------------
Log of earnings     The natural logarithm of the previous
                    calendar year earnings.

Education           The number of years of formal education
                    completed.

Experience          The number of years of potential work
                    experience.

Experience\b        The square of years of potential work
                    experience.

Black               A dummy variable equal to one if the
                    respondent is black and zero otherwise.

Black female        A dummy variable equal to one if the
                    respondent is a black woman and zero
                    otherwise.

Female              A dummy variable equal to one if the
                    respondent is a woman and zero
                    otherwise.

Federal             A dummy variable equal to one if the
                    respondent is a federal employee and
                    zero otherwise.

FS dummy            A dummy variable equal to one if the
                    respondent did not respond to the
                    establishment size question and zero
                    otherwise.

FG dummy            A dummy variable equal to one if the
                    respondent did not respond to both the
                    establishment size question and the firm
                    size question and zero otherwise.

GS dummy            A dummy variable equal to one if the
                    respondent did not respond to the firm
                    size question and zero otherwise.

Firm size 1         A dummy variable equal to one if the
                    respondent works in a private
                    establishment of 24 employees or fewer
                    and zero otherwise.

Firm size 2         A dummy variable equal to one if the
                    respondent works in a private
                    establishment of between 25 and 99
                    employees and zero otherwise.

Firm size 3         A dummy variable equal to one if the
                    respondent works in a private
                    establishment of between 100 and 499
                    employees and zero otherwise.

Firm size 4         A dummy variable equal to one if the
                    respondent works in a private
                    establishment of between 500 and 999
                    employees and zero otherwise.

Company size 1      A dummy variable equal to one if the
                    respondent works for a private multi-
                    establishment employer with fewer than
                    25 employees at all locations and zero
                    otherwise.

Company size 2      A dummy variable equal to one if the
                    respondent works for a private multi-
                    establishment employer with a total of
                    between 25 and 99 employees at all
                    locations and zero otherwise.\a

Company size 3      A dummy variable equal to one if the
                    respondent works for a private multi-
                    establishment employer with a total of
                    between 100 and 499 employees at all
                    locations and zero otherwise.\b

Company size 4      A dummy variable equal to one if the
                    respondent works for a private multi-
                    establishment employer with a total of
                    between 500 and 999 employees at all
                    locations and zero otherwise.\c

Black federal male  A dummy variable equal to one if the
                    respondent is a black male federal
                    employee and zero otherwise.

Black federal       A dummy variable equal to one if the
female              respondent is a black female federal
                    employee and zero otherwise.

White federal       A dummy variable equal to one if the
female              respondent is a white female federal
                    employee and zero otherwise.

Black private male  A dummy variable equal to one if the
                    respondent is a black male private
                    employee and zero otherwise.

Black private       A dummy variable equal to one if the
female              respondent is a black female private
                    employee and zero otherwise.

White private       A dummy variable equal to one if the
female              respondent is a white female private
                    employee and zero otherwise.

Intercept           The intercept for the regression.
------------------------------------------------------------
\a For regressions using the 1988 May CPS, this dummy variable equals
one if the establishment size was between 25 and 49 employees. 

\b For regressions using the 1988 May CPS, this dummy variable equals
one if the establishment size was between 50 and 99 employees. 

\c For regressions using the 1988 May CPS, this dummy variable equals
one if the establishment size was between 100 and 249 employees. 



                          Table V.2
           
            Basic Earnings Regression for May 1979

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.419      386.34
Education                                  0.066       45.70
Experience                                 0.035       33.44
Experience\2                              -0.001      -23.82
Black                                     -0.149       -9.46
Black female                               0.110        4.28
Female                                    -0.441      -54.61
Federal                                    0.134        8.96
Sample size                                           11,611
Adjusted R-squared                                    0.3893
------------------------------------------------------------
Source:  GAO analysis of May 1979 CPS data. 



                          Table V.3
           
            May 1979 Earnings Regression Adjusted
                      for Employer Size

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.670      364.42
Education                                  0.061       43.20
Experience                                 0.033       32.48
Experience\2                              -0.001      -23.05
Black                                     -0.159      -10.35
Black female                               0.104        4.17
Female                                    -0.437      -55.60
Federal                                   -0.032       -1.90
FS dummy                                   0.065        0.65
FG dummy                                  -0.189      -12.24
GS dummy                                  -0.129      -10.01
Firm size 1                               -0.139       -9.21
Firm size 2                               -0.105       -7.33
Firm size 3                               -0.126       -9.42
Firm size 4                               -0.085       -4.98
Company size 1                            -0.166      -11.08
Company size 2                            -0.129       -8.87
Company size 3                            -0.057       -4.19
Company size 4                            -0.029       -1.48
Sample size                                           11,161
Adjusted R-squared                                    0.4249
------------------------------------------------------------
Source:  GAO analysis of May 1979 CPS data. 



                          Table V.4
           
              May 1979 Earnings Regression With
           Sector-Specific Race and Gender Dummies

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.419      386.24
Education                                  0.066       45.76
Experience                                 0.035       33.48
Experience\2                              -0.001      -23.84
Black federal male                        -0.111       -2.19
Black federal female                      -0.436       -7.87
White federal female                      -0.363      -10.34
Black private male                        -0.153       -9.17
Black private female                      -0.485      -22.98
White private female                      -0.445      -53.70
Federal                                    0.107        5.64
Sample size                                           11,611
Adjusted R-squared                                    0.3894
------------------------------------------------------------
Source:  GAO analysis of May 1979 CPS data. 



                          Table V.5
           
            May 1979 Earnings Regression Adjusted
             for Race, Gender, and Employer Size
                           Effects

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.670      364.29
Education                                  0.061       43.24
Experience                                 0.033       32.51
Experience\2                              -0.001      -23.06
Black federal male                        -0.115       -2.35
Black federal female                      -0.442       -8.21
White federal female                      -0.369      -10.84
Black private male                        -0.163      -10.06
Black private female                      -0.498      -24.25
White private female                      -0.441      -54.60
Federal                                   -0.057       -2.80
FS dummy                                   0.066        0.66
FG dummy                                  -0.189      -12.21
GS dummy                                  -0.129       -9.97
Firm size 1                               -0.139       -9.23
Firm size 2                               -0.105       -7.34
Firm size 3                               -0.126       -9.41
Firm size 4                               -0.085       -4.97
Company size 1                            -0.165      -11.05
Company size 2                            -0.129       -8.85
Company size 3                            -0.056       -4.17
Company size 4                            -0.028       -1.46
Sample size                                           11,611
Adjusted R-squared                                    0.4250
------------------------------------------------------------
Source:  GAO analysis of May 1979 CPS data. 



                          Table V.6
           
            Basic Earnings Regression for May 1983

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.499      219.50
Education                                  0.087       37.51
Experience                                 0.038       23.46
Experience\2                              -0.001      -16.36
Black                                     -0.261       -8.60
Black female                               0.235        5.74
Female                                    -0.455      -41.90
Federal                                    0.070        3.73
Sample size                                            7,066
Adjusted R-squared                                    0.4498
------------------------------------------------------------
Source:  GAO analysis of May 1983 CPS data. 



                          Table V.7
           
            May 1983 Earnings Regression Adjusted
                      for Employer Size

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.684      207.64
Education                                  0.084       36.08
Experience                                 0.037       23.13
Experience\2                              -0.001      -16.11
Black                                     -0.278       -9.28
Black female                               0.235        5.84
Female                                    -0.453      -42.23
Federal                                   -0.048       -2.17
FS dummy                                   0.243        1.22
FG dummy                                  -0.039       -1.56
GS dummy                                  -0.093       -4.23
Firm size 1                               -0.154       -7.29
Firm size 2                               -0.109       -5.33
Firm size 3                               -0.082       -4.28
Firm size 4                               -0.039       -1.64
Company size 1                            -0.086       -4.27
Company size 2                            -0.042       -2.06
Company size 3                            -0.016       -0.91
Company size 4                            -0.009       -0.40
Sample size                                            7,066
Adjusted R-squared                                    0.4679
------------------------------------------------------------
Source:  GAO analysis of May 1983 CPS data. 



                          Table V.8
           
              May 1983 Earnings Regression With
           Sector-Specific Race and Gender Dummies

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.506      219.75
Education                                  0.087       37.49
Experience                                 0.038       23.51
Experience\2                              -0.001      -16.40
Black federal male                        -0.265       -3.94
Black federal female                      -0.301       -4.15
White federal female                      -0.299       -7.45
Black private male                        -0.248       -7.28
Black private female                      -0.507      -17.02
White private female                      -0.466      -41.65
Federal                                    0.003        0.12
Sample size                                            7,066
Adjusted R-squared                                    0.4512
------------------------------------------------------------
Source:  GAO analysis of May 1983 CPS data. 



                          Table V.9
           
            May 1983 Earnings Regression Adjusted
             for Race, Gender, and Employer Size
                           Effects

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.690      207.89
Education                                  0.083       36.06
Experience                                 0.037       23.19
Experience\2                              -0.001      -16.14
Black federal male                        -0.268       -4.05
Black federal female                      -0.307       -4.30
White federal female                      -0.304       -7.70
Black private male                        -0.269       -8.00
Black private female                      -0.524      -17.81
White private female                      -0.464      -41.93
Federal                                   -0.114       -4.08
FS dummy                                   0.247        1.25
FG dummy                                  -0.039       -1.58
GS dummy                                  -0.092       -4.17
Firm size 1                               -0.154       -7.28
Firm size 2                               -0.109       -5.31
Firm size 3                               -0.080       -4.23
Firm size 4                               -0.038       -1.58
Company size 1                            -0.086       -4.26
Company size 2                            -0.042       -2.04
Company size 3                            -0.016       -0.90
Company size 4                            -0.009       -0.38
Sample size                                            7,066
Adjusted R-squared                                    0.4693
------------------------------------------------------------
Source:  GAO analysis of May 1983 CPS data. 



                          Table V.10
           
            Basic Earnings Regression for May 1988

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.532      212.19
Education                                  0.096       39.40
Experience                                 0.041       24.17
Experience\2                              -0.001      -17.58
Black                                     -0.190       -5.93
Black female                               0.206        5.04
Female                                    -0.404      -36.71
Federal                                    0.020        0.95
Sample size                                            7,013
Adjusted R-squared                                    0.4158
------------------------------------------------------------
Source:  GAO analysis of May 1988 CPS data. 



                          Table V.11
           
            May 1988 Earnings Regression Adjusted
                      for Employer Size

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.758      210.09
Education                                  0.089       37.33
Experience                                 0.039       23.43
Experience\2                              -0.001      -16.90
Black                                     -0.202       -6.46
Black female                               0.180        4.50
Female                                    -0.403      -37.44
Federal                                   -0.095       -4.38
FS dummy                                  -0.018       -0.29
FG dummy                                  -0.193       -6.11
GS dummy                                  -0.135       -5.17
Firm size 1                               -0.169       -9.41
Firm size 2                               -0.066       -3.27
Firm size 3                               -0.084       -4.11
Firm size 4                               -0.103       -5.59
Company size 1                            -0.102       -5.00
Company size 2                            -0.054       -2.80
Company size 3                            -0.029       -1.60
Company size 4                            -0.035       -1.25
Sample size                                            7,013
Adjusted R-squared                                    0.4431
------------------------------------------------------------
Source:  GAO analysis of May 1988 CPS data. 



                          Table V.12
           
              May 1988 Earnings Regression With
           Sector-Specific Race and Gender Dummies

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.539      212.34
Education                                  0.095       39.43
Experience                                 0.041       24.17
Experience\2                              -0.001      -17.58
Black federal male                        -0.046       -0.63
Black federal female                      -0.119       -1.69
White federal female                      -0.272       -6.07
Black private male                        -0.213       -5.99
Black private female                      -0.426      -15.28
White private female                      -0.413      -36.51
Federal                                   -0.064       -2.25
Sample size                                            7,013
Adjusted R-squared                                    0.4174
------------------------------------------------------------
Source:  GAO analysis of May 1988 CPS data. 



                          Table V.13
           
            May 1988 Earnings Regression Adjusted
             for Race, Gender, and Employer Size
                           Effects

                                      Coefficien           T
Variable                                       t   statistic
------------------------------------  ----------  ----------
Intercept                                  8.766      210.36
Education                                  0.089       37.37
Experience                                 0.039       23.44
Experience\2                              -0.001      -16.90
Black federal male                        -0.055       -0.76
Black federal female                      -0.128       -1.87
White federal female                      -0.277       -6.34
Black private male                        -0.227       -6.52
Black private female                      -0.469      -17.10
White private female                      -0.412      -37.22
Federal                                   -0.180       -6.23
FS dummy                                  -0.013       -0.20
FG dummy                                  -0.190       -6.04
GS dummy                                  -0.134       -5.13
Firm size 1                               -0.170       -9.48
Firm size 2                               -0.067       -3.32
Firm Size 3                               -0.084       -4.14
Firm Size 4                               -0.103       -5.61
Company size 1                            -0.103       -5.04
Company size 2                            -0.056       -2.88
Company size 3                            -0.030       -1.65
Company size 4                            -0.036       -1.28
Sample size                                            7,013
Adjusted R-squared                                    0.4449
------------------------------------------------------------
Source:  GAO analysis of May 1988 CPS data. 



                          Table V.14
           
               Sample Statistics for Regression
                           Analysis

Variable                   1979 Mean   1983 Mean   1988 Mean
------------------------  ----------  ----------  ----------
Log of earnings                9.478       9.926      10.093
Education                     12.491      14.002      14.064
Experience                    19.484      18.629      17.668
Experience\2                 552.294     490.577     435.491
Black                          0.090       0.063       0.070
Black female                   0.034       0.035       0.043
Female                         0.314       0.447       0.498
Federal                        0.062       0.077       0.067
FS dummy                       0.001       0.001       0.007
FG dummy                       0.079       0.056       0.028
GS dummy                       0.106       0.062       0.042
Firm size 1                    0.273       0.288       0.282
Firm size 2                    0.192       0.191       0.118
Firm size 3                    0.189       0.189       0.097
Firm size 4                    0.064       0.069       0.122
Company size 1                 0.163       0.169       0.147
Company size 2                 0.108       0.104       0.118
Company size 3                 0.102       0.117       0.112
Company size 4                 0.038       0.053       0.034
Black federal male             0.006       0.006       0.006
Black federal female           0.050       0.005       0.006
White federal female           0.014       0.025       0.021
Black private male             0.050       0.022       0.022
Black private female           0.030       0.030       0.037
White private female           0.265       0.387       0.434
------------------------------------------------------------
Source:  GAO analysis of May 1979, May 1983, and May 1988 CPS data. 




(See figure in printed edition.)Appendix VI
COMMENTS FROM THE BUREAU OF LABOR
STATISTICS
=========================================================== Appendix V



(See figure in printed edition.)

See comment 4. 

See comment 5. 

See comment 6. 

See comment 7. 

See comment 8. 

See comment 9. 


The following is GAO's comments on the Bureau of Labor Statistics
letter dated January 31, 1994. 


   GAO COMMENTS
--------------------------------------------------------- Appendix V:1

1.  We are in general agreement with this interpretation of our
analysis.  We have expanded the text on pages 26 through 28 in
response to these comments. 

2.  We agree that we do not explicitly model differences in the
occupational composition of the federal and private sector in our
human capital earnings functions.  However, we do discuss the
importance of occupational differences in the two sectors on pages 5,
6, 26, and 27. 

3.  We agree that removing sales workers might have some effect on
our analysis.  Undoubtedly, private white-collar occupations were
included that are not represented in the federal government, just as
federal occupations were included that are not represented in the
private sector.  Although examining federal and private occupations
in the CPS for comparability may be valuable, such an exercise is
beyond the scope of this report. 

4.  We have modified the text on page 2 in response to this comment. 

5.  We have modified the text on pages 2 and 21 in response to this
comment. 

6.  We have added a footnote on page 49 in response to this comment. 

7.  We have modified the information provided on page 50 in response
to this comment. 

8.  We have modified the text on page 51 in response to this comment. 

9.  We have modified the text on page 53 in response to this comment. 




(See figure in printed edition.)Appendix VII
COMMENTS FROM THE OFFICE OF
PERSONNEL MANAGEMENT
=========================================================== Appendix V


MAJOR CONTRIBUTORS TO THIS REPORT
======================================================== Appendix VIII

Timothy J.  Carr, Project Director, (202) 512-4083
Gene G.  Kuehneman, Jr., Project Manager
Yesook S.  Merrill, Senior Economist
Paula J.  Bonin, Computer Systems Analyst

