Gross Domestic Product: No Evidence of Manipulation in First Quarter 1991
Estimates (Letter Report, 03/10/93, GAO/GGD-93-58).
Press reports alleged that the Department of Commerce inflated the first
quarter 1991 gross domestic product (GDP) to mask the true dimensions of
the economic downturn. GAO, however, found no evidence that the Bureau
of Economic Analysis manipulated first quarter 1991 personal income or
GDP for political purposes. GAO does make several suggestions for
improving the perceived integrity of the Bureau's data. For example,
GAO believes that the Bureau could more completely document and explain
its data. Additionally, the Bureau has issued no public response to the
press allegations. To determine user needs and to assure Congress and
the public of the integrity and credibility of its data, the Bureau also
needs to increase outside expert review and comment on its procedures.
Congress may want to request that the Bureau testify regularly when it
releases major national economic statistics.
--------------------------- Indexing Terms -----------------------------
REPORTNUM: GGD-93-58
TITLE: Gross Domestic Product: No Evidence of Manipulation in
First Quarter 1991 Estimates
DATE: 03/10/93
SUBJECT: Economic analysis
Reporting requirements
Statistical data
Statistical methods
Income statistics
Labor statistics
Gross National Product
Economic indicators
Ethical conduct
IDENTIFIER: California
Maryland
New Jersey
New York
**************************************************************************
* 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 the Chairman, Committee on Commerce, Science, and
Transportation, U.S. Senate
March 1993
GROSS DOMESTIC PRODUCT - NO
EVIDENCE OF MANIPULATION IN FIRST
QUARTER 1991 ESTIMATES
GAO/GGD-93-58
GDP Evaluation
Abbreviations
=============================================================== ABBREV
BEA - Bureau of Economic Analysis
BLS - Bureau of Labor Statistics
CES - Current Employment Statistics Survey
GDP - Gross Domestic Product
GNP - Gross National Product
NIPA - National Income and Product Account
UI - Unemployment Insurance Reports
Letter
=============================================================== LETTER
B-249775
DATE GOES HERE
The Honorable Ernest F. Hollings
Chairman, Committee on Commerce,
Science, and Transportation
United States Senate
Dear Mr. Chairman:
This report responds to your letter expressing concern about press
reports that alleged the Department of Commerce inflated the first
quarter 1991 gross domestic product (GDP)\1
to mask the true size of the economic downturn. The press reports
alleged that the Bureau of Economic Analysis (BEA)\2 did not
incorporate, for political purposes, a downward revision of original
employment levels into its October 1991 estimate of first quarter
1991 state personal income growth and its December 1991 estimate of
first quarter 1991 GDP growth. You asked us to determine whether
there was any political manipulation of these first quarter 1991
estimates. On November 19, 1992, we briefed the Committee on our
findings. This report documents and supplements the information we
reported at the briefing.
--------------------
\1 In December of 1991, the Department of Commerce began to use gross
domestic product (GDP) as the primary measure of economic
performance. While gross national product (GNP) measures output of
U.S. individuals and firms regardless of location, GDP measures
output of all individuals and firms located in the United States.
According to the Commerce Department, there is little difference
between the dollar levels of GDP and GNP.
\2 The Bureau of Economic Analysis is the office within the
Department of Commerce that calculates and publishes GDP and personal
income statistics.
RESULTS IN BRIEF
------------------------------------------------------------ Letter :1
We found no evidence that BEA manipulated first quarter 1991 personal
income or GDP estimates for political purposes. BEA generally
followed its standard procedures for using employment data in these
estimates and deviated from these procedures only when required by
what we believe were reasonable technical judgments.
The role employment data play in BEA's procedures varies for
conceptual and timing reasons. GDP is a measure of the total
production of the economy and is based almost exclusively on measures
of output rather than on income or employment. In contrast,
employment data are much more important in the estimation of personal
income than GDP. As a consequence, while BEA did account for the
employment revision in its December estimate of first quarter 1991
GDP growth, this did not affect the estimated growth in the December
or subsequent estimates of first quarter 1991 GDP. Additionally,
BEA's procedures call for employment and related wage data to be used
in different ways and at different times in state personal income
estimates. As a result of these procedures, the incorporation of the
revised employment data did not affect the estimated growth of BEA's
October estimate of first quarter 1991 state personal income growth
as much as was expected by those who made the allegations.
We found several ways that BEA could improve the perceived integrity
of its data. For example, we believe BEA could more completely
document and explain its data. BEA has not adequately documented or
explained to its users several key assumptions it used in its
methodology to incorporate employment data into these first quarter
1991 estimates. Additionally, BEA has issued no public response to
the allegations made in the press about the accuracy of first quarter
1991 data. To determine user needs and to assure Congress and the
general public of the integrity and credibility of its data, BEA also
needs additional mechanisms for outside expert review and comment on
its procedures. To enhance congressional and public assurance of the
integrity of BEA's data, Congress may want to request that BEA
testify regularly before Congress when BEA releases major national
economic statistics.
BACKGROUND
------------------------------------------------------------ Letter :2
In late September of 1991, the Bureau of Labor Statistics (BLS)\3
first realized that there was a difference of 650,000 positions, or
-0.6 percent, between the first quarter 1991 employment levels
measured by BLS. This difference was between BLS' monthly survey,
called the Current Employment Statistics (CES) survey, and employment
levels obtained by BLS through quarterly unemployment insurance
reports (UI) required of all nonagricultural establishments covered
under unemployment insurance laws. CES provides monthly data on
employment and earnings and UI provides monthly employment levels, as
well as quarterly wage and salary data. Preliminary UI data,
however, are not available until at least 6 months after the end of
the quarter. For that reason, BLS believes that using both measures
allows it to accurately determine the level of employment in the
country and the rate of change in employment growth. When BLS began
to receive first quarter 1991 UI employment data in September, it
compared the employment estimates provided by the CES and UI and
discovered that CES employment levels may have been overstated. For
more information on how BLS measures employment, see pp. 44 to 45.
By September, however, BEA had already issued several of its
regularly scheduled estimates of first quarter 1991 GDP and personal
income growth, using wage and salary data derived from CES employment
data.\4 This situation raised a number of questions about the
accuracy of national employment measures, as well as how the possible
overestimate affected the already published GDP and personal income
growth estimates. Since BEA was scheduled to release additional
revised estimates of first quarter 1991 GDP and personal income
growth in the next several months, there was also speculation about
how the revised employment data, once incorporated by BEA, would
affect the growth of the revised GDP and personal income estimates.
When BEA revised these data in October and December, however, neither
GDP nor personal income growth declined as was expected, causing some
press to question BEA's use of employment data in these estimates.
Several journalists alleged that BEA politically manipulated the data
by not adequately incorporating the revised employment data into
these estimates. (For a detailed description of these allegations,
see pp. 29 to 30. Additionally, for excerpts of selected articles
that raised questions about first quarter 1991 data, see pp. 32 to
40.)
State officials in California raised concerns as well. In October of
1991, California Department of Finance officials began to believe
that BEA's estimates for first quarter 1991 California personal
income growth were too optimistic, since BEA's data showed growth
when California UI data showed a decline for the first quarter of
1991. The California Department of Finance officials said they
believed BEA's procedures for calculating California personal income
growth produced misleading data because BEA's estimates were higher
than California's estimates of personal income growth.\5 (For a
detailed description of these concerns, see p. 30.)
In January 1993, the Deputy Commissioner of BLS announced that the
difference in reported employment levels was primarily due to a
one-time, noneconomic correction in how employment was reported on
the UI rather than an error in CES data. BLS has since contracted
with the American Statistical Association to do a full review of its
research findings and conclusions. The reporting change accounted
for 500,000 of the 640,000 difference and meant that the actual
difference between the two data sources was 140,000 positions, which
was well within recent historical ranges.\6
--------------------
\3 BLS is the agency within the U.S. Department of Labor responsible
for collecting and publishing national and state data on employment,
earnings, and wages.
\4 BEA does not collect its own data; instead, it receives data from
a number of sources in different formats. For that reason, BEA
regularly performs a number of revisions of estimates, with each
estimate based on more complete or accurate source data. For more
information on data used by BEA in its estimates, as well as BEA's
data revision and release schedule, see appendix IV.
\5 GAO contacted three other states that experienced large downward
revisions of employment in first quarter 1991 to determine whether
this position was shared by other state finance officials. We found
the finance officials in these other states--Maryland, New Jersey,
and New York--did not share California's belief that BEA's procedures
produced misleading data.
\6 According to BLS officials, because the UI correction was
noneconomic in nature, this announcement also meant that BLS'
original CES estimates of job loss during the recession were very
close to the final estimate of job loss for the recession.
OBJECTIVE, SCOPE, AND
METHODOLOGY
------------------------------------------------------------ Letter :3
As agreed with the Committee, our objective was to determine whether
the Department of Commerce manipulated first quarter 1991 GDP and
personal income estimates for political purposes. Since the
allegations of manipulation concerned BEA's incorporation of the
revised employment data into these statistics, to address this
objective we identified BEA's standard procedure for incorporating
employment data into GDP and personal income estimates. We then
reviewed BEA procedures used and decisions made for first quarter
1991 data and determined whether BEA deviated from this standard
procedure for first quarter 1991 data. We assessed the
reasonableness of any deviations through interviews with BEA
officials. We also reviewed the documentation and publication of
procedures BEA used for first quarter 1991 data, and we examined
existing organizational and procedural safeguards at BEA designed to
ensure the actual and perceived integrity of these data.\7 We did not
assess the adequacy of BEA's standard methodology. Appendix I
contains a detailed explanation of our objective, scope, and
methodology.
--------------------
\7 We agreed with the Committee that we would review other broader
issues relating to the accuracy and integrity of national economic
statistics after completing this examination.
NO EVIDENCE OF ALLEGED
POLITICAL MANIPULATION BY BEA
IN FIRST QUARTER 1991 DATA
------------------------------------------------------------ Letter :4
We found no evidence that BEA manipulated first quarter 1991 GDP or
personal income statistics for political purposes. BEA generally
followed its standard procedures for incorporating employment data
into these statistics and deviated from these procedures only when
required by what we believe were reasonable technical judgments.
(For a detailed description of how BEA estimates GDP, see pp. 54 to
56, and see pp. 57 and 58 for how BEA estimates personal income.) We
also found that BEA has many procedural and organizational safeguards
that help to ensure the integrity of BEA's data against political
manipulation. Finally, other knowledgeable federal officials said
they did not believe BEA's first quarter 1991 estimates were
manipulated for political reasons.
BEA FOLLOWED STANDARD
PROCEDURE FOR USING
EMPLOYMENT DATA IN STATE
PERSONAL INCOME
---------------------------------------------------------- Letter :4.1
In July of 1991, BEA issued its first scheduled estimate of first
quarter 1991 state personal income growth. For the first estimate of
state personal income, BEA uses CES employment data\8 to determine
wages and salaries for the quarter, which are the largest component
of state personal income.\9 (For more detail on how BEA estimates
state personal income, see pp. 62 to 63.) As shown in figure 1, the
estimated rate of growth for both the sum of all states and for
California\10 was shown to each be 1.2 percent from the prior
quarter.
Figure 1: Revisions in
Estimated Growth of State
Personal Income for First
Quarter 1991
(See figure in printed
edition.)
Source: BEA.
In October 1991, BEA released its first scheduled revision of
estimated growth for first quarter 1991 state personal income. This
estimate showed 1.3 percent growth for the sum of all states and 0.2
percent growth for California.\11 BEA was expected by various press
and California Department of Finance officials to use the UI data in
its October revision of first quarter 1991 state personal income
growth.\12 As a result of this expectation, both the press and
California officials were surprised when the October
revision actually showed better growth for the sum of all states than
had the previous estimate and still showed growth for California.
This unmet expectation resulted in the press allegation that BEA did
not use the UI data in state estimates and California officials'
concern that BEA's procedures produced misleading state data.
We found BEA did use the UI wage and salary data in its October
revision of state personal income in accordance with its standard
procedure. Under this procedure BEA takes the growth rate of state
wages and salaries as shown by the newly available UI data as a base
for estimating individual state personal income levels. It then
adjusts the growth of state UI wages and salaries so that the sum of
all states' growth is equal to the growth rate of wages and salaries
used in the most recently published estimate of personal income at
the national level (see description of procedure, pp. 64 and 65.
BEA follows this procedure because it allows BEA to introduce the
higher quality seasonal adjustments that are possible in the national
CES-based estimates. According to BEA, the ideal would be to use
quarterly UI wage data both nationally and for the states. The lack
of a suitable seasonal adjustment for the UI data so far has
precluded this, however. Additionally, UI data are not available
until at least 6 months after the end of a quarter.\13
When BEA began to receive preliminary UI data from BLS in September
of 1991, it observed a large discrepancy between the estimated growth
in first quarter 1991 UI wages and salaries and BEA's original
CES-based estimates of wage and salary growth for the quarter.
Despite this large discrepancy, BEA decided to adhere to its standard
procedure for several reasons. BEA was not sure at that time whether
UI data represented a real economic downturn, or the data were in
error. Also, because the data would continue to be revised by the
states and BLS for several more months, much of the states' UI data
was still subject to change. Finally, seasonal variations in the
quarterly UI data are difficult to adequately measure.\14
Following this procedure for first quarter 1991, BEA based its state
personal income growth estimates on UI wage and salary data. It then
adjusted the UI-based estimates of state wages and salaries upward to
equal the June 1991 estimate of national wage and salary growth,
which was still based on CES employment data.
While BEA has traditionally followed this procedure, in 1991 the
adjustment required was very noticeable because of the atypically
large discrepancy of -0.6 percent between CES and UI employment
levels\15 and the subsequent divergence between CES-based wage and
salary estimates and UI estimates of wages and salaries for first
quarter 1991.\16 For example, the growth in unadjusted wages and
salaries for the sum of the states for the first quarter 1991 showed
a decline of -6.4 percent. After adjustment, the rate of growth was
0.2 percent. For California, the decline in unadjusted wages and
salaries for first quarter 1991 was -7.2 percent, while it was only
-0.4 percent after the adjustment.
In December 1991, BEA accounted for the downward revision in
employment in its personal income estimates at the national level by
reducing the estimated growth rate of first quarter 1991 wages and
salaries. When BEA performed its scheduled January 1992 revision of
first quarter 1991 state personal income growth, it still adjusted UI
wage and salary growth to match the growth in national wages and
salaries, but the amount of adjustment required was considerably
lessened. As shown in figure 1, the January 1992 revision of state
personal income growth for first quarter 1991 showed only 0.4 percent
growth for the sum of all states' personal income, compared to the
October estimate of 1.3 percent growth. Additionally, for California
personal income, the January estimate showed a decline of -0.7
percent as compared to the October estimate, which showed 0.2 percent
growth for the quarter.
BEA's subsequent revisions in April and July resulted in still lower
estimates of state personal income growth, ultimately declining to
only 0.1 percent growth for the sum of all states and a -2.8 percent
decline for California. In October 1992, BEA issued its final
estimate of first quarter 1991 state personal income growth, which
incorporated four quarters of 1991 UI data as well as other data that
were available during the year. As shown in figure 1, the result of
this final revision was a �.1 percent decline for the sum of all
states, and a -4.1 percent decline for California. This final
revision showed that personal income for the sum of all states and
California did not increase by 1.2 percent over the prior quarter
(which was originally estimated in July 1991), but instead declined.
According to BEA officials, most of the difference between the
original estimate of California personal income growth and the
October 1992 estimate was due to information that became available
after the October-November 1991 period when BEA was reestimating
personal income at the state and national level.
Additionally, they said almost all of the difference between the
January 1992 estimate and the October 1992 estimate was due to
revisions California and other states made to their UI data, national
estimates from the comprehensive revision of the GDP accounts,
various annual 1991 data from the states for nonwage components of
personal income, and updated seasonal factors. BEA officials said
the larger revision in the California estimate than in that for all
states reflects large revisions in the California UI data that were
made after the October-November period.
BEA's decision to adhere to standard procedure was a reasonable one
for first quarter 1991. Nonetheless, while following standard
procedures assures users of the integrity of the data, following
these procedures may still produce results that could be misleading
to data users focusing solely on estimates of personal income growth
for individual states, especially if these users do not fully
understand BEA's procedures. This was the case for first quarter
1991 state personal income growth estimates, since both the press and
California officials did not understand how the UI data could have
been incorporated without an immediate effect on state personal
income growth. Additionally, because it took a year (from October of
1991 to October of 1992) for BEA's estimates of first quarter 1991
personal income growth to fully reflect the array of final source
data and show what some of the press and California finance officials
suspected in October 1991, it is imperative that BEA fully
communicate its procedures to users so that they understand how the
data are calculated and what the data do and do not include.
--------------------
\8 For manufacturing industries, BEA uses production hours and
earnings data as well.
\9 This is necessary because wage and salary data for states are not
available from the monthly CES survey. UI data, which provide this
information, are not available at the time of BEA's first estimate of
quarterly state personal income growth, which is released 4 months
after the end of each quarter. UI data do not begin to become
available until at least 6 months after the end of the quarter.
\10 California is highlighted here because of California Department
of Finance officials' concerns that BEA's estimates of first quarter
1991 personal income growth were too optimistic. The fact that the
growth for the sum of all states and California are both 1.2 percent
was merely a coincidence.
\11 Between the time of these two estimates, BEA had been told by BLS
that it was beginning to investigate the difference between CES and
UI employment levels for the first quarter of 1991. However, BLS did
not announce the 650,000 difference until November, after BEA had
released the state personal income revision.
\12 Because UI data begins to become available about 6 months after
the end of the quarter, BEA regularly incorporates UI data into its
first revision of state personal income growth, which it performs 7
months after the end of the quarter. BEA uses the wage and salary
data from UI.
\13 An alternative to this is the monthly collection of expanded wage
data through CES. Having expanded wage data earlier would reduce
BEA's reliance on employment change to estimate wage change in its
first estimate of state personal income. BLS plans to begin research
in this area later this year.
\14 The UI wage and salary data include bonus-type payments that vary
greatly in some industries in both magnitude and timing. In
addition, the quarterly data reflect a variety of payday patterns in
different industries. For such reasons, BEA prefers to use the UI
data on an annual basis.
\15 According to BLS officials, employment revisions have averaged +
0.2 percent and have ranged + 0.5 percent between 1980 and 1990.
\16 BLS now believes the discrepancy in employment levels between the
CES and UI was due to a one-time, noneconomic correction in how
employment was reported on the UI. For more information, see pp. 48
to 49.
BEA INCORPORATED EMPLOYMENT
REVISION INTO GDP AND
PERSONAL INCOME AS SOON AS
POSSIBLE
---------------------------------------------------------- Letter :4.2
BEA released three regularly scheduled estimates of first quarter
1991 GDP and personal income growth in the first 3 months after the
end of the quarter. The GDP estimates, released in April, May, and
June of 1991, showed a decline (from the prior quarter) in first
quarter 1991 GDP (at an annual rate), ranging from -2.8 percent to
-3.0 percent, as adjusted for inflation.
Figure 2: Revisions in
Estimated Growth of GDP for
First Quarter 1991
(See figure in printed
edition.)
Source: BEA.
In contrast, BEA's estimates of first quarter 1991 personal income
growth, released at the same time, showed growth--not adjusted for
inflation--ranging from 1.4 to 1.6 percent from the prior quarter.
Figure 3: Revisions in
Estimated Growth of Personal
Income for First Quarter 1991
(See figure in printed
edition.)
Source: BEA.
In early December 1991, BEA completed a comprehensive revision (which
it performs once every 5 years) of GDP and personal income estimates.
At this time BEA accounted for the lower estimate of employment based
on first quarter 1991 UI data (i.e., the employment revision). The
effect of the employment revision was evident in the December
estimate of first quarter 1991 personal income growth (see fig. 3).
However, the effect of the employment revision on the growth of GDP
was not evident, as the December estimate actually showed less of a
decline (-2.5 percent) for first quarter 1991 GDP growth than had the
prior estimates (see fig. 2). This unexpected result led to the
press allegations that BEA increased other components of GDP to make
up for the employment revision or did not use UI data at all in the
revised estimate. BEA did, in fact, incorporate the employment
revision into its December revision of personal income and GDP.
However, because employment data are not a major factor in
determining the growth rate or level of GDP, the incorporation of the
UI data did not affect the growth of GDP.
Two factors were at work in the fall of 1991 that allowed BEA to
incorporate the employment revision into the wages and salaries
components of GDP and personal income during the December revision.
The first was the unusually early public announcement by BLS of the
employment revision in 1991. In November 1991, BLS publicly
announced the employment revision; BLS would not have ordinarily
announced this revision until June of 1992. Second, BEA had delayed
its comprehensive revision of these estimates from November of 1990
to December of 1991 because it needed more time to complete selected
data analyses needed for the comprehensive revision. Ordinarily, BEA
would not have attempted to reflect the UI employment or wage and
salary data for the most recent quarter in a comprehensive revision;
however, because of the significant divergence between CES and UI
wages and salaries data for the first quarter 1991, BEA decided it
needed to account for this somehow in its national estimates.\17
As previously discussed, when BEA began to receive preliminary UI
data from BLS in September of 1991, it observed a large discrepancy
between the estimated growth in first quarter 1991 UI wages and
salaries and BEA's original CES-based estimates of wage and salary
growth for the quarter. BEA wanted to make a proper adjustment for
these new data, and from September through November BEA officials
assessed various ways to incorporate the UI data into the December
revision of GDP and personal income. BEA decided to account for the
decline in wages and salaries in its December revision by multiplying
the average yearly wage by the estimated size of the employment
revision (at that time estimated by BLS at 650,000). This
calculation resulted in a downward adjustment of $15 billion to first
quarter 1991 wages and salaries. According to BEA officials, the
main reason this methodology was chosen was that BEA simply ran out
of time trying to determine a better method.\18
Other components of wages and salaries experienced increases during
the December revision, however, so the actual reduction to wages and
salaries was $10 billion for first quarter 1991. This contributed to
about a $14 billion reduction in first quarter 1991 personal income.
As a result, the December estimate of first quarter 1991 personal
income growth, as shown in figure 3, was only .3 percent, which was
lower than the previous estimate of 1.5 percent in June 1991.\19
The reduction in wages and salaries and, subsequently, personal
income growth did not adversely affect the December estimate of GDP
growth, however because the growth rate and level of GDP are based on
the level of production in the country rather than the level of
income (see fig. 2). The income estimate provides a check against
the measure of production. According to BEA officials, the two
measurements should be fairly equal, although some difference in the
total exists. This difference is called the statistical discrepancy;
historically, the discrepancy has been fairly small, ranging from �.7
to .6 percent of GDP.
Wages and salaries data are used primarily in the income measurement.
The $15 billion decline contributed to a reduction in the income
measurement, which when combined with other changes in both the
income and product side resulted in a greater difference between the
product and income measurements than in the prior estimate. This
meant the statistical discrepancy increased from what it had been in
the prior estimate. Despite this increase, however, the statistical
discrepancy was still only 0.2 percent of GDP, which is well within
the historical range. The press allegation that BEA increased the
statistical discrepancy so the UI data would not have an impact
suggests a lack of understanding that CES and UI data do not directly
affect the GDP estimate. While it is true that the statistical
discrepancy increased due to the incorporation of the UI data, this
resulted from the standard procedure used for estimating GDP rather
than a purposeful attempt on the part of BEA to mask the effect of
the employment revision or UI data.
We believe BEA's decision to account for the employment revision in
its measurement of GDP and personal income reflects its policy to
incorporate all available data into its revisions. We also believe
the methodology for doing so was reasonable under the circumstances.
--------------------
\17 In July of 1992, BEA completed its regularly scheduled annual
revision of GDP, which incorporated a year's worth of 1991 UI data.
The result of this annual revision was that first quarter 1991 GDP
growth was -3.0 percent, which was only .1 percentage point lower
than BEA had originally estimated in April of 1991 (see fig. 2). We
compared this to similar changes over the past 11 years and found the
�.1 percentage point is well within the historical average and range.
Additionally, we found that over the past eleven years there has been
little upward or downward bias between the first estimate and the
first annual revision, as the first estimate of GDP growth is neither
consistently optimistic nor pessimistic when compared to the later
estimates, which are based on more complete source data. See
discussion on pp. 59 and 60.
\18 Additionally, this method is similar to how BEA estimates monthly
wages and salaries based on CES employment data. For a discussion of
this procedure, see pp. 60 to 62.
\19 In July of 1992, BEA completed its regularly scheduled annual
revision of personal income, which incorporated four quarters of 1991
UI data. The result of this annual revision for first quarter 1991
personal income growth was that personal income growth for first
quarter 1991 declined to only .1 percent, as compared to the 1.4
percent growth estimated by BEA in April 1991 (see fig. 3).
BEA HAS ORGANIZATIONAL AND
PROCEDURAL SAFEGUARDS
---------------------------------------------------------- Letter :4.3
BEA has several existing organizational and procedural safeguards
that can help protect the integrity of its data. These safeguards
are found in the composition of BEA staff and in the procedures used
for the calculation, release, and publication of its data. All of
BEA's 11 Senior Executive Service positions are reserved for career
civil servants, which helps to ensure that data preparation is not
influenced by political partisanship. Second, BEA has security
procedures that govern data preparation and release. For example,
there is no involvement by political appointees during the
preparation of GDP or personal income data. According to BEA
officials, BEA completes GDP and personal income estimates under
tight security and with only a small group of BEA officials present.
Additionally, key BEA officials' performance standards contain
schedules of dates for completion and review of specific data
elements.
Regarding BEA's release procedures, the Office of Management and
Budget announces release dates for quarterly data such as GDP and
personal income at the beginning of each calendar year. On the day
of the release of these data, BEA publicizes the estimates at 8:30
a.m. The Undersecretary of Commerce's office does not receive the
estimates until the morning of the release, in most cases, 1
half-hour before the estimates are made public. Additionally, no
Commerce or other policy official is allowed to comment publicly on
the estimates until at least 1 hour after the public release. On the
afternoon before the release, the estimates are hand-delivered to the
Chairman of the Council of Economic Advisers for transmittal to the
President of the United States. Career BEA officials told us that
there were no other contacts outside BEA in the calculation of first
quarter 1991 GDP or personal income statistics and that these
safeguards were followed.
BEA produces a monthly publication, the Survey of Current Business,
which describes standard methodology, source data, and key
assumptions used for various statistics and provides further
assurance of the integrity of its procedures. BEA publishes a
schedule for the release of statistical information, which it
followed for first quarter 1991 GDP and personal income estimates.
Each month BEA issues press releases on various economic statistics,
including personal income and outlays and indexes of leading,
coincident, and lagging indicators. Additionally, it has quarterly
press releases for quarterly data, such as GDP and corporate profits.
EXPERTS DO NOT BELIEVE BEA
ACTIONS POLITICALLY
MOTIVATED
---------------------------------------------------------- Letter :4.4
We found that knowledgeable federal experts from the Department of
the Treasury, the Federal Reserve, and BLS, and former staff of the
Council of Economic Advisers did not believe that BEA's actions, in
the first quarter of 1991 or in general, were politically motivated.
Federal Reserve officials even noted that BEA faces a number of
obstacles in producing accurate and timely statistics, and does an
admirable job despite these obstacles.\20
--------------------
\20 We also contacted an economic consultant who had been quoted in
the first Barron's article as saying that he had been told by a BLS
employee that BEA did not intend to incorporate UI data in the
October revision. The consultant told us he had been incorrectly
quoted in the article and that he had no knowledge that BEA was not
going to include UI data.
WAYS EXIST TO IMPROVE PERCEIVED
INTEGRITY OF DATA
------------------------------------------------------------ Letter :5
Despite their expressed confidence in BEA's integrity, even some of
these experts said they did not fully understand how BEA uses various
data elements in its specific calculations. This lack of knowledge
and some of the press allegations and California officials' concerns
about first quarter 1991 data (as discussed on pp. 29 to 30) are
indications that there was a misunderstanding of the procedures BEA
used for making first quarter 1991 estimates. Similar questions and
allegations arose about BEA's GDP growth estimates for third quarter
1992. BEA could do more to actively forestall misunderstanding and
increase the level of trust.
In October of 1992, the National Academy of Science's Committee on
National Statistics published guidelines, contained in Principles and
Practices For A Federal Statistical Agency, which call attention to
several principles and practices that can aid a statistical agency's
ability to maintain credibility for itself and its data. The Academy
suggests that a statistical agency should, at a minimum, (1) be open
about its data, including making its data widely available and
cooperating with users; and (2) have a strong measure of independence
to guarantee the integrity of the data. For BEA to ensure its users
of the integrity and credibility of its data, it needs to more
effectively document and publicize its methodologies and
assumptions--especially in situations like first quarter 1991. It
also needs to enhance its communication with users to achieve this
perceived independence.
INCOMPLETE BEA DOCUMENTATION
AND EXPLANATION OF KEY
PROCEDURES
---------------------------------------------------------- Letter :5.1
A 1990 report by the Department of Commerce's Inspector General\21
found BEA needed improvement in documenting estimating procedures,
revisions and adjustments. In the case of BEA's use of employment
data in first quarter 1991 GDP and personal income estimates, we also
found that more complete documentation was needed for revisions and
adjustments. BEA officials said BEA provides documentation of its
standard methodology in its monthly publication, the Survey of
Current Business. While that is true, we found the key assumptions
used in these methodologies for selected first quarter 1991 data were
not discussed in the Survey, nor was BEA able to provide us with
sufficient documentation of these assumptions.
For example, BEA officials were able to explain to us how and why
they estimated the $15 billion adjustment to wages and salaries.
They noted that when they were first determining how to account for
the decline in first quarter UI wages, they assessed other economic
data, such as federal tax receipts, to determine whether the UI wage
data represented an actual economic downturn. After assessing these
data, none of which was truly superior, however, they concluded that
the best course of action was to adjust the CES-based estimate by the
product of the 650,000 positions and average earnings. There is no
discussion of this in the Survey, nor were BEA officials able to
provide any written documentation showing these research efforts or
the review and approval process that determined this was the best
method available. In the January 1992 Survey, BEA announced that it
had reduced wages and salaries by $15 billion to account for the
employment revision, but this announcement did not explain how the
$15 billion was estimated.
Furthermore, BEA officials were unable to completely document-- nor
is there any explanation of this in the Survey--how factors offset
the $15 billion so the final reduction in wages and salaries was $10
billion, nor how this contributed to the $14 billion reduction in
personal income. According to BEA officials, BEA's standard
methodology clearly explains how a revision in wages and salaries
contributes to a revision in personal income, and there are limits to
how much additional detail the press or other users want or find
useful. Additionally, BEA officials said that to retrieve or
replicate data at each intermediate step of the process would require
an inordinate amount of effort. Nonetheless, in this instance, BEA
officials said they believe it would have been useful if BEA had been
able to complete and publish a methodology paper synthesizing
existing explanations of how personal income is calculated, as it may
have aided user understanding of this issue.
Regarding BEA's state personal income procedures, BEA officials said
they did not issue any warning about the large upwards adjustment of
the October estimates of state personal income because BEA followed
its standard procedure of adjusting state growth to equal national
growth. The officials said this procedure, as well as limitations
with the data, is fully explained in several Survey articles, most
recently in 1989 and 1990, and in a detailed methodology paper about
state estimates published in 1989. BEA officials said the 1990
article was prepared primarily to provide state users some idea of
the limitations of using BEA state personal income data in state
revenue estimates. The 1990 article said that
"...an important factor limiting the improvements in the
reliability of the estimates achieved by the introduction of the
[UI] data into the state wage and salary estimates is that the
second estimate of wages and salaries is controlled to
[national] estimates that have not yet incorporated the [UI]
data."
Additionally, BEA officials said they presented these findings to a
conference of state tax administrators and sent this article, as a
part of its regularly quarterly mailing, to its Users Group, which
includes one or more state offices from each state. Nonetheless, the
revision made to first quarter 1991 data was unusual in that UI data
were significantly lower than CES-based data (yet were still revised
upwards to match the CES-based national estimates) and BEA officials
said they knew the UI data were not stable. We believe that this
situation merited an additional explanation at the time of the
release. In the end, it turned out that the cumulative revision to
the original estimate of state personal income growth estimates (from
July 1991 to October 1992) was actually larger than it historically
had been (according to the 1990 article) for several states, and for
California the revision was almost twice as large.
California Finance officials told us they had known that BEA adjusted
data to the national total, and they were under the impression that
large states received a disproportionate share of the conforming
adjustments. However, they had not fully understood how these
adjustments were made nor that some nonwage income items were also
adjusted on the basis of CES-based wage and salary growth. As a
result, California Finance officials did not understand why BEA's
estimates were so much more optimistic than California data showed
they should be. Although the officials now are aware of BEA's
procedures, they still question the rationale of adhering to such a
policy when it produces state data they believe are misleading.
The central issue for California Finance officials is that this
procedure adjusts the growth in UI wages and salaries to match
CES-based estimates of wage and salary growth, when California
Finance officials do not believe CES adequately measures employment
in this state. California Finance officials said they believe BEA
should be willing to deviate from its standard procedures in unusual
circumstances such as these. They said BEA should at least be more
forthcoming about what BEA does in situations like the first quarter
1991. (For a detailed discussion of California Finance officials'
concerns with CES, see pp. 51 to 52.) According to California
Finance officials, the magnitude of the BEA personal income
revisions, and the amount of time it took for these revisions to be
reported, underscores California's problem with federal economic
statistics. The officials said procedural delays in incorporating
source data are not acceptable when the data are used for estimating
revenue, because accurate revenue estimating requires up-to-date,
correct source data.
Moreover, at no time, either before or after BEA released the revised
GDP estimates, was there an explanation either in the Survey or in a
special press release that stated that the employment revision would
have no effect on GDP levels, in spite of the press speculation to
the contrary. BEA officials noted that BEA held a press briefing
before it released the results of the comprehensive revision, and the
issue did not surface at that time which indicated that the press was
not concerned about this issue. After the release of the data,
however, BEA did receive several calls from reporters asking about
the effect of the employment revision. In that regard, BEA officials
acknowledge that it might have been helpful to place the announcement
of the $15 billion revision in the December rather than the January
Survey. Nonetheless, an additional public statement by BEA at the
time of the data release would have established its position and
might have focused the issue as a technical or procedural
disagreement rather than an allegation of political manipulation.
--------------------
\21 Review of Procedures for Developing GNP Estimates, Inspector
General, United States Department of Commerce (September 28, 1990)
pp. 4 to 5.
NO RESPONSE TO ALLEGATIONS
---------------------------------------------------------- Letter :5.2
In addition to incomplete documentation and explanation before or at
the time of the data releases, BEA issued no public response to the
press allegations or California's concerns. BEA has not issued any
kind of a public statement defending its first quarter 1991 estimates
or explaining that it followed standard methodology for making those
estimates. BEA officials said in October 1992 that in past years,
BEA used to respond to allegations but found the truth never caught
up with the allegations. As a result, they are not sure a public
explanation would have been effective for first quarter 1991 data.
Additionally, they said they believed the allegations made for first
quarter 1991 were from a small segment of the press and one state and
were not widespread. BEA officials said that BEA cannot successfully
respond to such uninformed views when they present BEA's methods
inaccurately.
In the last several months, a number of press reports have raised
questions and allegations about the integrity and accuracy of BEA's
estimates of 1992 GDP growth. The reports raising these questions
have been found in a wide variety of publications. Although we did
not evaluate the basis of these recent articles, the Commerce
Department and BEA have publicly acknowledged this undercurrent of
suspicion by responding to the allegations. On November 6, 1992, BEA
wrote a letter in its defense to the editor whose newspaper had
published an article implying political manipulation of third quarter
1992 GDP estimates by saying that the preliminary estimate of 2.7
percent GDP growth for the third quarter of 1992 (issued the week
before the presidential election) was not credible. Later, beginning
on November 18, 1992, BEA also held several briefings with
representatives of several newspapers in an effort to fully inform
the press about how BEA makes its estimates.
An article in The New York Times called the briefing a "highly
unusual step of inviting closer scrutiny of its [BEA's] methods for
compiling reports on gross domestic product." The article said that
the Commerce Department had ". . . responded with plausible,
generally persuasive explanations of suspiciously high numbers over
the last three weeks. But [Commerce] officials said they feared that
confidence was not fully restored in the Government data ...". A BEA
official quoted in the article said he felt the economic fraternity
might accept the figures and be reassured, noting, however, that
"Outside the Beltway, the notion that this [political manipulation of
the data] could happen again is not very far-fetched." BEA officials
subsequently told us that this does not represent BEA's official
viewpoint, which is that BEA believes today there is no question
among the mainstream press that BEA has high standards of integrity
and that estimates are free of political manipulation.
BEA officials said the criticisms raised with regard to the third
quarter 1992 statistics merited attention because they were raised by
more knowledgeable reporters and were, at the time, reasonable, as
the quarterly GDP showed more positive growth than many economic
experts had expected. Moreover, the allegations primarily concerned
methodological procedures, rather than allegations of political
manipulation. BEA officials said the continued upturn of the economy
supported BEA's early estimates.
BEA's response to the most recent allegations is a positive example
of how BEA should respond when its integrity and methods are
questioned. BEA's belief, however, that the allegations and
criticisms voiced about first quarter 1991 data represented merely
views of "uninformed" commentators and therefore did not merit a
response is faulty, for, as demonstrated in appendix II, one or two
allegations can have far reaching effects into a variety of other
sources. Regarding the allegations made about first quarter 1991
data, it would have greatly benefitted BEA to have issued a public
statement explaining the distant, or even nonexistent, relationship
of employment revisions to GDP, especially in light of the press
speculation about the issue. Additionally, BEA could have refuted
the allegations of manipulation sooner with such an explanation.
Moreover, as suggested before, if BEA had informed its state users
that the UI wage data did not appear to be stable at that time, and
might continue to be significantly revised either up or down, this
would have provided a warning to state users to use state personal
income estimates with more than the usual degree of caution.
BEA's current belief that some allegations require a response while
others do not also could present the appearance of inconsistency.
BEA needs to use appropriate criteria that allow it to make
consistent decisions on how it should respond to allegations, whether
they be allegations of political manipulation or methodological
procedures.
BEA CAN ENHANCE
COMMUNICATION WITH USERS
---------------------------------------------------------- Letter :5.3
As shown above, there can be much discussion, yet little
understanding, about how GDP and personal income are estimated and
how BEA uses employment data in these calculations. It is BEA's
responsibility to communicate more effectively with its users on how
it uses employment as well as other data in its calculations. BEA
seeks and receives considerable advice from federal, business, and
academic experts on conceptual issues involved in constructing the
framework of national economic accounting and identifying its
long-term direction. Additionally, last year BEA began discussions
with the Department of Commerce about creating a formal advisory
board for obtaining additional expert advice. We believe these
reviews can help BEA establish the general direction and concepts
used in national economic accounting. They are not necessarily
designed, however, to provide BEA with guidance or advice on specific
cases for which source data are questionable, such as in first
quarter 1991.
BEA does not believe outside advice can be sought during the
development of quarterly estimates because it would be impossible to
achieve a consensus of unbiased persons on a particular issue within
the required time frame. Nonetheless, when reliable or appropriate
source data needed to measure selected national economic statistics
are missing or are otherwise questionable, BEA's estimates can be
said to be part science and part art. In these situations, BEA is
forced to make judgments, as it did for the first quarter 1991 in
deciding whether and how to incorporate the employment revision and
UI data. In instances such as this, when reliable or current data
are not available and disputes about estimating methods exist, BEA
needs mechanisms for outside expert review and comment on its
procedures, including discussions with users to determine what would
best suit users' needs. Such reviews could be done by the formal
advisory board BEA seeks to establish or by other means. By
demonstrating openness about its procedures and a willingness to
explain its judgments, BEA could use these expert reviews to reduce
suspicion and increase the level of trust. The reviews would be
especially helpful in times such as first quarter 1991, or even more
recently, when controversy and distrust exist about the perceived
integrity of BEA's estimates.
We understand that BEA must meet its data release schedules and,
therefore, must use its judgment as to when and how it can use
outside peer review and comment. For that reason, these expert
reviews could be done after the data are released--particularly when
serious allegations are made about the integrity and credibility of
BEA's estimates. If later published, these reviews also would serve
to inform users how BEA uses various data elements in its
methodologies, as well as ensure the public and Congress that BEA's
judgments are based on sound technical considerations and not
political motivations.
It is not easy to explain or understand the methods used to produce
economic statistical data such as the GDP or personal income. For
this reason, it is BEA's responsibility to reassure users at all
levels of technical proficiency that the economic data released by
BEA are credible and produced with the greatest professional
integrity. As a result, even if BEA fully documents and widely
explains its calculations, issues public statements defending its
estimates when questioned, and obtains independent outside expert
review of its procedures and methods, more regular testimony may be
needed to help assure Congress and general users of the integrity of
BEA's data.
It is of some note that the head of BEA does not customarily speak
before Congress or public bodies to present the results of BEA's data
analysis, as this is usually done by the Undersecretary of Economic
Affairs. The head of BLS currently testifies each month when it
releases labor force statistics; this could be a model for BEA to use
as well. Congress may want to request that BEA regularly testify
before the Joint Economic Committee, perhaps on a quarterly basis,
when BEA issues its sensitive economic statistics, such as GDP. This
would serve to enhance the perceived integrity of BEA's data for
general users and the public.
NATIONAL ACADEMY OF SCIENCES
EMPHASIZES IMPORTANCE OF
INDEPENDENCE
---------------------------------------------------------- Letter :5.4
The recent report by the National Academy of Sciences emphasizes that
the organizational independence of a statistical agency can affect
the perceived integrity of the data it produces. It notes that a
perception of independence is necessary in order for the agency to be
effective. As a result, the Academy suggests several features and
practices that help to establish a statistical agency's independence.
It notes that not all these aspects are necessary, as circumstances
may govern what form independence takes.
BEA exhibits several of the aspects of independence suggested by the
Academy. For example, BEA has a professionally qualified agency
head. It has procedures that control the release of data, and it
adheres to data release schedules. BEA also strives to separate data
release from policy analysis by making all of BEA's press releases
and the Survey publications factually based and consistently
formatted. According to BEA officials, BEA leaves the policy
analysis of the data to the users.
BEA does not exhibit other aspects of independence suggested by the
Academy, such as organizationally defined independence and
presidential appointment and Senate confirmation of the agency head
who reports directly to the Secretary and has primary authority for
the selection and promotion of BEA's senior professional staff.
Additionally, the Academy recommends that the head of an agency have
the authority to present the results of data analysis before public
bodies and Congress, which, as previously stated, the head of BEA
does not regularly do.
Just because BEA lacks some aspects of independence does not mean it
is insufficiently independent or less independent than those agencies
that exhibit independence in a different way. The appropriate form
and degree of independence necessary to ensure not only the fact but
the perception of BEA's statistical integrity is a complex issue
outside the scope of this review. It should be assessed in a context
larger than just that of the circumstances surrounding the issuance
of first quarter 1991 economic data.\22
--------------------
\22 In that regard, GAO is currently examining the issue of BEA's
independence as a part of a larger review of how several federal
statistical agencies, also including the Bureau of the Census, the
Bureau of Labor Statistics, and the National Center for Health
Statistics, compare in their policies and practices with many of the
guidelines recommended by the National Academy of Sciences.
CONCLUSIONS
------------------------------------------------------------ Letter :6
We found no evidence that BEA manipulated first quarter 1991 personal
income or GDP estimates for political purposes. BEA generally
followed its standard procedures for using employment data in these
estimates and deviated from these procedures only when required by
reasonable technical judgments.
The role employment data play in BEA's procedures varies for
conceptual and timing reasons. GDP is a measure of the total
production of the economy and is based almost exclusively on measures
of output rather than income or employment. Employment data are more
important in the estimation of personal income than GDP. As a
consequence, while BEA did account for the employment revision in its
December estimate of first quarter 1991 GDP growth, this did not
affect the estimated growth of the December or subsequent estimates
of first quarter 1991 GDP. Additionally, BEA's procedures call for
employment data to be used in different ways and at different times
in state personal income estimates. As a result of these procedures,
the incorporation of the revised employment data did not affect the
estimated growth of BEA's October estimate of first quarter 1991
state personal income growth as much as was expected by those who
made the allegations.
We found several ways, however, that BEA could improve the perceived
integrity of its data. For example, we believe BEA could more
completely document and explain its data. BEA has not adequately
documented or explained to its users several key assumptions it used
in its methodology to incorporate employment data into these first
quarter 1991 estimates. Additionally, BEA has issued no public
response to the allegations made about the accuracy of first quarter
1991 data. To determine user needs and to assure Congress and the
general public of the integrity and credibility of its data, BEA also
needs additional mechanisms for outside expert review and comment on
its procedures. To enhance congressional and public assurance of the
integrity of BEA's data, Congress may also want to request that BEA
testify regularly before Congress when it releases major national
economic statistics.
RECOMMENDATION
------------------------------------------------------------ Letter :7
We recommend that the Secretary of Commerce instruct the head of BEA
to formulate a strategy to provide better explanation and
documentation of BEA's procedures to general users and assure
Congress and the general public of the integrity and credibility of
BEA's estimates. This strategy should include a review of the
principles and procedures for documenting and explaining its
methodology and calculations, responding to allegations of political
manipulation, and creating additional mechanisms for outside expert
review and comment.
MATTER FOR CONGRESSIONAL
CONSIDERATION
------------------------------------------------------------ Letter :8
Congress may want to consider requesting that BEA testify at regular
intervals when it releases major national economic statistics.
AGENCY COMMENTS
------------------------------------------------------------ Letter :9
We discussed a draft of this report with BEA, BLS, and California
Department of Finance officials and incorporated their comments as
appropriate. BEA officials agreed with our description of how they
used employment data for first quarter 1991 estimates, while BLS
officials agreed with our description of their procedures and efforts
undertaken to assess the causes of the 1991 employment revision.
BEA officials said BEA generally ensures complete documentation and
explanation of BEA's procedures to its users, but BEA is always
willing to enhance the effectiveness of its communication.
Nonetheless, staff and time resources limit BEA's ability to
participate in additional types of communication. BEA officials said
they would be willing to testify regularly before Congress in an
effort to keep Congress and the public more informed of the results
of BEA's data analysis.
BLS officials said they continued to believe BLS' employment measures
are adequate based on the results of their research on the 1991
benchmark revision and the small size of the 1992 benchmark revision.
However, BLS officials also recommend that users of BLS employment
data should exercise caution when making wage or revenue estimates
since there is not any simple, direct, relationship between
employment and wage movements. They said they would welcome broad
outside review specifically addressing CES and UI employment data not
only to assure users of the integrity and adequacy of BLS' employment
data, but also to provide users with more information on the
limitations in the direct use of employment data in revenue and
income estimates. BLS officials said the efforts required to combat
the misinformation supplied to the press during this time, including
preparing testimony for Congress, contacting policy makers and data
users, developing special public information packages, making
follow-up calls and holding briefings with journalists, was a severe
drain on BLS' limited resources.
California Finance officials said they agreed that neither BLS nor
BEA politically manipulated first quarter 1991 personal income or GDP
estimates. However, the officials said they continued to have
concerns about the procedures used by BLS and BEA, since there
appears to be an emphasis by both agencies on following standard
procedure, even when deviating from that procedure may provide more
appropriate data for state users. They said better communication and
documentation would enhance states' ability to effectively use
federal data in state revenue forecasting.
---------------------------------------------------------- Letter :9.1
As arranged with the Committee, unless you publicly announce its
contents earlier, we plan no further distribution of this report
until 15 days after its issue date. At that time we will send copies
to the Secretary of Commerce, the Undersecretary for Economic Affairs
of the Department of Commerce, the Director of the Bureau of Economic
Analysis, the Deputy Commissioner of the Bureau of Labor Statistics,
and other interested parties.
The major contributors to this report are listed in appendix V. If
you have any questions about this report, please contact me on (202)
512-8676.
Sincerely yours,
L. Nye Stevens
Director, Government Business
Operations and Information
Issues
OBJECTIVE, SCOPE, AND METHODOLOGY
=========================================================== Appendix I
Our objective was to determine whether the Department of Commerce
manipulated the first quarter 1991 GDP and personal income figures
for political purposes. We identified and examined the allegations
made by press and by state officials about first quarter 1991
estimates, and we interviewed cognizant officials at BEA, BLS, and
four states that experienced large employment revisions. We found
that the allegations of political manipulation concerned whether BEA
incorporated a downward revision of employment into its estimates of
GDP as well as personal income.
As a result, we identified BEA's standard procedure for incorporating
employment data into GDP and personal income. To determine whether
BEA deviated from this procedure for first quarter 1991 GDP and
personal income estimates, we reviewed BEA procedures and decisions
made for first quarter 1991 data and discussed these with appropriate
BEA officials. On the basis of this evidence, we determined when
procedures used for first quarter 1991 data represented a deviation
from standard procedure and whether the reasons for these deviations
were reasonable. Additionally, we compared the magnitude of
historical revisions in these data series to the revisions made in
first quarter 1991. We did not examine in detail the procedures and
methods used to incorporate other source data into GDP or personal
income, nor did we assess the adequacy of BEA's methodology or
independently verify data used in this report.
We also assessed how well BEA documented and publicized key decisions
and procedures used for first quarter 1991 data, especially when
there were any deviations from standard procedures. We also examined
existing procedural and organizational safeguards at BEA designed to
ensure the public of the integrity of BEA's data.
We interviewed cognizant officials in BEA and BLS and key officials
in organizations that provide source data to BEA and those that use
BEA data for forecasting, such as the Internal Revenue Service, the
Council of Economic Advisers, the Department of the Treasury, and the
Federal Reserve Board. We also obtained and examined procedures and
methodology at BLS for measuring and analyzing employment growth
rates and trends.
As agreed with the Committee, we obtained informal agency comments
and incorporated them as appropriate. We did our work in Washington,
D.C., from June to October 1992 in accordance with generally accepted
government auditing standards.
CHRONOLOGY OF EVENTS FOR FIRST
QUARTER 1991 DATA
========================================================== Appendix II
BLS' realization in the fall of 1991 that original employment levels
estimated by CES may have been overestimated for the first quarter of
1991 caused several journalists and California state officials to
raise a number of concerns about how well employment is measured.
The suggestion of a large possible downward revision in employment
also raised questions about how BEA's next scheduled revision of
first quarter 1991 GDP and personal income growth would be affected,
since BEA's original estimates of these data had used the original
employment levels. When BEA revised its original estimates in
October and December of 1991, however, the new estimates of GDP and
personal income did not appear to be affected as much as the press
and California officials expected they would have been if BEA had
used the revised employment data.
The allegations about the accuracy of employment measurement, as well
as the integrity of BEA's data and its methodology, then began. The
allegations and concerns have continued throughout 1992; recent
articles now question BLS and BEA's procedures and results for 1992
data. Discussed below are the press allegations and California
officials' concerns with regard to first quarter 1991 GDP and
personal income. The concerns raised by California officials with
regard to BLS' 1991 employment revision are also discussed in
appendix III. Figure II.1 depicts the chronology of events for first
quarter 1991 data, including when BLS and BEA released original and
revised estimates of employment, GDP and personal income growth, the
statements and concerns made by California Department of Finance
officials, and excerpts of many of the subsequent news reports about
and interpretations of first quarter 1991 data.
POLITICAL MANIPULATION
ALLEGED BY THE PRESS
------------------------------------------------------ Appendix II:0.1
In October 1991, when BEA released its revised figures for first
quarter 1991 state personal income growth, the estimated growth for
the sum of all states' personal income actually increased from the
growth rate shown in the original estimate, which had used CES-based
wages and salaries. An October 1991 Barron's article alleged that
BEA purposely did not incorporate the UI data (which showed a lower
level of wages and salaries than did the CES-based data). According
to the article, a private consultant had been told by a BLS employee
that the employment revisions for large states would probably not be
reflected in the revised estimates. According to the article, the
slightest appearance of a weaker labor market would be too much of an
embarrassment to the President, who had vetoed the unemployment
benefits extension bill.\1 After this article was released, a number
of other articles followed that continued to question BEA's
procedures and calculations.
The December 1991 estimate of GDP (adjusted for inflation) actually
showed a better rate of growth than had the previous estimate that
had been released in June 1991. This occurred despite BEA's
incorporation of the revised employment data. In December 1991,
another Barron's article appeared, which reported that while BEA did
incorporate the revised employment data into its GDP estimate, it
nonetheless increased the statistical discrepancy so there was no
economic impact of the employment revision. In May 1992, an article
in the San Francisco Examiner alleged that the revised employment
data were simply not included in this revision of first quarter 1991
GDP growth.
--------------------
\1 GAO contacted this consultant, who said he had been misquoted in
the article. According to the consultant, he had no knowledge that
BEA did not plan to incorporate the UI data into the revised
estimates.
CALIFORNIA OFFICIALS SAID
1991 BEA STATE DATA
MISLEADING
------------------------------------------------------ Appendix II:0.2
A procedural concern about BEA's estimates originated in the
California Department of Finance in October 1991 when BEA released
its October revision of first quarter 1991 state personal income
growth. Because there was a 320,000 difference between first quarter
CES and UI employment levels as well as a decrease in state revenue
withholding for the quarter, California Finance officials believed
the UI wage data for the quarter were accurate. As a result, they
expected to see California personal income growth become negative
once the employment revision was incorporated. Instead, BEA's
published estimates showed California personal income at a growth of
about 0.2 percent, which, while it was a decrease from the estimated
growth shown in the original estimate, was still too optimistic for
California Department of Finance officials. This estimate caused
California Finance officials to question how BEA was able to make
personal income growth remain positive when available California data
showed a definite decline from the prior quarter. For that reason,
California officials have said that BEA's procedures produced
misleading data.
Figure II.1: Chronology of
Events for First Quarter 1991
Data
(See figure in printed
edition.)
(See figure in printed
edition.)
(See figure in printed
edition.)
(See figure in printed
edition.)
(See figure in printed
edition.)
(See figure in printed
edition.)
(See figure in printed
edition.)
(See figure in printed
edition.)
Note: All data in the figure are at seasonally adjusted annual
rates.
\a GDP is measured in dollars adjusted for inflation.
\b Personal income is measured in dollars not adjusted for inflation.
\c State personal income is measured in dollars not adjusted for
inflation.
\d Estimates from California ranged from 1 to 2 million.
MEASURING EMPLOYMENT: QUESTIONS
AND ISSUES
========================================================= Appendix III
BLS is the agency within the United States Department of Labor
responsible for collecting and publishing data on national and state
levels of employment, hours, earnings, and wages. BLS collects these
data through two measures: (1) a monthly sample of establishments,
which is called the Current Employment Statistics (CES) survey; and
(2) quarterly reports required by unemployment insurance laws (UI).
These data are used in BEA's estimates of GDP, personal income, and
state personal income.\1
Once a year, BLS determines a benchmark level of employment by
reconciling the difference in the employment levels estimated by
these two measures. In 1991, the difference was unusually large at
-640,000 positions, or �.6 percent, as the average revision over the
past decade has been under 200,000 positions.\2
Historically, a downward revision in employment such as this has
generally been thought to mean original CES employment levels were
overestimated.
BLS publicly announced the expected revision in testimony before the
Joint Economic Committee in November 1991. After making several
public statements about how it determined the magnitude of the
revision in 1991 and the potential causes of the revision, BLS
completed its research in January 1993. In January 1993, in
testimony before the Joint Economic Committee, BLS testified that it
had completed its research. It announced that 500,000 of the
original employment revision (640,000 positions) was due to a
one-time, noneconomic correction in how employment was reported in
the UI. According to BLS officials, BLS now believes the 1991
benchmark revision was actually 140,000 positions, which is well
within the historical average. BLS also announced at that time that
it expects a relatively small national employment revision for
1992--under 100,000 positions.
Because of the results of their research, which showed the original
1991 CES estimates to be more accurate, BLS officials continue to
believe CES is an adequate measure of monthly employment levels.
They are not sure, however, whether the use of CES employment data in
wage and revenue estimates is the most effective, as they do not
believe there is a simple, direct relationship between changes in
employment and changes in wages and revenue.
--------------------
\1 BEA uses other employment data in its calculations, albeit in a
minor way. See appendix IV for more information on other employment
data used by BEA.
\2 BLS originally estimated the revision at -650,000 positions; this
was later revised to -640,000 positions.
BLS USES TWO METHODS TO
ESTIMATE EMPLOYMENT LEVELS
------------------------------------------------------- Appendix III:1
Because it would be prohibitively time-consuming and costly to survey
all industries and all establishments on a monthly basis, BLS
measures employment through a monthly sample survey as well through
data obtained from quarterly reports required by unemployment
insurance laws. CES is BLS' monthly sample of employment and
earnings; it produces very timely estimates of monthly changes in
employment and earnings levels and provides BLS the ability to
control the data quality of individual reports. It is necessary,
however, to periodically recalibrate these sample estimates against
full population counts. That is the role of the UI reports, which
collect employment and wage data from practically the entire universe
of the country's nonagricultural establishments subject to
unemployment insurance laws. However, the data are collected only
once a quarter and are not reported to BLS until about 6 months after
the end of each quarter.
CURRENT EMPLOYMENT
STATISTICS SURVEY (CES)
FOCUSES ON MONTH-TO-MONTH
CHANGE
----------------------------------------------------- Appendix III:1.1
Every month state employment security agencies, through a contract
with BLS, collect data on employment and earnings from over 380,000
nonagricultural establishments for the CES survey.\3
These data, which are provided voluntarily by the establishments,
include monthly employment levels, hours, and earnings based on the
number of employees who worked during any part of the pay period that
included the 12th of the month.\4 CES data are collected from a
sample representing about one-third of the jobs in U.S.
establishments subject to unemployment insurance laws. The smaller
sample size allows BLS to more tightly control the quality of the
monthly data. BLS chooses the CES sample by stratifying the universe
of establishments into more homogeneous groups based primarily on
industry and size. These groups are called estimating cells. This
sampling method practically ensures that larger establishments will
be included in the sample, but there is less assurance that smaller
establishments will be included. Nonetheless, over half of the
establishments sampled in CES have fewer than 50 people.
To determine employment and earnings estimates from this sample, BLS
first derives employment and earnings for each estimating cell based
on the data provided by the sampled establishments in that cell.
Because there are numerous estimating cells for each industry, BLS
sums the totals of each of these industry estimating cells to achieve
an employment and earnings total for a particular industry. BLS
aggregates the totals for each industry to derive the employment and
wage totals for all establishments sampled. BLS then adjusts the
totals to compensate for any sampling biases in CES or its inability
to capture the entry of new establishments. Although CES data are
collected monthly, some establishments in the sample report late and
may take up to 3 months after the reporting month to submit and edit
their data. As a result, BLS publishes three estimates of each
month's data, with each successive estimate based on a greater rate
of response from the establishments sampled. Table III.1 shows the
release dates and names of the CES estimates, as well as the percent
response upon which each estimate was based in 1991.
Table III.1
CES Release Dates, Names, and Response
Percentages
Date of Name of Percent of
release release response
-------------------------------- ------------ ------------
First Friday after end of Preliminary 54
reporting month
30 days later Revised 82
60 days later Final 90
------------------------------------------------------------
Source: BLS.
Every month BLS publishes three estimates of employment--a
preliminary release for the previous month, a revised release for 2
months prior, and a final release for 3 months prior.
--------------------
\3 An establishment is an economic unit that produces goods or
services, such as a factory, mine, or store. Establishments included
in CES are classified by industry on the basis of their major product
or activity as determined by the establishments' percent of total
sales.
\4 The monthly earnings data from CES are currently limited to only
production and nonsupervisory workers and do not include nonwage cash
payments. From these data, BEA calculates monthly and quarterly
estimates of wages. The quarterly estimates are used in BEA's
calculation of GDP and national and state personal income.
UNEMPLOYMENT INSURANCE
REPORTS (UI) PROVIDE
UNIVERSAL COVERAGE
----------------------------------------------------- Appendix III:1.2
BLS' quarterly estimates of employment levels and wages are obtained
through the use of UI reports. At the end of each quarter, state
employment security agencies, through a contract with BLS, collect
data on monthly employment levels and total wages paid in the quarter
from over 6.5 million reporting units.\5
The units report data for the universe of the nation's
nonagricultural establishments subject to unemployment insurance
laws. As these laws cover about 98 percent of employees in the total
nonfarm sector and 97 percent of those in the private nonfarm sector,
UI data are considered to represent the universe of nonfarm employees
working in the United States. In UI reports, as in CES, respondents
are asked to report the number of employees employed during any part
of the pay period that includes the 12th of each of the months in the
quarter. However, these respondents are also asked to report total
wages paid during this quarter.
BLS does not begin to receive UI data, however, until about 6 months
after the end of the quarter. For example, although the first
quarter of a year ends in March, UI data are generally not submitted
to BLS until September and are not finalized until December. BLS
publishes annual UI data once a year.
--------------------
\5 A reporting unit is the economic unit for which an establishment
submits reports of employment and wage information required by
unemployment insurance laws. Establishments must have different
reporting units for each location of service and for each different
good or service they provide.
BLS COMPARES CES AND UI TO
DETERMINE EMPLOYMENT LEVEL
------------------------------------------------------- Appendix III:2
The use of both data instruments allows BLS to set employment trends
according to CES, the more accurate gauge of monthly change, while
using the UI as a fixed, point-in-time anchor to set the levels of
employment. By reconciling the differences in estimated employment
between the two, BLS believes it is able to provide an accurate
assessment of the annual level of employment (by using the UI count)
while maintaining accurate month-to-month changes in employment (by
using the monthly CES changes). This is called the benchmark
process; the benchmark process for 1991 is illustrated in figure
III.1.
Figure III.1: BLS' Benchmark
Process
(See figure in printed
edition.)
Source: BLS.
(See figure in printed
edition.)
BLS receives preliminary UI employment data in September for the
first quarter of the particular year. It compares the level of
employment defined by UI for March of that year to the level
estimated by CES for the same time period. As the UI is a universe
count, its employment level for March is maintained, while the CES
level is adjusted. The difference between the two surveys--called
the benchmark revision--has averaged + .2 percent and ranged + .5
percent between 1980 and 1990.
BLS then revises monthly CES estimates for the preceding year (from
the current March to the prior April) following the establishment of
this benchmark level of employment. The benchmark difference for the
current year (in 1991 it was 640,000 positions for March) is "wedged
back" incrementally to the previous April. For example, the prior
April receives 1/12 of the difference, May receives 2/12, June
receives 3/12, and so on, so the current March will amount to 12/12
of the difference. According to BLS officials, the wedging procedure
reflects BLS' assumption that the difference occurred incrementally
during the year. Also on the basis of this benchmark level of
employment, BLS revises CES monthly estimates for the succeeding year
(from the current March to the next March). BLS then has a
benchmarked level of employment for the current March, which will be
used as a starting point for the next year's benchmark process. BLS
normally releases the results of its benchmark revision in June of
the next year.
BLS ORIGINALLY REPORTED
AN ATYPICALLY LARGE
REVISION IN 1991
--------------------------------------------------- Appendix III:2.0.1
In November 1991, BLS reported that the benchmark revision was
atypically large, at -640,000 positions (-0.6 percent). The downward
adjustment affected all major industries except manufacturing. It
especially affected the services industry, which showed a decrease of
1.6 percent. BLS officials attributed much of the decrease in the
services area to changes in the definitions of what constituted
services positions. Historically, the CES survey estimates and the
UI universe counts have tracked very closely. Figure III.2 shows a
pattern of relatively small annual revisions since 1981. It was the
break in this pattern in 1991 that caused the extensive media
attention.
Figure III.2: Benchmark
Revision, 1981-1991
(See figure in printed
edition.)
Note: The revision indicates the percent difference between original
CES levels of employment and UI levels of employment.
Source: BLS.
BLS first realized there would be a large revision in 1991 when it
began receiving preliminary UI data in September 1991 for the first
quarter of that year. At that time, BLS noticed that UI employment
levels for first quarter 1991 were significantly lower than CES
estimates for the same time period. BLS immediately began research
to determine the cause of the large discrepancy; however, according
to BLS, the analysis was made difficult by the fact that the
difference was widespread across industries. In November of 1991,
the Commissioner of BLS publicly announced the likely downward
revision (at that time estimated at 650,000) in her regular monthly
testimony before the Joint Economic Committee. The early
announcement of the expected revision represented a deviation from
standard procedure, since BLS ordinarily does not announce the
benchmark revision until June of the following year. According to
BLS officials, however, BLS believed it needed to inform Congress and
the public about the expected revision in order to combat erroneous
estimates in the press about the expected size of the revision.
These statements of erroneous estimates had been widely cited in
press reports beginning in September of 1991 and were estimated at as
high as 2 million positions. The primary source of these erroneous
estimates was the California Department of Finance.\6
Between November 1991 and June 1992, in an effort to accurately
inform BLS' data users and to combat misinformation in the press, BLS
issued a number of press releases and testimonies about the expected
revision and BLS' efforts to determine the causes of the revision.
For example, in February 1992, the Deputy Commissioner of BLS
testified before the Joint Economic Committee and discussed the
continuing allegations made by the press and California officials
that the employment revision could be as much as 2 million positions.
He said BLS still estimated the revision would be about -650,000
positions. He also noted that the unemployment count would be
unaffected by this adjustment.
In February BLS also issued a set of detailed questions and answers
in response to continued speculation in news reports about the
expected revision. In June 1992 BLS issued the final results of the
revision, along with a public explanation of how it performed the
benchmark and current efforts it was undertaking to assess the causes
of the revision. At this time BLS announced that the final downward
employment revision was actually 640,000 positions, or about -.6
percent. At this time, BLS also explained in detail that the
employment decline in 1991 occurred during a 1-month time frame. As
shown in figure III.2, the CES estimates and UI records tracked very
closely with each other for all months except for January 1991.
Figure III.3: Estimated Change
in Monthly Employment Levels by
CES and UI, March 1990 to March
1991
(See figure in printed
edition.)
Source: BLS.
At that time, BLS also discussed the preliminary findings of its
research, which suggested that a significant portion of the 1991
revision could be attributed to changes in UI employment reporting
procedures put in place in January 1991. In addition, BLS also ruled
out a number of possible CES survey causes that had been suggested,
and it reported that other major national employment-related
indicators had not shown a significant worsening of the economy
during January 1991 as suggested by the UI data.
BLS officials believe the atypically large revision in 1991 brought
to the forefront many long-standing concerns about CES' ability to
adequately measure employment. Research completed by BLS, however,
has allowed it to discount many possible sources of the benchmark
revision due to potential inadequacies in the CES. For example,
there has long been a question of whether CES can adequately measure
business births and deaths on a timely basis. However, BLS
investigated the possibility that business deaths contributed
disproportionately to the 1991 revision and found that nearly all of
the January employment decline was in firms that are still viable.
BLS also ruled out another concern about CES that CES is biased
towards larger and more stable establishments due to how the CES
sample is drawn. It is possible that the sample, which covers more
employees in large establishments than in small ones, could have
prevented detection of a sudden employment decline concentrated among
smaller and less stable establishments. BLS officials said, however,
that there was no evidence of systematic survey bias, as the
discrepancy occurred all in 1 month, and BLS believes such a bias
would build gradually over the course of many months.
--------------------
\6 California officials said their original estimates were the total
of the benchmark revisions in the largest states; they did not
realize there was such a wide discrepancy between the total of the
states' estimates and the national estimate. They claim there was a
lack of explanation or communication between BLS and the public about
this. BLS officials denied this, providing documentation that showed
they had briefed the Finance staff on numerous occasions between
November 1991 and February 1992 that the projected national
employment revision would be approximately 650,000 positions.
According to BLS officials, despite these briefings, California
officials continued to provide misinformation to the press.
UI REPORTING CHANGE PRIMARY
CAUSE OF 1991 REVISION
----------------------------------------------------- Appendix III:2.1
In January 1993, in his testimony before the Joint Economic
Committee, the Deputy Commissioner of BLS discussed the final results
of BLS' research into the causes of the 1991 revision. He said that
500,000 of the 640,000 revision resulted from a one-time, noneconomic
improvement in the UI employment counts and was not due to an error
in CES employment data. As a result, the actual benchmark revision
for 1991 was 140,000, which is well within the historical average and
range. Additionally, BLS found that approximately 200,000 of the
benchmark revision in California for 1991 (-320,000 positions) was
due to the correction.
According to BLS, the drop in the January 1991 UI employment counts
was due to the implementation of new employment reporting procedures.
As a result of a BLS effort to clarify employment definitions, UI
respondents realized that they were reporting employment levels on
the basis of the number of paychecks issued rather than the number of
employees on the rolls. This practice resulted in an overestimate of
employment levels.\7 Other overstatements resulted from reporting
employees who worked at any time during the month or quarter, or from
counting individuals who were still on an employer's file but no
longer employed.
According to BLS, the effect was magnified beyond what might have
been expected to result from individual respondent corrections
because many UI reports are prepared using standard payroll or tax
processing industry software. As software was corrected, the effect
was reflected in corrected employment counts for a large number of
firms.
The remaining issue for BLS is how to adjust the historical, pre-1991
employment series to correct for historical overreporting in the UI
data. While the precise methodology is still being developed, BLS
may incorporate the 500,000 overcount into the series by assuming
that the overcount occurred incrementally over the last 10 years,
culminating in the 500,000 error in 1991. According to BLS
officials, however, this will not significantly affect historical
data; existing month-to-month movements will essentially be preserved
at a slightly lower total employment level.
--------------------
\7 The overestimation of employment could result if someone receives
two checks--for instance, a regular payroll check and a bonus check.
While this represents only one employee, it represents two checks and
the payroll firm may have reported it as two employees.
BLS ANTICIPATES A TYPICAL
REVISION FOR 1992
----------------------------------------------------- Appendix III:2.2
In November 1992, the Deputy Commissioner had told the Joint Economic
Committee that BLS expected the 1992 benchmark to produce a typical,
relatively small revision of less than 100,000 positions, which is
well within the historical revision over the past decade. He
confirmed this in his January 1993 testimony. Again, BLS officials
said they believed such an announcement was necessary to preempt and
address potential questions about the size of the current year's
benchmark revision.
BLS officials said all states except for California will have a
typical revision in 1992. California will have a larger than normal
revision, but one smaller than last year. According to BLS
officials, a typical national revision and a large revision only in
California is further evidence that there is not an ongoing systemic
problem in either data series used to measure employment. BLS
officials said the larger than typical revision in California was the
result of more reporting improvements being made by UI respondents in
California.\8
--------------------
\8 BLS officials also said that California's CES estimates have been
fairly accurate over the last decade.
IMPLICATIONS OF THE 1991
REVISION
------------------------------------------------------- Appendix III:3
The atypically large employment revision in 1991 caused many
questions and allegations to be raised about the measurement of
employment and its use in national economic statistics.
Additionally, despite BLS' continued public announcements about the
revision and its recent announcement that the expected 1992 revision
will be typical and its more recent announcement that the revision
was due to a one-time, noneconomic improvement in UI reporting,
questions and allegations continue over the adequacy of employment
measures.
BLS officials said they are aware of the concerns and questions about
the accuracy of employment data; they said they believe there
continues to be repeated confusion in the press about the various
measures of employment, and especially, its relationship to wages and
revenue. They also believe this confusion contributed to some of the
press allegations concerning BLS' ability to adequately measure
employment. BLS has already asked for an outside review of its
research and methodology used for first quarter 1991 data, and would
welcome a broader review to not only assure users of the integrity
and adequacy of BLS' employment data, but also to provide information
for BLS users on the limitations in the use of employment data wage
and revenue forecasts.
CALIFORNIA DEPARTMENT OF
FINANCE HAS CONCERNS WITH
CES ADEQUACY
----------------------------------------------------- Appendix III:3.1
While BLS officials maintain CES is a valid measure of monthly
employment and earnings, the California Department of Finance
continues to express its concerns about CES for several reasons. For
several decades the Department has used the employment levels
estimated by CES and the average wages as identified by UI to project
taxable wages for the state budget. In 1991, the projections made
from CES called for a slight reduction in wages; however, the wages
reported by UI showed a further decline of about $10 billion.
According to California officials, this decline in wages caused about
one-third ($900 million) of the state's revenue shortfall for the
1991-1992 fiscal year. Because of this real decline in taxable
wages, California Finance officials do not believe the employment
decline in 1991 represented merely a reporting change. They believe
that if it had only been a reporting change, wages from the UI would
not have decreased significantly from the original CES-based
estimates. Although California will experience another large
employment revision in 1992, Finance officials said they do not
expect to see any additional decline in estimated wages similar to
first quarter 1991 because they did not use CES to estimate wages for
1992.
The fundamental issue for the Finance Department is the adequacy of
CES for estimating employment and subsequently, state wage revenues.
California Finance officials said they believe that UI is inherently
a better measure of employment and wages because UI data represent
virtually a census of nonagricultural employees working in the
country and UI wage data have historically tracked very closely with
California tax return data. The officials admit, however, that
perhaps CES may be adequate for other states, but not California.
The officials said they believe one of the major problems could be
California's sheer size; it alone makes up 12 percent of the
population and 14 percent of personal income. As a result, if there
are deficiencies in CES, it is likely California would experience
them to a greater degree than a smaller state. Another problem could
be that the recession in California has hit hardest in those areas
CES may not measure adequately, such as services and construction.
BLS DOES NOT BELIEVE
EMPLOYMENT DATA BEST
INDICATOR OF REVENUE AND
WAGES
----------------------------------------------------- Appendix III:3.2
BLS officials said the California Department of Finance has not
demonstrated the relationship between employment and wages and
revenue; as a result, it may be using CES data incorrectly in its
revenue forecasting model. According to BLS officials, there may not
be a simple, direct link between employment and wage movements, so
CES employment data may not be the best basis for estimating state
revenue. To determine the relationship between the employment change
and wage change, BLS compared the historical relationship between the
two as reported in UI reports, which are generally considered the
most accurate measure of employment levels. BLS' analysis for the
last two decades indicated that there was no direct relationship
between a change in employment and a change in wages. For example,
while employment dropped about 4 percent between fourth quarter 1990
and first quarter 1991 (this includes the 640,000), wages dropped
almost 10 percent. As a result, BLS officials question whether
employment data reported by either CES or UI should be used in
revenue estimates. BLS' analysis also causes it to question the
relative importance of employment in these estimates, when wages seem
to change with a greater volatility than does employment. BLS
officials pointed out also that the limited scope of CES data (CES
covers only production nonsupervisory workers and excludes cash
payments) limits BEA's ability to construct wages and salaries for
all employees. BLS has proposed research, which is scheduled to
begin this year, to expand the scope of CES to obtain wage data.
BLS WOULD WELCOME A REVIEW
OF EMPLOYMENT
----------------------------------------------------- Appendix III:3.3
BLS officials said BLS has made extensive efforts to fully inform the
public of how employment is measured and to ensure the public of BLS'
integrity. BLS officials said they believe the press and state
allegations are based on incorrect data or faulty assumptions. The
officials believe the allegations made by the California Department
of Finance may have been explicit efforts to blame other sources for
California's poor revenue forecast performance in 1991. California
Finance officials deny this, saying that California's fiscal
condition merely reflects the state's economy; the problem, however,
was that the economy was far better reflected in state UI data than
in official BEA or BLS data.
Although BLS continues to believe CES is an adequate measure of
employment, as the recent announcement about the 1991 benchmark
indicates, BLS officials acknowledge that a great deal of confusion
continues about how employment data is measured and how it should be
used. BLS is interested in assuring the public that its data
collection methods produce accurate data. To that end, BLS has
contracted for an outside review by the American Statistical
Association of its research findings and conclusions about the cause
of the 1991 benchmark revision. BLS officials said BLS would also
welcome an additional outside study to prove not only the integrity
and accuracy of BLS' data, but also to help it obtain information on
how others use its data, and potential limitations in using
employment data in wage and revenue estimates.
MEASURING GDP AND PERSONAL INCOME
AND THE USE OF BLS EMPLOYMENT DATA
========================================================== Appendix IV
BEA is the agency within the U.S. Department of Commerce responsible
for calculating and releasing a vast array of statistics designed to
measure U.S. economic performance and competitiveness. At the heart
of these statistics are the National Income and Product Accounts
(NIPAs), which summarize the production, distribution, consumption,
and saving undertaken in the United States. One of the NIPA accounts
measures gross domestic product (GDP), the total production of goods
and services in the United States. Another NIPA account measures
personal income, which includes income received by persons regardless
of whether the income earned contributes to national production.
Using similar concepts, BEA also estimates personal income for each
state and the District of Columbia. BLS employment, hours, earnings,
and wage data from CES and UI are among the many sources used by BEA
in these estimates.
ESTIMATES OF NATIONAL
PRODUCTION AND INCOME
-------------------------------------------------------- Appendix IV:1
The NIPAs provide a statistical depiction of the production,
distribution, consumption, and saving undertaken in the U.S.
economy. Two of the accounts within the NIPAs measure GDP and
personal income.\1 As shown below, while CES and UI data are used in
the NIPA account measuring GDP, they are not critical to determining
the actual level of GDP. CES and UI data, however, are more
important in the NIPA calculation of personal income, as wages and
salaries account for over half of the level of income.
--------------------
\1 The other accounts within the NIPA structure measure government
receipts and expenditures, foreign transactions, and gross saving and
investment.
ESTIMATING THE GROSS
DOMESTIC PRODUCT
------------------------------------------------------ Appendix IV:1.1
GDP measures the total market value of all goods and services
produced by labor and property located in the United States. It is
used extensively to track the economy's cyclical fluctuations and to
monitor long-term growth. As shown in table IV.1, one of the
accounts of the NIPAs measures GDP. The right side of this account,
commonly referred to as the "product side," measures GDP as the sum
of goods and services sold to final users. The largest component of
the product side is personal consumption expenditures. Other major
components of the product side are government purchases of goods and
services, gross private domestic investment, and net exports (exports
less imports).
Table IV.1
National Income and Product Account
Structure, 1991
(In billions of dollars)
Goods and services
Income payments and sold to final
other costs users
-------------------- -------- ------------------ --------
Compensation of $3,390.8 Personal $3,887.7
employees consumption
expenditures
Wages and salaries 2,812.2
Other labor income 288.3
Social insurance 290.3
contributions
Proprietor's income 368.0 Gross private 721.1
domestic
investment
Rental income -10.4 Net exports of -21.8
goods and
services
Corporate profits 346.3 Government 1,090.5
purchases of
goods and
services
Net interest 449.5
National income 4,544.2
Other costs and 1,111.4
charges
Total charges 5,655.6
against GDP
Statistical 21.9
discrepancy
============================================================
Gross domestic $5,677.5 Gross domestic $5,677.5
product product
------------------------------------------------------------
Source: BEA.
The left side of the account, commonly called the "income side,"
measures the sum of income payments and other costs incurred in the
actual production of GDP, i.e., it is the sum of charges against GDP.
The largest component on the income side is compensation of
employees, the wages and salaries part of which is based primarily on
BLS' CES and UI data. Other labor income--social insurance
contributions--is based on several sources, including tabulations by
the Internal Revenue Service. Other components of the income side
are proprietors' income, rental income of persons, corporate profits,
and net interest. The income side also includes other costs and
charges, such as government subsidies, capital consumption, and the
flow of capital for U.S. production.
For the most part, the product and income side of this account can be
measured independently of each other, and this independent
measurement allows BEA to check for accuracy. For example, it is
possible to measure not only the sales of new cars but also the wages
and salaries and the profits earned by the automobile manufacturers
and their suppliers in making those new cars. Conceptually, the
income side should add up to the product side of the account; in
practice, however, there is always some difference between the two
sides. The balancing item that reconciles the two sides is called
the statistical discrepancy, which is always placed on the income
side and is small in percentage terms. According to BEA officials,
when the statistical discrepancy becomes too large, BEA looks for
errors and may modify its estimates. The statistical discrepancy, as
shown in figure IV.1, ranged from �.7 to .6 percent of GDP over the
last 11 years. Even in the first quarter 1991 when the statistical
discrepancy increased $10 billion after BEA accounted for the first
quarter 1991 UI data in December 1991, the discrepancy was still only
.2 percent of GDP, which is within the historical range.
Figure IV.1: Statistical
Discrepancy as a Percent of GDP
1981-1991
(See figure in printed
edition.)
Note: A negative percent means the income side of NIPA is greater
than the product side. A positive percent means the product side of
NIPA is greater than the income side.
Source: BEA.
The official measure of GDP is based on the product side because
there is wide agreement that the product side measurement is more
accurate than the income side measurement.\2 CES and UI data from BLS
play only a minor role in the product side and thus have little
effect on the level or growth of GDP.\3
--------------------
\2 Developments in recent years have improved the accuracy of the
product side over the income side. For example, there have been
improvements in the quarterly estimates of several product side
components, such as business inventory component of gross private
domestic investment, while the growth of taxpayer misreporting may
have diminished the accuracy of the income side, given that the
adjustments for this misreporting are rough estimates.
\3 There are some income data from BLS, such as for physicians, that
are used to impute components of the product side. Additionally,
there are some additional wage data that are used to estimate both
output and income of certain sectors of the economy (such as the use
of wages of government workers and domestic workers of nonprofit
institutions). These data are not derived from CES or UI.
ESTIMATES OF PERSONAL INCOME
------------------------------------------------------ Appendix IV:1.2
Personal income, unlike the income measured as charges against GDP,
includes transfer payments, which is income received by persons whose
services did not contribute to production. Transfer payments can
include Social Security, welfare, unemployment insurance, and
Medicare. However, wages and salaries--based on BLS' CES and UI
data--are the largest component; in 1991, they accounted for almost
60 percent of personal income.
Figure IV.2: Components of
Personal Income, 1991
(See figure in printed
edition.)
Note: Social insurance contributions are always negative components
of personal income.
Source: BEA.
RELEASE SCHEDULE OF GDP AND
PERSONAL INCOME
------------------------------------------------------ Appendix IV:1.3
BEA strives to make its estimates both timely and accurate; however,
because BEA does not collect its own data but rather receives data
from a number of other sources, it does not have a great deal of
control over when or how it receives its source data. To respond to
the competing demands for timeliness and accuracy, BEA releases
initial estimates for a specific period based on preliminary source
data, then releases several revisions that incorporate better and
more complete source data.
BEA revises and releases GDP and personal income estimates on the
same quarterly and annual schedule; personal income estimates are
also available monthly as a part of BEA's personal income and outlays
release. Personal income data (including monthly wages and salaries)
are released about 30 days after the end of the reference month. BEA
then uses these monthly estimates to derive the quarterly estimates
of wages and salaries used in personal income and GDP quarterly
estimates.
The first estimate of quarterly GDP and personal income (called the
advance estimate) is released about 30 days after the end of the
reference quarter. BEA then revises this estimate and releases a
second estimate (called the preliminary estimate) approximately 30
days later. BEA revises the estimate again and issues a third
estimate (called the final estimate) of GDP and personal income
approximately 30 days later. The final estimate remains fixed until
the following July, when BEA performs an annual revision of quarterly
and annual GDP and personal income.
The annual revision incorporates new and revised source data from
annual surveys and administrative records, along with updated
seasonal factors. For example, the annual revision incorporates
annual UI wage and salary data. The annual revision typically covers
GDP and personal income annual estimates for the past 3 years, and
quarterly estimates for the prior 12 quarters. It also covers the
first quarter of the current year of the July revision.
As shown in figure IV.3, the degree of revisions in estimated growth
between the advance estimate and the first annual revision for GDP
has ranged from 2.8 to -3.2 percentage points over the last 11 years,
with an absolute average value of 1.1 percentage points (i.e., the
average change without regard to sign). This means that the advance
estimates of growth have been between 2.8 percentage points lower or
3.2 percentage points higher than the growth shown at the first
annual revision. For first quarter 1991 GDP growth, the difference
between the advance estimate and the first annual revision was �.1
percentage point, which is well within the historical average and
range. As also shown in figure IV.3, in the last 11 years there has
been little upward or downward bias in the revisions, as the advance
estimates of GDP are neither consistently optimistic or pessimistic
when compared to the later estimates, which are based on more
complete source data.
Figure IV.3: Degree of
Revisions in GDP Growth
Estimates From the Advance
Estimate to the First Annual
Revision, 1981-1991
(See figure in printed
edition.)
Source: BEA.
Approximately every 5 years, BEA performs a comprehensive revision of
both the quarterly and annual GDP and personal income estimates. The
comprehensive revision differs from the annual revision in that in
addition to incorporating new data, such as the results of the latest
economic censuses, the comprehensive revision also incorporates
definitional and statistical changes and covers more years of data.
USE OF BLS DATA TO ESTIMATE
WAGES AND SALARIES FOR GDP
AND PERSONAL INCOME
------------------------------------------------------ Appendix IV:1.4
BEA uses the monthly CES data to estimate monthly wages and salaries,
which is the largest component of monthly estimates of personal
income. For most industries, BEA determines wages and salaries by
multiplying CES employment levels times average hourly earnings and
weekly hours.
Total wages of production and nonsupervisory workers, and the total
salaries of supervisory, clerical, and other nonproduction employees
for manufacturing industries are estimated separately. Wages are the
product of production worker employment, average weekly hours for
which pay is received, and average hourly earnings (including
overtime). BEA estimates total salaries by applying the estimated
percentage change in the product of the employment level and the
straight time hourly earnings of production workers to the prior
period total salaries estimate. This procedure reflects the
assumptions that (1) the average weekly hours for which salaried
workers are paid remains constant, and (2) the straight-time hourly
earnings of wage earners and the hourly earnings of salaried workers
both change by the same percentage during the year.
For those employees not covered by CES, BEA uses other information to
estimate monthly wages and salaries. These include the wages and
salaries of farm employees, which are derived from USDA information;
federal civilian employees, which incorporate information provided by
the Office of Personnel Management; military wages and salaries,
which are based on Department of Defense strength figures and BEA's
average earnings estimates; and private households, which are
prepared by using data from the Current Population Survey (a separate
household survey).
BEA sums the monthly estimates of wages and salaries to derive an
estimate for total wages and salaries for the quarter.\4 These
estimates are then used in BEA's advance, preliminary, and final
estimates of GDP and personal income.
About 6 months after the end of each quarter, BEA begins to receive
UI data from BLS that shows total employment and wages and salaries
for the entire quarter. UI data represent a census of those
employees covered by CES, so BEA only needs to estimate the wages and
salaries for the small number of employees not covered by UI
(generally the same as those not covered by CES, discussed above).
With the 6-month lag in UI data, the first chance BEA normally has to
incorporate UI wage data into its quarterly GDP and personal income
estimates is the following July as part of the first scheduled annual
revision of quarterly GDP and personal income estimates.
By this time BEA has received four quarters of UI wage and salary
data. BEA then adjusts the average annual level of CES wages and
salaries\5 to equal the annual average level of UI wages and
salaries. The quarterly estimates are then revised so that the level
of wages and salaries for the year reflects the level of wages shown
by the UI. The estimates, however, still maintain the quarterly
pattern of change identified by CES data.
--------------------
\4 BEA estimates the quarterly values of other labor income (i.e.,
employer contributions) included in personal income by assuming that
a direct relationship exists between employer contributions and wages
and salaries in each type of industry division (manufacturing, retail
trade, services, etc.).
\5 BEA derives its annual estimates of other labor income from the
Internal Revenue Service, the Health Care Financing Administration,
and private sources, such as the U.S. Chamber of Commerce and trade
associations.
ESTIMATES OF STATE PERSONAL
INCOME
-------------------------------------------------------- Appendix IV:2
In addition to its national measures of income, BEA also estimates
state personal income, the largest component of which is wages and
salaries based on BLS' CES and UI data. State personal income
estimates have a different revision and release schedule than do the
national estimates, and the sum of the 50 states' and the District of
Columbia's personal income is slightly different from personal income
because of different source data used at the national level. BEA
methodology, however, calls for the rate of growth in the wages and
salaries used to estimate state personal income to be adjusted to
equal the rate of growth in the wages and salaries used to estimate
personal income at the national level.
ESTIMATING STATE PERSONAL
INCOME
------------------------------------------------------ Appendix IV:2.1
State personal income includes all income received by, or on behalf
of, all residents of a state; it does not, however, include the
earnings of federal civilian and military personnel stationed abroad
and of U.S. residents employed abroad by U.S. firms. In
calculating state personal income, BEA ties the actual rate of growth
of wages and salaries in state personal income to the rate of growth
of the wages and salaries used to estimate the most recent personal
income estimates at the national level. In this fashion, BEA uses
national wage and salary growth estimates to act as a control for the
state wage and salary growth estimates. Specifically, standard
procedure calls for BEA to determine the difference, by industry,
between the rate of wage and salary growth for the sum of all states
and the rate of wage and salary growth for national estimates used in
the most recent estimate of personal income at the national level.
This difference is then allocated across all states, with each state
receiving a proportional share of the gap by industry. In so doing,
BEA adjusts the sum of all states' industry wage and salary growth to
equal the growth of the wages and salaries in that industry at the
national level. For example, if there were a 5-percent gap between
the construction wages and salaries at the national level and the
construction wages and salaries for the sum of all states, BEA would
adjust each state's construction wages and salaries by 5 percent.
This then yields an additive system wherein state growth sums to the
national growth.\6
After adjusting the wages and salaries, BEA then adds in other labor
income, which it receives from other sources often on an annualized
or national basis. It allocates other labor income based on the
proportion of wages and salaries by industry for each state. BEA
then allocates nonwage data based on other available and relevant
state data.\7
--------------------
\6 According to BEA, different source data and estimating schedules
for the national and state estimates can still result in temporary
statistical differences between the growth in the two estimates.
\7 For example, a nonwage income item such as UI benefits paid to
individuals is allocated by state on the basis of unemployment
benefits reported by the Department of Labor.
RELEASE OF STATE PERSONAL
INCOME ESTIMATES
------------------------------------------------------ Appendix IV:2.2
BEA publishes quarterly and annual estimates of state personal income
and its wages and salaries component. The first estimate of
quarterly state personal income is released 4 months after the end of
the reference quarter and is based on the wage and salary data BEA
estimates from CES and other sources discussed above. Three months
later (7 months after the end of the reference quarter), BEA issues a
revised state personal income estimate that incorporates available UI
data. These estimates are then revised at successive 3-month
intervals as improved UI data become available.
Every year in October, the quarterly estimates for the prior 13
quarters are revised on the basis of annual estimates released 2
months prior in August. Additionally, every April, the prior 15
quarters of estimates are also revised on the basis of updated state
data. In January and July of each year, at least the quarter
immediately preceding the current quarter is revised according to
updated state data.
BEA also publishes annual estimates of state personal income. The
preliminary annual estimates, which are published in April of the
year following the reference year, are derived primarily from
quarterly data, and most private wages are based on three quarters of
a year's worth of UI data. Four months later, in August, after BEA
has received the final quarter of UI data, BEA publishes annual
estimates of state personal income. These are developed
independently of the quarterly series in greater component detail,
mainly from federal and state government administrative records. In
April of the next year (16 months after the close of the reference
year), revised annual estimates are published that incorporate newly
available information. After these three steps, the annual estimates
are subject to further revision in April and August for several
succeeding years as additional data become available. Routine
revisions of the state personal income estimates for a given year are
normally completed with the fourth yearly April publication.
Thereafter, further changes result only from a comprehensive revision
of all the national income and product accounts (approximately every
5 years) or to improve the estimates through the incorporation of
additional or more current state and local area data.
USE OF BLS DATA TO ESTIMATE
WAGES AND SALARIES FOR STATE
PERSONAL INCOME
------------------------------------------------------ Appendix IV:2.3
BEA uses BLS CES employment data to determine the first estimate of
state wages and salaries (the largest component of state personal
income). For most industries, the wage and salary data produced in
the first estimate of state personal income (released 4 months after
the end of the reference quarter) are determined using the growth in
CES-based employment change for those industries. For manufacturing,
wage and salary estimates are made by using the growth rate of the
product of employment and production workers' average earnings.
After BEA begins to receive state UI employment and wage data about 6
months after the end of the reference quarter, however, the procedure
differs. Because BEA has access to the UI data for that particular
quarter before it makes its first revision (7 months after the end of
the quarter), BEA is able to incorporate quarterly UI data into its
first revision of state personal income. Standard procedure at BEA,
however, calls for BEA to adjust the states' wage and salary growth
in various industries as shown by the UI data so that the growth is
equal to the rate of wage and salary growth for that industry that
was used in the most recent estimate of personal income at the
national level. Because the previous national estimate of wage and
salary growth at this time is always based on CES, the amount of
adjustment required depends on the difference between CES-based wages
and salaries and UI-based wages and salaries.
In 1991, the adjustment was particularly noticeable, as UI wages and
salaries were significantly different from the original CES-based
wages and salaries estimates. As shown in figure IV.4, the original
decline (for all industries) of first quarter 1991 total wages and
salaries (before adjusting the UI-based industries to the national
growth rate) for the sum of all states and California was -6.4 and
-7.2 percent, respectively. However, after adjustment upwards to
match the original CES-based estimates of wages and salaries used in
the national data, wages and salaries growth was 0.2 percent growth
for the sum of all states and only a -0.4 percent decline for
California.
Figure IV.4: Growth Rate of
First Quarter 1991 Wages and
Salaries Before and After
Adjustment by BEA
(See figure in printed
edition.)
Source: BEA.
State personal income estimates continue to be revised every 3 months
to incorporate updated UI data. The growth rate in UI wages and
salaries, however, continues to be adjusted to equal those used in
personal income at the national level. Even after BEA has completed
its annual revision of personal income at the national level in July
and its state annual revision in August (when national and state
wages and salaries are based on a year's worth of UI data), the need
for adjustment continues. Although at this point all wage and salary
levels of annual estimates for personal income at the national and
state level are based on a year's worth of UI data, the national wage
and salary quarterly estimates still maintain the original CES
pattern of estimated quarterly growth in wages and salaries. As a
result, the quarterly rate of change for wages and salaries at the
state level is still adjusted to the original CES-based estimates of
the quarterly rate of change used in national data.
MAJOR CONTRIBUTORS TO THIS REPORT
=========================================================== Appendix V
GENERAL GOVERNMENT DIVISION,
WASHINGTON, D.C.
Bruce Johnson, Assistant Director
Government Business Operations and Information Issues
Lori Rectanus, Evaluator-in-Charge
Carolyn M. Peake, Evaluator
OFFICE OF THE CHIEF ECONOMIST,
WASHINGTON, D.C.
Richard S. Krashevski, Senior Economist