The Employment Cost Index: What Is It?
The Employment Cost Index: What Is It?
The Employment Cost Index: What Is It?
he Employment Cost Index (ECI) is a quarterly measure of the change in the price of
labor, defined as compensation per employee hour worked. Closely watched by many
economists, the ECI is an indicator of cost pressures within companies that could lead to price
inflation for finished goods and services. The index measures changes in the cost of compensation not only for wages and salaries, but also for
an extensive list of benefits. As a fixed-weight, or
Laspeyres, index, the ECI controls for changes
occurring over time in the industrial-occupational
composition of employment.
This article provides a broad overview of the
ECI. Beginning with how the data for the index are
collected and how the index is calculated, the discussion draws attention to some of the underlying
challenges that are involved in calculating such a
complex statistic: What types of data should, ideally, be collected? What data are collected under
nonideal conditions? and How are infrequent payments handled? Then, the article addresses a variety of questions that have been raised about the
behavior and efficacy of the ECI: How does the
index behave over the business cycle? Is it, like the
Consumer Price Index (CPI), affected by substitution bias? Does the ECI capture emerging forms of
compensation, such as hiring and retention bonuses and stock options? and, finally, How does
employer cost relate to employee value?
business establishments and government operations. (In what follows, business establishments
and government operations will collectively be
called establishments.) The data are weighted
to represent the universe of establishments and
occupations in the nonfarm private sector and in
State and local governments. The ECI sample,
and hence the estimates derived from it, exclude
Federal, private-household, and unpaid family
workers, as well as self-employed individuals
and owners of establishments.
The ECI sample is currently drawn in three stages
as part of the larger National Compensation Survey. First, sample geographic areas are selected by
dividing the United States into primary sampling
units.1 Second, a sample of business establishments and State and local government operations
is selected from within each primary sampling unit
that is chosen. Third, a BLS data collector visits
each establishment in the sample (the first visit is
termed initiation), asks for a list of employees,
and then collects a sample from this list, using predetermined rules. The employees making up the
sample represent jobs that enter into calculations
of the ECI. Techniques in which the probability of
being selected for the sample is proportional to size
are used at all stages of sampling, which means
that larger geographic areas, larger establishments,
and jobs with more employees have a higher probability of appearing in the survey. However, smaller
areas, establishments, and jobs appear as well. For
a fixed sample size, variances of estimates tend to
be smaller under this kind of sampling than under
simple random sampling.
Establishments and jobs within them remain in
Monthly Labor Review
September 2001
the ECI sample for approximately 5 years, contributing data every quarter for the pay period that includes the 12th day of the
survey months: March, June, September, and December.2 Data
on the cost of compensation are collected for all employees in
sampled jobs. After the initial personal visit, quarterly reports
are normally collected by mail or telephone by economists located in BLS regional offices. During the time a job remains in the
ECI sample, data are collected on all incumbents in the job, even
through changes in incumbency. Because the ECI does not follow changes in compensation costs for individual workers, the
average wage and salary of a job may vary over time as the
composition of incumbents varies (for example, when the tenure
of incumbents changes with the business cycle).
Due to business closings, the elimination of jobs, and the
refusal of respondents to participate further in the survey, some
establishments and some jobs drop out of the sample, an event
termed attrition. To reduce the burden on respondents, rebuild the attrition-depleted sample, and keep the sample current
with the changing economy, establishments in the sample are
replaced in a procedure termed sample replacement. Replacement of ECI samples was begun in 1981, and the method for
replacement has differed over time. From 1986 until 1997, all of
the establishments within designated groups of industries were
replaced at the same time, with different industry groups replaced in different years. This approach had the disadvantage
that the samples for some industries were older than those for
other industries, which was a problem because the sample of
jobs can become unrepresentative over time.3
Since 1997, when the Bureau began integrating the ECI into
the National Compensation Survey, replacement samples have
become cross-area and cross-industry samples, meaning that
each replacement sample is now nationally representative.
The sample is divided into five approximately equal groups
that are replaced every 5 years. Such a replacement scheme
has an advantage over the previous scheme in respect of
maintaining the currency of the sample. The new replacementgroup data contain information about the changing workforce
that may be used to adjust the sampling weights of the older,
less representative, replacement groups. Alternatively, it may
be possible to weight more heavily the data from the more
recent and more representative replacement groups. The Bureau will conduct research to determine which approach holds
more promise for maintaining an up-to-date survey.
The sample size at any time depends on the size of the initial
sample, its age, the rate of sample attrition, and sample replacements. The size of the ECI sample has varied over time. Recently,
the sample has begun to grow from a realignment of compensation survey resources. The sample is expected to continue to
grow, both from this realignment and from a budget increase. As
of June 2001, 7,365 private-industry establishments provided
data on about 31,100 occupations, while 790 State and local
government operations afforded data on about 3,800 occupa-
September 2001
tions. Current plans call for expanding the gross sample to 18,000
units, although the usable sample is expected to be at least 25
percent smaller, as some units fail to respond and others are
found to be out of business or out of the scope of the survey.
in addition to the regular work schedule (for example, overtime pay and pay for working weekends and holidays),
shift differentials, and nonproduction bonuses, such as
lump-sum payments provided in lieu of wage increases;
Insurance benefitslife, health, short-term disability, and
long-term disability insurance;
Retirement and savings benefitsemployers payments
into defined-benefit and defined-contribution plans, including Employee Stock Ownership Plans (ESOPs);
Legally required benefitsSocial Security, Federal and
State unemployment insurance, workers compensation insurance, and Medicare;
Other benefitsseverance pay and payment into supplemental unemployment plans.
All costs of benefits are converted to an hourly basis by dividing the annual cost of benefits by annual hours worked.
The information needed to calculate the cost of benefits
according to rate and usage depends on the specific benefit plan. The discussion that follows shows how rate and
12 $120 . 9
= $ 0 . 65 per hour worked.
2 , 000
For vacations and health insurance, information on eligibility
and participation in benefit plans is collected at initiation. The
information includes the distribution of workers by length of
service (used to determine the average number of vacation days
taken by employees) and the fraction of workers participating in
health insurance. When costs per hour worked for these benefits are calculated in subsequent quarters to measure the
change in the cost of the benefits, the same eligibility and participation rates are assumed as at initiation. Holding these values constant for a sampled job eliminates the effects of shifts in
the composition of the workforce on the measurement of cost
changes. (For example, it eliminates the effect of a changing
distribution of length of service, as might occur over a business
cycle.)
The policy of holding usage of benefits fixed over the period that a job remains in the ECI extends to all benefits for
The
The rate-and-usage approach usually permits the calculation of separate costs for each occupation in an establishment. An expenditure may yield just one cost for the
establishment, requiring the costs to be prorated among
occupations. Note, however, that it may be possible to obtain expenditure data for the specific jobs sampled.
Expenditures may include unwanted costs that can be difficult to exclude from the survey because the respondent
does not know whether they are included and what the
amounts may be. For example, a life insurance expenditure
might include life insurance costs for retirees.
The presumption, then, is that collecting data in the form
of rates and usage renders the data more likely to be specific
to the sampled job and to pertain to the current period. In
reality, BLS data collectors sometimes cannot obtain rate and
usage information for the sampled job. In that case, they
must either fall back on rate and usage information for a
broader occupational group or obtain expenditure data for the
job or for a broader occupational group.
Monthly Labor Review
September 2001
ECI,
a given job, if rate and usage data for that job were mixed with
expenditure data for an aggregation of jobs, then the benefit
cost would be coded as coming from expenditure data. The category titled other sources of data includes both cases in
which data were estimated and a small percentage of cases in
which the data source was not recorded.7 Estimated data
represent situations in which at least one data element used to
calculate a cost had to be estimated by the respondent. Estimated data may still be high in quality, as hard data might
account for the majority of the elements in a cost calculation.
Table 2 also shows that rate and usage data for the specific
job were most often available for holidays (93.0 percent) and
vacation leave (85.9 percent). In contrast, rate and usage data
for the specific job were available only 33.7 percent of the
time for sick leave, which often comes from other sources. A
closer examination of the data indicates that sick leave data
are often estimated.
The central point is that BLS data collectors attempt to obtain cost information that is as close to the sampled job and as
close to the reference period as possible. However, limitations
in the data available from the respondent necessitate compromises in what is collected.
Infrequent payments
Many forms of compensation are paid out relatively smoothly
over time or exist as part of a well-specified benefit package,
so that their costs can be easily associated with the reference
period. The most obvious example of this is hourly wages,
which are paid for labor services in the reference period. Even
a schedule of paid holidays (which are not necessarily taken
during the reference period) can be viewed as part of a compensation package that exists during the reference period,
and its annualized hourly costs can be attributed to that period. But some components of compensation, such as bonuses, are paid infrequently (less than quarterly), and whether
and how much will be paid in the future is uncertain. This
uncertainty raises the question of how these payments should
[In percent]
Cost data
Definedcontribution
pension
Definedbenefit
pension
Health
insurance
30,269
100.0
51.3
30.1
13.4
5.1
30,269
100.0
59.1
26.9
8.9
5.1
30,269
100.0
22.8
54.2
18.1
5.0
September 2001
Holiday
leave
30,269
100.0
16.9
68.3
10.2
4.7
Vacation
leave
Sick
leave
30,269
100.0
20.9
58.5
16.0
4.6
30,269
100.0
23.2
49.1
22.8
4.9
Table 2.
ECI,
[In percent]
Definedcontribution
pension
Definedbenefit
pension
Health
insurance
11,256
9,108
17,407
21,224
17,549
14,237
100.0
100.0
100.0
100.0
100.0
100.0
44.8
7.8
58.2
1.9
51.5
10.7
93.0
.3
85.9
1.6
33.7
9.9
.3
31.4
15.7
.7
32.9
6.3
.4
23.3
14.1
.0
.9
5.9
.0
1.7
10.7
.0
2.5
53.9
Holiday
leave
Vacation
leave
Sick
leave
September 2001
September 2001
September 2001
10
September 2001
September 2001
11
12
September 2001
September 2001
13
cal series for measuring inflation in the economy. Understanding its characteristics is helpful for interpreting how it measures cost pressures that may lead to inflation in the price of
goods and services.
Notes
1
A primary sampling unit consists of a county or a number of contiguous counties. Thirty-three primary sampling units are selected with
certainty. (That is, they would appear in any sample that was drawn.)
Others are selected with a probability proportional to their employment. For more information about samples from the National Compensation Survey, see Kenneth J. Hoffman, New sample areas selected
for BLS National Compensation Survey program, Compensation and
Working Conditions, spring 1997, pp. 2731.
2
CPI
12
14
September 2001
ing the sample weights fixed. Sample attrition may lead to some
within-cell reweighting. Further, as samples are replenished, the
within-cell weights may shift across jobs, reflecting a change in the
employment distribution within cells. Thus, the ECI does reflect some
within-cell substitution.
13
That is, the Fisher ideal is the square root of the product of the
Laspeyres and the Paasche indexes.
14
Ana M. Aizcorbe and Patrick C. Jackman, The commodity substitution effect in CPI data, 198291, Monthly Labor Review, December
1993, pp. 2533.
15
ECI
sensitive.
16
David Lebow, Louise Sheiner, Larry Slifman, and Martha StarrMcCluer, Recent Trends in Compensation Practices, Board of Governors of the Federal Reserve System, Finance and Economics Discussion Series no. 199932, working paper, 1999.
17
Recently, the Bureau conducted a quality control review of the
data on retention bonuses to confirm that data were being captured
correctly.
18
ECI data are collected from employers, so capturing exercise cost
data might be easier in the case of the more prevalent nonqualified
stock options.
19
Recall that premium pay for overtime appears in the benefit
portion of the ECI; the wage and salary measure includes only straighttime pay.
20
This section borrows heavily from Melissa Famulari and Marilyn
E. Manser, Employer-provided benefits: employer cost versus employee value, Monthly Labor Review, December 1989, pp. 2432.
21
Ibid., p. 25.
22
That is, suppose that the value to the employee of the last dollar
spent on a benefit was less than one dollar. Then the employer could
reduce expenditures on the benefit by a dollar and give that dollar to
the worker as cash compensation instead, making the worker better
off.
23
Taxation of benefits varies. Cash payments for paid leave, overtime, and nonproduction bonuses, included as benefits in the ECI , are
generally taxable in the year in which they are paid. Contributions to
retirement plans are generally tax deferred until payments are made to
the employee upon retirement or some other kind of withdrawal from
the labor force. Insurance premiums are generally not taxed.
24
Suppose instead that the compensation package were such that the
value of benefits equaled one dollar before taxes, and suppose that the
tax rate were t. Then the employer could spend one less dollar on cash
compensation (costing the employee only 1 t dollars after taxes) and
give the employee benefits equal to one additional dollar. The employee would then be better off. This substitution of benefits for cash
continues as the value of additional dollars of benefits declines, to the
point where the value of an additional dollar of benefits equals a dollar
of pay after taxes.
25
In economic theory, a normal good is defined as a good whose
quantity demanded increases with income.
26
Stock options are included in compensation to the extent that
they are reported as wages for unemployment insurance tax purposes,
a principal source of compensation income in the national income
accounts.
APPENDIX A:
(1) ECI t =
W
i Wit 100,
i0
Wit
= (1 + ri1 )(1 + ri 2 )...(1 + rit ) .
Wi 0
(4)
L
i
where
(2) i
E i 0Wi 0
.
Ei 0Wi 0
(5)
E itWi 0
E itWi 0
itp =
(3)
1 + ri =
s
jI
s
jI
where
sij =
(6)
Ft = Lt
1/ 2
Pt
1/ 2
T
j
j=1
ij
ij
Wij
Wij 1 =
sij Wij 1
s
jI
ij
Wij 1
s
jI
ij
Wij
Wij 1
100,
where
W
k =1
k0
of economic activities.
The estimation cell is the nexus of employment in a major occupation group (MOG) and an industry group (SIC); that is, the estimation cell is an item in our shopping basket of labor services.
The base-period employment weight is the number of employees
in any estimation cell estimated by the Occupational Employment Survey (OES) for the base period. The use of constant
base-period employment weights is what makes the ECI a
Laspeyres index construction.
The establishment selection weight is the inverse of the sample
establishments chance of having been selected from the universe of
establishments. (For example, if the chance of having been selected
is 5 out of 20, or 5/20, the inverse is 20/5, for a weight of 4.)
The occupation sample interval is the number of employees in
the sampled establishment that is represented by each occupation quote sampled from the establishment; that is, the occupation sample interval is the establishment employment divided by
the number of quotes selected.
September 2001
15
Assume that the OES base-period employment for these occupation groups and industries (or estimation cells) were the following for the base period:2
15
0
10,000
30,000
MOG
17
50,000
0
50,000
SIC
E .............................................
G ............................................
H ............................................
SIC
1
2
1
2
1
2
1
2
G
H
G
H
E
H
E
H
$20.00
10.00
15.00
7.50
25.00
10.00
20.00
11.00
10,000
10,000
10,000
10,000
25,000
25,000
25,000
25,000
Calculation steps
1. Calculate the weighted average hourly wage rate for the four (MOGSIC) estimation cells in the base period, using observed wage
rates, establishment selection weights, and occupation sample
intervals:
a. For each estimation cell, sum the products of each quotes
average hourly wage and its final weight.
b. For each estimation cell, sum the final weights over all quotes.
c. For each estimation cell, divide a by b to get the average
hourly wage.
Estimation
cell
a
MOG G, SIC 15 ............... $350,000
MOG H, SIC 15 ...............
175,000
MOG E, SIC 17 ................ 1,125,000
MOG H, SIC 17 ...............
525,000
b
20,000
20,000
50,000
50,000
Average hourly
wage
$17.50
8.75
22.50
10.50
2. Calculate the wage cost weight for the estimation cell by multiplying the average hourly wage by the OES employment for the
base period:
16
September 2001
Estimation
cell
Average hourly
wage
employment
Wage cost
weight
$17.50
8.75
22.50
10.50
10,000
30,000
50,000
50,000
$175,000
262,500
1,125,000
525,000
OES
3. Sum the wage cost weights over all estimation cells in blue-collar
occupations in construction: $2,087,500
$2,087,500.
4. Calculate the weighted average hourly wage rate for each estimation cell in the quarter after the base period, thereby reflecting
new wage rates (boldface type denotes a change from the base
period):
Estimation
cell
MOG G, SIC 15 ...........
MOG H, SIC 15 ...........
MOG E, SIC 17 ............
MOG H, SIC 17 ...........
Average hourly
wage
$350,000
175,000
1,175,000
550,000
20,000
20,000
50,000
50,000
$17.50
8.75
23.50
11.00
Estimation
cell
MOG G, SIC 15 ........
MOG H, SIC 15 .......
MOG E, SIC 17 ........
MOG H, SIC 17 .......
Percent change
in average
hourly wage
0.00
0.00
4.44
4.76
Previous
quarter's
wage cost
weight
$175,000
262,500
1,125,000
525,000
Current
quarter's
wage cost
weight
$175,000
262,500
1,174,950
549,990
6. Sum the wage cost weights for the current quarter over all estimation cells in blue-collar occupations in construction: $2,162,440
$2,162,440.
7. Compute the current quarters index to equal 100 (current
quarters aggregate wage cost weight/base quarters aggregate wage
cost weight), rounded to 0.1: 100 (2,162,440/2,087,500) =
103.6 for blue-collar occupations in construction.
8. Calculate the 3-month percent change equal to [(current quarter's
index/previous quarters index) 1] 100, rounded to 0.1 (in this
example, the previous quarter just happens to be the base quar3.6.
ter): [(103.6/100.0) 1] 100 = 3.6
9. Calculate the 12-month percent change in a similar fashion.
The preceding methods work for each succeeding quarter if one follows steps 49.
Notes to Appendix B
1
This description simplifies the calculation of the final weight in
this example. In the actual ECI , the final weight is the product of the
area weight, establishment weight, occupation sample interval, establishment nonresponse adjustment, occupation nonresponse adjustment,
documentation factor, and rotation factor.
2
In normal operations, there would never be estimation cells with
zero OES base-period employment. These zeros appear only for simplicity in this example.