July 2026

Employment changes in jobs and their effect on the Employment Cost Index

The Employment Cost Index (ECI) for wages and salaries is based on a comparison of the average wage rates for the same set of jobs across a 3-month interval. Employment in most of the jobs remains the same from one reference period to the next. However, employment does increase for some of the sampled jobs, while it decreases for others. Wage growth among jobs with an employment increase is typically lower and with an employment decrease is typically higher than wage growth among jobs in which employment remains the same. The high growth rate of wages among jobs with an employment decrease largely offsets the low growth rate of wages among jobs with an employment increase, so the overall wage growth in the ECI is usually similar to the wage growth among jobs in which employment remains the same.

The Employment Cost Index (ECI) measures the change in the hourly labor cost to employers over time.¹ The ECI uses a fixed “basket” of labor to produce a pure cost change. A fixed "basket" of labor is free from the effects of workers moving between occupations and industries and includes both the cost of wages and salaries and the cost of benefits. The ECI uses a Laspeyres index to calculate changes in compensation. A Laspeyres index tracks the change in the average price for a set of goods over time, in which the quantity for each good, and hence its weight in the average, is held constant. In the ECI, a job serves the role of the good, and the average compensation per hour among the workers in the job serves the role of its price.² It is important for users of the ECI to understand how the ECI obtains its sample of jobs to measure the growth in compensation over time. This article begins by describing how the sample of jobs for the ECI is constructed. Next, the article shows how increases and decreases in employment in particular jobs coincide with slower or faster wage growth in these jobs. Last, instances of job and wage changes from 2006 to 2025 are analyzed with the aid of ECI data. In addition, an appendix presents an overview of the formulas used in the calculation of the ECI.

The sample of jobs

Data for the ECI is collected as part of the National Compensation Survey (NCS). There are two stages to the sampling procedure for the NCS. The U.S. Bureau of Labor Statistics (BLS) Handbook of Methods describes the initial stage that determines which establishments are selected for the NCS sample.³ Of primary interest for the focus of this article is the second stage, which determines the jobs within the selected establishments for which the NCS will collect compensation data.

Typically, a BLS field economist begins the second stage of the sampling by obtaining a list of an establishment’s employees. From this list, a small number of names are randomly chosen, usually four, six, or eight depending on the size of the establishment. Then, for each of the names selected, the chosen employee is grouped with other employees from the establishment who hold the same job. This job becomes a unit of observation in the NCS sample. For example, suppose a retail store has been chosen to be in the NCS sample, and John Smith is one of the four workers randomly chosen from the employee list. John Smith is classified as a cashier, and there are two other cashiers at the company. These three cashiers collectively will form a unit of observation in the NCS, and the average compensation data of these three workers will contribute to the calculation of the ECI.⁴

The NCS program has developed a detailed procedure for how workers should be grouped into jobs. The BLS field economist uses the most narrowly defined job level recognized by the establishment when grouping the workers into jobs. The duties and responsibilities of the job should be defined broadly. For example, if the selected job is a machine operator, no distinction should be made between the types of machines operated unless the job description specifies a particular machine. However, workers who differ in full-time or part-time status, union or nonunion status, or time-paid or incentive-paid status are not to be classified as being in the same job. Continuing the example above, if John Smith is a full-time cashier, the other two cashiers will be considered to have the same job as him only if they also work full time. Also, longevity steps are not considered when defining the job unless the job duties differ between steps.

The BLS field economist needs to make sure the job will be identifiable from one collection period to the next. Index numbers require that a piece of information––often the price––is collected for the same unit in consecutive periods. Therefore, the unit of observation for the ECI must exist for at least two consecutive reference periods that are 3 months apart. Jobs from private establishments selected for the NCS are typically scheduled to remain in the ECI production sample for about 3 years, so there must be a reasonable expectation that the unit will remain viable for that length of time. An individual worker might quit, be promoted, or otherwise change job duties, so the labor services that the worker provides would not be comparable from one period to the next. Therefore, the NCS uses the job as its unit of observation instead of a specific worker.

Data from the NCS provide an opportunity for a better understanding of a job as a unit of observation and its potential effect on the ECI. Although the ECI relies only on the average compensation among workers in the job for its calculation, the BLS field economist attempts to collect the individual wage rates for all workers in the job. Thus, the unit of observation provides information about the number of workers in a job at a single point in time and any change in the number of workers in a job over time.

For the 78 quarters from March 2006 to June 2025, the average number of workers in a job ranges from 30 workers per job to nearly 60 workers per job. However, a few of the jobs in the sample have an extremely large number of workers, which drives the average up and causes it to fluctuate from quarter to quarter. Most jobs have a much smaller number of workers than the average. Nearly a quarter of the jobs have only 1 worker, and the median number of workers in a job is consistently 5 workers. The 75th percentile for the number of workers in a job is typically close to 20 workers, and the 90th percentile ranges from 55 to 75 workers, approximately.

Changes in jobs over time

The estimation of an index number depends on the ability to collect the price for the same good over time. The goal of ECI data collection is to record the employer’s compensation costs for the same labor input in periods 3 months apart. A change in the number of employees indicates that the job has changed at least along this one dimension.

Chart 1 shows the percentage of jobs with no change in employment, with an increase in employment, and with a decrease in employment. The three percentages are calculated for all quarters from March 2006 to June 2025, but chart 1 shows a four-quarter moving average of the percentages starting with December 2006. For example, the value for December 2006 equals the average of the percentages for March 2006, June 2006, September 2006, and December 2006. This strategy will be used throughout the article to smooth out seasonality and other short-term fluctuations in the ECI to identify longer term trends.

The percentage of jobs with the same employment in the current and previous periods stays within a relatively narrow band from just above 60 percent to just below 50 percent. The percentage does show a downward trend since 2010, before turning back upward in 2023. (See chart 1.)

Jobs with an employment decrease consistently outnumber jobs with an employment increase, usually from about 2 to 7 percentage points. The percentage of jobs with an employment decrease range from approximately 21 percent to 27 percent, while the percentage with an employment increase range from approximately 17 percent to 24 percent. The largest differences occur in early 2009 and late 2020. The first period corresponds to the recession of the late 2000s, while the second period corresponds to shutdowns during the COVID-19 pandemic.

To users of the ECI, the important issue is whether the change in employment for a job systematically affects the ECI as a measure of compensation growth. Chart 2 shows the 12-month percent growth in wages from December 2006 to June 2025 for groups defined by the job’s employment in the current period relative to its employment in the prior period.⁵ As with the previous chart, the 12-month changes smooth out some of the variability of the 3-month changes. The overall series in chart 2 equals an unrounded version of the published ECI for wages and salaries for all private workers.

Chart 2 includes two vertical lines to designate three periods. The first period runs through the end of 2019. The analysis of the results will focus initially on this period of relative stability, when there was a consistent ordering to the 12-month wage growth among the categories of employment change. Analysis later in the article will focus on years since 2019, particularly on the second period, from March 2020 to June 2022, when the pattern of wage growth showed the greatest differences from the earlier period.

From 2006 to 2019, wage growth among jobs with the same number of workers in both the current and prior period (no change) closely followed the pattern in wage growth overall. In contrast, wage growth among jobs with an employment decrease was always higher than wage growth among jobs with no change, while the wage growth among jobs with an employment increase was always lower. The average annual wage growth rate from December 2006 to December 2019 equaled 4.6 percent among jobs with an employment decrease. In comparison, the average annual wage growth rate among jobs with an employment increase was just 1.1 percent. For reference, the average annual wage growth rate over the same period equaled 2.2 percent for jobs with constant employment, 2.4 percent for jobs with imputed wage data (labeled as "missing" in chart 2), and 2.4 percent for jobs overall.

A change in employment for the job does not affect the weight that the job receives in the ECI. The weight for a job is determined when the establishment first enters the ECI sample. The ECI holds this weight constant for all subsequent periods, even if employment for the job increases or decreases. All that matters is the change in the job’s average wage rate. For example, suppose a job initially has 3 workers and its weight in the ECI equals 4,000, so it represents 4,000 workers in the population. If employment for the job increases to 4 workers in the next period, then the average wage rate will now be calculated among 4 workers rather than 3. However, the weight will remain the same at 4,000. The new employment value has no effect on the ECI calculation other than in the numerator and denominator in the calculation of the average wage rate. The only additional use for the new employment value is to sort the jobs into the categories of employment change.

The results in chart 2 from 2006 to 2019 suggest a consistent composition effect. When jobs add workers, the new workers earn less on average than the incumbent workers. When jobs lose workers, the exiting workers earn less on average than the workers who remain in the job. This composition effect leads to lower wage growth for jobs with an increase in their number of workers and higher wage growth for jobs with a decrease in their number of workers.

Chart 3 shows the percentage of jobs with the same average wage rate in both the current and previous periods by the type of employment change. Like the previous charts, chart 3 shows a four-quarter moving average of the percentages starting with December 2006.

Among jobs with the same employment in both periods, at least 60 percent also had the same average wage in both periods. And, for some periods, over 70 percent of jobs with the same employment had the same average wage. In contrast, only about 10 percent of jobs with either an employment increase or decrease had the same average wage rate in both periods. For all types of employment changes, the percentage with the same average wage rate, in both periods, dipped slightly in 2007 and 2008, and the percentage dipped again after 2020. These periods coincide with generally high wage growth.

Chart 4 shows the percentage of jobs with a decrease in the average wage rate between the current and previous period: that is, the job’s average wage for the current period is lower than its average wage 3 months prior.

 

Less than 10 percent of jobs with the same employment showed a decrease in the average wage rate, whereas about 25 to 30 percent of jobs with an employment decrease showed a decrease in the average wage rate. And approximately 35 to 50 percent of jobs with an employment increase showed a decrease in the average wage rate.

The same employment in both the current and prior periods for a job does not necessarily mean that the two average wage rates refer to an identical group of workers in chart 2, chart 3, and chart 4. For example, the job may have had one worker in the prior period and one worker in the current period, but the new worker could have replaced the former worker during the intervening 3 months. The BLS field economist does not attempt to track individual workers. Nonetheless, the results suggest a lesser composition effect for jobs with no change in employment relative to jobs with either an employment increase or an employment decrease.

ECI wage growth from 2020 to 2022

The ordering of the 12-month wage growth by type of employment change was consistent from 2006 to 2019. Jobs with a decrease in employment showed the highest wage growth, while jobs with an increase in employment showed the lowest wage growth. The pattern was less consistent during 2020, 2021, and 2022. Table 1 shows the 3-month rates of change for wages for the quarters from March 2020 to December 2022 by type of employment change. The 3-month changes are more variable than the 12-month changes, but the 3-month changes help to pinpoint deviations from the consistent pattern by employment-change type that was previously seen. Jobs with a decrease in employment continued to show relatively high rates of wage growth, which is consistent with the pattern in the ECI before 2020. The break in the pattern occurred most notably for the jobs with no employment change in 2020 and for jobs with an employment increase from 2021 to 2022.

Table 1. Type of employment change, 3-month change for 2020, 2021, and 2022 (in percent)

Date	Employment decrease	No change	Employment increase	Missing	Overall
Mar 2020	2.2	0.9	1.9	1.0	1.2
Jun 2020	1.9	-0.5	0.5	0.5	0.4
Sep 2020	1.0	0.4	0.3	0.5	0.5
Dec 2020	1.0	0.5	0.7	0.7	0.7
Mar 2021	1.7	1.3	2.9	1.0	1.4
Jun 2021	1.1	0.8	-0.1	1.3	0.9
Sep 2021	1.8	1.9	1.0	1.3	1.5
Dec 2021	1.4	0.3	1.4	1.3	1.0
Mar 2022	1.3	1.3	1.0	1.7	1.4
Jun 2022	1.7	1.9	1.3	1.4	1.6
Sep 2022	1.4	0.9	0.2	1.4	1.1
Dec 2022	1.2	1.0	0.6	0.9	0.9
Source: U.S. Bureau of Labor Statistics.

The most noticeable deviation is during 2020 for jobs with no employment change. The 12-month growth rate for wages was 1.4 percent for December 2020 among jobs with the same employment, compared with a growth rate of 2.8 percent for jobs overall. The 12-month change for December reflects an aggregation of the four 3-month changes from March 2020 to December 2020. The 3-month wage change for June 2020 was particularly low at –0.5 percent. For June 2020, the percentage of jobs with no wage change and the percentage of jobs with a wage decrease were similar to the percentages of jobs for the periods prior to the COVID-19 pandemic. However, there was a particularly large negative impact on the ECI when the average wage rate decreased for jobs with the same employment from March to June 2020. Documentation from BLS field economists for the most impactful decreases mentions lower incentive payments and wage reductions because of the COVID-19 pandemic.

The other aberration from the pattern up until 2019 was the relatively strong wage growth among jobs with an increase in employment from 2020 to the first half of 2022. As shown in chart 4, the percentage of jobs with a wage decrease and an employment increase dipped to about 35 percent during this period; it had previously ranged from 40 to 50 percent. Fewer wage decreases will lead to a higher average wage increase.

Also contributing to the high and variable wage growth among jobs with an employment increase during this period was the estimate of the 3-month change for March 2021 of 2.9 percent. For March 2021, the published ECI estimate for the 3-month change in wages and salaries among private industry workers in the finance and insurance industries was also extremely high at 6.1 percent.⁶ Employment increased for several of the most impactful observations from the financial sector. These observations had a substantial impact on the estimate for jobs with an employment increase. A few of these observations again showed employment increases in June 2021. However, their incentive payments in June were reduced from their incentive payments in March. This led, in part, to the low estimate of the 3-month change among jobs with an employment increase in June of –0.1 percent.

Implications for the ECI

In a Monthly Labor Review article, John Ruser discusses the possible effects of the business cycle on the ECI.⁷ Ruser postulates that less experienced and lower paid workers are more likely to exit jobs during economic downturns, which would add a countercyclical component to the ECI. The higher growth rate for wages among jobs with a decrease in employment aligns with this scenario. (See chart 1.) However, the greater likelihood of less experienced and lower paid workers to exit jobs appears to occur during most periods since 2006 and not just during economic downturns.

Some degree of change to the labor input used for the ECI is inevitable because of the continual movement of workers in and out of jobs in the labor market. One cannot always hold the labor input constant between periods in strict accordance with the Laspeyres formulation. The NCS procedure for selecting and updating the average wages for jobs gives BLS field economists the flexibility to adapt the unit of observation to the structure of the establishment’s workforce, and this flexibility maximizes the amount of data that can be collected and the number of workers that can be kept within the scope of the ECI. However, as this article shows, this flexibility also leads to variation in the number of workers in jobs across the establishments in the sample and leads to changes in the number of workers within a job over time. These features are important for users of the ECI to be aware of.

During most of the years since 2006, the low growth rate in wages among jobs with an employment increase has been largely offset by the high growth rate in wages among jobs with an employment decrease. So, the overall ECI growth rate for wages and salaries generally aligns with the growth rate among jobs with no change in the number of employees. These are, presumably, the jobs least affected by the change in the composition of their employees. Therefore, these results do not indicate a substantial distortion in the ECI growth rate for wages and salaries because of changes in the number of workers in jobs in the ECI sample.

The main exceptions to the general pattern were during the COVID-19 pandemic and its immediate aftermath. During 2021 and 2022, wage growth for jobs with all types of employment changes was high. The typical ordering of wage growth by employment type was changed during this period.

Summary

This article showed how the Employment Cost Index (ECI) for wages and salaries is based on a comparison of the average wage rates for the same set of jobs across a 3-month interval. The ECI uses a Laspeyres index to calculate changes in compensation, and the growth rate of the ECI is determined by the 3-month change in compensation for a matched sample of jobs. This article documented that, from 2006 to 2019, increases and decreases in employment in particular jobs coincided with slower or faster wage growth in these jobs. But this pattern was less consistent during 2020, 2021, and 2022. In particular, there was strong wage growth among jobs with increases in employment from 2020 to the first half of 2022; and jobs with no employment change had 12-month wage growth that was 1.4 percent for December 2020, while jobs overall had 2.8-percent wage growth. That is, the general pattern observable in the ECI between wage growth and changes in the labor market was less pronounced during the COVID-19 pandemic and its immediate aftermath. 

Appendix

This appendix will review the Employment Cost Index (ECI) formula for wages and salaries. The ECI rates of change by type of employment change are calculated with the same method used to calculate the published ECI estimates for regional, union and nonunion, and excluding-incentive workers indexes. The description of the method in this appendix builds on formulas for the calculation of the ECI from the U.S. Bureau of Labor Statistics (BLS) Handbook of Methods.⁸

In the notation from the section on calculation of the National Compensation Measures from the BLS Handbook of Methods, the wage bill for private industry occupational cell i is defined as follows:

 $W_{0,i}$ is the estimated wage bill for the cell i in the base period 0.

$M_{t,i}$ is the multiplicatively-accumulated weighted average wage change in the cell i from period 0 to period t. Therefore, $W_{0,i}{M}_{t,i}$ is the estimated wage bill for cell i in the quarter t.

ECI estimates defined by ownership, industry, or occupational groups are calculated using aggregations of these wage bills. However, for indexes not defined by ownership, industry, or occupation, such as the published ECI indexes for regional, union and nonunion, and excluding-incentive workers, the ECI uses the distribution in the sample for the characteristic to apportion the wage bill for the cell. The same method is used in this article to calculate indexes by type of employment change, and calculating these indexes requires the following additional estimates:        

${\bar{S}}_{t-1,i}^c$ is the weighted share of the sampled quotes with characteristic c for cell i in period t – 1 using the matched sample between period t – 1 and period t.

${\bar{Y}}_{t-1,i}$ is the weighted average wage of the sampled quotes for cell i in period t – 1 using the matched sample between period t – 1 and period t.

${\bar{Y}}_{t-1,i}^c$ is the weighted average wage of the sampled quotes with characteristic c for cell i in period t – 1 using the matched sample between period t – 1 and period t.

Using these definitions, one can write this equation $\left({\mathrm{W}}_{0,i}{M}_{t-1,i}\right)\left[\left({\bar{S}}_{t-1,i}^c\right)\left(\frac{{\bar{Y}}_{t-1,i}^c}{{\bar{Y}}_{t-1,i}}\right)\right]$ that equals the estimated wage bill for characteristic c for cell i in the period t – 1. The wage bill for an ownership-industry-occupation cell is determined by its employment and the level of its wage rate, so the amount of a cell’s wage bill apportioned to the characteristic is determined by its share of employment for the cell, which is measured by ${\bar{S}}_{t-1,i}^c$ and the characteristic’s average wage rate relative to the overall average wage rate for the cell.

Updating the wage bill for period t – 1 to period t requires the following additional estimate:

${\bar{Y}}_{t,i}^c$ is the weighted average wage of the sampled quotes with characteristic c for cell i in period t using the matched sample between period t – 1 and period t.

Thus, $\left({\mathrm{W}}_{0,i}{\mathrm{M}}_{t-1,i}\right)\left[\left({\bar{S}}_{t-1,i}^c\right)\left(\frac{{\bar{Y}}_{t-1,i}^c}{{\bar{Y}}_{t-1,i}}\right)\left(\frac{{\bar{Y}}_{t,i}^c}{{\bar{Y}}_{t-1,i}^c}\right)\right]$$\left({\mathrm{W}}_{0,i}{\mathrm{M}}_{t-1,i}\right)\left[\left({\bar{S}}_{t-1,i}^c\right)\left(\frac{{\bar{Y}}_{t-1,i}^c}{{\bar{Y}}_{t-1,i}}\right)\left(\frac{{\bar{Y}}_{t,i}^c}{{\bar{Y}}_{t-1,i}^c}\right)\right]$$\left({\mathrm{W}}_{0,i}{\mathrm{M}}_{t-1,i}\right)\left[\left({\bar{S}}_{t-1,i}^c\right)\left(\frac{{\bar{Y}}_{t-1,i}^c}{{\bar{Y}}_{t-1,i}}\right)\left(\frac{{\bar{Y}}_{t,i}^c}{{\bar{Y}}_{t-1,i}^c}\right)\right]$equals the estimated wage bill for characteristic c for cell i in period t. The above equation can be simplified to $\left(W_{0,i}{M}_{t-1,i}\right)\left[\left({\bar{S}}_{t-1,i}^c\right)\left(\frac{{\bar{Y}}_{t,i}^c}{{\bar{Y}}_{t-1,i}}\right)\right]$.

The 3-month rate of change in the index for characteristic c then equals the sum of the estimated wage bills for characteristic c across the 531 private cells for period t divided by the sum of the estimated wage bills for characteristic c across the 531 private cells for period t – 1. The 12-month rate of change equals the product of the four 3-month rates of change over the 12-month period.