Removing workers’ compensation costs from the National Compensation Survey

Workers’ compensation (WC) is one of 18 employer-provided benefits for which information is collected by the U.S. Bureau of Labor Statistics (BLS) under the National Compensation Survey (NCS) program. Workers’ compensation is a key part of social insurance in the United States, providing medical care, rehabilitation, and cash benefits for workers who are injured on the job or who contract work-related illnesses and is mandatory in most states.¹ While this fact implies that almost all employers have positive expenditures for workers’ compensation, the costs are relatively small, particularly as a share of total compensation. As of December 2022, the average cost to employers for workers’ compensation was $0.46 per employee hour worked across all civilian workers, as measured in BLS’s Employer Costs for Employee Compensation (ECEC), about 1 percent of total compensation. Workers’ compensation’s share of total benefits was about 3 percent, and its share of legally required benefits (which also include employer costs for Social Security, Medicare, federal unemployment insurance, and state unemployment insurance) was roughly 15 percent. By comparison, the benefit with the highest cost, health insurance, averages $3.02 per hour, making up 25 percent of the cost of total benefits and about 8 percent of total compensation.

After a 2011 report by the National Academy of Social Insurance (NASI),² and because of the ongoing efforts of BLS to maintain the quality of all its data, some questions have been raised in recent years about the quality of the WC data collected as part of the NCS. After providing some background on workers’ compensation, this article reports on these questions, along with the research and other efforts undertaken by the NCS program to address them. Unfortunately, the issues raised turned out to be intractable, as potential solutions proved not to be feasible within the context of respondent burden and the resources available to the NCS. As a result, after a long period of deliberation, the NCS program is proposing that the collection and publication of WC costs should be ceased and will work with the U.S. Office of Management and Budget on this effort. The remaining sections of this article present an analysis on the effects of removing workers’ compensation on the NCS program. The analysis is in three parts: first on the ECEC; then on the Employment Cost Index (ECI); and then on the linked ECI, which is a variant of the ECI being evaluated for publication.³ For the ECEC, the analysis suggests dropping WC costs will have very little impact on compensation estimates, some effect on total benefit estimates, and an unavoidably large impact on estimates of legally required benefits. For the ECI and the linked ECI, growth rates for published series were virtually unchanged after the exclusion of WC costs. Finally, removing WC costs will not have any appreciable effect on the precision of estimates produced by the NCS program.

Issues in collecting and estimating workers’ compensation costs

There is wide variation across states in how WC programs are administered because each state regulates its own WC program, with no federal agency mandating standard reporting requirements.⁴ Workers’ compensation programs also vary across states in terms of which injuries or illnesses are compensable and the level of benefits. Workers’ compensation is financed almost exclusively by employers. Employers either self-insure or purchase insurance. Generally, state laws require employers who wish to self-insure to obtain approval from the state regulatory authority after demonstrating financial ability to carry their own risk. For those employers who purchase insurance, the premiums are based in part on industry and the occupation classifications of their workers. Many employers are also experience rated, which results in higher (or lower) premiums for employers whose past experience—as evaluated by actuarial formulas that consider injury frequency and aggregate benefit payments—is worse (or better) than the experience of similar employers in the same insurance classification.

As will become apparent, these institutional features complicate the task of collecting reliable WC data. Besides the NCS, a second source of WC data is the National Academy of Social Insurance (NASI), which publishes annual reports showing the benefits, costs, and coverage of workers’ compensation by state and for the nation.⁵ As with all other benefits in the NCS, BLS collects workers’ compensation such that it can calculate the employer cost per employee hour worked. This approach contrasts with that of NASI, which computes employer costs per $100 insured payroll when standardizing its cost estimates. Converting the NCS numbers to costs per $100 insured payroll, a NASI report in 2011 found that BLS cost estimates were usually higher over the preceding three decades and that, while general trends were similar, there were clear differences in peak and trough years. A 2015 BLS report examined nonfederal WC costs per $100 payroll for both NASI and NCS and found that, over the 2004–13 period, the two series largely moved in tandem but that there was a large level difference.⁶ The report then took on the difficult task of trying to reconcile the two series, examining both the numerator—WC costs—and the denominator—payroll—for clues as to why NCS estimates were systematically higher than those from NASI. For the denominator, it was suspected that differences in the components included as part of payroll were making the NASI payroll numbers higher than the NCS payroll numbers, leading to a lowering of NASI WC costs per payroll relative to NCS WC costs per payroll.

The numerator (the actual WC costs themselves) is affected by the different approaches NCS and NASI take to collect the information. The NCS program uses an establishment survey. How the NCS collects data on WC costs from each establishment largely depends on how an establishment chooses to finance the benefit. As noted earlier, an employer will typically finance workers’ compensation by purchasing insurance (usually through a private carrier, but sometimes through a state fund), self-insuring by assuming liability, or some mix of both. When an employer pays for WC costs by purchasing insurance, the data on employer costs are obtained by collecting the rate for the sampled occupations along with an experience modifier and premium discount.⁷ For self-insured firms, WC costs are usually obtained by collecting the employer’s total expenditures including items such as medical claims, indemnity claims, and administrative costs. Regardless of the type of data available, every effort is made to try to collect data on employer costs that are directly related to each occupation sampled because tracking at the job level is important for computing the ECI and the ECEC.

In contrast to the survey approach of NCS, NASI’s data for workers’ compensation come from a variety of different sources, depending upon the type of insurance being considered and the state. For those who do not self-insure, the premium data are provided by either a state agency or AM Best, a credit-rating agency that focuses on the insurance industry. In addition to the premiums, NASI needs to estimate the costs of deductibles employers pay directly.

For the self-insured, NASI estimates the costs of WC benefits plus administrative costs. The costs of benefits are reported directly by some states, and in other states are estimated based on the percentage of payroll covered by the self-insured. The administrative costs are estimated at the same percentage as in private insurance. These administrative costs include the direct costs of managing claims as well as litigation, cost containment, taxes, licenses, and fees.

When the NCS and NASI numerators (the WC costs themselves) were compared by the workers' compensation cost team for 2004–13 in the BLS report, the NCS costs were considerably higher. Although the series generally followed a similar time trend, business cycle movements seemed more pronounced in the NCS data. When comparing WC costs per $100 of payroll, NCS numbers were 36 percent higher per year averaged over the period. As noted, some of this difference is due to higher NASI payroll numbers. When comparing NASI WC costs (excluding federal) with NCS WC costs (i.e., without dividing by payroll), NCS costs were 20 percent higher per year, on average, over the period. The size of the difference decreased over the period, from a high of NCS being 26 percent larger than NASI in 2009 to NCS being 9 percent larger than NASI in 2013, the most recent year examined in the report. While NCS and NASI numbers might be closer when comparing only the WC costs themselves (the numerator), the report concluded that the remaining differences required further exploration of NCS data and procedures.

The BLS report noted some key features of NCS data collection but was unable to uncover which of them, if any, might be responsible for the discrepancy in costs. As noted, a key goal of the NCS is to collect and report on the compensation costs of employees at the occupation level. In the pursuit of this goal, the ideal method of data collection involves gathering the actual cost of each compensation component for each sampled occupation at the establishments selected to be in the survey. Capturing the cost of workers’ compensation by occupation is straightforward when a firm finances workers’ compensation via insurance through a private carrier. In these cases, the NCS collects the rate for the sampled occupations along with an experience modifier and premium discount charged by the private insurance carrier for each occupation class code. This type of information collection is known in NCS terminology as “rate-and-usage”.

When an employer self-insures and assumes the risk of WC costs, the NCS is usually able to collect WC data only in the form of an establishment expenditure. The expenditures are an aggregate of an establishment’s total costs associated with all WC claims and not a detailed account of the costs for each occupation. The respondent is unlikely to provide the breakout of those costs by occupation. The NCS has little control over whether occupation-specific costs can be collected for workers’ compensation, since the collection method is driven by how establishments finance WC costs. Although collecting expenditures is more straightforward and closely related to NASI’s collection methods, it eliminates the occupational variation in costs. This variation is especially large for workers’ compensation in comparison with other benefits collected by the NCS.⁸

Prompted, in part, by the investigation into NCS–NASI differences, the NCS program undertook an intensive examination of WC collection. Item nonresponse, which results when a respondent refuses or is unable to answer a particular question, is a common problem in sample surveys such as the NCS. Missing responses have always been an issue for WC collection, but the situation has worsened over the last two decades. In 2004, there were missing WC costs for 42 percent of the establishments that responded to the survey. This share rose to 70 percent in 2014 and was 84 percent by 2023.⁹ There is no easy way to reverse this trend, given the rise in item nonresponse for other benefits as well.¹⁰ Pushing respondents harder for a response to workers’ compensation, a relatively small benefit, would risk worsening cooperation for other benefits that make up a larger share of compensation costs.¹¹

Besides underscoring the item nonresponse issue, the examination discovered systematic differences between rate-and-usage and expenditure, the two main methods of collection for the NCS. For each March reference month, from 2005 to 2014, the rate-and-usage estimates are 12 to 38 percent higher than the expenditure estimates. If the two methods are estimating similar outlays for similar workers, this discrepancy should not be present. The gap between the two measures seems to result because the rate-and-usage method targets the specific occupations that were sampled in the NCS, while the expenditure estimates typically use an employer-wide average for each of the selected occupations. In addition, construction, production, and other occupations with higher risks and higher premiums tend to be overrepresented in the rate-and-usage data. Although the larger sample size of rate-and-usage data for this group of occupations is useful to the NCS, the higher premiums of the occupations tend to raise costs in the rate-and-usage sample relative to the expenditure sample. Occupation-specific costs are not collected for nearly half of the sample.

Efforts to address the issues affecting workers’ compensation

In light of these issues—low response rates for workers’ compensation and the collection of data at the employer level rather than the occupation level—much consideration was given to finding possible solutions. The most straightforward solutions—redoubling efforts to collect WC data and attempting to shift collection toward rate-and-usage data and away from expenditure data—are, unfortunately, not practical. Increasing respondent burden for workers’ compensation would potentially lower response rates for other benefits that make up a larger share of compensation costs. Moreover, when employers self-insure, it is usually not possible to collect rate-and-usage data because these employers do not have the rate sheets to provide the required information.

To address the limitations of expenditure data, NCS investigated whether an additional step could be taken to make adjustments across occupations within a given employer. For example, consider the example of a logging company that has three different occupations: fallers, environmental engineers, and administrative assistants. Each of these occupations has a different risk of injury, so one would expect WC costs to differ by occupation accordingly. Yet, if the company self-insures or simply reports expenditures for the full company, current NCS procedures will lead to an apportionment of these costs that ignores the differing risks of injury. In theory, however, one can change the procedures so that costs are allocated in proportion to risks, but one needs a measure of that risk. Several possibilities were investigated including total cases and median days away from work by occupation from the Survey of Occupational Injuries and Illnesses, state-level WC claims by occupation (which are publicly available from only four states), and state-level loss cost rates by classification code from the National Council on Compensation Insurance (NCCI), which is the WC ratings bureau for the majority of states.¹²

Each of the aforementioned approaches has limitations. For instance, total cases and claims measure the frequency of injuries but not their severity, while median days away from work is an index of severity but not frequency. State-level cost rates account for both frequency and severity, and, partly as a result, an approach using them was considered the most promising way to adjust expenditures. This method also has limitations, however, including the need for a complex mapping between classification codes and Standard Occupational Classification (SOC) codes,¹³ the assumption that the loss-cost rates are appropriate for employers who reported expenditure data, and more.¹⁴

The final approach investigated to improve the quality of WC data was to use an alternative data source to either substitute for or augment collected values. The alternative source considered was the National Council on Compensation Insurance. Using data from this source requires a detailed mapping between SOC codes and the classification codes in the NCCI. A key limitation of this approach is that these data, which are most appropriate for employers who do not self-insure, represent averages across employers. They do not take into account such factors as premium discounts and experience modifiers. The NCS program conducted an experiment in which selected classification codes from the NCCI were mapped to SOCs and the data from NCCI compared to collected NCS data. On average, the NCCI imputations were reasonably close, but not close enough to meet the standards of the National Compensation Survey.¹⁵

In the end, therefore, it was not possible to arrive at a feasible solution that would halt and reverse the gradual erosion of the quality of the WC data. With respondents seemingly less and less willing to spend time on voluntary surveys, efforts to improve item nonresponse rates for WC costs risk reducing response rates for other benefits. Finding alternative data sources also did not prove fruitful. Thus, though a difficult decision, the NCS program decided to plan to remove WC costs from the NCS. The rest of this article is devoted to showing the effects this removal will have.

Employer Costs for Employee Compensation

As mentioned earlier, workers’ compensation is one of the 18 benefits collected in the NCS. Besides its appearance in the ECEC series for workers’ compensation itself, workers’ compensation is also included in the following three aggregates: legally required benefits (which also includes employer costs for Social Security, Medicare, federal unemployment insurance and state unemployment insurance), total benefits, and total compensation (which is the sum of total benefits plus wages and salaries). To provide background, chart 1 shows ECEC estimates over time (December 2013 through September 2022) in the averages for civilian WC costs and the three aggregates, while chart 2 displays the ratio of WC costs to the three aggregates over the same time. Chart 1 shows that, in nominal terms, WC costs have been flat, legally required benefits have gone up slightly, total benefits have risen some, and total compensation has risen more quickly. These patterns imply, as shown by the shares in chart 2, that WC costs have lost ground to the three aggregates, particularly to compensation.

For each of the three aggregates, there are 331 series published each quarter, with each series being a combination of ownership, industry, occupation, and other employer or worker characteristics. These combinations have a total of four dimensions because if other employer characteristics are specified, then worker characteristics are not, and vice versa.¹⁶ The other employer characteristics are establishment size class and area, while the other worker characteristics are collective bargaining status, full-/part-time status, and basis of pay. Repository tables 1–3 under the source-data heading display each of these dimensions and their subcategories. It is not generally the case that the dimensions are broken into mutually exclusive and exhaustive categories. For instance, industries and occupations are generally at the sector level, but there are some categories with more detail. For areas, both the four Census regions and the nine Census divisions within these regions are included. The series are chosen on the basis of what is of interest and what meets statistical standards and, thus, it is not the case that all combinations of each dimension are published.

For the analysis that follows, we examine the three aggregates containing workers’ compensation for 37 periods, which cover the span from December 2013 through December 2022.¹⁷ As a result, there are a total of 36,741 estimates for the three aggregates.¹⁸

Where are workers’ compensation costs the largest?

Will it make an important difference if workers’ compensation is excluded from the ECEC and, if so, to which estimates? As a first step to addressing these questions, for each of the dimensions noted above, we calculate the average of WC costs in each category of the dimension over all the estimates in which that category appears.

The dimension of ownership is shown in table 1. There is a total of 12,247 WC estimates, and, of these, 10,730 are for private ownership, 925 for civilian, and 592 for state and local government. The final column of the table suggests that WC cost estimates vary little by ownership, with the mean cost per hour worked for civilian workers being $0.51, for state and local government being $0.49, and that for private industry being $0.47.¹⁹

The dimension of industry is shown in table 2. There is considerable variation along the industry dimension. Estimates for construction, transportation and warehousing, utilities, public administration, and goods-producing industries are considerably higher than the average of $0.53 for estimates when all industries are included. These are industries in which one might anticipate the risk of injury being higher than average. Most remaining industries have costs that are below average, some substantially so.

The dimension of occupations is similar in that a handful of occupations have very high WC costs while most occupations are below average. (See table 3.) The higher cost occupational groups are led by farming, fishing, forestry, construction, and extraction occupations; natural resources, construction, and maintenance occupations; and transportation and material moving occupations.

Turning to employer characteristics, there tend not to be wide differences in cost estimates by the number of employees, with the gap between the smallest and largest establishment size class being about $0.09. (See table 4.) There is a wider gap by Census division, with the average cost for estimates for the Pacific ($0.68) exceeding that for those for the East south central ($0.33) by about $0.35. The difference between union and nonunion series is about $0.42, attributable to the fact that union jobs are more likely to be in industries or occupations with higher costs, while the difference for full time versus part time is nearly $0.16.

The preceding analysis revealed when WC costs are large in an absolute sense, but it is also useful to see when they are large in a relative sense. Table 5 displays statistics on WC cost as a percentage of total compensation, total benefits, and legally required benefits. As a share of total compensation, WC costs average about 1 percent, so for most series, its exclusion will not make a profound difference. For benefits, the average share of workers’ compensation is a more noticeable 5 percent, reaching as high as 25 percent, and as low as 1 percent. For legally required benefits, the average share is 17 percent, and it exceeds 22 percent in one-fourth of the cases.

Does removal of workers’ compensation costs lead to “large” differences?

To answer the question heading this subsection, we define a large difference as whether the removal of WC costs changes the estimate by an amount at least equivalent to what would be a statistically significant change at the 10-percent level.²⁰ In table 6, the three panels show by ownership and for each type of estimate—total compensation, total benefits, and legally required benefits—the share of the estimates in which the exclusion of workers’ compensation leads to a change of this size. As shown in the first panel, none of the compensation estimates for state and local government are changed significantly when WC costs are excluded, while 1 percent of the private estimates and 5 percent of the civilian estimates are. For benefits, more than 40 percent of the estimates are statistically different for civilian and state and local versus 30 percent for private. For legally required benefits, virtually all estimates, irrespective of ownership, are statistically different.

This pattern—in which few of the compensation estimates are materially affected, virtually all of the legally required benefit ones are, and the total benefits fall somewhere in the middle—holds for the other dimensions: industry, occupation, and employer and worker characteristics. (See tables 7 to 12.) What is of primary interest in these tables, therefore, is the middle panel, the one for benefits. By industry, public administration, construction, and retail trade stand out as having higher than average rates of large differences. By occupation, there is a large difference in the benefits estimate more than 28 percent of the time for all occupations, with the share being higher than 75 percent for categories such as farming, fishing, forestry, construction, and extraction occupations; natural resources, construction, and maintenance occupations; production, transportation, and material moving occupations; and production occupations.

Table 7. Share of estimates with “large” differences after exclusion of WC costs, by estimate type and industry group (ECEC)

Table 8. Share of estimates with “large” differences after exclusion of WC costs, by estimate type and occupational group (ECEC)

Table 9. Share of estimates with “large” differences after exclusion of WC costs, by estimate type and establishment size class (ECEC)

Table 10. Share of estimates with “large” differences after exclusion of WC costs, by estimate type and area (ECEC)

Table 11. Share of estimates with “large” differences after exclusion of WC costs, by estimate type and collective bargaining status (ECEC)

Table 12. Share of estimates with “large” differences after exclusion of WC costs, by estimate type and work status (ECEC)

Users of ECEC estimates should be aware that the removal of WC costs will constitute a change to the definition of any series of which it is a part. For compensation estimates, the impact will be minor and equivalent to random fluctuations inherent in the NCS sample. But for total benefits, and especially for legally required benefits, the effects will be more evident.

Does the removal of WC costs change ranks?

A final and complementary way to examine the impact of excluding WC costs is to use a tool known as correlation analysis. Correlation analysis addresses the question of whether the removal of WC costs affects which of the industries, occupations, or areas are considered high compensation. Let us say a person calculates the mean of all compensation estimates by industry category and then ranks these categories by average compensation. Then let us say one subtracts WC costs from compensation, recalculates the mean of all the compensation series by industry category, and then reranks the industry categories by average compensation minus WC costs. One can then calculate two correlation coefficients. The first will be a correlation of the actual mean compensation by industry category, with and without WC costs. The second will be a correlation of the ranks of all 32 industry levels, where 1 is the highest rank and 32 is the lowest. The basic idea behind calculating both correlation coefficients is the same: if WC costs are large and if there is much variation in them by industry, then their removal will lead to significant changes in values and many changes in ranks, and the correlation coefficients will be low. If, however, WC costs are small or if their removal leads to equal or proportionate changes by industry, then there will be only small changes in values and few changes in ranks, leading to high correlation coefficients.

The first panel of table 13 displays the result of this exercise by industry dimension and total compensation. An examination of the mean costs shows mainly small changes. In terms of ranks, most are the same, with the biggest switch being one position in either direction. As a result, the correlation coefficients are quite high, nearly the maximum of 1 for both, which is consistent with what we have seen so far (in that subtracting WC costs does not have a big impact on ECEC compensation estimates).²¹ For total benefits, the picture is similar. Most means do not change substantially and, except for construction, which falls three places once WC costs are subtracted, any shift in rank is limited to one position. Both correlation coefficients are again very close to the maximum of 1. In terms of legally required benefits, it is not surprising that the removal of WC costs makes more of a difference. Industries with relatively large WC costs, such as construction and transportation and warehousing, see their ranks fall by 10 or more positions. The correlation coefficients still suggest a fairly high degree of stability, though, at 0.93 for the means and 0.90 for the ranks.

Finally, for Census divisions and regions, there is also much stability in levels and ranking for total compensation and benefits and even for legally required benefits, given that WC costs generally vary less by geography than by industry or occupation. (See table 15.) The correlation coefficients are nearly one for the first two aggregates and 0.98 and 0.95 for legally required benefits.

Impact on standard errors for ECEC

Does the removal of WC costs affect the precision of estimates for total compensation, total benefits, and legally required benefits? To address this question, we calculated the standard errors for the three types of estimates with WC costs removed and compared these to the original standard errors for the period from the second quarter of 2018 through the third quarter of 2022, a span of 18 quarters. There are a total of 17,874 estimates with their corresponding standard errors.²² One way to make this comparison is to calculate basic statistics such as the mean for the two sets of standard errors, as well as to calculate a correlation coefficient. Of interest is not only the standard error but the relative standard error (RSE) as well.²³ Panel A of table 16 shows statistics for all the estimates. The original standard errors for the estimates average 0.63. This amount is close to the 0.62 standard error average for the estimates without WC costs. After normalizing for estimate size, the relative standard errors increase in size, but only by a small amount, from an average of 3.5 percent to an average of 3.6 percent. To make sure that the close averages are not hiding big differences elsewhere, we examine the correlation coefficient, which is nearly 1 for the standard errors and 0.99 for the RSEs.

To see if the same patterns hold (that is, little change in the standard errors) for each estimate type as for the three types of series combined, the panels present these same statistics for total compensation, total benefits, and legally required benefits, respectively. For total compensation, any difference in relative standard errors is hard to find, because the average for these is the same (3.3 percent), with or without WC costs. The two sets of standard errors and RSEs are also almost perfectly correlated with each other. For benefits, the two sets of standard errors and RSEs are almost perfectly correlated with each other as well, but there is a tiny difference in the average RSEs: 4.4 percent for the set of benefits that include workers’ compensation and 4.6 percent for the benefits that exclude workers’ compensation. Finally, looking at legally required benefits, we again see a small loss of precision—the average RSE goes from 2.7 percent to 2.8 percent—but the correlation coefficients, though somewhat lower at 0.97 and 0.98, still show a close alignment between the two sets of standard errors and RSEs.

Compensation percentile estimates

Every March, the NCS program publishes compensation percentile estimates (CPE) using the same microdata that produce the ECEC estimates. The CPE provide a sense of how compensation and its components are distributed, showing, at the 10th, 50th, and 90th percentiles of the compensation distribution, the mean values for selected compensation categories. As with the ECEC estimates, the removal of WC costs from the NCS program would affect CPE involving total compensation, total benefits, and legally required benefits. To see how much of an impact such a change would have, we calculate CPE with and without WC costs for 2020, 2021, and 2022 for the three different ownership categories (civilian, private industry, and state and local government). This calculation leads to a total of nine sets of estimates. For total compensation, total benefits, and legally required benefits, and for each of the three percentiles, we check whether the exclusion of WC costs changes the estimate by an amount at least equivalent to what would be a statistically significant change.²⁴ For compensation, the change was never significant. For benefits, the change was significant in four out of nine cases at the 10th percentile, three out of nine at the 50th percentile, but never at the 90th percentile. For legally required benefits, the change from excluding WC costs was always significant. This pattern is consistent with the rest of ECEC results, where compensation is little affected, legally required benefits usually have statistically significant differences, and total benefits fall somewhere in the middle. Because WC costs do not rise with percentile as much as many other forms of compensation, the net effect of removing them is to make the estimated distribution of compensation more equal, but only by a very small amount, given that WC costs are such a small portion of total compensation.

Employment Cost Index

In this section, we consider the impact of the removal of WC costs on ECI estimates. A total of 132 different estimates for total compensation and 14 different estimates for total benefits are published each quarter for the ECI, each one showing the percent change over the preceding quarter or the quarter one year ago. We also consider the 132 legally required benefit series defined in the same way as the compensation series. These series for legally required benefits are not published, so we have moved their detailed analysis to an appendix that is available upon request and to appendix tables that are available under source data. As with the ECEC series, we recompute all 278 series by subtracting out WC costs. Just like in the ECEC series, the removal of WC costs could make more of a difference if the level of workers’ compensation is high relative to the other items. In addition, the relative growth of such costs will play a role in whether the recomputed ECI series will be different.

Impact on 3-month percent changes

Because there are 37 periods of data, there are a total of 10,286 percent changes with WC costs and the same number without them.²⁵ One summary measure is the share of these percentages in which the difference between the two corresponding values is large enough that a movement equal to that difference would be statistically significant.²⁶ As shown in the first row of table 17, it turns out that in 441 out of 10,286 cases, or 4 percent of the time, the difference is of a size consistent with a statistically significant change. Not surprisingly, however, it matters which type of series is being considered. In no cases are differences of a statistically significant size noted in the compensation series or the benefit series. Thus, the 3-month percent changes are only affected, as measured by this metric, for the legally required benefits, and only in 9 percent of the cases.

Impact on 12-month percent changes

In this subsection, we consider 12-month percent changes in the index. While there are still 278 series each period, there are four fewer observation periods (only 33 instead of 37), because the series assessed begin with the change from December 2014 relative to December 2013, rather than the 3-month change from December 2013 relative to September 2013. Out of 9,174 estimates, in 1,040 of the cases, or 11 percent, the removal of workers’ compensation makes a difference whose value equals or exceeds that of a statistically significant difference.²⁷ (See table 18.) This frequency is higher than the 4 percent for 3-month percent changes. This increase can be explained, in part, by the longer time span for divergence to occur. Once again, though, none of the compensation estimates are in this category and only 2 out of 462 benefit estimates are. Some 24 percent of legally required benefit series have the equivalent of statistically significant differences.

Impact on index values

In this section, we turn to an examination of the index values themselves, rather than the percent changes. As standard errors are not calculated by the NCS program for the index values, to see whether there is an important difference when WC costs are subtracted out, we use a threshold of 1 percent. That is, we flag those cases in which the difference between the two index values is greater than or equal to the absolute value of 1 percent. Going back to the set of 10,286 estimates, 1,371, about 13 percent, achieve this threshold. (See table 19.) Once again, it is primarily legally required benefits that are affected. No compensation estimates differ by 1 percent or more in magnitude. Among the benefits, 18 out of 518 have such a difference, or about 3 percent. For the legally required benefits, it is 1,353 out of 4,884, or 28 percent.

Impact on standard errors of ECI

To assess any impact on the precision of the ECI estimates, we compare the standard errors, for both 3-month percent changes and 12-month percent changes, before and after the removal of WC costs from the index. The fact that the estimates of the growth rates themselves are sometimes close to zero causes relative standard errors to become erratic. Thus, the focus here is just on the actual standard errors. We calculate 10 quarters of standard errors for the 3-month percent changes and 6 quarters of standard errors for the 12-month percent changes.²⁸ For both timeframes, the removal of WC costs has only a minor impact on the average size of the standard errors. (See table 20.) The size of the correlation coefficients, moreover, is nearly 1 when aggregates other than legally required benefits are considered, suggesting close alignment between the two sets of standard error estimates. Thus, there would not be any notable change in precision in removing WC costs.

Linked Employment Cost Index

As discussed by Allamani, Aysheshim and Righter in 2022, BLS is exploring the publication of a new index of employment cost change, called the Linked Employment Cost Index.²⁹ This index is calculated with the use of a linking Laspeyres approach, and it has some advantages over the modified Laspeyres ECI already produced by BLS that we analyzed in the previous section.

Impact on 3-month percent changes

The analysis in this subsection is very similar to that in the subsection for 3-month percent changes for the ECI that is currently being published. To measure whether the removal of WC costs makes an important difference in ECI growth rates, we once again check to see whether the growth rates with and without WC costs differ by an amount that would make the difference statistically significant.³⁰ As standard errors are available for 9 years or 36 quarters, from the first quarter of 2014 to the fourth quarter of 2022, there are a total of 10,008 sets of estimates to examine.³¹ It is the case that the subtraction of WC costs makes a statistically significant difference in 367 cases, or about 4 percent. (See table 21.) Examining by type of estimate, we find that there are no cases where there is a significant difference for the total compensation and total benefit series; the 367 cases are all in legally required benefits and account for about 8 percent of those estimates.

Impact on 12-month percent changes

Again, we note the similarity of the analysis here to that in the section on 12-month percent changes for the current ECI. As in that subsection, there are a total of 9,174 estimates to consider.³² Of these, in 965 of the cases the removal of WC costs makes a significant difference to 12-month percent changes, which is somewhat fewer than the 1,043 cases for the current ECI, but still around 11 percent overall. (See table 22.) Once again, the only impact is on legally required benefit estimates, as only one total benefit series estimate is affected and no total compensation series estimate is.

Impact on standard errors of linked ECI

As with the current ECI, we are interested in what impact removing WC costs from the index has on the standard errors of linked ECI estimates, for both 3-month percent changes and 12-month percent changes. As with the current ECI, the estimates of the growth rates themselves are sometimes close to zero, which causes relative standard errors to become erratic. Thus, the focus here is just on the actual standard errors. We calculate 8 quarters of standard errors for the 3-month percent changes and 4 quarters of standard errors for the 12-month percent changes. For both timeframes, the removal of WC costs has only a minor impact on the average size of the standard errors. (See table 23.) The size of the correlation coefficients, moreover, is always above 0.9, and close to 1, when aggregates other than legally required benefits are considered. This suggests close alignment between the two sets of standard error estimates and, therefore, no notable loss of precision in removing WC costs.

Summary and conclusions

Prompted by a NASI report and the ongoing efforts by the NCS to maintain data quality, more attention has been paid to the collection of the legally required benefit workers’ compensation. Workers’ compensation suffers from a low item response rate of only about 16 percent, which is substantially lower than that for the other NCS benefits. The NCS is hindered in its ability to improve the currently low response rate for this benefit, given that such an effort could potentially lead to lower response rates for other benefits, which make up a larger share of total compensation. Other proposals to improve the quality of the collected data or to find alternative sources of data have limitations as well. Unfortunately, it was not possible to arrive at a feasible solution that would halt and reverse the gradual erosion of the quality of the workers’ compensation data in the NCS. As a result, BLS decided the best path forward is to remove workers’ compensation data from the NCS.

This article explored the impact that a cessation of collection of workers’ compensation would have on NCS products. For the ECEC, WC costs are part of the aggregates for total compensation, total benefits, and legally required benefits. The removal of workers’ compensation costs constitutes a change in the definition of these series. For compensation estimates, the impact will be marginal, but it will be more evident for total benefits and legally required benefits. For the ECI and the linked ECI, the compensation and benefit series are not generally affected in a statistically significant way. In terms of the precision of the estimates, there is no indication that removing WC costs would have a notable impact on standard errors for the ECEC, ECI, or linked ECI estimates. Overall, BLS is comfortable removing workers’ compensation costs from the NCS and believes doing so will improve the quality of NCS data.