A new measure of the cost of compensation components.

Wood, G. Donald

A New Measure of the Cost of Compensation Components

THIS paper describes and evaluates a new measure of employer costs--that is, cents per hour measures--for the components of employee compensation. The new measure is estimated from data collected for the Employment Cost Index (ECI), which has provided, since 1980, index numbers of the change in compensation costs. It was decided to use ECI data to prepare cost-level estimates, since these estimates could be generated from the ECI without increasing in any way the reporting burden on establishments and at only a fraction of the cost of a separate survey.

The first cost-level estimates, for March 1987, were published in the October 1987 Monthly Labor Review and are presented in tables 1-5 (Nathan 1987). Beginning this year, cost estimates with a March reference date will be published annually by the Bureau of Labor Statistics in a news release issued in June.

Data collected for one purpose are rarely ideal for other purposes, and cost-level estimates from the ECI are no exception. However, evidence presented in this paper indicates that these estimates are very reliable.

Summary of Results

In March 1987, compensation for all private industry workers averaged $13.42 per hour worked. Wages were $9.83, or 73.2 percent of total compensation. Benefit costs were $3.60, or 26.8 percent of total compensation. The largest component of benefits was legally required benefits, which was dominated by social security costs and which accounted for 32 percent of benefit costs. Other major components of benefits are paid leave (26 percent), insurance (20 percent), retirement benefits (13 percent), and supplemental pay (9 percent).

There is considerable variation in levels of compensation and proportions of benefits to compensation among the broad industrial and occupational groups for which estimates are available. White-collar workers received $15.56 an hour, which is 16 percent more than the $13.43 received by blue-collar workers. The highest paid white-collar occupational group is composed of executive, administrative, and managerial workers who received $23.81 per hour worked, which is 3.7 times the pay of workers in the lowest paid occupational group--service workers--who received $6.43 per hour.

Workers in the goods-producing sector received $15.86 an hour, which is 28 percent more than the $12.41 received by workers in the service-producing sector. Workers in the highest paid industry--transportation and public utilities--received $20.24 an hour, which is 2.6 times the pay of workers in the lowest paid industry--retail trade--who received $7.85 an hour.

The proportion of total compensation that is accounted for by wages decreases as the level of compensation by industry increases. Wages and salaries as a proportion of total compensation ranged from 77.3 percent for retail trade to 68 percent for transportation and public utilities. This inverse relationship between the level of compensation and the proportion accounted for by wages should be expected. Most benefits have a high income elasticity of demand, and social security has become less regressive, because the 1987 earnings ceiling of $45,000 is well above most annual wage and salary incomes.

However, for any level of compensation, blue-collar workers tend to have a lower proportion of total compensation accounted for by wages than do white-collar workers, even though blue-collar workers earn less. Wages and salaries for white-collar workers average $11.61, and wages and salaries for blue-collar workers average $9.38. Benefit costs for blue-collar workers of $4.05, however, are slightly higher than the benefit costs of $3.95 for white-collar workers.

Even when blue-collar and white-collar worker groups are considered separately, the expected inverse relationship between the level of compensation and the proportion of compensation accounted for by wages and salaries does not appear. Wages account for 70.6 percent of total compensation for the lowest paid blue-collar group--laborers--and 70.8 percent for the highest paid blue-collar group--precision workers. Wages account for 72.3 percent for the lowest paid white-collar group--clerical workers--and 75 percent for the highest paid white-collar group--executive, administrative, and managerial workers.

There are a number of other relationships to be found among the data in the tables; some of these are expected, and some are perplexing. But the purpose of this paper is not to analyze the data; instead, it is to provide information to aid potential users in properly interpreting and analyzing these new data from the ECI program.

ECI Survey Design

The 1987 cost-level estimates were based on data collected from about 3,200 establishments in the price nonfarm sector of the economy. The establishments were selected with probability proportional to employment from the Bureau of Labor Statistics (BLS) Unemployment Insurance (UI) File. The file lists every establishment with one employee or more covered by State unemployment insurance. About 98 percent of all private industry workers are employed by establishments listed in the file.

The ECI sample is replaced on a 4-year cycle, with about one-fourth of the establishments replaced each year. Replacement is by industry or by groups of industries. Each selected establishment is visited by a BLS economist from one of the eight regional offices. The first task, after explaining the survey and securing the cooperation of the establishment, is to select the jobs for which wage and benefit data are to be collected.

Four, six, or eight narrowly defined jobs are selected with probability proportional to the number of workers employed in each job. The number of jobs selected depends on the size of the establishment, but, on average, about five jobs are selected for each establishment. For these jobs, initial wage and benefit data are collected and then updated each quarter. The March 1987 estimates are based on about 16,000 jobs selected in 3,200 establishments.(1)

Job selection is crucial to the index. It plays the same role in the ECI that item specification plays in the Consumer Price Index. Each job selected in an establishment must be defined narrowly enough that its incumbents carry out the same tasks at roughly the same skill level. If the selected jobs are not narrowly and clearly defined, then the changes in cost over time might reflect, not the change in the cost to the employer of having specific tasks performed, but rather the collection of cost for a higher or lower skilled group of workers.

Once the jobs are selected, BLS economists collect the information necessary to compute the employer's cost, by kind of benefit, and change in cost when it occurs. Before describing the data collected, it is necessary to discuss the concept of cost used in the ECI.

ECI Costs

In estimating the cost of the compensation package (wages, salaries, and the employer's cost of employee benefits), the ECI measures the cost per hour worked as a rate at a point in time--the time of data collection. The rate is what the cost per hour worked would be if the wage or salary, the benefit package, and the employer's cost of each benefit were unchanged for a long enough time period for the employee to receive all benefits and for the employer to make all payments for benefits provided.

Such a concept of a rate at a point in time is necessary for the ECI, which attempts to measure change in labor cost in a timely fashion. The concept makes it easy to identify when a change occurs and to compute the cost of the change. A change in labor cost occurs when the wage or salary changes, when the benefits provided change, or when the cost of providing benefits with the same provisions changes.

A simple example will illustrate the difference between the ECI rate of cost at the time of collection and the actual expenditures over an interval. Suppose that, for the selected occupation, the wage rate is $10.00 per hour, the work schedule is 52 weeks a year, the workweek is 40 hours, and the only benefit is 2 weeks, or 80 hours, paid vacation per year.

The wage cost is $10.00 per hour worked. Leave is also paid at $10.00 an hour. The cost of leave is $800 (80 hours x $10.00) per year. Of the 2,080 scheduled hours, 2,000 are worked per year. Then the cost of leave per hour worked is $800 divided by 2,000 or $0.40. The total cost per hour worked is the wage plus the cost of leave or $10.40.

If the wage rate rises to $11.00 on July 1 with no change in the vacation plan, then the cost of paid leave will rise by the same percent, 10 percent, as wages to $0.44, and the total cost per hour worked will rise to $11.44.

The $10.40 used for the ECI in the first half of the year and the $11.44 used in the second half will not equal the actual expenditures per hour worked for incumbents in the job over, for example, a year. The cost per hour worked at a point in time would equal the actual expenditures per hour worked over a year only if the wage remains unchanged over the entire year and if the number of paid vacation days granted per year remains unchanged.

Data Collected

The wide range of wage and benefit practices precludes a thorough discussion of what is collected or of how costs are calculated for every possible situation for every benefit (Nathan 1987). Instead, a few examples will be given, which should be sufficient to provide the user with the information required to properly use the estimates. Hours worked

Hours worked are also considered as a rate at the time of collection and not as actual hours worked over any calendar time period. They are the number of hours that would be worked if conditions at the time of collection were to remain unchanged for a long enough period for the entire schedule to be worked--usually a year.

Scheduled hours per day and per week and scheduled weeks per year are collected. They are used to determine scheduled annual hours and the number of hours not worked. Hours paid but not worked--based on the vacation, holiday, sick leave, and other leave plans at the time of collection--are deducted from scheduled annual hours; scheduled hours not worked and not paid are also deducted. Overtime hours are added to scheduled work hours. Wages and salaries

For ECI purposes, wages and salaries exclude shift differentials, premium pay, and nonproduction bonuses. All costs for these items are included in benefit cost, not in wages and salaries. In addition to straight time pay, wages and salaries include a number of add-ons--such as incentive pay, sales commissions, hazard pay, on call pay, deadhead pay, and cost-of-living allowances, which are not paid as part of the straight time rate.

Wages.--If all workers receive the same hourly wage rate, then the wage rate is collected and used. If different incumbents within the same job receive different wage rates--for example, because of length-of-service premiums or differences in commissions earnings--then the average wage is used.

The use of the average wage highlights the importance of the definition of the job selected for the index. If differences in wages among workers in a selected job reflect differences in duties or skill levels, then a change in the mix of duties or skill levels would introduce error in the index.

Salaries.--Salaries are labor payments that are not quoted on an hourly basis. In most cases, there is a work schedule for salaried employees. When this is true, the salary is divided by scheduled hours to obtain a salary cost per hour worked. Scheduled hours, not hours worked, are used to compute salary per hour worked, because the salary includes pay both for hours worked and for hours not worked but paid.

When the salary is not related to a work schedule--for example, for executives, teachers, sales workers, truckdrivers, and airline flight crews--the field economists try to get the respondent to supply a reasonable estimate of the employee's work schedule or hours paid. If this is not possible, the predominant work schedule for similar occupations in the establishment is used. Benefit costs

If all incumbents receive the same benefit, then data collection and calculation of benefit cost are fairly straightforward. In the previous example in which all employees received 2 weeks paid vacation a year, the hours of vacation are multiplied by the wage rate and divided by hours worked, yielding vacation benefit as cents per hour. Other plans would be treated similarly. For example, if all employees received the same basic health plan, then only the price of the plan is collected. The price is divided by the number of hours worked per employee to obtain cents per hour worked.

Usage.--Frequently, employees doing the same job do not all receive the same benefits. For example, the amount of paid vacation received may depend on the employee's length of service--1 week the first year, 2 weeks the second through the fifth year, and so on. In these cases, it is necessary to determine how much of each benefit is received by each employee, and the result is termed the "usage" of the benefit.

Determining usage greatly increases the collection burden and the complexity of evaluating the cost. In the previous example, the length of service of each incumbent in the job is collected. The length-of-service distribution and the benefit plan determine how much vacation each employee receives. Average hours of vacation per year is then calculated. The average number of hours is multiplied by the wage rate, and the product is divided by hours worked to get vacation costs per hour worked.

If no vacation is given until the employee has worked for a minimum period of time, then those employees that fail to meet the eligibility requirements are included with no benefit This procedure is used for any benefit with an eligibility requirement.

Another usage issue arises in measuring health insurance. Suppose the employer pays the total cost for a health insurance plan in which married employees receive a family plan and single employees receive a self-only plan. The proportion of employees that receive each type of plan is collected. The price of each plan is multiplied by the proportion of employees receiving the plan, and then the products are summed and divided by the hours worked per employee.

Expenditures.--In the previous examples, costs have been computed by multiplying the number (proportion) of employees that receive a benefit by the price (cost) of the benefit. In the ECI, this is the preferred method, which is termed "rate times usage."

When rate and usage cannot be collected, expenditures data can be collected and used. Perhaps the most important use of expenditures data occurs when the employer is self-insured and there is no rate. For example, employees might receive health insurance, but the establishment itself pays for covered expenditures. Expenditures over the previous year per employee are collected. The expenditures per employee are divided by hours worked.

Quarterly Change

Once all the required data have been collected and the cost of all benefits for each occupation computed, the initiation of the establishment is completed. Once the establishment is initiated, it is requested (usually by mail or telephone) to provide quarterly the information required to update the costs. Prior to each quarter, the establishment is sent a wage "shuttle" form and a summary of benefits form. The wage shuttle identifies the selected jobs and the wages of the previous quarter. The summary of benefits summarizes the benefits provided for each job the previous quarter.

The Employment Cost Index is published the month after the reference month. The only reason why the data can be collected and processed so quickly is that the ECI concept of change does not, in general, require that usage be collected each quarter. For the ECI, as noted, a change in labor cost occurs for a benefit only when the benefit changes or the price of an existing benefit changes. For example, if an establishment provides family and self-only health coverage and if there is no change in the plans or their cost, the ECI cost remains unchanged; it remains unchanged even though the distribution between married and single employees in the job may have changed.

Furthermore, even if the benefit or its price changes so that there is a change in ECI costs, usage is not updated unless the change in the benefit or its price will induce (cause) a change in usage. In general, new usage data will be collected only if new plans are added, if old plans are dropped, or if the cost or provisions of contributory plans change.

For example, suppose the employer provided family and self-only health plans to the employees the previous quarter, but this quarter employees are also given the option to join a health maintenance organization (HMO). Since some employees would join the HMO, the proportion of workers receiving the other plans would fall. Then, new usage data, as well as the price of the HMO membership, would be collected, and a new cost would be calculated (Bureau of Labor Statistics 1986).

This treatment of usage is appropriate for an index number that measures the change in cost over time. Cost levels, however, should have current usage, which does raise a question concerning the accuracy of the ECI cost levels (Scheifer 1975). Empirical analysis of this question is presented later in this chapter.

Employment Weights

In March 1987, the ECI survey yielded estimates of wage and benefit cost for about 16,000 occupational observations, where the costs, as noted, are estimates of a rate at a point in time. These costs are estimated with usage ranging from the current year to 4 years old.

These 16,000 occupational observations are aggregated in the ECI using the 1980 census of population (because the ECI is a "fixed weight" index). Data for 1980 are clearly not suitable for weighting cost estimates for March 1987.

The weights for 1987 cost levels are obtained in two steps. First, the ECI sample provides an estimate of the occupational employment distribution within each industry at the time of initiation. Second, these occupational distributions are used to apportion the employment by industry for March 1987 from the BLS Current Employment Statistics program. The industry employment distribution is current, but the occupational distribution by industry varies from the current year to 4 years old (depending on the phase of the ECI initiation schedule).

There is evidence suggesting that labor cost indexes are not very sensitive to variation in employment weights (Schwenk 1985), but the use of occupational employment distributions by industry that range from the current year to 4 years old does raise a question about the accuracy of the ECI cost levels. Empirical analysis of this question is given later.

Evaluation of Estimates

The usual concerns about data quality are sampling and nonsampling errors. Sampling error

Standard errors of the estimate are calculated for each estimate using a balanced, repeated replication method with 64 psuedoreplicates. A detailed description of the method used for calculating the variance for the index, which is the same method used for calculating the levels, will be published in early 1989 in the Monthly Labor Review. Nonsampling error

There are no measurements of nonsampling error for the ECI, but all BLS surveys, including the ECI, have a wide range of quality management programs that are designed to hold nonsampling errors within acceptable bounds. ECI survey procedures designed to control nonsampling error include clear documentation and instructions for each survey activity, quality control to ensure that the instructions are followed, the collection of data by personal visit by professional field economists, regular training on program procedures, professional review of all data collected, and machine edits and review at each stage of processing. The only sources of nonsampling error that will be explicitly discussed here are nonresponse and noncurrent distribution, which is the error introduced because usage and the occupation distributions within industry are not current.

Refusal to participate in survey.--Refusals are eligible establishments that refuse to provide data. The ECI survey has one of the highest refusal rates of all BLS wage surveys. In March 1987, the response rate--that is, responding establishments divided by responding establishments plus eligible nonresponding establishments--was 73 percent.

There are a number of reasons for the high nonresponse to the ECI survey relative to other wage surveys. The ECI is a length interval survey, so it has refusals not only at the time of initiation but also over time, as respondents drop out of the program. The ECI sample includes establishments of all sizes in all industries, whereas other wage surveys usually exclude small establishments, which tend to have higher nonresponse rates. The ECI is relatively new and, therefore, is not widely known; there are establishments that will continue to respond to surveys that they have responded to in the past, but they will not participate in new surveys. Other establishments will only respond to a survey if they use the survey or if they are already familiar with the survey.

Every effort is being made to reduce nonresponse rates, and we expect them to decline, though slowly, over time, as the ECI becomes more widely known and used. But whatever the future holds, the response rate for the March 1987 estimates was 73 percent.

In deriving cost-level estimates, the weight of nonrespondents is allocated to similar (same industry, establishment, size, area, and so on) responding establishments. The accuracy of the estimates depends on how close the data of the nonrespondents are to the data of the respondents that carry their weight. It is not possible to determine how accurate the nonresponse adjustments are, because, by definition, data from the nonrespondents are not available. The only way to reduce the potential error caused by nonresponse is to reduce the proportion of nonrespondents.

Noncurrent distribution.--The other potential sources of nonsampling error that have been identified are the use of dated occupation distributions and usage. The potential bias of these sources of nonsampling error can be analyzed because of the ECI's sample replacement program.

Whenever the sample for an industry is replaced, there are two estimates of labor cost for the same industry for the same time period. One estimate is for the sample to be dropped, which has a dated occupational distribution and usage pattern that is 4 years old. The other estimate is for the sample to be added, which has a current occupational distribution and usage pattern. If the impact on the estimates of the dated occupational distribution and usage pattern is large, it should be possible to reject the hypothesis that the two estimates are based on samples drawn from the same universe.

At the time the analysis was carried out, data on two replacement samples were available only for wholesale trade. The wholesale trade sample was replaced first in 1982 and then in 1986. The difference of the means of the two sample estimates would reflect differences in both usage and occupational mix. A test of the differences between two means could not reject the hypothesis that the means are equal. Being unable to reject the hypothesis does not imply that the estimates of the old sample are unbiased, because they almost certainly are biased. It simply implies that whatever bias exists is small relative to the variances.

Dated occupational structure and fixed usage can be tested separately. The impact on the estimates of the dated occupational weights was evaluated by reweighting the costs from the 1982 wholesale trade sample using the weights of the 1986 sample. The only difference between the 1982 estimates with the 1986 weights and the actual estimates of the 1982 sample is the occupational weights. The differences were small relative to sampling errors. Dated usage was evaluated by developing a distribution of the number of days paid but not worked by benefit plan and occupation from the 1986 sample. For each establishment in the 1982 sample, usage was selected from the 1986 distribution. The only difference between the 1982 sample with the 1986 distribution and the actual estimates of the 1982 sample is the usage. The differences were small relative to sampling errors.

Comparisons with Other Data

The ECI estimates will be compared with estimates prepared by the Bureau of Economic Analysis (BEA) and with estimates of average hourly earnings prepared by the Bureau of Labor Statistics from the Current Employment Statistics survey. BEA compensation

The Bureau of Economic Analysis provides estimates of expenditures on total compensation and wages for the private nonfarm economy. The BEA definition of total compensation is roughly the same as that used for the ECI. The BEA definition of wages, however, includes the payments for hours paid but not worked and supplemental payments--that is, payments for shift pay, premium pay, and nonproduction bonuses.

For 1987 the BEA estimates showed that wages (which are equivalent to gross earnings from the ECI) were 84.9 percent of compensation. In the ECI data for March 1987, wages, paid leave, and supplemental payments are 82.5 percent of compensation. The difference between 84.9 percent and 82.5 percent is reasonable given the differences in the reference periods and the data sources. Average hourly earnings

The Average Hourly Earnings (AHE) series covers all production workers in goods-producing industries and non-supervisory workers in service-producing industries. Similar coverage can be obtained from the ECI by excluding all white-collar occupations from the goods-producing industries and the manager, executive, and administrator occupations from the service-producing industries.

Wages from the AHE include overtime and shift pay. Average earnings from the AHE for March 1987 were $8.92. The ECI wage rate for the AHE occupational coverage was $8.72, overtime was $0.18, and shift differential was $0.04. Thus, the value from the ECI, comparable with the AHE wage, was $8.94. The difference between $8.92 and $8.94 is well within sampling error of the ECI.

Another interesting comparison with the AHE uses wages for production workers in manufacturing, the only industry for which the AHE has a separate estimate for overtime. In addition to the impact of the occupation distribution and nonresponse, this comparison gives an indication of the impact of dated usage on overtime. The ECI estimate of overtime cost was $0.43 per hour, and the AHE estimate of overtime cost was $0.41. The ECI wages, plus shift differentials, was $9.34, and the comparable value from the AHE was $9.44.

Although the AHE has only limited coverage of benefits, the above comparisons are of interest, since the dated occupational distribution and the nonresponse affect wages as well as benefit costs.

Uses of Cost Levels

The new measures do not place any additional burden on respondents, nor are they expensive. The estimates are not free, however; resources are required for their preparation and publication. In these days of very scarce resources for general economic statistics, it is necessary to justify any expenditures. The justification can only be in the use made of the statistics.

The ECI costs, as rates at a point in time, are the statistics of choice for wage and salary administration, labor negotiations, and comparisons of compensation among groups of workers. Because of these uses, cost levels were the most requested statistics by the BLS's Business Research Advisory and Labor Research Advisory Committees. The Employment Standards Administration wanted ECI cost levels in order to carry out their responsibilities under the Service Contract Act. It is their responsibility to set minimum wages and benefits for employees of firms that have contracts to provide services to the Federal Government.

Other users may be interested in cost levels for different purposes. For example, benefits might be measured as income to employees rather than as cost to employers.

I cannot recommend without qualification the use of the ECI cost levels for analysis that requires expenditures rather than rates or employee income rather than employer cost. But if the ECI data and their limitations are clearly understood, it seems to me that the data could prove of value in research in these areas as well.

References Bureau of Labor Statistics (1988). BLS Handbook of Methods. Chapter 8. Bulletin No. 2285. Washington, DC: GPO. Nathan, Felicia (1987). "Analyzing Employers' Cost for Wages, Salaries, and Benefits." Monthly Labor Review (October 1987): 3-11. Scheifer, Victor J. (1975). "Employment Cost Index: A Measure of Change in the Price of Labor," Monthly Labor Review (June 1975): 22-27. Schindler, Eric (1988). "Statistical Development to the Employment Cost Index," Proceedings of the Section on Business and Economic Statistics, American Statistical Association. Schwenk, Albert E. (1985). "Introducing New Weights for the Employment Cost Index," Monthly Labor Review (June 1985): 22-27.

Table : 1.--Employer Cost for Employee Compensation for Private Industry and Major Occupational Categories

Table : 2.--Employer Cost for Employee Compensation for Selected Major White-Collar Groups

Table : 3.--Employer Cost for Employee Compensation for Major Blue-Collar Groups

Table : 4.--Employer Cost for Employee Compensation by Major Industrial Sectors

Table : 5--Employer Cost for Employee Compensation for Major Service-Producing Industries