Longitudinal data from the Occupational Employment Statistics survey

The Occupational Employment Statistics (OES) survey of the U.S. Bureau of Labor Statistics (BLS) is a large survey of employers in the United States. The data from this survey are used to produce estimates of employment and wages by occupation at industry, state, and metropolitan area levels. They are used by the Office of Foreign Labor Certification of the U.S. Department of Labor in setting pay rates for workers in the United States on certain work visas. They are also key inputs to the Employment Projections program, which is one of the most widely used products at BLS. However, the OES data are little known by most researchers.

While the published estimates of the OES program are not designed for making comparisons over time, the confidential microdata underlying these estimates do allow for following certain employers over time, and these data have several features that make them particularly useful for research purposes. Most notably, the microdata are a very large set of data containing full staffing patterns, including occupation and wage information for all employees within surveyed establishments.

This article describes the longitudinal features of the OES data and the composition of the establishments that are observed longitudinally. We begin with a discussion of changes in the OES survey that affect the probability of sample selection over time for any particular establishment. We then show how the sample design shapes the composition of establishments sampled repeatedly for the OES program. Next, we show how differences in response rates also shape the composition of establishments observed repeatedly. Finally, we compare the distributions of employer and employee characteristics in the longitudinal OES data with the distribution of these characteristics in the cross-sectional data.

An overview of changes to the OES survey affecting the data available for longitudinal study

BLS began conducting surveys of employment by occupation in the 1960s.¹ It was in 1971 that the Occupational Employment Statistics survey began, and it was a collaboration between BLS and 10 State Employment Security agencies. The OES initially was a survey of 50,000 manufacturing establishments, asking these establishments about the distribution of their employees by occupation. Similar surveys were soon developed for other industries. Beginning in 1977, surveys were conducted in every state and the District of Columbia. Since then, major changes to the design of the OES survey have included the following:²

• In 1988, the OES program began collecting employment data by industry in a 3-year cycle, with manufacturing and hospitals surveyed in the first year; mining, construction, finance, insurance, real estate, and service in the second year; and trade, transportation, communications, public utilities, and government establishments in the third year of each cycle.
• In 1991, 15 states began to collect wage information along with the occupational employment information.
• The year 1996 saw the greatest changes in the history of the OES program. First, the sample was redesigned to collect data from all industries every year so that estimates for all industries could be produced annually, although the program retained some aspects of the 3-year cycle. Establishments were generally included in the survey no more than once in each 3-year cycle. Because some establishments are large enough to affect estimates of employment and wages in their industry or geographic area, 3 years of microdata continue to be used in the production of all program estimates (despite the annual release of updated estimates). Second, the sample was redesigned in order to produce estimates at the Metropolitan Statistical Area (MSA) and Balance of State (BOS) levels. This involved increasing the sample size from approximately 250,000 establishments per year to 400,000 establishments per year. Third, the OES program began collecting wage information in every state from every establishment responding to the survey.
• In 1997, the scope of the survey expanded to include very small establishments.
• In 1999, the OES program began using the Standard Occupational Classification System when collecting data on occupations. Although the program did not switch from using the Standard Industrial Classification System (SIC) to the North American Industry Classification System (NAICS) in the sample design until 2002, nearly all OES data collected in the 1999, 2000, and 2001 panels was recoded from SIC to NAICS to produce estimates for NAICS-defined industries in combination with the 2002 panel.
• In 2002, rather than 400,000 establishments being asking about their employment and wages in October, November, and December, 200,000 establishments were asked about their employment and wages in May and another 200,000 establishments were asked the same questions in November. This was done so that the OES data would better reflect year-round employment patterns. Employers are still surveyed no more than once in each 3-year (six-panel) collection cycle. The rural portion of the sample was redesigned to provide geographic estimates for more than one nonurban area in large states. These nonurban areas are known as Balance of State or BOS areas.
• Over the period of 2005–09, several changes were made to the OES sample design, affecting the probability that particular establishments would be selected for the OES sample. These changes included
  • simplifying the sample design to sample establishments in only two size classes per industry and geography area, rather than seven size classes
  • moving to a sample design closer to probability proportional to employment size within industries and geographic areas
  • allocating more of the sample to industries with more variation in occupational employment between establishments
  • moving to a Power Neyman allocation of the OES sample between industries and geographic areas³
  • lowering the employment level at which establishments from particular states were certain to be selected into the OES sample in each 3-year cycle (in 2009, these levels returned to the same levels used in other states)
  • more flexibly allocating the sample between states in order to decrease the variability of estimates between states.⁴

The OES sample design is a complex stratified random sample, which differs in several ways from a simple random sample of employers. The goal of the sample design is to produce estimates of employment and wages for each occupation with as little sampling variability as possible given the overall sample size. Portions of the sample are allocated to each industry, each geographic region, and (before 2005) each employer size class to reduce the variation of estimates of employment and wages by occupation for detailed industries and detailed geographic areas. Since 2006, the occupational variability of industries has also been a factor in the sample design—to increase the precision of the estimates, the sample size is larger in industries with more variation in the composition of occupations, and these industries are more likely to be selected for the OES sample.

In addition, the sample design considers the size of employers as a criterion for selection. Large employers, which affect overall average employment and wage levels in their industry or geographic area, are always selected for the OES sample during the six-panel cycle. Smaller employers are selected with a probability that varies by the number of employers in their stratum (industry and geographic area) and by the number of employees in their establishment.⁵ Current methodology selects small- and medium-sized employers within each stratum with a probability proportionate to their employment levels. However, even very small employers are almost certain to be selected for the OES sample in each six-panel cycle if there are few other employers in their industry and geographic area. When there are fewer than four establishments in a stratum, all are selected for the OES sample. When there are 4–12 establishments (4–18 establishments in Idaho, Montana, or Wyoming until 2009) in a stratum, at least 3 are selected for the OES sample. When there are additional establishments in a stratum, at least six are selected for the sample.⁶

To reduce the burden of survey response for smaller employers, BLS assigns each known establishment a permanent random number (PRN) between 0 and 1. Within each stratum, establishments are selected for the OES sample beginning with those having a PRN value of 0.4, and establishments with PRN values greater than 0.4 have an increased probability of sample selection in every wave of the OES sample. Other BLS surveys begin selecting their random samples at different PRN values, making it unlikely that small establishments will be selected for multiple BLS surveys.⁷

The large number of employers in the OES sample and the stratified design of the sample mean that hundreds of thousands of employers have been included in the OES sample multiple times, albeit usually at least 3 years apart. This feature allows researchers with access to the OES microdata the opportunity to study changes in the occupational employment and wage patterns of the repeatedly sampled employers.

Repeated sampling in the OES survey

As described above, the sample design of the OES survey leads to an increased probability of selection for large employers, employers in geographic areas with relatively few employers, employers in industries with a greater variability of occupational employment, and employers with a BLS internally assigned PRNr greater than 0.4. Such employers are particularly likely to be sampled repeatedly in the OES survey.

Repeated sampling magnifies differences in sampling probability that occur in each wave of a survey. Consider a much simpler example than the complex OES survey: large employers have a 50-percent chance of being sampled, and small employers have a 10-percent chance of being sampled. To make population estimates for a single wave of data collection of this simple survey, weights of 2 should be applied for large employers and weights of 10 should be applied to small employers. However, if there are two independent samples drawn for two waves of this survey, the chance that a large employer will appear in both waves is .5 × .5, or 25 percent, while the chance that a small employer will appear in both waves is .10 × .10, or 1 percent. In this example, the weights used for a single wave of data collection will be inappropriate to use in weighting the data observed twice to population levels. Furthermore, the sample of smaller employers observed twice may be too small to make conclusions about this subpopulation.

We examine how differences in the OES cross-sectional sampling probabilities across establishment sizes, geographic area types, and industries translate into differences in the probability that establishments with different characteristics appear in longitudinal OES data. For our analyses, we begin with private sector data from 1999, as these data are comparable to later data in their industry and occupational coding systems. (We drop 4,238 establishments from our analyses because they were not used for the 2002 estimates and so their industries were not recoded using the NAICS.) We omit the large number of public sector establishments included in the OES survey (such as publicly owned schools and hospitals), as well as establishments sampled in Guam, Puerto Rico, and the Virgin Islands. The frame for this sample is the Quarterly Census of Employment and Wages (QCEW), assembled from the reports to the unemployment insurance system made by nonfarm establishments in every state, and supplemented with a list of rail transportation employers (not covered by the unemployment insurance system).⁸

Table 1 shows the number of private sector establishments selected for sampling by the OES program from fall 1999 through May 2014, by establishment size. In all, more than 2.8 million establishments have been selected for the OES sample, of which nearly 1.3 million (46 percent) have been sampled more than once and nearly 80,000 have been sampled five or more times during this period.⁹ This table uses the employer sizes reported in the QCEW, as these are the estimates of employer size used to select the sample. Establishments sampled more than once are classified in this table by their largest size class.

As shown in table 1, large establishments are more likely than small establishments to be sampled repeatedly by the OES program. Only 23 percent of the establishments with fewer than five employees that are ever sampled for the OES survey are sampled more than once (7 percent are sampled at least three times). However, 80 percent of the establishments with 250 or more employees that are ever sampled for the OES survey are sampled more than once, and 61 percent of these establishments are sampled at least three times.

Table 2 shows the number of establishments selected for sampling by the OES program during this same period by area type. (Some area definitions change over time, and so establishments sampled more than once are classified in this table by the type of area in which they are located the final time they are sampled.) Approximately 46 percent of establishments ever sampled by the OES program in urban (MSA) areas and 45 percent ever sampled in rural (BOS) areas are selected for sampling more than once. Within urban (MSA) areas, establishments located in areas with less employment are more likely to be repeatedly sampled. About 48 percent of establishments ever sampled by the OES program in MSAs with employment of less than 100,000 are sampled more than once, while 45 percent of establishments ever sampled by the OES program in MSAs with employment of more than 1,000,000 are sampled more than once.

Table 3 shows the number of establishments selected for the OES sample during this period, by industry. (Establishments may change industry classifications over time, and so establishments sampled more than once are classified in this table by the industry they report to the QCEW the final time they are sampled.) The frequency with which establishments are sampled more than once by the OES program varies by industry. The industry with the least amount of repeated sampling is accommodation and food services (NAICS code 72), in which 38 percent of establishments sampled are sampled more than once. This relatively low percentage is due to a combination of little variation by occupation between establishments, a large number of establishments, a large number of small establishments, and a large number of establishment openings and closures. The industries with the greatest amount of repeated sampling are manufacturing (NAICS code 31–33) and utilities, in which, respectively, 56 percent and 57 percent of establishments sampled are sampled more than once. Establishments in these industries are sampled more frequently because of their large size.

View Chart Data

Figure 1 shows the number of establishments sampled by the OES program and those sampled more than once, by the panels in which they are sampled. This figure shows that establishments sampled more than once for the OES program come from all waves of data collection. It also shows the clear drop in sample size per panel in 2002, when the sample collection was split from one panel each fall to two panels, in May and November, of each year. Otherwise, the distribution of the potential sample for longitudinal observations of the OES data is remarkably similar from panel to panel, aside from an unusually small overall sample size in May 2008 because of budget constraints¹⁰ and an unusually small number of the establishments sampled multiple times that were sampled in November 2006.¹¹

By far the most common interval between sample dates is exactly 3 years. Intervals of exactly 3 years represent 47 percent of all intervals between successive sample dates. Intervals of whole years are more common than intervals, such as 3 ½ years, that are off by half a year (because all establishments of very large companies are sampled at regular 3-year intervals), but the frequency with which each interval is observed is otherwise decreasing in interval length—that is, there are more intervals of 3 years than of 4 years, more intervals of 4 years than of 5 years, and so forth. The greater the number of times an establishment is sampled, the shorter the average interval between sample dates. Less than 1 percent of intervals between sample dates are spaced less than 3 years apart. About 3 percent of intervals between sample dates are 10 years or longer.

Nonresponse in the OES survey

The composition of longitudinal data from the OES survey is affected not only by the sample design but also by differential response rates to the OES survey. The OES survey is primarily conducted by mail, in a collaboration between BLS and State Workforce Agencies. There is an open-ended form to collect data from small employers, and, until November 2015, there were 97 different industry-specific forms used for collecting occupational employment and wage data from medium- and large-sized employers. After the initial mailing of forms, three industry-specific followup mailings are sent to nonrespondents at intervals of 3–4 weeks. The State Workforce Agencies also conduct followup by telephone, with timing varying by state.¹² These agencies collect some data for large establishments via personal visits. Large numbers of respondents choose to respond to the survey in electronic format.

Establishments that are sampled more than once may respond to the survey in some panels but not in others. Table 4 shows the number of times that establishments are sampled, by the number of times they respond. There were 2,840,923 establishments sampled for the OES survey between fall 1999 and May 2014, and over this period, these establishments were sampled 5,283,824 times. The probability of always responding declines as the number of times sampled increases: 78 percent of establishments sampled only once responded to the survey (some went out of business in the year between the sample frame date and the survey date), while 65 percent of those sampled twice responded twice, 57 percent of those sampled three times responded three times, 50 percent of those sampled four times responded four times, and 38 percent of those sampled five times responded five times. However, enough of those sampled responded each time that 76 percent of establishments responding exactly twice were sampled only twice, 76 percent of those responding exactly three times were sampled only three times, and 85 percent of those responding exactly four times were sampled only four times.

View Chart Data

Figure 2 shows response rates for private sector establishments sampled by the OES program were steady at about 77 percent from 1999 through 2010 or 2011. There was a noticeable decline in response rates in 2013 because of the federal government shutdown in October 2013.¹³ State Workforce Agencies are required to follow up with employers (except when funding for this work lapses in a federal government shutdown) until the overall response rate for the survey reaches at least 75 percent overall (including the establishments of state and local governments) and 75 percent in each MSA. Data are collected from state governments only in the November panels, and so state analysts do not need to follow up with as many private sector establishments to reach these response targets in the November panels as in other months. Response to the OES survey has been mandatory in North Carolina since 1995, Oklahoma since 2000, South Carolina during the 2002–05 period, Georgia since 2005, Vermont since November 2012, Wyoming since May 2013, the District of Columbia since November 2014, and Hawaii and Oregon since May 2015. Private sector response rates are about 85 percent in these states when response is mandatory.¹⁴

Table 5 shows that small establishments generally have higher response rates than large establishments. Polly A. Phipps and Carrie K. Jones report that this is due to the preference of state analysts for focusing followup work on the smallest establishments.¹⁵ When analysts contact employers by phone, it is easiest both to find someone who can report on the occupational employment and wage information and to code occupational data over the phone for the smallest establishments. Hence, in working to reach the required 75-percent response targets across all establishments, state analysts find it easiest to focus on these small establishments.

Table 6 shows that rural areas (BOS areas) have higher response rates than urban areas (MSAs), and among urban areas, response rates are higher in smaller MSAs than in larger MSAs. Phipps and Jones report that establishments in larger MSAs are both less likely to respond by mail to the survey forms and more difficult for state analysts to reach by telephone; these establishments often state that they are too busy to respond to the OES survey.¹⁶

Table 7 shows response rates by industry. Response rates are particularly high in establishments within the agriculture, forestry, fishing, and hunting sector (farms are outside the scope of the survey) and in some service industry establishments. Response rates are particularly low for company headquarters establishments; administrative and support and waste management and remediation establishments; mining, quarrying, and oil and gas extraction establishments; and information establishments.

When sampled establishments do not respond to the OES survey, data for these establishments are imputed in order to produce estimates of occupational employment and wages by industry and geographic area. These imputations proceed in two steps. First, occupational staffing patterns are taken from establishments similar in industry, overall employment size (as reported to the unemployment insurance system), and geographic area. Second, wage distributions are taken from the distribution of wages for the same occupation among employers in the same geographic area, in the same industry, and with the same overall employment size.¹⁷ In assembling the longitudinal OES data described in the next section, we do not include this imputed data for establishments that do not respond to the OES survey.

Establishments observed longitudinally in the OES data

The combination of sampling and response patterns described above shape the composition of establishments that respond more than once to the OES survey, sometimes in opposing ways. Figure 3 shows the composition of the longitudinal data by panel and by the number of times establishments appear in the longitudinal data. Establishments are disproportionately likely to be observed in the fall 1999, fall 2000, and fall 2001 panels, which had larger sample sizes than more recent panels. Because of an error in the selection of the sample for the November 2006 panel, establishments in the longitudinal data are less likely to appear in that panel.¹⁸ Establishments appearing five times in the longitudinal data are less likely to appear in the November 2011 or November 2008 data. We believe this pattern will disappear once the data of the November 2014 panel are added to the longitudinal data, because so many establishments are sampled at regular 3-year intervals.

View Chart Data

The distribution of establishment and (weighted) employee characteristics in the longitudinal sample is remarkably similar to the distribution of each characteristic within the cross-section of OES data as a whole. OES employment estimates use establishment weights based on the inverse of the probability of sample selection. These estimates are then benchmarked to total employment levels by industry and geographic area.¹⁹ We use these same weights in tables 8 through 11.²⁰ Overall, 973,857 of the 2.8 million private sector establishments ever sampled by the OES program, or 34 percent, appear in the longitudinal data. These 973,857 establishments are all observed multiple times in the longitudinal data, and establishment observations total 2,562,195. These 2,562,195 establishment observations in the longitudinal data represent 48 percent of the 5,283,824 private sector observations and imputed observations in the OES data. Summed over all panels from 1999 through May 2014, weighted employment in the longitudinal sample is 268 million, half the weighted private sector employment in the OES data. Estimation weights in the OES data are designed to represent the U.S. workforce at the end of each 3-year data collection period: reweighting to an average of these employment levels over the fall 1999–May 2014 period gives an employment level in the longitudinal data of 57 million workers out of a total private sector workforce of 113.8 million (not shown in the tables below).

Table 8 shows the distribution by establishment size class, categorizing establishments in the longitudinal sample by their size when they last responded. The smallest establishments are the size class that appears least frequently in the longitudinal data relative to the number of times these establishments are observed in the OES data as a whole. Such establishments represent 12.6 percent of all establishments (16.5 percent of observations) in the longitudinal data and 21.5 percent of all establishments (18.7 percent of observations) in the OES data as a whole. Because the probability that small units will be sampled is low, they are assigned higher weights for calculating estimates; their weighted employment is 6.3 percent of the employment in the longitudinal data, which is remarkably close to their 6.7 percent of weighted employment in the OES data as a whole. After estimation weights are used to calculate employment totals, the most overrepresented size class in the longitudinal data is the class of establishments with 10–19 employees, which has weighted employment representing 12.7 percent of all employment in the longitudinal data and 11.9 percent of employment in the OES data as a whole.

Table 9 shows a similar comparison by area type. Establishments and employment in rural areas are slightly overrepresented in the longitudinal data: rural establishments comprise 21.5 percent of establishments, 21.5 percent of establishment-observations, and 14.0 percent of weighted total employment in the longitudinal data, compared with 20.9 percent of establishments, 20.6 percent of establishment-observations, and 13.3 percent of weighted total employment in the private sector OES data as a whole. The area type most underrepresented in the longitudinal data is those with employment greater than 1 million: these establishments comprise 20.9 percent of total establishments and 29.9 percent of weighted total employment in the longitudinal data, compared with 22.7 percent of establishments and 31.8 percent of weighted total employment in the OES data as a whole.

Table 10 gives a similar comparison by industry. Again, the similarity of distributions in the composition by industry of establishments, establishment observations, and weighted employment levels between the longitudinal sample and the OES data as a whole is remarkable. The industry most overrepresented in the distribution of establishments (and establishment observations) in the longitudinal data is manufacturing (NAICS code 31–33), which contains 12.5 percent of establishments (13.0 percent of establishment observations) in the longitudinal data and 10.0 percent of establishments (11.5 percent of establishment observations) in the OES data as a whole. This is not surprising, given the large establishment sizes in manufacturing described above. The industry most underrepresented in the distribution of establishments is accommodation and food services (NAICS code 72), which contains 4.6 percent of establishments in the longitudinal data and 5.9 percent of establishments in the OES data as a whole. This is not surprising, given the low variation in the occupational composition of establishments in this industry, reducing sampling probabilities. The industries in which the fraction of weighted employment in the longitudinal data differs most from the fraction of weighted employment in the OES data as a whole are health care and social assistance (NAICS code 62), which contains 15.9 percent of weighted employment in the longitudinal data and 14.0 percent of weighted employment in the OES data as a whole, and administrative and support and waste management and remediation (NAICS code 56), which contains 5.7 percent of weighted employment in the longitudinal data and 7.1 percent of weighted employment in the OES data as a whole. Administrative support and waste management and remediation is somewhat underrepresented in the weighted employment of the longitudinal data because of the low response rate among establishments in this industry, as described above. Health care and social assistance is somewhat overrepresented in the weighted employment of the longitudinal data because of a combination of a large industry size, a higher-than-average rate of repeated sampling and a slightly-higher-than-average response rate.

Table 11 makes a comparison (only for weighted employment) of longitudinal sample data and all OES data by the occupations of employees.²¹ Once again, the distributions of occupations are remarkably similar. The most underrepresented occupations in the longitudinal data is the management, business, and financial occupations group, which comprises 8.7 percent of weighted employment in the longitudinal data and 9.4 percent of weighted employment in the OES data as a whole. The most overrepresented occupations in the longitudinal data are the healthcare practice occupations, which comprise 6.1 percent of occupations in the longitudinal data and 5.4 percent of occupations in the OES data as a whole. Given the small overrepresentation of employment in the healthcare and social assistance industry in the longitudinal data, some overrepresentation of the occupational category associated with this industry is not surprising.

The OES program collects information on wages from each establishment surveyed. Employers are asked to report the number of employees in each occupation by wage range; employers select among 12 wage ranges. These wage ranges were constant from 1999 through May 2005 and were then adjusted in November 2005, November 2008, and November 2013.²² Figure 4 shows nominal wage distributions for private sector establishments in the longitudinal sample compared with wage distributions from all private sector establishments in the OES survey (including those with imputed wage data), in groups of panels with common wage ranges. For example, wage data for all panels from fall 1999 through May 2005 are combined in the upper left wage density plot, including multiple observations from establishments in the longitudinal sample if they responded more than once during this period.

Overall, the wage distributions for private sector establishments in the longitudinal data are quite similar to wage distributions in the private sector of the OES data as a whole. However, the patterns of differences between these distributions vary over time. In the panels of 1999–May 2005, establishments in the longitudinal data had a larger percentage of workers earning $6.75–$16.99 per hour and a smaller percentage of workers earning less than $6.75 per hour than the OES data as a whole. Similarly, in the panels of November 2005–May 2008, establishments in the longitudinal data having a larger percentage of workers earning $9.50–$30.99 per hour and a smaller percentage of workers earning less than $9.50 per hour than the OES data as a whole. Overall, for the panels of fall 1999–May 2005 and November 2005–May 2008, there was less employment in the very bottom of the wage distribution and more employment in the middle of the wage distribution in the establishments of the longitudinal sample than in the establishments of the OES data as a whole. Later panels show more similarity between the longitudinal data and the OES data as a whole, and the patterns of differences change. For the panels of November 2008–May 2013 and November 2013–May 2014, there was more employment in the lower portion of the wage distribution and less employment in the upper portion of the wage distribution in the establishments of the longitudinal sample than in the establishments of the OES data as a whole.

Research efforts and opportunities

Although the OES program is designed to produce estimates of occupational employment and wages for specific industries and geographic areas, assembling the microdata collected by this program longitudinally permits the study of many additional topics. Zachary Warren used data from establishments that responded at least twice to the OES survey from 2000 to 2005 to examine changes in employment by occupation in growing and shrinking establishments.²³ Dina Itkin used data from establishments that responded exactly twice to the OES survey from 2000 to 2006 to examine changes in employment and wages by occupation in establishments that changed ownership between their responses.²⁴ Itkin and Laurie Salmon used data from establishments that appeared in the Mass Layoff Statistics Survey and responded at least twice to the OES survey from 1999 to 2008 to examine changes in employment and wages by occupation in establishments that were affected by mass layoffs.²⁵

Those interested in using longitudinal microdata from the OES program should be aware that the establishments responding two or more times to the OES survey are not designed to be a representative sample of establishments or of employees in the United States. The OES sample is not a simple random sample but is designed to minimize the variance of estimates produced for the nation as a whole, for each industry, and for each geographic area. Therefore, larger establishments, establishments in certain industries, and establishments in geographic areas with fewer establishments are more likely to be sampled repeatedly. These differences in the probability of selection for different types of establishments will be compounded when researchers examine establishments that are selected for the survey more than once. However, the use of permanent random numbers in the OES sample design increases the likelihood that establishments will be repeatedly sampled for this survey. Furthermore, response to the OES survey is not random, and patterns of nonresponse act in opposite ways to the intent of the sample design. Smaller establishments and establishments in geographic areas with fewer establishments are more likely to respond to the survey. Overall, the characteristics of longitudinal microdata from the OES are remarkably representative of the OES data as a whole regardless of whether one examines establishment size, geographic area type, industry, occupation, or wage range.

Thus, it is possible to use data collected from the OES to examine employment and wage patterns for establishments observed longitudinally—not only for the largest establishments that are most likely to be selected for this survey but for a broad range of establishment types.