© Copyright 2003 by the Wyoming Department of Employment, Research & Planning
Methods for Imputing Work Status
by: Mark Harris, Sociologist, Ph.D., and Krista Gerth, Economist
contributors to study: by: Tony Glover, Senior Research Analyst; Sylvia Jones, Statistical and Research Analyst; and Dave Bullard, Senior Economist.
This article presents a methodology for imputing full- and part-time work status for individuals attached to firms appearing in the Employee Benefits in Wyoming: 20001 study. We first report on a cross-check validation of our data with other published sources. We then report on the results of our imputation strategy. Our results indicate that Research & Planning's (R&P) methodology appears relatively reliable for imputing work status. This allows us to present the profile of full- and part-time individuals who have exposure to benefits in Wyoming (see "Who Has Access to Employer-Provided Benefits in Wyoming?").
Data Information for this project comes from five different sources. Data on firms come from the Employee Benefits in Wyoming: 20002 (Benefits Survey) and the Bureau of Labor Statistics' Covered Employment and Wages (ES-202) program. Employment data for individuals working in these firms are from Wyoming Unemployment Insurance Wage Records (Wage Records). Demographic information on these individuals are from the Driver's License database provided to R&P by the Wyoming Department of Transportation. Data utilized in the study for weighting and cross-check validation are from the Current Population Survey (CPS)3. The CPS is a valuable comparison tool that provides estimates of work status, age, gender, and industry distributions for Wyoming's working population. We removed self-employed individuals from the CPS data to make it more comparable to our data. CPS data for the years 1997 through 2000 are combined for this study, although firm-specific data are restricted to those surveyed for the 2000 Benefits Survey.4
The 2000 Benefits Survey requested information from a total of 1,600 firms (400 employers each quarter of 2000). For a survey response to be considered valid in this study, employers were required to complete a survey and report at least one employee on quarterly Unemployment Insurance (UI) contributions reports for the survey reference quarter. For example, if an employer was surveyed during first quarter 2000 (2000Q1), we matched any employees reported in Wage Records for 2000Q1.5 This process, otherwise referred to as the Merged Benefits Data (MBD), resulted in 890 employers (55.6%) returning usable surveys. Together, these firms employed 50,341 workers.
Government agencies were over-sampled in the 2000 Benefits Survey (Table 1), resulting in overrepresentation of the MBD compared to the CPS (a difference of 16.4 percentage points). Additionally, a few of the large Retail Trade firms did not respond to the survey, reducing reported employment in this industry. Thus, Retail Trade is underrepresented (a difference of 8.5 percentage points). To correct for this, the data were weighted so that the MBD employment distribution would reflect the CPS distribution. The resulting data set, referred to as the Weighted Merged Benefits Data (WMBD), consisted of 50,6056 employees (Table 1).7
Comparison of Demographic Distributions
To validate the resulting age and gender characteristics for WMBD employees, we compared our distributions to those of the CPS. Both gender and age distributions were consistent with CPS data (Tables 2 and 3). CPS estimates show 52.5 percent of Wyoming workers are male and 47.5 percent are female (Table 2). The 2000 WMBD data produce similar results with 51.9 male and 45.2 female.8 The age distributions reported in Table 3 for the CPS and WMBD are also very similar. The two largest differences in WMBD age group data when compared to CPS data are an overestimation of the number of workers younger than 25 (1.9 percentage points) and an overestimation of the number of workers between 35 and 44 years of age (2.9 percentage points).
Given the favorable results for demographic comparisons, we are reasonably confident that the WMBD data are an adequate representation of employment in Wyoming by gender and age. The next step is to develop an imputation strategy for work status (full- or part-time) and verify its results against CPS estimates.
Employers report the number of full- and part-time employees as part of the Benefits Survey. We know that employees' work status often dictates whether they will have access to certain benefits. We also know the demographic profile of full-time workers is quite different from that of part-time employees. The crux of the project, therefore, was the problem of how to determine the full- and part-time status of each individual in the MBD.9 Unfortunately, Wage Records does not contain information such as the number of hours worked, so the data had to be imputed.
Imputation Strategy #1
The first imputation strategy was based on the number of full- and part-time workers each employer reported on the Benefits Survey. The majority of employers responding to the survey (89.0%) provided the data allowing us to calculate the distribution of full- and part-time workers in each firm. From Wage Records, we know the average quarterly wage for all employees based on each employee's entire work history (including the survey quarter) with the surveyed employer. Employees within each firm were ranked from highest to lowest average quarterly wage.
Using the percentage of full-time employees the firm reported on their survey questionnaire, employees were imputed as either full- or part-time (i.e., if a firm with 100 employees reported that 75 percent were full-time, then the 75 employees with the highest wages were assumed to be full-time while the 25 employees with the lowest wages were assumed to be part-time). The weakness of this strategy is that it assumes all full-time employees will have higher average quarterly wages than part-time employees. Although this may generally be true, there are certainly situations where part-time workers earn more than full-time employees (e.g., a part-time computer programmer may earn more than a full-time clerk).
Imputation Strategy #2
The second imputation strategy was used for employers (11.0%) that did not report the number of full- and part-time workers on the Benefits Survey questionnaire. For this strategy, we calculated the quarterly wage that would be earned by an employee working 35 hours per week at $5.15 per hour.10 This resulted in a quarterly wage of $2,343.11 Anyone who earned more than that amount was assigned to the full-time group.12 Again, the weakness with this strategy is that workers who earn more than $2,343 may not necessarily be full-time, but instead may work fewer hours at a higher rate of pay.
Work Status Imputation Results
As shown in Table 4, the CPS data indicate that our imputation strategies produce reasonably reliable estimates of the numbers of full- and part-time employees in Wyoming.13 The CPS estimates that 77.3 percent of Wyoming workers are full-time, and our imputation strategy estimates that 73.6 percent work full-time. Likewise, the CPS and our WMBD report the distribution of part-time workers in Wyoming at 22.7 and 26.4 percent, respectively.
Data presented here indicate that the WMBD data are a reasonably accurate representation of the gender, age, industry, and work-status profile of Wyoming workers. The Benefits Survey and our imputation strategies produced results strikingly consistent with other published data for Wyoming. The next step in the study is to analyze the characteristics of the employees exposed to benefit offerings. This will allow us to answer the question, "Who Has Access to Employer-Provided Benefits in Wyoming?"
Department of Employment, Research & Planning, Employee
Benefits in Wyoming: 2000
2Wyoming Department of Employment, Research & Planning, Employee Benefits in Wyoming: 2000 <http://doe.state.wy.us/LMI/benefits/bentoc.htm>.
3The Current Population Survey, a monthly household survey conducted by the Bureau of the Census for the Bureau of Labor Statistics, provides a comprehensive body of information on the employment and unemployment experience of the nation's population, classified by age, sex, race, and a variety of other characteristics. For more details, see <http://stats.bls.gov/cps/>. For this article, we use the March supplement.
4The Benefits Survey is an ongoing research activity at R&P. For the most recent data, see <http://doe.state.wy.us/LMI/benefits/bentoc.htm>.
5The survey reference quarter was the primary quarter for which we matched employees. However, because a few surveys were returned late, and employers did not always provide data for the quarter requested, we matched employees in the following quarter.
6The total changes slightly across the four tables presented here due to rounding.
7Weights were developed in the following manner. First, CPS proportions were calculated for the ten major industries. Second, the expected major industry distribution of employment for MBD data was calculated using CPS proportions for the total MBD (50,341). Third, expected MBD industry proportions were then divided by actual MBD proportions to arrive at a weight factor. To illustrate, the proportion of Mining in the CPS is .08. The expected n for Mining in the MBD, given a total of 50,341 and .08 CPS Mining proportion, is 3,983 (.08 * 50,341 = 3,983). The actual n for Mining in the MBD is 4,999. The weight adjustment factor is calculated as the expected n for Mining in the MBD divided by actual n for Mining in the MBD (3,983/4,999 = .797). This same process was repeated for all major industries.
8The WMBS data have 1,485 cases with missing demographic information. These individuals do not have a Wyoming Driver’s License and they have not worked in Wyoming long enough to be eligible for demographic imputation. For more details, see Tony Glover, “Enhancing the Quality of Wage Records for Analysis Through Imputation: Part One,” Wyoming Labor Force Trends, April 2001, pp. 9-12; Tony Glover, “Enhancing the Quality of Wage Records for Analysis Through Imputation: Part Two,” Wyoming Labor Force Trends, June 2001, pp. 1-6.
9Work status imputation strategies were conducted on the unweighted data. Data were then weighted for comparison to the CPS work status distribution.
10The 35 or more hour work week is consistent with the CPS’s definition of full-time status (see <http://stats.bls.gov/bls/glossary.htm#F>). However, this may or may not be consistent with specific employer definitions of full-time status. The current minimum wage is $5.15.
1135 hours x 13 weeks per quarter x $5.15 per hour = $2,343 per quarter.
12Tips are not accounted for in Wage Records. Tipped employees are, therefore, more likely to be among those imputed part-time.
13Full-time employees who worked only part of a quarter are more likely to be among those imputed part-time.
Table of Contents | Labor Market Information | Wyoming Job Network | Send Us Mail
designed by Julie Barnish.
Last modified on by Susan Murray.