Longitudinal Employer-Household Dynamics

Methodology

Veteran Employment Outcomes (VEO) are experimental tabulations developed by the Longitudinal Employer-Household Dynamics (LEHD) program in collaboration with the Department of Defense and state agencies. The most recent v2.0 release of VEO data provides earnings and employment outcomes for veterans of the Air Force, Army, Coast Guard, Marine Corps, and Navy. Earnings and employment outcomes are calculated by rank and military occupation, as well as veteran and employer characteristics. VEO are currently released as a research data product in "experimental" form.

The VEO provide data on earnings and employment for recently discharged military veterans. Earnings are available at the 25th, 50th, and 75th percentiles, one, five, and ten years after separation from active-duty service, by service branch, rank, occupation, and discharge cohort. Selected tables include industry and location of employment for veterans. These statistics are generated by matching veteran records with a national database of jobs.

The VEO use cutting-edge differential privacy methods to protect the confidentiality of the underlying data, a protection method developed in computer science to bound the privacy risk to individuals from multiple queries to the same database. Differential privacy methods allow the Census Bureau to release detailed tabulations on veteran outcomes while minimizing the privacy risk to individuals in the data.

top

Data Sources

The source of veteran information in the VEO is administrative record data from the Department of the Defense, Defense Manpower Data Center (DMDC) and the Military-Civilian Transition Office (MCTO). This personnel data contains fields on service member characteristics, such as service start and end dates, occupation, pay grade, characteristics at entry (e.g. education and test scores), and demographic characteristics (e.g. sex, race, and ethnicity). Once service member records are transferred to the Census Bureau, personally-identifying information is stripped and veterans are assigned a Protected Identification Key (PIK) that allows for them to be matched with their employment outcomes in Census Bureau jobs data.

Veteran Population Coverage

The VEO data covers all service members who have been honorably discharged between 2002 and 2021. We include service members from the Air Force, Army, Coast Guard, Marine Corps, and Navy. Space Force service members are not included due to the availability of only a small number of veteran cohorts. We additionally restrict our sample to service members who have completed active-duty service as enlisted service members (not as commissioned officers or warrant officers) and have final ranks E1-E9. Discharge rates for more senior military personnel are not sufficiently large to produce VEO statistics without adding significant statistical noise to protect those records.

Note that veterans are omitted from the earnings and employment outcome statistics when they have insufficient labor market attachment in the reference year. For example, a veteran with zero earnings for three quarters of the calendar year but positive earnings in a single quarter will not be included in the earnings statistics or employment counts. These individuals are omitted as the VEO are intended to reflect earnings and employment for veterans who work throughout the year. More specifics on the labor force attachment restrictions are provided in the earnings section.

Employment Coverage

The LEHD data at the U.S. Census Bureau is a quarterly database of jobs covering over 96% of employment in the United States. The core jobs data is state unemployment insurance (UI) wage records collected via a voluntary federal-state data sharing partnership. These job records are then supplemented with Census Bureau surveys and other federal agency administrative records to supply additional information on the characteristics of workers and employers. This linked employer-employee data for the U.S. is the source data for Census Bureau's Quarterly Workforce Indicators (QWI), LEHD Origin-Destination Employment Statistics (LODES), and Job-to-Job Flows (J2J). More information about the LEHD data is available in Abowd et al. (2009).

Private-industry employment: : Covered private-industry employment in the LEHD data includes most corporate officials, all executives, all supervisory personnel, all professionals, all clerical workers, many farmworkers, all wage earners, all piece workers, and all part-time workers. Workers on paid sick leave, paid holiday, paid vacation, and the like are also covered. Workers on the payroll of more than one firm during the period are counted by each employer that is subject to UI, as long as those workers satisfy the definition of employment (see below). Workers have UI wages filed in every quarter they are covered.

Notable exclusions from UI coverage among private sector employers are independent contractors, the unincorporated self-employed, railroad workers covered by the railroad unemployment insurance system, some family employees of family-owned businesses, certain farm workers, students working for universities under certain cooperative programs, salespersons primarily paid on commission, and workers of some non-profits. States have some leeway in designating coverage; for a complete list, see the coverage section of the most recent Comparison of State UI laws.

State and local government employment: Covered employment in the LEHD data includes most employees of state and local governments with the exception of elected officials, members of a legislative body or members of the judiciary, and some emergency employees.

Federal government employment: Federal government workers are not covered by state UI. LEHD uses data from the Office of Personnel Management (OPM) to generate earnings and employment histories for federal workers. The OPM data covers most federal employees but excludes White House officials, members of Congress, and certain national security agencies, which are excluded for security reasons. Members of the armed forces and the U.S. Postal Service are also not covered in OPM data. The OPM data has coverage for 2000-2022.

UI coverage across years: Availability of state UI data in the LEHD system varies by state. LEHD has data for only ten states in the early 1990s, expanding rapidly to 40 states by the late 1990s.Massachusetts was the last state to enter the system in 2010.

Earnings, and Employment Concepts

Earnings are total annual earnings for attached workers from all jobs, converted to 2022 dollars using the CPI-U. For the annual earnings tabulations, we impose two labor force attachment restrictions. First, we drop veterans who earn less than the annual equivalent of full-time work at the prevailing federal minimum wage. Additionally, we drop veterans with two or more quarters with no earnings in the reference year. These workers are likely to be either marginally attached to the labor force or employed in non-covered employment.

While most VEO tabulations include earnings from all jobs, tabulations by employer characteristics only consider the veteran's main job for that year. Main jobs are defined as the job for which veterans had the highest earnings in the reference year. To attach employer characteristics to that job, we assign industry and geography from the highest earnings quarter with that employer in the year. For multi-establishment firms, we use LEHD unit-to-worker imputations to assign workers to establishments, and then assign industry and geography.

top

Employment and earnings outcomes are tabulated by the following veteran characteristics: discharge cohort, sex, race and ethnicity, education at enlistment, AFQT score range, pay grade at separation, and military occupation. They are also tabulated by the state and industry of each worker's dominant employer. All outcomes are additionally reported one, five, and ten years post separation from the military.

In order for changes in veterans' labor market outcomes to be evaluated over time, we group veterans into cohorts based on their year of separation from active-duty service. For most tables, we use 2-year cohorts. However, for some tables, data quality constraints limit the extent to which we can provide detailed cohort groups. As a result, these tables are aggregated to 4-year and 8-year cohorts. Statistics for the following separation cohort bins are therefore generated:

• Two year bins: 2002-2003, 2004-2005, 2006-2007, 2008-2009, 2010-2011, 2012-2013, 2014-2015, 2016-2017, 2018-2019, and 2020-2021.
• Four year bins: 2004-2007, 2008-2011, 2012-2015, and 2016-2019.
• Eight year bins: 2008-2015.

Worker demographic information allows for comparison of employment and earnings outcomes by age, sex, race, and ethnicity. Age is defined as age at separation from service and is taken from DOD records. We then create two age groups: age <25 and age 25+. Sex, race, and ethnicity are taken from the LEHD data to allow for representation of race and ethnicity information in a manner that is consistent with the format used by the U.S. Census Bureau, presented in the LEHD schema available at https://lehd.ces.census.gov/data/schema/latest/lehd_public_use_schema.html. In the small number of cases where demographic characteristics are not available from the LEHD data, we use DOD records and recode variable values as necessary to maintain consistency. Sex has two bins: Male and Female. Race has six categories: White Alone, Black or African American alone, American Indian or Alaska Native Alone, Asian Alone, Native Hawaiian or Other Pacific Islander Alone, and Two or More Race Groups. Ethnicity has two bins: Hispanic and Not Hispanic.

Eligibility for military enlistment depends on meeting certain education thresholds. As a result, nearly all service member records include their education level at time of enlistment. We use DOD administrative data to generate three categories of education level: General Educational Development (GED) Test, High School Diploma, and Some College or Higher. In some service branches, education attainment is missing for some earlier cohorts. In these cases, we create a separate "unknown" category.

All recruits take the Armed Forces Vocational Aptitude Battery (ASVAB), which is used to assess their skills at time-of-entry and their fit for any particular military occupation. A subset of the ASVAB is used to calculate the Armed Forces Qualification Test (AFQT) score. This score is reported as a percentile that is relative to a reference group and is comparable across time. We report employment and earnings outcomes by AFQT score ranges, with 1 corresponding to scores 0-33, 2 to 34-66, and 3 to 67-100. In some service branches, AFQT score is missing for some earlier cohorts. In these cases, we create a separate "unknown" category.

We use pay grade at separation to capture each service member's performance during active-duty service. Due to sparsity of cells, some pay grade categories are aggregated into larger bins. Reported pay grade bins include: E1, E2, E3, E4, E5, E6, and E7-E9, with E1 being the pay grade for Privates and E7-E9 being the pay grades for senior non-commissioned officers. Note that when labor market outcomes are reported by pay grade crossed with military occupation, we use a higher level of aggregation and report two grouped pay grade bins, E1-E5 and E6-E9.

We use three bins to capture the distribution of tenure for active-duty service at year of separation: 0-5, 6-19, and 20+ years. Note that most enlisted service members serve less than five years and career personnel are eligible for retirement at 20 years of service.

Occupation for enlisted personnel is typically defined by a Military Occupation Specialty (MOS) code. MOS code usage varies across service branches and over time as new occupations are created and old ones are eliminated or reorganized. To maintain consistency across service branches and time, we aggregate MOS occupation codes to the Department of Defense's Military Occupational Specialty Classification codes at the 2- and 3-digit levels.

Employment and earnings outcomes are available for each of the 50 states and the District of Columbia. A worker is assigned to a given state if their dominant employer for the calendar year paid UI compensation for that worker in that state. For federal employees, we use the location of the government agency to establish employer geography. States are identified by their Federal Information Processing Standard (FIPS) state code.

We provide statistics by the industry of the dominant employer, at the North American Industrial Classification System (NAICS) Sector level with federal government employment reported as a separate category.

top

The protection system for VEO takes into account that external parties have access to a large portion of the data. To address these disclosure issues, the VEO data product uses differential privacy techniques to protect individual confidentiality. We follow the procedure described in Foote, Machanavajjhala, and McKinney (2019) and Graham et al (2024).

In the VEO tabulations, we release three percentile values (25th, 50th and 75th) and a cell count. To protect the earnings percentiles for a given cell, we categorize the earnings of all individuals into pre-defined histogram bins. We then add noise according to the geometric mechanism to each bin. We use these counts to construct an empirical CDF, from which we calculate the percentiles. We also calculate the protected cell count from the sum of the bin counts. For cells with a protected count of less than 50, we suppress output (due to low quality) and indicate the suppression in the data with a status flag = 5.

top

[1] John M. Abowd, Bryce E. Stephens, Lars Vilhuber, Fredrik Andersson, Kevin L. McKinney, Marc Roemer, and Simon Woodcock. The LEHD Infrastructure Files and the Creation of the Quarterly Workforce Indicators. In Producer Dynamics: New Evidence from Micro Data, NBER Chapters, pages 149-230. National Bureau of Economic Research, Inc., September 2009.

[2] Foote, Andrew David, Ashwin Machanavajjhala, and Kevin McKinney. 2019. Releasing Earnings Distributions Using Differential Privacy: Disclosure Avoidance System for Post-Secondary Employment Outcomes (PSEO). Journal of Privacy and Confidentiality 9 (2).

[3] Graham, Matthew, Andrew Foote, Lee Tucker, and Hubert Janicki. 2024. Differential Privacy Implementations. In J. Drechsler, D. Kifer, J. Reiter, A. Slavkovic (Eds) Handbook of Sharing Confidential Data (pp. 259-278). Chapman & Hall/CRC.

top