Latent Class Analysis of Worker Knowledge of Their Employment Status
Stanislav Kolenikov
Abt Associates
Kelly Daley
The Worker Classification Survey conducted by Abt Associates for the Department of Labor in 2014 measured several concepts that are difficult for respondents to report accurately. These concepts include their own employment status as well as details about their employer, benefits, and tax forms. The survey also asked questions that some respondents may purposefully answer inaccurately. For example, certain response patterns may indicate that respondents are "working under the table" and not paying taxes on their income. We analyzed measurement error in this survey using latent class analysis (LCA), a statistical technique that links discrete latent variables, or discrete classes, with discrete responses. We have been able to identify respondents who clearly are employees of their companies; those who clearly are non-employees (e.g., contractors or consultants), as well as those whose status is unclear despite detailed collected self-reports on their documented status with the company, as well as their economic relations and job responsibilities