Using contingent worker supplement for comparison

I would like to understand who is currently working in contingent employment, as opposed to more traditional types of work. I understand the questions from the contingent worker supplement were only asked for a subset of respondents. Is it reasonable to compare those in the CWS to the rest of the respondents in the May 2017 sample. More specifically, can I assume the rest of the sample are not contingent workers, or were they simply not asked those questions so we do not know if they have similar working arrangements?

The documentation from BLS and the Census Bureau isn’t explicit about who is NOT included in the contingent worker supplement. To be included, respondents must be a civilian, aged 16 and older, in the right rotation group, and employed for pay. People without responses to the supplement include both persons with “Not interviewed” and “Unsure if eligible” values for CWSUPPINT; however, it isn’t necessarily clear why “unsure if eligible” respondents were not interviewed for the supplement as they appear to meet the stated requirements.

Due to this uncertainty it seems that counting people without responses to the supplement as “non-contingent workers” will not give accurate results. You should be safe in generating statistics about contingent workers by dropping those without responses to the supplement. The exceptions are 1999 and 2001 where there is still some question about the proper weighting. So I would be cautious with those samples.

The best measures of contingent workers we have are three possible definitions provided to us by CPS. These are the variables CWDEFNARR, CWDEFMED, and CWDEFBRD. There is a narrow definition, a medium definition, and a broad one. These variables identify who is considered a contingent worker in the CPS based on certain criteria laid out in the variable descriptions. More info on these definitions and the supplement can be found on the BLS website at this page.