One question that could not be answered from the CPS and SIPP analysis is the impact of an economic downturn on the temporary help industry in general, and at-risk workers within that industry in particular. However, other data sources, particularly the CES, provide some clue.
A crude analysis of quarterly CES data from 1982:1 to 2001:2 suggests that temporary help employment is extremely responsive to GDP changes--a simple regression of the log of temporary help employment against the log of GDP suggests that the elasticity exceeds three. This is supported by a graphical analysis: the plateauing or downturns in temporary help employment exactly match downturns in economic activity. In addition, somewhat ominously, employment in this industry has taken a very clear downturn in the latest two quarters for which data are available.
These results are not unexpected. A study by Mangum, Mayall and Nelson (1985) noted that the temporary help industry was quite procyclical long ago, and this has been confirmed by Abraham (1990) and Houseman (2001).
What does this imply about the impact of an economic downturn on at-risk workers in temporary work? Several factors work together to suggest that they will be the first to be laid off. First, the low education levels of this group make them much more vulnerable than other workers. Second, the very nature of temporary work means that there are likely to be very low rates of job-specific skill acquisition, and hence that there are minimal firing costs to employers. Third, since all temporary workers are outside the standard employment relationship (Benner, 1997), there are few penalties associated with layoff. Finally, employers explicitly use non standard work arrangements to buffer against changes in the economy, so it should be expected that this sector will be disproportionately affected by an economic downturn.
All of this taken together suggests that there are likely to be substantial adverse consequences to at-risk temporary help workers in an economic downturn, although the data do not permit this to be quantified.