Large impact of cross-sectional weight in employment changes

Hello, I am currently comparing employment changes with changes in its components over a 10-year period 2004 - 2014 using CPS data. I compare the change in total employment for each industry between April 2004 and April 2014 with the sum of the net changes of each of its components summed across all months between the two points in time.

Components include e.g., all individuals returning from a 8-month break minus all going on a break, all transitions to an industry minus all transitions from an industry or all changes from unemployed to employed minus changes into the other direction, among others (for each industry). In the end, total employment change in the two points in time is equal to the sum of its components/net changes summed across all months.

My goal is to analyze the effect of net changes from unemployment or not in labor force to employment ,the effect of industry transitions and the impact of young individuals entering the sample as well as old people exiting it on employment change. For all the calculations, I use the cross-sectional variable wtfinl. However, I also noticed that by far the largest impact comes from changes in the cross-sectional weight variable wtfinl for individuals that are employed in two consecutive months and remain in the same industry (this is one of the components). For most individuals, the cross-sectional variable declines from month-to-month. Thus, for most industries, the sum of this component across all months is very large and negative (in many cases, around ± 1000% of actual employment change).

Is there a reason why cross-sectional weight tend to decline from month-to-month and why they play such a crucial role for employment change, when summing it up across all months and individuals for each industry? Does it have to do with the sampling methodology?

Shouldn’t changes in industry employment be only explained by changes from unemployment/nilf into employment, by industry transitions as well as by sample changes (new individuals replacing those exiting the sample)? How can I explain the impact of those factors if by far the largest influence on employment change is just the change in weight for those that do not exhibit any change?

If I use headcounts and no weight, the change in employment over a decade is equal to the sum of monthly changes of all components. In that case of course, there are no weight changes as before and it works too.

Thank you very much!

I took a look at WTFINL and found both cases where the weight increases and where it decreases month-to-month. Chapter 2-3 of Technical Paper 77 describes the process of creating person weights. In summary the five steps include producing base weights under ideal survey conditions, adjusting for non-response, reducing variance for non-self reporting units, benchmarking estimates of the population, and composite weighting. This process is used to produce person weights for each survey month and for this reason they are likely to vary month-to-month due to nonresponse, demographic changes, seasonal effects, and other factors.

I’m not exactly sure how you’re finding that change in industry employment is driven by these weights. Perhaps you’re running a type of decomposition analysis? If so, a change in weights would in fact suggest a change in other factors you’ve mentioned such as shifts from unemployment/nilf into employment or industry transitions. The job of the weights is to make your results representative of the population. It could be that there’s no change in the number of people in the sample who have transitioned industries, but if these people are becoming more representative of the general population then it’s likely that the increasing number of people similar to them who are also transitioning industries are driving your results.