May 1, 2015

Which Unemployment Survey is Better?

As a follow up to my post about the two employment surveys, it is often asked which report is better. The CES seems to be conventionally preferred. And research by George Perry, among others, finds evidence of a preference among traders and central bankers for the CES estimates. The CES surveys businesses, rather than households, to measure non-farm payroll employment, and has a larger sample size and less monthly volatility than the CPS. 

But the CES has the added complication of needing a representative sample of businesses, which are created and fail all the time. The BLS gets around this by adjusting the sample periodically using Unemployment Insurance tax data, which is collected for nearly all payroll employees. They also use models to predict business creation and failure to inform sample adjustment. The answer of which one is better is “it depends”.

Both surveys overlap in that they produce employment, hours, and earnings data, but they also measure things that the other survey does not. The CPS measures the size of the labor force, and its composition. It also collects demographic information on the employed and unemployed, whereas the CES does not. That data is needed for calculating the unemployment rate. The CPS does not exclude the self-employed or farm workers.

It also depends on how quickly data is needed. Much of the volatility of the CPS diminishes when averaging across quarters or years. Perry finds that when using quarterly employment data to track changes in GDP, an average of the two series performs better than either on its own. William Wascher uses the first month’s employment report for each quarter to get at monthly accuracy[1]. He finds that an average of the two series does as well as the CES on its own, and both do better than CPS on its own[2].

So it depends on what data is needed, over what time period, and how soon after release it is needed. But if I had to pick which survey is best for estimating what last month’s change in employment was, I’d argue for the CES. If you want a more holistic look at the labor market over any longer period of time, either survey by itself leaves you with an incomplete picture.






1. The output data used, which needed to be collected independently of the CES and CPS were only available quarterly


2. Scroll down to the bottom of the Perry paper to find Wascher's comments

No comments: