A brief note on polling and unemployment data analysis
posted at 1:51 pm on October 5, 2012 by Ed Morrissey
I’m noticing a couple of odd reactions to today’s jobless data and the analysis it has produced so far today. On one hand, we have people proclaiming the difference between the household and establishment data as proof of a conspiracy to re-elect Barack Obama at the BLS. On the other, we have a number of voices calling anyone who applies any kind of critical analysis to these figures as “truthers” or “deniers.” Tom Elia has collected some of the latter at The New Editor, but I’ll focus on explaining what exactly the figures are, and why critical analysis does not amount to “trutherism.”
The BLS conducts two surveys each month to determine employment data. The first is the establishment survey, which polls 410,000 businesses each month. The second is the household survey, which polls 60,000 households each month. As one might expect, the larger survey provides a more stable series and more reliable data. The smaller one is still a very significant sample, but in a nation of around 150 million households, it’s hardly an exact science.
Any poll series can produce an outlier result. That’s true even of larger sample surveys, even when the sample is properly balanced, and even without malicious intent to tweak the results. It’s more likely to happen with smaller samples than larger samples, but can happen any time in surveys. That’s why it’s important to look at trending more than a single result within polling series, although some applications (jobless rates, elections) are intended more for single-result reporting.
In today’s case, the establishment survey showed a result that corresponds closely to other economic trends and that doesn’t deviate much from the intraseries trend. The household survey, from which the jobless rate is derived, showed a very large deviation from its own trending and from the growth data in the economy. The last time we had that many added in the household survey, the GDP growth rate was around 9%, and it’s currently 1.5%.
That’s why people who understand data and surveys look skeptically at the result of the household survey. It doesn’t mean a conspiracy is in place; it does strongly suggest that this month’s sample of 60,000 households threw an outlier, especially when compared with the establishment survey and other economic data. If so, it will likely correct itself in the next report. That’s not “trutherism” or denial, but straightforward data analysis.
Update: Just to remind everyone, the next report comes out before the election — the Friday before, actually. And if the BLS wanted to cook the numbers, I’m pretty sure they’d have cooked the establishment survey, too, to show more than +114K.