I don’t mean to get all statistical on you this early in morning, but I always like to think about the confidence intervals (statistical reliability) around these employment numbers.

For the payroll survey, the 90% confidence interval for the monthly change in jobs is about 100,000; for the Household survey, it’s about 400,000(!).

The means that monthly numbers at or below those levels are statistically indistinguishable from no change at all (more precisely, 90% of the time the true value will be plus or minus 400K of the estimate; the fact that the interval is larger than the estimate means that the estimate is statistically indistinguishable from zero (hattip: S)).

I raise this today because I suspect we’ll hear some people get all jiggy about the growth of employment in the Household survey of 277,000. But that’s well below the confidence interval of 400K (the sample size is a lot smaller in the Household survey; ergo, the wider confidence interval).

That doesn’t mean there’s no information in the headline numbers for months like October, where the change in both employment numbers is statistically insignificant (as is the change in the unemployment rate, btw, from 9.1% to 9%; on the other hand, the increase in private sector payrolls of 125K is significant). If you average over a bunch of months, you’re essentially increasing the sample size and that gives a more reliable read (which confirms the slog I discuss in my first post).

But let’s not forget our stats 101, America!

(BTW, kudos to EB for correctly predicting the unemployment rate today…and she’s only 12!).

Why a 90% confidence interval rather than the much more common 95% interval?

Thank you. It’s nice of you to cover statistical aspects every now and then.

One small point re interpretation of conf intervals. It’s not correct to say that “90% of the time the true value will be plus or minus 400K of the estimate…” (see 3rd para). The true value is fixed (unless we talk Baynesian theory)so the true value is +/- 400K or it is not, that is, the chance is either 100% or 0%. Rather, the correct interpretation is that when we construct a large number of these intervals using different samples we can say that 90% of the time the interval will cover the true value. It’s a common misinterpretation of the meaning of classical conf intervals.