Monday, November 5, 2012

Did the Jobless Rate Go Up, Down or Sideways?

Recently, the Bureau of Labor Statistics reported that the September 2012 jobless rate had fallen to 7.8% from 8.1% and the Obama administration pointed to the decline as proof that their economic policies were working. That’s seems only fair, since Romney’s camp claimed that the increase from 8.1% in April to 8.3% in July proved the opposite.
Monthly Jobless Rates from the Bureau of Labor Statistics Website

Then, in October, though the reported jobless rate increased from 7.8% to 7.9%, the BLS stated, “Both the unemployment rate (7.9 percent) and the number of unemployed persons (12.3 million) were essentially unchanged in October, following declines in September.”

So, when do jobless rates actually go up or down? The answer is that the BLS doesn’t know for sure, because jobless rates are very hard to measure.

What does the BLS actually mean when they report October joblessness with a single figure, like 7.9%?

If you research the details and read the fine print on their website, http://www.bls.gov/, you’ll find that what they actually mean is something like this:

We estimate the jobless rate by inferring it from random samples because it isn't practical to go ask every American every month if they have a job. We believe the jobless rate is somewhere around 7.9%. 

We’re 90% confident1 that the actual rate lies somewhere between 7.6% and 8.2%. In fact, we would bet that it wasn’t exactly 7.90%  and we would usually win that bet, if we could actually know with certainty what the jobless rate was (we can’t). We picked 7.9% because it’s in the middle of the range and so it minimizes our potential error.

10% of the time, we will be wrong and the actual jobless rate will be outside this estimated range, simply because inferring data from samples is imperfect. We could give you an estimate that was only wrong 5% of the time, or even 1% of the time, but the range of our guesses would be progressively and substantially larger.

We will modify our guess of the October jobless rate in November and December (after you have stopped paying attention to it) as additional data comes in. It wouldn’t be unprecedented for us to reduce our guess in the next 60 days by 0.7% or increase it by 0.6%2, so we might end up guessing somewhere between 7.2% and 8.5%.

When we finalize our estimate after 60 days, we will still only be guessing at the actual jobless rate within a range of about +/- 3%.

And remember that the number you’re comparing it to from September is also subject to this amount of uncertainty, so when the average drops 0.1%, we can’t say with any certainty whether that is more, less or the same as September’s rate.


For some reason, it's difficult to find the confidence interval for BLS' national jobless estimates. You'd think that would be an important piece of information to convey, like, "This election poll has a margin of error of +/- 3%." But, that would expose the broad range of uncertainty in the estimates.

BLS does, however, specify confidence intervals for regional jobless estimates as in the following table for September 2012. It's easier here to see that the BLS is actually estimating a range of possible jobless rates and not one number:

Let's put it in football terms, the universal metaphor. Before the play (September), the ball was somewhere between your team's own 30-yard line and your 35. After the play (October), the ball was somewhere between your team's own 31-yard line and your 36. Was there a gain or a loss on the play?

I'm not ragging on the BLS for its inaccuracy. They're providing the best information they can within the limits of statistical tools.

But, remarkably, when BLS estimates that in one month the jobless rate was somewhere between 7.5% and 8.1%. . .probably. . .and in the next month it was somewhere between 7.6% and 8.2%. . .probably. . . the news media shouts, “Unemployment grew!”

Maybe it did. But maybe it didn’t. We’ll never know for certain.





[1] The BLS website states that they commonly use a 90% confidence interval for all estimates.
[2] BLS-reported range of absolute corrections in the past decade.

No comments:

Post a Comment