Unemployment rate drops to 9.0 as employment population data skew US jobs numbers

The U.S. jobs data this month is a bit screwy. I wanted to flag something I saw because I think it is significant. 

First, here is the headline. The U.S. Bureau of Labor Statistics reported an addition to non-farm payrolls of 36,000 in January and a reduction in the unemployment rate to 9.0%. These seasonally-adjusted data come from two separate surveys. The 36,000 number comes from the establishment survey i.e. from businesses and the 9.0% unemployment rate is measured via the household survey i.e. from workers.

The surveys are telling two stories. The establishment story is one of too few jobs created: 36,000. Expectations were for about 150,000 jobs added. And we really need at least 200,000 jobs added per month over the long-term to keep up with population growth. A robust number is 300,000-400,000 then. So 36,000 doesn’t really get it done.

The household survey is based on the Department of Labor calling people up and asking them if they worked, whether it was part-time or full-time. If they weren’t working, they were asked whether they were actively looking for a job. The result is that this sample is extrapolated to the full population, determining both the numerator (employed or unemployed) and the denominator (the total active labour force). In the last month, the number of employed increased by 117,000 on a seasonally-adjusted basis while the number of unemployed went down more than 600,000 and those not in the labour force went up over 300,000. This combination of factors led to the unemployment rate dropping dramatically to 9.0%.

I should point out that on an unadjusted basis, January is a weak month for employment so the actual unadjusted data showed a loss of over 1 million jobs after population control adjustments.

I know this sounds a bit flaky. The truth is that the stats guys need to seasonally-adjust the numbers because the labour market is very seasonal. It makes a July to January comparison possible when it isn’t using the unadjusted data.

Here’s the thing though. The month of January adds a new kink that I have highlighted, namely that the population controls change every January.  So December to January is not a very good apples-to-apples jobs data comparison. Take a look here:

The total estimated U.S. population went down nearly 100,000 in January under the change. (It’s same number unadjusted and adjusted). That has an impact since the denominator in the unemployment rate comes from the labour force population. Note, originally when I talked about this on Twitter, I erroneously used the January 2010 total population number of 236.8 million as the comparison to December’s number. The total population number for January 2011 is lower than December 2010, but not that much lower. Apologies.

The National Post’s Paul Vieira says the "[s]ame thing happened in [Canada] last week with jobs data revised downward due to pop count. But appears all recovered with Jan data". He recommends waiting until February revisions and discounting this month’s data. I agree. Paul also says that CIBC World Markets noted the lower population estimates used for this month’s survey were not backdated into December results. If December population was adjusted in a similar way, it would show a 589,000 job gain, according to CIBC.

Bottom line: the 9.0% number is a statistical aberration. In fact, the unemployment rate went down because of people dropping out of labour force. That’s not good at all since the economy is in a recovery. Combined with a lower than expected addition of 36,000 to non-farm payrolls, it says the jobs picture is still weak. Let’s look to February for more information.