Were the September Unemployment Numbers Cooked?

Man, when the BLS announced on Friday that the U3 unemployment rate had declined from 8.1% to 7.8%, you’d have thought it was the show  of the century. The left hailed it as some momentous event, and the right either dismissed it, or even suggested the BLS was cooking the books.  So what’s the truth?

Well, it’s not momentous and it's not suspicious, either.  Useless, yes, as I explained in a previous post, but not momentous or suspicious.

1. It’s not  odd or unusual

• Since January 1948, the U3 unemployment rate has had a -.3% or greater decline 64 times.  That’s 8.2% of the time (64 of 777 months).  Yeah, some really rare event.  See accompanying chart.
• For some trivia:
  • The biggest increases and declines in U3 during this period came in consecutive months of 1949, when Harry Truman was president.  The rate went up from  6.6% in September to 7.9% in October (+1.3) and down again in November to 6.4% (-1.5).
  • The runner-up increase in U3 was under Eisenhower, when it went from 3.5% in November 1953 to 4.5% in December, 1953 (+1.0)
  • The runner up decline in U3 was under Reagan, when it declined from 10.1% in June, 1983 to  9.4% in July (-.7)

2. The U3 unemployment rate is determined by a SURVEY, and like other surveys, it is subject to sampling error, which is the error that exists in any survey that does not sample the entire population (well, actually sampling the entire population is termed a census, not a survey).

For reasons that truly escape me, the media does not report the confidence interval (or “margin of error”, as journalists call it) for the BLS survey statistics.  Why, I have no idea.  They do it all the time for public opinion polls.  Do they not understand that a survey is a survey?  What’s wrong with these people?

But fear not, we can easily approximate the “margin of error” ourselves.   As I discussed in some detail in a previous post, the BLS numbers are based on a survey of 60,000 randomly sampled households, with the employment status of each age 16+ person in the household being queried. With about two age 16+ people per household on average, this means about 120,000 people.  Now, if you randomly sample 120,000 people and ask them if they are employed, and 58.7% answer “yes”  (the September 2012 number, see here)  then the standard error for that sampling distribution will be approximated by the square root of ((.587 * (1-.587))/120,000), which happens to be .00142, (about .142%).  The 95% confidence interval will lie 1.96 times that number (.279%, I’m rounding here) either side of the observed value;  so the “margin of error”  for the survey is plus or minus .279%.   That may not sound like a lot, but when you consider the size of the number being estimated (total employment of  142,974,000 for September, 2012, see here) it works out to plus or minus 398,000 people. 

I should probably note that not every household sampled can be contacted and some refuse to answer and so forth, so the actual number of households polled is less than 60,000, but the point here is to show that like all survey estimates, the BLS data is subject to sampling error, the error is significant, and media people are remiss for not informing the public of it.

Also should observe that you can apply the formula above yourself to approximate the “margin of error” for public opinion polls.  So if you have a poll of 600 people that shows  50% of the people like Fig Newtons, the “margin of error” will be +- 4%.  For a poll of 1067 people, it will be +- 3%.  You may have noticed that the “margin of error” does not fall in a linear way as sample size increases, which explains why huge samples are rarely used – the extra cost does not justify the modestly increased precision.

3. Regardless of the employment and unemployment numbers, the fact remains that the employment to population ratio – which is the REAL employment rate, and the one preferred by anyone who studies economic growth – is still just where it was in the depths of the Great Recession.    This is a big problem, as worker productivity and the proportion of the population employed are what drive per capita income.   

4. Various people have suggested that the decline in the employment-population ratio (and also the related labor force participation rate) results from retiring baby boomers.  Several problems with this explanation:

• Being forced into early retirement is not a blessing.   These people are saying, “if you are old and want to work, you don’t count.”
• Labor force participation has fallen among all age groups except the elderly.  Ezra Klein had a good article on this a few months ago and no, the numbers have not changed notably since then. 
• While a decline in labor force participation has been expected owing to an aging population, the decline has been much more abrupt than anybody expected,  alarmingly so, in fact. 
• And to make matters even worse, the percentage of the workforce working part time is at sustained historic highs.  Yes, the BLS counts you as "employed" even if you are working lean part time.  See second chart.
• In the final end, it really doesn’t matter what the reason is anyway – it’s bad news.  Unless, like some people, you want the United States to end up like Greece or Spain, in which case it is good news.