Archive for October 24, 2011

How To Cherry Pick Data

In his post “Senate Republicans Block Targeted Jobs Relief for Teachers And First Responders“, Matthew Yglesias points out that “during the Obama years” private employment has rebounded while government employment has seen a “sharp contraction”.

Yglesias points to a couple of charts, but I’ve helpfully replicated his data set into a single chart, because that’s just the kind of guy I am.

As you can see, using January 2009 as our point of reference, private jobs have rebounded from a drop of 3.79% in 2010 to a drop of 1.63% in August (my data is slightly out of date, but good enough for gov’t work… get it?!?). Local gov’t employment has fallen 3.6% in that same time frame. I also added federal gov’t employment (which has fallen 2.75% since January 2009) for the heck of it.

In the comments section, Peter Schaeffer complains that Yglesias is cherry picking the data and points out that gov’t employment saw +10% gains in the decade leading up to the crash and 3-4% losses from the peak while the private sector saw slightly less than 5% gains in that time period and slightly more than 5% losses from the peak.

I thought that Schaeffer had a good point, but needed some visuals to drive it home, so I thought I’d show Yglesias’ jobs data in Schaeffer’s context.

As you can see, Yglesias’ data starts at a really handy place for his argument, since it begins measuring job losses and growth at a time when we had already seen drastic private sector losses, but no public sector losses.

Of course, the funny aspect to this data is that one could use it to say that President Obama is reigning in the public sector that George W. Bush let grow out of control. I think the only reason no one is saying this is because everyone on President Obama’s side would consider that a bad thing and everyone who opposes President Obama would consider that a good thing. Neither side really wants to attribute this trend to President Obama. In fact, President Obama is working actively to reverse this trend.

Ah, the little ironies of life.

Note: In the spirit of “never attribute to malice what can be explained by incompetence”, I wouldn’t be surprised if Yglesias unwittingly cherry-picked the data. “The Obama years” is a perfectly rational place to start looking at data and, if that was the only data you looked at, it would support his conclusion. On the other hand, Yglesias has always had a better grasp of the data than this particular post suggests, so I suspect he kind-of-sort-of knew that this was a cherry picked sample set but was OK with using it because it bolstered his argument.

How To Read Unemployment Reports

Every time a national unemployment report comes out, I tweet the many details from @politicalmath. Frequently I get a lot of the same questions, so I thought I’d jot down a quick summary on unemployment reports and numbers and where they come from.

There are 2 kinds of employment numbers, summarized here:

  1. Establishment Data (Current Employment Statistics or CES) – this survey covers 400,000 businesses and counts the number of payroll positions that are filled.
  2. Household Data (Current Population Survey or CPS) – this survey covers 60,000 households and counts the number of people who are employed and unemployed.

When an employment report comes out from the Bureau of Labor Statistics (BLS), they usually report:

  1. The unemployment rate, which is calculated using household data
  2. The number of jobs added, which comes from the establishment data

Sometimes this data can seem contradictory. For example, between March and  June 2011, we gained 290,000 jobs but the unemployment rate went up .4% (from 8.8% to 9.2%).

There can be a couple reasons for this. The first one is that, the “jobs added” number comes from subtracting last month’s establishment jobs number from this month’s establishment jobs number, but we never use either of those numbers to calculate the unemployment data.

Why?

Because the essence of the establishment jobs number is asking employers: “How many people work for you?” It gives a nice accurate number, but it doesn’t tell us anything about how many people don’t work for them. We don’t have any number on the unemployed, only a number for jobs.

For unemployment, we have to go to individuals and ask them: “Are you employed or unemployed?” Then we take the unemployed number and divide it by the total number of people who are in the labor force, which counts both the employed and the unemployed.

But even the differences between the establishment jobs number and the household jobs number can be big. According to the household jobs number (which is supposed to exclude farm workers and the self-employed), we had 139.6 million jobs in August 2011. According to the establishment jobs number, we had 131.1 million.

That’s a difference of 8.5 million jobs, and that kind pf spread is pretty normal. The variation changes a little month-to-month, but we could get a report of  jobs created from the household number and jobs lost from the establishment number. In fact, we saw something similar in August where the household number said we gained 331,000 jobs, but the establishment number said we gained 0.

So why is the establishment number reported?

Because the establishment survey is so much larger, more reliable and gives more consistent results. In the graph below , we can see that even though the establishment data counts fewer jobs, it is a less erratic count.

So… that is a quick explanation of the employment report. I dig into this data once a month, so I’m pretty familiar and I’m delighted to answer questions or explain in greater detail in the comments.