Collecting the data on jobs and unemployment is a complex and inexact science, but one thing they’re not is corrupt
Friday morning, the Bureau of Labor Statistics (BLS) reported that the United States added 114,000 new jobs in the month of September and the unemployment rate dropped to 7.8%.
The addition of a little over 100,000 jobs would not be enough to move the unemployment rate down from 8.1%, but the 7.8% rate was based partly on a total of 873,000 jobs added. So how is that we have these two jobs figures? And why is it that the 873,000 figure is what is counted for the unemployment rate?
The charge from some quarters of the conservative blogosphere is that the figures were being manipulated by the Obama administration to lower the headline unemployment rate? But let’s look at the facts.
The Current Employment Survey is where we get the 114,000 number of new jobs created. With the CES, it’s possible that people who already have one job might have gotten a second one. It’s a survey of businesses, so we shouldn’t expect the businesses to know about what their employees do outside of each firm’s work hours. That’s why it isn’t used for the unemployment rate.
The BLS asks businesses to voluntarily take part in the survey. Some businesses will refuse. Either way, it’s important to note that not all businesses are included in the survey. Every month, therefore, the survey must be weighted to ensure that the businesses that do submit reports are representative of all businesses. This is done from statistics collected yearly from hard data reported by employees to the government.
The problem, of course, is that any survey, no matter how well done, will always have survey error. That’s the reason that we have initial reports and then revisions. The errors have two-fold causes. Not all businesses report their data even if they are part of the survey. As more businesses report and more hard data is acquired by the BLS, the bureau gets a better idea of where the true employment figures stand. We also have a yearly revision, as a new benchmark is created when all businesses, regardless of whether they take part in the survey, are required to report their employment statuses to state governments. This can be combined with a seasonal adjustment to take into account jobs added because of, say, ski resorts in the winter.
These errors are typically not large in percentage terms, but the numbers themselves can look large. The 2011 change in the benchmark of +162,000 jobs (spread over the 12-month period) was only a change of +0.1%. For the monthly survey, the margin of error is a seemingly staggering 100,000 jobs at 90% confidence. Put another way, the 114,000 new jobs could actually be 14,000 or 214,000, and either would fall within the margin of error. That’s why, when I see an August revision of 96,000 to 141,000 jobs, I don’t blink an eye.
I also wouldn’t be terribly surprised if that 114,000 turned into a net job gain of -1,000 or +230,000 in the months to come. Ninety percent confidence means that 10% of the time, we can expect the first report to fall outside the margin: it’ll happen.
The other part of the BLS report is the unemployment rate. The unemployment rate is derived from the Current Population Survey. As the name implies, this is a poll of people, not businesses. Every month, 60,000 households or about 110,000 individuals are surveyed. This sample is weighted by geography, urban/rural, and industrial/farm divisions. Additionally, each person in the sample is weighted to account for differences in age, race, sex, Hispanic ethnicity, and state.
The unemployment rate is determined by the number of people looking for jobs (that is, the people not employed, but in the labor force) according to the CPS. It’s very possible to see a drop in the unemployment rate because people stop looking for jobs. We have seen that in past months, but that ‘s not what happened in September 2012.
According to the CPS, the civilian work force jumped by 418,000, with the employment level in September rising by a staggering 873,000 – the largest gain since January 2003. Meanwhile, the unemployment level dropped by 456,000 during the same period. A seasonal adjustment is then applied to this data to reach an unemployment rate.
But how is it possible that the two surveys show such vastly different results in terms of jobs gained?
First, the CPS includes people whom the CES excludes, and vice versa. The CPS includes the self-employed, private household workers (for example, nannies), unpaid family workers (for instance, the 17-year-old son working at the hardware store), and workers on leave. The CES can give a higher jobs report because an employee with two jobs can be counted twice, and because workers under the age of 16 are also included.
Second, the CPS has a far higher margin of error. The CPS is only interviewing 110,000 people out of the 240 million people potentially employed: that’s about 0.05% of the population it is trying to report on. The margin of error at 90% confidence is +/- 400,000. The CES, by contrast, is interviewing about 410,000 worksites and 140,000 businesses, when there are only about 30 million businesses in the United States.
When you take into account the differences in data collection and apply a historical adjustment, the CPS jobs gain is estimated at about 294,000 for September 2012 if adjusted to match CES criteria. That’s well within any margin of error.
Keep in mind that when we want to talk about the overall unemployment rate, we refer to the CPS because it “provides a broader picture of employment, including agriculture and the self-employed”. The unemployment rate itself only has a margin of error of +/- 0.2 of a point.
But is it possible that all of these numbers are a fraud? Can Obama really be getting at the BLS? That would require one of three things to happen.
First, Obama would to have to be in cahoots with businesses across this country to fabricate the number of employees reported all over the US – including many who reportedly didn’t like him two months ago. These businesses would also have to change their books if Romney were elected. But why the heck would a business want to promote a president if he was supposedly hurting their bottom line?
Second, people would have to be lying in the CPS. Again, this requires that people misreport their status to promote a president who is keeping them out of work. This would also imply that minorities, who typically support Obama in larger numbers, would also misreport their employment status in larger numbers, but we know that, in fact, they suffer the highest rates of unemployment.
Third, the BLS itself has been corrupted by the White House. If you think that’s plausible, I would suggest you read this piece by Eli Saslow from March of this year. The people who are compiling the jobs data are a bunch of dedicated, data-crunching nerds who live inside their own black box. These folks get the data from computers, close the blinds, and talk with no one before the jobs report is fully compiled.
There are simply no facts to support the thesis that the BLS data is cooked. Are we just not going to trust a number anymore because we don’t like it?