Should the unemployment survey be updated?
posted at 1:01 pm on October 6, 2012 by Jazz Shaw
Once upon a time – back in the “good old days” when jobs were considerably more plentiful – the monthly unemployment figures released by the Bureau of Labor Statistics (BLS) were barely noted by the majority of Americans. (Assuming, that is, that they were aware of them at all.) If unemployment is down around five percent and it goes up to 5.2 in July and down to 4.8 in October, nobody notices. As Ed Morrissey pointed out yesterday, such shifts can be accounted for by seasonal variations in labor demand, localized effects in particular industries or, as we probably saw this week, an outlier in the BLS home survey.
But these days the statistics draw a lot of attention – and heat – particularly in the run up to a national election. Depending which way they shift, the published number becomes a headline for one party and a headache for the other. So can they be be improved to more accurately reflect reality on a consistent basis? And perhaps more importantly, would it be worth the effort to do so?
For a quick review, this Policymic article provides an excellent breakdown of how the government arrives at the published number each month.
The government releases two big employment surveys on the first Friday of every month. The first is the Current Employment Statistics survey, normally called the “payroll survey.” CES surveys 140,000 business and government agencies nationwide (except Puerto Rico and other U.S. territories). It notes number of employees, hours worked, and salaries earned. It also logs employment by gender and whether the positions are full or part time with some specificity.
Also important is what the CES doesn’t count. CES omits not only contract workers – like me – but also farm workers, many of whom are migrants. Interestingly, CES also excludes workers on strike through the 12th of a month. For example, the latest unemployment number for Chicago could be nearly 30,000 people higher because the Chicago Teachers Union went on strike between September 10th and 19th.
The “household survey” (or Current Population Survey) is conducted by the BLS and the Census Bureau, together with state government agencies. The survey calls or visits around 60,000 households and asks to speak with the head of household (over 16 years old). After surveying whether a person is working or seeking work, CPS asks supplementary questions covering things such as tobacco use and voting patterns.
As noted, the payroll survey may be a little better over the long run, but it has some problems of its own. It ignores a number of segments of employees entirely and is still, after all, a survey, with its own possibilities for anomalous numbers. Further, it is only able to speak to the number of people actually employed at any given time, regardless of population figures, how many are looking, retired, ill, etc.
But the household survey is clearly fraught with even more problems, and improving the quality of the data appears to be a daunting task. First of all, as Ed pointed out, they are only surveying 60K households in a nation containing hundreds of millions of people. And we’re not just talking about a slice of the population the size of “likely voters” here. They need to estimate statistics for every adult in the nation who has or is looking for a job. From my recent interviews with pollsters, I can tell you that this sample size and methodology would have to be classified has having a huge margin of error in any honest analysis. But how could it be improved?
First of all, the survey is not just done by BLS workers. It’s conducted in conjunction with the census bureau. And how many people would they need to interview to produce more statistically meaningful results? Ten times as many? Polling professionals will tell you that it’s hard to get people to complete even a moderate size survey. Take a look at the current survey being used. The labor statistics section alone is more than 20 pages long. If you wanted to get a vastly larger sample, the government would need to be calling millions of people every month. It could turn an already bloated bureaucracy into an unmanageable behemoth.
And in the end, what would it get us? Let’s face it… even in the current climate, the only people who really care about this figure are interested in it for either politics or journalism. For the average American, how many other people do or don’t have a job pales in comparison to the question of whether or not they have one. Do you really want to pay that high a price for better data points?
Let’s face it… America isn’t a police state. (Not yet, anyway.) The government has no ability to track each and every adult and see who is getting up each day to go to work, where they go and how much they earn. And I doubt many of you would want them to have that power. Maybe we just need to accept the facts and educate voters to know that these numbers involve a lot of guesswork, smoke and mirrors. We can watch the long term trends and probably get a general idea as to whether employment is getting better or worse over time, but the month to month numbers are, in the end, little more than fodder for yet another political parlor game.