Unemployment Calculations (Updated)
I‘ve gotten some questions about how I do the unemployment calculation every month, and the wide variance between my rate and the official rate. It’s quite simple, although there are some caveats to the data, which I’ll cover in a methodology discussion below.
First of all, the data is all available from the Bureau of Labor Statistics, here. This is the retrieval page for the historical “A” tables of the employment report. You only need to retrieve historical data, in the following series: Civilian noninstitutional population, Participation rate, and Employed.
You have a choice, by the way, of choosing seasonally adjusted data or not. Seasonal adjustments smooth the numbers a bit from month to month, but not enough to be a major concern. There are pros and cons to the seasonal adjustments, but I’m happy either way. I use non-seasonally adjusted, so there’s more month-to-month variation, but it smooths out over longer time horizons anyway.
The BLS actually creates the employment/unemployment series from two different statistical surveys. One is the Household Survey, which asks households who is employed, who’s looking for work, and who has dropped out of the labor force. This is the series used to calculate the unemployment rate. The second series is the Establishment survey, which asks businesses how much hiring and firing they’ve done. This gives us the number of non-farm payroll jobs that have been created. It generally leaves out the self-employed, agricultural jobs, households, etc., so it doesn’t tell you much about unemployment. It mainly tells you about the rate of job creation. So, to calculate unemployment, we really only need to look at the Household survey’s historical data.
The first step is to calculate the historical labor force participation rate. This is complicated, conceptually, although not technically. All you have to do, technically, is download the participation data in excel, and run an average between two dates. Conceptually, you have to try and figure out what good dates are. There are…issues with this.
You don’t want to go back too far in time, because you are trying to capture the current labor force’s participation, not the participation of your dad’s generation. Labor force participation rates change over time, so the numbers need to be relatively current. I really didn’t want to project the numbers back into the 90s, for example, when the participation rate was solidly above 67%.
You also want the time frame to reflect at least one full economic cycle, so you can capture all the variation between an expansion and recession. But, you don’t want to choose an end date in the current economic cycle, because that skews the data up or down depending on where you are in the current economic cycle.
The dates I chose are January 2000 to Dec 2009. That takes data from right after the peak of the 90’s expansion, to right before the steep decline in labor force participation in the current recession. That’s where I get the 66.2% historical labor force participation rate. I could now include 2010 in that rate, which would introduce a slight downwards bias to the historical rate, but not much, yielding a participation rate of 66.1%. If I drop the rates from 2000, and go with a 10-year moving average (2001-2010), it drops to 66%. But, of course, that means that we’re including the current decline in participation, which hides, to an extent, how steep the decline actually is.
Now, there is a big question mark that is really impossible to address at the current time, which is whether or not the current decline in labor force participation is skewed by the Baby Boomer retirements which have begun as the first-year cohort of the Baby Boom hits 65 this year. The logical supposition is that such a large bolus of population retiring and passing out of the system will cause the participation rate to decline. How big of a decline? I dunno. We’ll really only know the answer to that question at the next peak of economic expansion, when the participation rate hits a new cycle high. I think it’s already started, though, and, indeed, started in the early 2000s, when the participation rate dropped a full percentage over several months, and then stayed in the 66% range, vice the 67% range of the 1990s. I think–though I can’t be sure–that we’re seeing a fair amount of early retirements among Baby Boomers who are affluent enough to do so.
The upshot of all this is that the selection of dates for calculating the historical average participation rate is very subjective. My calculation is, therefore, arbitrary, although, I think, logically reasoned out. It has a long enough time-line to be a reliable average across an economic cycle. It is not so far in the past that it skews the data. It is not so recent that current declines–or advances–skew the data. But it is arbitrary, and I’m sure others could come up with other ones. And, of course, when we hit another economic peak, the whole thing will have to be recalculated again to catch all those Baby Boomer early retirements.
In any event, once you’ve got the historical labor force participation rate, then all you need to do is multiply that by the civilian adult non-institutional population to derive the size of what should be the current labor force.
You then divide that into the size of the “Employed” population to come up with the unemployment rate.
The equation for all this would be:
1. Population x Participation Rate = Labor Force,
2.- (Labor Force/Employed) + 1 = Unemployment Rate
So, using this last month’s unemployment figures:
238704 * .662 = 158022
-(139323/158022 ) + 1 = 11.8%
And that’s how it’s done…assuming you correctly set up your Excel spreadsheet. As I was writing the formulas above, I noticed that the Excel spreadsheet had the division backwards, and was inflating the unemployment rate. I’ve corrected the post below on the Jan Unemployment Situation.
While I learned about figuring the average, all it does is pique further interest in determining the number of individuals who are underemployed or have simply given up and have dropped off the statistical graph.
How would you address those two questions and solve for both?
First, we already have a measure on unemployment that covers under-employment/part-time work, called the U-6, which already part of every monthly release. The BLS already does that work about as well as can be expected, so there’s no reason for me to reproduce it. As for those who’ve just dropped out of the labor force, we already see that on a monthly basis via the participation rate, and the labor force size that the BLS reports.
What we don’t know is why people are dropping out of the labor force. The “A” tables keep tabs on the number of discouraged workers, but we don’t have any read on who just decided to retire, or whose spouse got sick and needs them home to care for them, or who just thought taking up mainlining China White would be a good way to pass the time, or whatever.
Ultimately, if they’re out of the labor force, they’re no longer of concern to the statisticians at the BLS. What it really boils down to is how you define the “labor force” and that’s a pretty subjective measure, no matter how you cut it. The BLS has decided that if you’re not employed or actively looking for work, you’re not in the labor force. I’m sure there’s any number of people who would be in the current labor force under different circumstances, or who’ve been in it in the past, and who will be in it in the future. But I don’t know how you’d inquire into those myriad of reasons people aren’t in the labor force this month, and come up with a way to quantify that.