In praise of the unemployment rate

It's not perfect, but still an extremely useful macroeconomic barometer

My first “real” post! I’m still experimenting with this format, so feedback is very welcome. - Guy

TL;DR: This post explains why the government-published unemployment rate is a useful barometer for the US labor market and economy, and a solid candidate for anyone who wants to follow “just one metric.”

In praise of the unemployment rate

A lot of times I get asked by folks some variant of “what indicator should I follow to understand what’s going on in labor markets and the economy?”

And the answer I give, almost invariably, is “the unemployment rate”.

It’s not a perfect indicator by any means – indeed, it has some well-known drawbacks – and sophisticated consumers of the data can do a lot better looking at a holistic range of metrics.  But the truth, unless you’re committed to becoming a macro data nerd, the unemployment rate will tell you most of what you need to know.

In the rest of this essay, I’ll cover 5 main areas: (1) what the unemployment rate is (2) how to use it (3) its drawbacks (4) its strengths (5) possibly superior alternatives.

1. What it is

The unemployment rate is a ratio of two numbers.  In the numerator, you have the number of Americans who are unemployed.  In the denominator, you have the sum of people who are employed and people who are unemployed.

“Unemployed” has a pretty specific definition (and maybe one that doesn’t perfectly align with how normies think about unemployment).  The definition: people who don’t have a job, but are actively looking for one; or people that have been temporarily laid off, but expect to be recalled to their job.  (Think of a town that’s hit by a natural disaster like a storm or earthquake, so people temporarily can’t work; this type of unemployment became very material during the early months of the COVID pandemic, but matters less in normal times.)  The Bureau of Labor Statistics has a specific set of criteria for what “actively looking for a job” means.  As of September 2023, there were 6.36 million unemployed people in the United States (and 161.6 million employed people).

That’s a definition that excludes the vast majority of folks who don’t have a job: students, retirees, stay at home parents, and also (importantly) people who want a job but aren’t actively looking for one.  As of September 2023, there were almost 100 million people in this “non-institutionalized civilians over the age of 16 who are neither employed nor unemployed” category.

For the unemployment rate’s main use case (“how well is the labor market doing”) excluding most of those folks makes sense, but obviously “people who want a job but are discouraged from looking for one” is an important phenomenon too and people have tried to get around its omission in the unemployment rate by constructing alternate metrics.  (I say more about this below.)

The data is collected by a government phone survey called the Current Population Survey (CPS); 59,000 Americans are called by phone on a monthly basis and asked a series of questions, including “do you have a job” and “if you don’t have a job, are you actively looking for one”.  The Census Bureau conducts the survey, and the Bureau of Labor Statistics (BLS) tabulates and publishes the labor force data from the survey.

One final point – sometimes folks erroneously assume the unemployment rate is related in some way to unemployment insurance.  (i.e. “the unemployment rate is artificially low because folks have run out of unemployment insurance benefits.”)  The two are distinct!  You do not need to be a recipient of unemployment insurance benefits to count as “unemployed”, though actively looking for work is (mostly) a necessary condition for both.

2. How to use it

I think of the unemployment rate primarily as a barometer.  It’s useful to know both the level (where it is) and change (where it is going to/from, and how quickly).

It’s good to have an unemployment rate that’s low (in level terms) and falling (in change terms), and bad to have an unemployment rate that’s high (in level terms) and rising (in change terms).  There are some subtleties here (you’ll occasionally hear folks saying something like “the unemployment rate went up, but for good reasons”) but if you’re a casual consumer of this data they don’t matter too much.  Look at the chart I published at the beginning of this piece, and how closely the data’s behavior aligns with historical NBER recessions and expansions.

It's important to remember that the level and change can point in different directions, and this difference can be meaningful.  For example, at the beginning of a recession the unemployment rate is usually low but rising.  (This leads to some not-useful arguments like “how can we be in recession?  The unemployment rate is low.”)  And after a recession ends, we’ll frequently see the opposite: the unemployment rate is high, but falling.

Finally, it’s useful to remember that the unemployment rate is noisy just like any other economic indicator.  Sometimes it wiggles around on a month-to-month basis and those wiggles are meaningless.  Not every small decline indicates an improvement in the labor market, and not every small increase suggests a recession is about to begin.

If you were to ask me what the unemployment rate is saying right now, I’d say it’s a mix of good (very low in terms of levels) and so-so (neither rising nor falling, though some small increases may be on the way).

There are also other more sophisticated uses of the indicator (you may have heard of the Phillips Curve, the Beveridge Curve, the Taylor Rule, or the natural rate of unemployment) but for most casual users it’s probably not necessary to worry about them.

3. Drawbacks

The unemployment rate has become a bit of a punching bag in the last 15 years.  It’s not perfect!  It misses a lot of important trends in the labor market and economy, and if you want a sophisticated/rich understanding of what’s going on rather than a basic/adequate one, you’ll want to look at other data.

For example, sometimes the unemployment rate overstates the labor market’s strength.  One obvious dimension I mentioned earlier is it doesn’t include folks who want a job, but aren’t actively looking (perhaps because they’re discouraged).  This was a pretty meaningful problem in the early-to-mid 2010s.

More recently, during the period dubbed as “the Great Resignation” (2021-22), you could make a strong argument that the unemployment rate did the opposite: it underrated how strong/hot the labor market was.  Other indicators like wage growth, job openings, hires and quits all suggested the unemployment rate “should” have been much lower than it actually was.

Another quirky case of the unemployment rate being an imperfect barometer came during the most acute period of the COVID pandemic.  There was an unusually high share of folks who didn’t have jobs because public health restrictions shut down their places of work; however, a vastly expanded safety net replaced labor income for many unemployed people, meaning that the “economic distress” aspect of unemployment was less relevant than usual.

The unemployment rate can also behave, in the short run (think 1-3 months), in ways that are counterintuitive.  It can “go down for bad reasons”: folks who were previously actively looking for jobs stop looking without finding a job.  It can “go up for good reasons”: folks without a job who weren’t actively looking for work, start looking.  In my experience, these kinds of counterintuitive fluctuations wash out in the data pretty quickly and tend to be small – i.e. a sustained increase in the unemployment rate is always bad, and a sustained decline is always good.

Another critique I sometimes hear about the unemployment rate is that it undercounts “underemployment”.  Underemployment is a pretty amorphous concept – people mean a lot of different things by it.  “People are working jobs for which they are underqualified”; “people are being underpaid”; “people are working fewer hours than they want” (the BLS publishes that last one on a regular basis).  I think this gripe has truth to it for sure, but I also think the unemployment rate is correlated enough with these phenomena that you’re not missing that much by using it.  (i.e if the unemployment rate is falling, underemployment is probably becoming a less severe problem; if the unemployment rate is rising, underemployment is probably becoming a more severe problem).

Finally, it’s worth mentioning that the unemployment rate is suffering from the same falling survey response rate problem as many other economic indicators collected by survey. This raises concerns about statistical bias.

4. Advantages

OK, I talked about the drawbacks!  But nevertheless I think this is a very valuable indicator.

First off, it’s fairly intuitive.  Employment is good and unemployment is bad because most families depend on labor income to support themselves.  You can use its level as a reasonable proxy of human well-being, and its change as a proxy for how the economy is trending.

Second, it’s widely used.  The BLS publishes a monthly report that includes it.  It’s covered widely by nearly every major American news entity.  The Federal Reserve uses it to set monetary policy, and includes it in its quarterly forecasts.  As a casual reader of the news, you don’t have to spend much time digging for stats in obscure BLS tables. (I spend a lot of time digging in obscure BLS tables, as you can imagine!)

Third, it tracks overall trends in the economy quite well.  It’s usually low and/or falling when times are good, and high and/or rising when times are bad.  A large, rapid rise is one of the best recession indicators we have – indeed, one of the best-known recession indicators, the Sahm Rule, relies solely on changes in the unemployment rate.

Fourth, it’s timely.  The unemployment rate is usually released on the first Friday of the subsequent month.  (The reference period for each data point is roughly mid-month, so this translates into a functional publication lag of about 3 weeks.) There are some extreme situations when that is too slow, but they are very rare - I can think of just two brief ones in the past 20 years.

Fifth, it has limited revisions.  Revisions are the bane of data consumers everywhere – other data in the monthly BLS report, like nonfarm payroll employment, experience fairly substantial ones over time.  And don’t even get me started on GDP revisions, which can be very large!  Whereas with the unemployment rate, “what you see is (mostly) what you get” – aside from small wiggles generated by revised seasonal adjustment factors and the annual population controls, unemployment rate estimates stay fixed over time.

Sixth, it has a long history and is mostly (though not perfectly) “apples to apples” over that history – the BLS’s estimates go back to 1948.  (There are also estimates by other sources going back even further, to 1929.)

Seventh, it can be sliced in a fairly granular fashion: by age, by sex, by race and ethnicity, by industry, by education level, by geographic location… I’m sure I’m missing a bunch of dimensions.

5. Alternate indicators

As stated above, while useful, the unemployment rate does have drawbacks.  So for the casual consumer, are there any good alternatives?

Let’s start with a personal favorite, the prime working age employment-population ratio.  This just tallies the share of the age 25-54 American over-16, civilian non-institutional population who have a job.  It’s straightforward and avoids the “how should we think about people who want a job but aren’t actively looking” question (though not the “underemployment” question); it also relatively uncontaminated by changes like the baby boomer retirement and rising post-secondary educational enrollment.

I love this indicator; it’s one of the first things I look at when the BLS publishes its monthly report (table A-8a for those who are interested), and I agree it’s generally superior to the unemployment rate.  It does have a few drawbacks, though.  First, it’s not very useful for pre-1990 historical comparisons as it was driven upward by entry of American women into the workplace (pre-1990 numbers are artificially depressed by women being socially constrained from entry).  Second, it’s buried in a fairly obscure BLS table and doesn’t always get covered by media outlets; unless you’re a dedicated data nerd, it might be too much effort to follow. Third (and this is not quite a drawback), as you can see in the chart below, it behaves roughly the same as the unemployment rate - its marginal value added is positive but not that big, at least to casual data consumers.

Another indicator that used to get a lot of attention in the 2010s, when the unemployment rate was widely understood as an understatement of true labor market distress, is U-6.  U-6 is part of a collection of indicators the BLS publishes (the main unemployment rate is U-3, and the numbers run up to 6).  U-6 includes two groups beyond the conventionally-defined unemployed: people working part time for economic reasons (i.e. people who work part-time hours, wish they could work full-time hours, but are prevented from doing so by a weak economy) and folks who are “marginally attached” – they want a job, have not looked actively in the past month, but did look in the past 12 months.  A subcategory of marginally attached workers is “discouraged workers” – those who aren’t actively looking specifically because they don’t think they can find a job.

One indicator that frequently gets pitched as an unemployment rate substitute but I would avoid is the labor force participation rate.  I have nothing against the participation rate – it’s a useful measure, with caveats – but “as is”, it’s a terrible substitute for the unemployment rate as it is massively distorted by big demographic trends (like the baby boomers retiring post-2008).  At the very least, you want to restrict it to prime-working-age folks (ages 25-54).

Conclusion

So to wrap up: the unemployment rate is a super-useful indicator and if you are a casual consumer of business/economy news, you could do much worse than relying on it alone. You could also do a lot better, but only by spending a lot more time! Whether that’s a good use of your time is a personal decisions.