Technological Unemployment

Speed Read This
Posted by on June 3, 2015

Will automation displace workers and eliminate jobs? For many jobs, yes. Whether this will cause unemployment, however, is a controversial and open question. When technology automates away jobs, there are several possibilities for what happens to the people who would otherwise have been in those jobs. The total number of jobs worked might drop, or they might just shift to doing jobs that are hard to automate. If the number of jobs worked does go down, this could lead to people being unemployment and impoverished, or it could instead result in people entering the work force later, retiring earlier, working part time, and taking more long vacations.

Predicting what will happen is hard. Noticing what already happened, however, is more straightforward. Automation is not a new force in the world, and technology-driven unemployment has been a concern at least as far back as the Luddite movement in 1811. With the benefit of hindsight, nineteenth century Luddism was clearly incorrect; there was a massive amount of important work left undone for lack of people to do it. But what about more recent trends? Here from the Bureau of Labor Statistics is the United States unemployment rate, for everyone aged 16+.

Bureau of Labor Statistics Unemployment Rate, 1948-2014

This graph displays a pattern best described as “glaringly absent”. It’s basically noise. Why? It turns out that the definition of unemployment is fairly complicated. Unemployed refers to people who are jobless, available for work, and looking for work; it excludes people who have tried to find work recently but aren’t currently trying (marginally attached workers), people who have given up on finding work (discouraged workers), students not looking for work, and people unable to work because they are ill or disabled. The “discouraged workers” category is particularly unfortunate; it means that the unemployment rate is measuring a sort of residual, the people who can’t find work but haven’t realized it yet. A much more straightforward number is the employment-population ratio. This is simply the number of employed people divided by the population. So here, also from the Bureau of Labor Statistics, is the United States Employment-Population Ratio, for the civilian noninstutitional population, age 16+. (Excludes people on active duty in the armed services and prisons, nursing homes, and mental hospitals.)

Bureau of Labor Statistics graph of employment-population ratio, 16+, 1948-2014

This graph gyrates a little less wildly. If you squint hard enough, you might fool yourself into seeing a pattern. It turns out there is one more confounder to separate out.

Paul Samuelson famously criticized GDP by observing that, if a man married his maid, GDP would fall. In fact, it’s not just GDP that would fall; employment would fall, too. Employment only counts work that is done for wages; there is a separate category, “household activities”, for the rest. This includes things like cooking, laundry and child care. Unfortunately, this time the BLS doesn’t have a nice graph to give us, but we can get a few point estimates. The American Time Use Survey estimates that household take an average of 9.5 hours per week for men, 15.5 hours per week for women. By contrast, Valerie Ramie in the Journal of Economic History reviews twelve estimates from 1924-1953 and finds that in that time period, homemakers spent 47-63 hours per week on household production – a time expenditure comparable to or greater than that of full-time employment. This household labor fell primarily on women, and has now been substantially reduced by inventions such as laundry machines, microwaves, and robotic vacuum cleaners, as well as by declining fertility. This combined with changing social norms caused many more women to enter the labor force, as shown by the employment-population ratio for women:

Bureau of Labor Statistics graph of employment-population ratio for women 16+, 1948-2014

Here we see womens’ employment increases until some time around 1990-2000, then either stops changing or starts falling. This comes partially at the expense of mens’ employment, but I think it’s mostly at the expense of household activities. Meanwhile, the employment-population ratio for men looks like this:

Bureau of Labor Statistics graph of employment-population ratio for men 16+, 1948-2014

This is a steady decline of about 2.7% per decade. To see whether this trend extends past the time range covered by the BLS time series, I also checked the 1910 census and found that the employment-population ratio (male 16+) was 91% (Final Report Vol. 4 Ch. 1 Pg 69).

Based on this data, I expect both the US male and female employment-population ratios to decline at about this rate in the future. This will manifest as a mix of people entering the work force later, retiring earlier, working less, being declared legally disabled at a lower threshold, and sharing income within families. Depending on policy and social norms, we could have more leisure for everyone, or more poverty, or some mix of the two. It’s up to us to choose wisely.

2 Comments on Technological Unemployment

  1. Robin Hanson says:

    You title this “technological” and your first sentence talks about “automation” but your post doesn’t give reasons to think that is what is causing these trends.

    • jimrandomh says:

      True. I do have reasons for thinking that automation drives down employment, but they’re independent of what I’ve written here. If there are other forces that might explain these trends, I’d be interested to hear about them.

Leave a Reply

Your email address will not be published. Required fields are marked *