Every month when the Bureau of Labor Statistics releases its jobs report, people home in on one particular metric: the unemployment rate. But there are a lot of other interesting and potentially significant data in the report, though interpreting them appropriately can be tricky.
Take, for example, the duration of unemployment. There’s little doubt that more people are remaining jobless for longer than has historically been the case. Last month, according to the BLS, the seasonally adjusted average duration of unemployment was 35.6 weeks, not far below the post-crisis high of 40.7 weeks in late 2011. Before the financial crisis struck full-force, average duration was around 16 or 17 weeks.
But is average duration the best measurement? As the chart to the right shows, unemployment duration is strongly skewed: While most people are clustered toward the shorter durations, there are some people who’ve been out of work a very long time — 3, 4, 5 years or more — and those cases pull the average higher. When data are distributed like that, averages can be very misleading as to what the “typical” value is.
In such cases, most experts would instead use the median — the midpoint of a distribution, with half of the values falling above it and half below it. The median is much less affected by skewed distributions and outliers (data points that are unusually large or small compared to the rest of the dataset). The median duration of unemployment in June, seasonally adjusted, was 16.3 weeks — still well above the typical pre-recession median duration of 8 to 9 weeks.
Unemployment duration is hardly the only data distribution that can be overly affected by a relative handful of outliers. Wealth and income data, for example, are notoriously skewed: According to New York Times columnist Floyd Norris, median household wealth in 2010 (as calculated from the Federal Reserve’s triennial Survey of Consumer Finances) was about one-sixth of average household wealth.
As for income, as Stephanie Coontz argued recently in The New York Times, if Warren Buffett and Oprah Winfrey took it into their heads to move to Steubenville, Ohio, average household income there would soar overnight from $46,341 to $75,263. Coontz’s piece gives several other examples of skewed distributions and misleading averages.
Looking at both averages and medians, and at how close they are to one another, can tell you much about the underlying data. (In a perfectly normal distribution, they’d be identical.) The continuing wide gap between the average and media unemployment duration (though not quite as wide as it was a year or so ago) indicates that, while most jobless Americans are finding work in 20 weeks or less, a core of long-term unemployed remain out of work for longer and longer periods. And bear in mind that, since to be counted as unemployed a person has to have recently searched for work, these data don’t include the estimated 7.2 million Americans who want a job but haven’t looked for one in the past year.
