These tools provide ways to look far more deeply into unemployment data — current data and data over time. You have the ability to compare and contrast up to four demographic composites, each with a race/ethnicity, gender, age, and educational attainment element.
June 9, 2026 — It’s fairly routine to come across unemployment data disaggregated by race/ethnicity or gender. Slicing of the data by age or educational attainment is sometimes done, too. But combining multiple dimensions into a single composite and then comparing different composites is less accessible.
There’s some fine print to read in a moment, but we think these tools will be a great aid for reporters, researchers, and advocates. One can mix and match in different ways. You can create demographic composites that vary along each dimension, or you can, for example, hold the age and educational attainment elements constant and vary only by race and/or gender.
The data come from IPUMS, which is part of the University of Minnesota. IPUMS gets the data from the Current Population Survey of the Census Bureau, specifically the Basic Monthly Survey. These data are not seasonally adjusted.
What we report are 12-month moving averages, so single-month spikes are smoothed out. We do not report a composite for any 12-month period where the standard error — the statistical measure of how much an estimate is likely to bounce around due to sampling alone — is equal to or greater than 10 percent of the estimate. [More about standard error.] That decision has consequences. Only quite reliable data are reported. The flip side is that there are many composites that are suppressed from reporting either in the current snapshot or in the historical charting. A counterintuitive result: more composites are suppressed during periods of low unemployment. See why.
There are 375 composites in all. 230 of them have 100 percent reporting for each of the 184 rolling 12-month-window points we cover. Another 32 are reported for at least 75 percent of those points; 24 at least 50 percent of the instances; 53 less than 50 percent but more than zero; and 36 none of the time. You can view, sort, and filter the full list of composites here. (The default setting is from most reporting to least reporting.) So, more than three quarters of the composites are reported at least 50 percent of the time, with the bulk (61 percent of all 375) reported for all available rolling 12-month-window points.
You will notice that a relatively high proportion of composites always suppressed are those where “Asian” is chosen, a function of small sample size, as already discussed.
These visualizations are genuinely interactive — you have to make selections before you’ll see anything of interest. The composites you select on top control both the snapshot data and the historical data. Also notice that, for the historical data, you can choose the time-frame that is displayed. In the historical data graph, you can hover over any point and see the full report for the composites for a particular 12-month moving average.
As with any data visualization, please let us know of any errors you spot or suggestions for improvement. And please read this AI caution.