Did Unemployment Really Increase In April?

In its Friday news release the U.S. Bureau of Labor Statistics (BLS) reported that the unemployment rate in April "edged up from 8.8 to 9.0 percent over the month." Yet, according to the Bureau's own criteria for measuring statistically significant change, the unemployment rate did not rise in April.

As the Employment Situation Technical Note accompanying the news release states, at the 90 percent confidence interval used by the BLS to determine statistical significance, a monthly change in the unemployment rate within the range of plus or minus 0.19 percentage point is not significant, i.e., there is no change. The standard language the Bureau uses when a small variation in monthly data occurs that does not pass the statistical significance test is it was "little changed." The term "edged up" is used when a monthly increase is statistically significant, but not large. (The Bureau's chosen level of confidence means that there is a 90 percent chance that sample estimates will be different by no more than 1.6 standard errors from the so-called "true" value because of sampling errors.)

Taken out to two decimal places to be consistent with the BLS reported error interval, the unemployment rate for March was 8.83 percent and for April it was 8.96 percent. The difference of 0.13 is well below the 0.19 point required for the statistical significance, hence implying no change. Only if the unemployment rates for March and April are measured at one decimal place, which yields a 0.2 point monthly difference, could it be argued that the change represents a meaningful increase since it is less that the 0.19 point cutoff for statistical significance. But that would be inconsistent with a more accurate comparison based on the Bureau's own reporting of a two decimal point error range.

The BLS reported that the level of unemployment was 13,542 thousand in March and 13,747 thousand in April, a difference of 205,000. The Bureau's error range for monthly changes in the unemployment level is 280,000, so the April change was not statistically significant. In this case the BLS press release used the appropriate standard language by stating that the number of unemployed persons "changed little."

This would seem to buttress the case for concluding that the unemployment rate was also essentially unchanged instead of edging up as reported. Within the unemployment total, the number of long-term unemployed (those unemployed for more than six months) declined significantly in April, by 283,000, lowering their share of total unemployment to 42 percent, from 45 percent in March.

Civilian employment, enumerated from the same household survey as the unemployment rate, was within its monthly error range in April, and so was little changed.

However, the generally preferred measure of employment - nonfarm payroll jobs collected from employer survey - rose strongly in April, by a statistically significant 244,000, further supporting the interpretation that the unemployment rate didn't actually increase last month. Private sector jobs rose even more than nonfarm payrolls, by 268,000, with employment gains broadly based.

So large an increase in April payrolls would be expected to draw down the unemployment rate further. On the other hand the unemployment rate dropped a full percentage point in the previous five months, from November to March, which is a sharper decline than the employment gains experienced would ordinarily have supported. It may be that the interruption in the downtrend in unemployment last month is in part a statistical correction for the recent imbalance between the jobs and jobless data.

All in all, there's good reason to believe that the reported rise in the April unemployment rate should be taken with a large grain of salt.