In his column this weekend the Numbers Guy at Wall Street Journal, Carl Bialik, notes* how uncertain the monthly statistics for unemployment and the like can be. For example, the Census Bureau reported that sales of new single-family homes fell to record low last month. However, if anyone (other than Bialik) read the fine print, they’d see that the upper end of 90 percent confidence interval estimates an increase in sales!
“Most of the month-to-month changes are not only nonsignificant in a statistical way, but they are often straddling zero, so you can’t even infer the direction of the change has been accurately represented.”
– Patrick O’Keefe, economic researcher
The uncertainty stems for the use of sampling as a cost-saving measure for government agencies and ultimately us taxpayers. For example, field representatives covering 19,000 geographical units throughout the U.S. only sample 1 out of 50 houses to see whether they’ve been sold.
The trouble with all this uncertainty in statistics is that it ruins all the drama of simply reporting the point estimate. ; )
*(See “It Is 90% Certain That Unemployment Rose. Or Fell.” and a related blog on “What We Don’t Know About the Economy” )