Tuesday, October 10, 2006
Statistical Behavior of Statistics Matters!
by Tom Bozzo
Today, Barry Ritholtz is doing yeoman's duty on a more active front by explaining how the same statistical considerations make the larger employment growth shown in last month's "household survey" employment, relative to the "establishment survey," effectively meaningless, with appeals to the former the "last refuge of bullish employment scoundrels." Indeed, both the household survey and establishment survey month-to-month employment changes were in the ballpark of one standard error (a little less for the latter, a hair more for the former) — it's just that the standard error of the household survey's month-to-month change is some 4.5 times larger than that of the establishment survey.
Given that the "true" employment increases are small relative to the sampling standard error of the household survey, good old noise will naturally give scoundrels frequent spin opportunities. Barry is dead right in saying "case closed."
A couple weeks ago, I had noted that the business economics press and the commentariat that gets interviewed basically struck out in interpreting a "surprise" increase in new home sales. (Correct answer: the month-to-month change could not be distinguished from statistical noise and therefore was No News; the more precise year-to-date and year-over-year changes remained ugly.)
Today, Barry Ritholtz is doing yeoman's duty on a more active front by explaining how the same statistical considerations make the larger employment growth shown in last month's "household survey" employment, relative to the "establishment survey," effectively meaningless, with appeals to the former the "last refuge of bullish employment scoundrels." Indeed, both the household survey and establishment survey month-to-month employment changes were in the ballpark of one standard error (a little less for the latter, a hair more for the former) — it's just that the standard error of the household survey's month-to-month change is some 4.5 times larger than that of the establishment survey.
Given that the "true" employment increases are small relative to the sampling standard error of the household survey, good old noise will naturally give scoundrels frequent spin opportunities. Barry is dead right in saying "case closed."
