Thursday, January 19, 2006

Lies, D****d Lies, and Carefully Selected Statistics

With the support of number-crunching experts, Knight-Ridder News Service is standing by its recent finding that MSHA penalties are down.
[Inserted 1/20: The Herald-Leader published only a softened version of the KR Washington bureau's followup.]

"It's really wrong for them (MSHA) to say you're incorrect,'' said John Grego, professor of statistics at the University of South Carolina in Columbia. "When you aggregate the whole thing, everything has gone down.'

The Department of Labor's "rebuttal" claimed that the reporters looked at the government data the wrong way. But when the reporters ran the government data the government's way, they still found that fines were down. For instance,
the average fine paid during the last five Clinton years was $20,832, while the Bush average is $8,746.
The government took the unusual step of posting its "rebuttal" on a federal holiday, Monday.

Even though Knight-Ridder's conclusions apparently have been vindicated, the government may count on the "rebuttal" raising enough doubt in people's minds to muddle the issue, since the average reader rarely has time to delve into statistical arguments. It's an old P.R. technique, sadly.

The original story is here.

The "rebuttal" is here.

One of the classic ways to "lie with statistics" is to run available data according to a number of possible parameters, take the best-looking charts and come up with with some ex post facto rationalization for why those parameters are especially meaningful.

By the way, no one should assume this has anything to do with the civil-service "numbers" people at MSHA. All the time I worked at MSHA myself (up to 2004) these professionals were competent, careful and as helpful to the public as they were permitted to be. Any manipulations came from higher up.

Also on the statistical front, a year ago, an op-ed in the Herald-Leader was commenting on MSHA's ways with death statistics.

Today, MSHA still hasn't counted the death of underground coal miner Forrest Riley in October 2004. The West Virginia medical examiner found that Mr. Riley had a heart attack, but at the same time, he was pinned in the cab of his machine by the end of a bent pipe -- which might have contributed to his death. The ME classified the death as accidental, but MSHA went the other way: "natural causes." This contrasted with the past, when the agency used to count deaths if the cause was uncertain.

Not counted in 2005 was the death last October of a worker removing trees to ready the ground for a new waste dump at a Virginia coal preparation plant (covered by federal mine safety law). This contrasts with other cases, where MSHA has taken responsibility for mine workers who cut trees to clear the ground for mining or for a haul road within mine property. There could be more uncounted deaths that deserve a second look.


Post a Comment

Links to this post:

Create a Link

<< Home