I just submitted a guest column to the Seattle Times

[kurtyboy]

I also posted this in GDP.

RE: The Gregoire-Rossi election contest, which drags on in court this month.

http://seattletimes.nwsource.com/html/localnews/2002278...

In my column, I point out the hypocricy of the GOP when they want to apply "proportional analysis" to the election outcome.

http://www.king5.com/topstories/stories/NW_050205WABele...

My submission follows:

The Republicans in Washington State, particularly Slade Gorton, have short memories when it comes to using statistical methods for estimating likely outcomes of governmental counting tasks such as elections and the census. While today they zealously defend the accuracy of “proportional analysis” to estimate the way felons would (probably) have voted in the 2004 governor’s race, they forget their own zeal in a similar federal counting issue just six years ago.

In 1998 and 1999, Republicans in the House of Representatives and the Senate hotly disputed the Clinton administration’s proposed use of “statistical sampling” to reconcile “undercounting” problems anticipated for the 2000 census. Undercounting problems (to the tune of tens-of-millions) were most pronounced in geographic areas with high concentrations of minorities—generally believed to lean Democratic.

Under the Dept. of Commerce’s proposed statistical sampling scheme, geographic areas with undercounted populations would be assigned a computer-generated population derived statistically from known-good information—in this case, live, on-site interviews with selected households in the given area. The Commerce number-crunchers would then extrapolate general sorts of demographic details from the area and apply the same details to the estimated non-responding households in the same area—broad details like minority status, age demographics, and so on—but nothing as definitive as behavioral predictions.

But these broad, general extrapolations were still too shaky for the GOP. House Republicans, led by Speaker Newt Gingrich (GA), vociferously and violently objected to statistical sampling with the following arguments:

• Sampling is unreliable.
• Real people may be replaced by virtual people.
• Sampling is polling, and polling is not accurate.
• Sampling could be influenced by partisan politics.

The GOP, worried that inherent error rates would falsely alter census counts by hundreds of thousands or even millions in each state, took their impassioned battle all the way to the Supreme Court, where they finally prevailed in a divided decision. The Court’s decision protected the Republicans’ gravest political concern—that under statistical sampling, population estimates from minority districts would skew population numbers in such a way that decennial Congressional redistricting would favor Democrats, even if only in a tiny way.

But take a look at the GOP arguments above, and apply the same arguments to proportional analysis in the Washington election contest. Clearly, proportional analysis has flaws similar to statistical sampling; and where a mere few-dozen votes are at stake, the GOP’s proportional analysis offers not one scintilla of assurance that its presumed accuracy will provide a definitive, believable final answer to the 2004 governor’s election.

Nevertheless, the GOP would have the state’s voters (and Judge Bridges) believe that a convicted felon from some given precinct is (statistically) likely to behave in the voting booth in a manner similar to all of the law-abiding voters in the same precinct. This similarity seems, frankly, implausible to any reasonable observer. Do any of us think we behave the same way as felons?

So, what can we now surmise from all of this? We can see that the GOP fought statistical adjustments tooth-and-nail when such adjustments might have benefited the Democrats six years ago, and yet they now coddle and adore these adjustments when they might benefit the GOP. Whether or not citizens are properly served is simply not of consequence to the Republicans: Political outcomes and goals seem to be all that matter.

The Republicans (including Dino Rossi’s front-man Slade Gorton) in this state—based upon their own historical objections to statistical sampling—are showing themselves to be, in this case at least, hypocrites of the highest order. It is a shame that so few among us notice or remember.

This was an unsolicited opinion, but I hope that the piece has enough clarity to get published this weekend, since the trial, which the Seattle Weekly calls the "political trial of our new century" will begin on Monday. Their excellent coverage of the story was the inspiration for my piece, along with an insight from (conservative) Professor Butch Kemena of Western Washington University.

http://www.seattleweekly.com/features/0520/050518_news_...

If I don't hear from the Times in a couple of days, I'll send it to a couple of other papers. What do y'all think?

[LisaM]

1. Very nice

I remember the statistical analysis argument for the census, but, being no statistician myself, could not have summed this up so nicely. I hope they publish it. And thanks, also, for pointing out that using statistical analysis for the census would have changed voting districts.

I hope the judge is smart enough to know this.

[jody]

2. Excellent but consider reducing to 200 words from its present 595.

Also when one makes inferences by extrapolating statistical results, the results are suspect because the variance is essentially unknown, hence the standard error is also essentially unknown. Note that in statistics, extrapolation usually means going beyond the range of observed data.

Perhaps another word might capture your thought.