Wednesday, October 20, 2010

Polling distortion: lies, damn lies and statistics

Nothing more surely presages defeat than lying to oneself.  Whether its Adolf Hitler crouching in his Berlin bunker moving pins on maps to represent the movement of armies that no longer existed or Baghdad Bob broadcasting American defeat as the American juggernaut rolled across Baghdad International Airport, knowing self delusion is rarely followed by victory.

The recent spate of public released polling data that shows the Republican polling advantage in Senate races to be shrinking shares a level of credibility with Baghdad Bob.

Some polling services, Quinnipiac, Survey USA and especially Rasmussen and Gallup, for various reasons, have historically closely tracked elections, with projections closely predicting results.  Your side may not always look good in those polls but those polls usually look good relative to the results.

Recently, the reliable liberal polling groups have begun distorting the running averages so as to create the impression that Democrats are closing across the country.

Take Colorado; the only poll showing Democrat Bennett leading is a two week old PPP-an openly Democrat polling group-that premises its sampling on Democrats turning out at same rate as 2008.  Nothing in the current environment evidences the liklihood of such a heavy turnout in a midterm election, and nothing supports the assumption that Democrat turnout will exceed Republican turnout, much less reach the stratospheric Obama 2008 surge.

Pennsylvania is worse.  Non partisan Rasmussen, who continuously updates its turnout model to reflect current partisan registration and voter intensity shows Pat Toomey up by a well beyond the margin of error of 10%.  PPP, using the same unrealistically weighted turnout model places Joe Sestak up by one.  The even stranger, and more secretive turnout projections of the Muhlenberg College Morning Call survey appears to assume turnout based on the last three elections rather than current registrations.  Since '04, '06 and '08 saw progressively better Democrat turnout the model literally cherry picks the three best cycles possible for Democrats.  That would be like predicting Republican turnout in Texas on 2000, '02 and '04.  There's a word for that, ridiculous I think.

West Virginia might well be the most laughable of all.  Rasmussen's state of the art turnout modeling shows Republican John Raese up 7%, beyond the margin of error.  Two openly Democrat groups, Orion Strategies and the afore indicted PPP show Democrat Joe Manchin up ten and three respectively.  Orion's methodology is unpublished and we already know the problems associated with the PPP model.  Even CNN/Time-which has yet to err in a Republican direction although not openly partisan, shows a tie.

Baghdad Bob has nothing on these guys.

1 comment:

RF said...

Sporer, don't get your bloodpressure up because of these polls. This D is here to tell you, your side will be just fine when the real votes are counted in a couple of weeks.