Generic Bias

Rep. Jack Kingston's office points out this Washington Times article, in which the Gallup Poll is cited as being aware of (and not particularly put off by) the fact that their polling consistently overstates the Democratic vote.

The Gallup Poll has been issuing [survey results suggesting a 10 point Democratic lead] over several months that are drawn from what the pollsters call "generic ballots" that do not name the specific congressional candidates, but ask only which party voters will support when they go into the voting booth.

The latest USA Today/Gallup poll, conducted from April 7 to April 9, found that the Democrats lead Republicans by 52 percent to 42 percent among all registered voters surveyed. "The Democratic lead on this measure has been in the double digits in each of the last three Gallup Polls, starting in late February/early March," wrote Gallup analyst Jeffrey M. Jones.

But Gallup and other election pollsters who conduct generic polls acknowledge that the turnout rate for registered voters is much less than it is for Americans who say they are "regular voters." When Gallup polled a subgroup of people who say they vote on a regular basis, the Democratic lead fell to seven points, 51 percent to 44 percent.

But why, Gallup?  Why?

The reason has to do with a different turnout for various groups of voters: registered voters, regular voters and, the most accurate of all, "likely voters." Notably, "In midterm elections, fewer than half of eligible voters usually turn out to vote and Republicans are more likely to turn out than Democrats," Mr. Jones said.

Hmmph.  Seems like something you could control for a bit better if, you know, your whole business were polling.

But let's abide Gallup's limited ability to discern respondents' likelihood to vote.  It's easy enough to acknowledge that the ranks of the polled may always be overly steeped in Democrats and simply remember to 1) adjust, 2) discount, and/or 3) ignore the results accordingly.  After all, if the results are consistently biased, though it speaks poorly for Gallup's accuracy and perhaps their integrity, that very consistency would mean a reliable adjustment could be applied to the results.

But as Kingston blogged last month, the bias is not consistent.

Party ID Breakdown Republican Independent Democrat

January 34 34 31

February 33 34 30

March 28 36 33

Jan – March -6 +2 +2

What's worst about this shifting supersaturation of one party within the polling set is that it will tend to give evidence of a trend where one may not exist.  So while Gallup couldn't realistically continue shifting the balance by one percent each month, even if it were to do so for 2 or three months, its resulting trending data would suggest a broadly shifting sentiment among the greater population.

Note to Gallup: you can begin shifting me from the "adjust and discount" to the "ignore" column.