Why turnout models are so important in polling

One of the untold stories of this election is the turnout modeling used by pollsters. Each has his own model and thus each one has a chance at being right or wrong. The first thing you need to know is that any poll of “registered voters” is bogus. That’s because registered voters don’t necessarily vote. In presidential elections you have higher voter turnout than in other elections but still you don’t get anywhere near full participation.
The Florida Division of elections shows that only 74% of Florida’s voters cast a vote in the 2004 general election. With all of the money being spent on “get out the vote” efforts and voter registration drives we can assume that turnout will be greater in 2008 than 2004 but how much greater? And who is most likely to actually go out there and cast a ballot.
Well I was looking at some 2004 general election statistics from the Miami-Dade Elections Supervisor and found them enlightening.
When we register to vote in Florida we basically put ourselves into one of four racial/ethnic buckets: Black, White, Hispanic and Other.
And we also put ourselves into one of three party affiliation buckets: Republican, Democrat and Other.
That’s 12 different combinations. And in 2004, each one had a distinct rate of voter turnout:
voterturnout.jpg
As you can see from the table, the combination that followed through and voted at the highest rate for the 2004 general election in Miami-Dade was Hispanic Republicans followed by White Democrats. Hispanic Democrats and Hispanic Other had much lower turnout rates as did Black Democrats. No wonder “community organizing” is so important to the Dems. Their constituencies apparently don’t seem to do too well when left to their own devices.
Now this election has a lot imponderables to it. For one thing the top of the ticket for the Democrats is Black (or portrays himself as Black despite the differences between him and most Black voters, White mother, African father, Harvard education, etc.). This should effect the historically lower turnout rates for Black Democrats. But those historical differences are not insignificant.
You can easily see how a pollster is faced with a dilemma when he gets a respondent on the phone. He needs to determine what the likelihood of that respondent actually following through and voting is.
I’ll leave you with this last bit of cheerful news. There’s a blogger out there that is claiming that the election is a lot closer than many of the polls are letting on because of the turnout models being used. That blogger mentions the Investors Business Daily/TIPP poll as being the most accurate. And in 2004, the IBD/TIPP poll was indeed the most accurate in predicting the popular vote. On Sunday IBD/TIPP had Obama ahead by only 3.2% with 10.1% undecided.
By the way, the blended turnout rates by party in 2004’s general election for Miami-Dade County were:
Republicans: 78.81%
Democrats: 72.64%
Other: 64.85%