Most pollsters go into creating a poll with a demographic model. The population has X% of 65 and older and Y% of them vote traditionally. This is a reasonable method if things are the same from one election to another.
Selzer says she goes in without any model. She allows the numbers tell the story. In 2016, her method predicted Trump was going to win by 7. (He won by 9.) Her poll was the canary in the coal mine. Iowa is demographically similar to Wisconsin and Michigan. If Trump wins Iowa by 7 then there is a decent chance he could win those states as well, which he did.
I came appreciate Selzer as a genius in the 2008 Iowa caucus.
Take yourself back to the cold days of January 2008. Democratic top tier candidates are all Senators: Joe Biden, Hillary Clinton, Chris Dodd, John Edwards, and Barack Obama. By December it's down to 3 - Clinton, Edwards, and Obama.
Who is going to win is answered by answering, "How many Iowans will the winning candidate need to win?"
Previous caucus attendance:
- 2000 Caucus: 61,000 Iowans
- 2004 Caucus: 124,000 Iowans
All campaigns agreed attendance would increase in 2008. Perhaps 160,000 or maybe even 180,000 will show up. Clinton's well run campaign was confident they had 60k voters locked up. In a 3 way race this would be enough to win. Edwards campaign was equally confident. (Although I was on Obama's team I had good friends the other camps and we were honest with each other about what we were hearing.)
Then came the final Selzer poll that jolted all of the campaigns. Here two big predictions: Obama is going to win by 7 points and the turnout will be around 220,000.
My Clinton friend said they had no problem disregarding the poll as a whale of an outlier: The data didn't match anything their field operation was seeing and it is preposterous to believe 100,000 more Iowans would caucus in 2008 than in 2004.
Selzer's premise was that other pollsters were polling those like me who had caucused before. Two massive groups were being missed: Young people and disaffected Republicans. (There is a saying in politics, "Candidates who rely on the youth vote cry on election night.") Caucuses are a serious time commitment of 2-3 hours and most young people won't do it.
Because it is a lot of work to caucus there was no reason to believe Republicans and Independents would make the effort. It was bold of Selzer to predict these groups would turn out in massive numbers.
Result?
239,000 Iowans came out to caucus in 2008. Nobody - not even the Obama campaign - predicted turnout that high.
- 38% Obama
- 30% Edwards
- 29% Clinton
Selzer not only blew every other pollster out of the water but she had a better feel for the electorate than the internal campaign numbers.
******
Selzer is hardly a Democrat pollster. Her 2016 prediction of a 7 point Trump win was a massive wake-up call that Clinton was in danger in blue wall states of Wisconsin and Michigan. In 2014, there was a close Senate race to replace Tom Harkin. Selzer's last poll accurately predicted that Republican Ernst would win it comfortably.
******
Because Selzer randomly calls every number she is able to get a pool of voters other pollsters won't get. In 2008 pollsters only dialed registered Dems. Some only dialed registered Dems with a history of going to the caucus. Quite simply, they missed the new/young voters and the Republicans and Independents. Her method costs quite a bit more but has proven to be more accurate.