A quick perusal of my past diaries and comments should easily reveal that I'm a strong Edwards supporter. But I'm also trying to get a Ph.D. in statistics, so maybe, just maybe, I know a thing or two about analyzing polling data. ;-) And after looking more closely at the two polls that came out yesterday, one showing good news for Edwards, the other for Obama, I can say that BOTH polls have their flaws.
Follow me below the fold for a detailed explanation of where each poll goes wrong.
First up, the ABC News poll. Things seemed OK, until I got to question #14.
14. Have you attended any previous Iowa caucuses, or will this be your first caucus?
Have attended First caucus No opinion
12/17/07 55 45 0
11/18/07 63 36 *
7/31/07 68 31 *
Now, it's unclear how much impact that shift will have on the results. The previous two polls showed about a 2:1 ratio of those who had attended a caucus before versus first-time caucus-goers. Now, it's much closer to 50-50.
Since most agree that those who've caucused before generally lean more towards Edwards, Obama's counting on younger voters to show up, and Hillary's looking to shore up women, especially single working women, it's not surprising that with the decrease in their sample of previous caucus-goers, Obama and Hillary's numbers have both improved, while Edwards' number has fallen a bit.
12/17/07 11/18/07 7/31/07
Barack Obama 33 30 27
Hillary Clinton 29 26 26
John Edwards 20 22 26
Bill Richardson 8 11 11
Joe Biden 4 4 2
Dennis Kucinich 1 2 2
Chris Dodd 1 1 1
Other (vol.) * 1 0
None of these (vol.) * * 2
No opinion 3 3 4
If I had the cross-tabs of who each of those two groups in question #14 support, I could re-weight the sample to show what the numbers would be if you still had about a 2:1 ratio of previous caucus-goers to newcomers. But from eyeballing the data, the shift in their support may have come solely from sampling more first-time caucus goers.
So let's think about this. Is a 55:45 ratio likely of previous to new caucus goers on January 3rd? Turnout in 2004 was only about 5.5% of all registered Democrats, about 124,000 people. And that was an all-time high when that happened. Other than from moving, health reasons, or passing away, I don't really foresee someone who caucused in 2004 as sitting this one out. After 8 years of Bush, they're just gonna stay home in the first battle to appoint his successor? Eh..... I don't think so.
True, the exit polls showed in 2004, 45% of the attendees were first-time caucus goers. But they would now be grouped with the previous attendees group. Those newcomers, you could say, were part of that Dean group. Does there still exist another 45% of newcomers this time around? A 55:45 ratio would then assume that about 10% of all Iowa Democrats will participate in the caucus, or over 225,000 people. Is this a realistic number for two days after the New Year? A 2:1 ratio, FWIW, as the two earlier Iowa ABC polls show, would roughly translate to over 8% already.
Now for the Insider Advantage poll. Yes, Edwards fans were jumping on this poll in joy yesterday, but look carefully at their sample. Of the 977 likely caucus goers they sampled, almost 43% of them are over 65 years old. And ONLY 3.5% were under the age of 30. Sure, the older ones seem to be the ones who generally tend to show up. But ONLY 3.5% of the younger crowd?
Again, look at the exit polls from 2004 in Iowa. 17% of caucus-goers were under the age of 30 in 2004. Only 27% were of the over 65 crowd. And with Obama energizing young people, are we to believe they'll only make up 3.5% of the caucus goers this time around?? And whereas 54% of the 2004 caucus goers were female, this poll has a sample where 60% are female. Now, that may sound reasonable, given what people are saying about Hillary pulling in single working women. But the youth vote doesn't seem to jibe with the past data.
The poll then goes on to have a "highly likely voter" category. There, young people only make up 2.7% of those most likely to show up. Is this realistic??
Update: OK, so Mark Blumenthal has shown that the numbers I was looking at was the unweighted data. Without seeing the weighted cross-tabs, I can't comment on whether the numbers make sense or not. Argh. So... ignore everything I just wrote about the Insider Advantage poll numbers.
OK, so I've more or less refused to answer my own questions, because I do not know what will happen on January 3rd. Maybe 45% of the caucus goers that day will again indeed be newcomers, and over 225,000 people will participate in just the Democratic caucuses alone. Then, the ABC News poll numbers may indeed hit the mark.
Maybe since school is not in session yet, those college students won't have their peers to go caucus with, and will stay home instead and watch the Orange Bowl. Will they care about Virginia Tech versus Kansas more than the caucus? Or maybe they show up in their home precinct, but since their Obama support is more diluted, their vote alone may not make an impact. Who knows? If very few young people actually show up, then the Insider Advantage poll numbers may indeed hit the mark.
So both polls, once you look at the details, have things that make you go Hmmmm...... Are their scenarios realistic? We can only go by so much past history and anecdotal evidence for now to determine what's realistic or not. None of us know for sure what the turnout will be like on January 3rd.
But wait!
I'll leave you with another piece of data from these two seemingly contradictory polls where they do seem to strangely cross paths. Putting aside their possible flaws for the moment, both polls asked supporters of the NON-frontrunners who they'd support among Hillary, Obama, and Edwards, if their candidate didn't get the 15% viability threshold. What's interesting are the results here.
Edwards Obama Clinton
Insider Advantage 42 29 28
Insider Advantage(H) 42 31 27
WaPo/ABC News 41 29 16
The (H) stands for the smaller "highly likely voter" screen Insider Advantage applied.
Huh. The Edwards and Obama second choice numbers match up almost perfectly. Yes, Hillary's differ wildly, but it's uncanny to see those first two match up almost perfectly. This says something; I'm just not sure exactly what it says. :-\
Hopefully this will make people settle down a bit (not likely) over the latest polls.