Despite all of the discussion in the last 18 hours about this poll, I have not seen any comments on the point I am about to make (my apologies if someone has already mentioned this). A couple of observations about this poll:
1. They have 37% already voted -- that seems reasonable. (Final 2010 early votes were 31% of 1.13 million total votes; final 2012 early votes for a higher turnout comparison was 44% of 1.57 million votes).
2. The remainder of the poll (63%) only counts those who will "definitely vote". If you will "probably vote", you won't get counted in this poll. I am against this in general from a statistical point of view, but note that prior DMR polls have used the same methodology, so a change compared to the last DMR poll (Ernst+1, I believe) cannot be blamed on this approach.
3. Among Democrats, the poll claims that 54% of Democrats have already voted (vs. 46% who say they will definitely vote by Tuesday. However, among Republicans, they say that only 33% have voted, vs. 67% who say they will definitely vote. (Among other voters, the model counts 28% who have already voted vs. 72% who "definitely will"vote.
In my opinion, this is where the poll's model becomes problematic. Basically, it is modeling the electorate as if the Democrats have had a great early vote turnout that will cannibalize election day votes, while assuming that the lack of Republican banked votes means that they will have a big election day turnout. My initial reaction was that this is a ridiculous model, but wanted to try to quantify it to see how ludicrous it really is. Because they didn't give many cross-tabs (and no demographics) it is difficult to quantify, but I will try anyway.
To start, let's compare it to 2010. This was a great year for mid-term turnout, even though the IA senate race was not competitive. Early votes made up about 31% of all votes. Democratic early votes accounted for about 39% of their total vote (and 30% for Republicans and 23% for others).
In 2012, a competitive and presidential year, early votes were up to 44% of all votes. For Dems, 55% of their total votes were early votes. For Rs, it was 40%, and for others, it was about 37%.
Note that the 37% is between the 2010 and 2012 number, so it seems reasonable at face value. However, if 54% of expected Democrats had already voted, is it reasonable to think that only 33% of Republicans have voted? It's kind of like saying Democrats have a 2012-like turnout, while Republicans are having a 2010 turnout. It doesn't make sense.
Let's suppose that total EV turnout will be 39% (since we were at 37% in the poll). If the total electorate is like 2010, based on the Iowa SoS records, it means that overall, 40% of votes will be Republican, 35% Democrat and 25% other. What this would mean is that the election day composition of voters would be D: 26%, R: 44%, Other: 30%, or an R+18 election day advantage.
What was the actual election day turnout in 2010? D: 31%, R: 40%, Other: 28%, or R+9.
How about 2012: D: 27%, R: 37%, Other: 36%, or R+10.
An R+18 election day turnout just seems crazy in comparison. Note that I am not saying that Ernst won't win, but the evidence would suggest the DMR poll model is simply way off on its turnout model. I think it is incorrect in assuming that many (54%) Democrats have already voted and will therefore not turn out on election day, while 67% of Republicans have not voted yet, but will definitely show up on election day. Either one of these facts may be true, but I am skeptical that both are simultaneously true.
So put me in the outlier poll camp -- I think the race will be within 2 points in either direction (my hunch, no modeling implied)! Actually, even though the DMR did not proivde the cross-tab, the implication in their poll is that Braley has banked a lot of early votes relative to Ernst.