Hi everybody. I haven't posted in awhile, and generally I avoid it, but I wanted to put in my two cents about how I'm seeing political polls being interpreted here by a lot of people.
Before I begin, however, I'd like to put two caveats out there:
- I am not trying to make any point whatsoever about what may or may not happen next week.
- I also do not think that my opinion about what will happen next week matters at all, so I'm not going to discuss it.
First, I want to make a general point about statistical analysis that is simple, but so often overlooked: survey research cannot provide accurate enough point estimates to be entirely reliable. Some leads are more plausible than others, but the point estimate of any given poll is almost more like a statistical "suggestion" of the state of the race than anything else.
Second, let's please make the distinction between likely and registered voters. I don't mean to say that there is or isn't a big enthusiasm gap out there, but there isn't enough evidence to suggest that a poll of registered voters is highly suggestive or analogous to the turnout seen in this midterm election or any other midterm election. Next cycle (or any presidential cycle, for that matter), it probably will be, so the polls referenced by the diaries on the rec list showing leads for Dems or good approvals for Obama amongst registered voters are very positive leading indicators for 2012, not for 2010.
Third, as Nate Silver has repeatedly pointed out, most elections are decided by structural factors and do not truly swing wildly at the last minute (except in rare cases). His models correctly show reduced sensitivity to individual polls and short-term dynamics. It's obviously not accurate to say that voting doesn't matter, but candidates clearly don't matter as much in terms of partisan performance as larger structural factors.
Fourth, and perhaps most importantly, estimates are, in the end, a range of potential outcomes. The House may statistically appear to be out of reach for Democrats and the Senate may statistically appear to be out of reach for the GOP, but this does not control for all of the potential outcomes in which the Democrats could hold the majority (or ones in which the GOP could control the Senate).
There's a reason why good forecasters know that they're just giving things their best, most educated guess - prediction is based on assumptions, ranges, and other things that are incompatible with people who want to see meaning in a range of outcomes derived from data.
In other words, the conclusion that an election was "stolen" because it turned out differently than you expected is a narrow-minded conclusion insensitive to uncertainty.
For the sake of neither raising our hopes or crushing them, let us embrace uncertainty and GOTV!
Thanks.