One final thought before bed on election eve...
By CosmicConservativeThe mythical "+D bias" in all the polls.
I've been skeptical of this argument from the start. I honestly don't think it means what most people think it means. I listened to a Hugh Hewitt interview with Scott Rasmussen on this subject and Scott basically said that all raw polls are biased, it's impossible not to be. They are not merely biased according to party affiliation, they are also biased demographically.
Polling companies attempt to "fix" that with normalizing algorithms that are supposed to massage the poll results from the original sampling to a projected turnout based on party affiliation and certain demographics. Rasmussen has a fairly complex algorithm that factors in not just the basic demographic information (meaning they try to map the ratios of demographic sets into something that resembles the population), but they also attempt to factor in demographic turnout.
This is the "sausage making" aspect of polling and is where it can be more art than science. But I've had enough statistics in school to think I understand the basic principles. The whole point of the algorithm is to adjust so that the sample bias is factored out. Some polls are better than others. Gallup and Rasmussen seem to do it best.
If, for example, a poll assumes a different ratio of party turnout than actually occurs then the algorithm itself is biased. This is one of the key things that partisan polls tend to do. Democrat pollsters tend to predict higher Democrat turnout, and Republican pollsters do the reverse.
This is why poll aggregaters claim to have accurate results. They assume their results have contrasting bias which balances out.
Now, in a situation like our current situation where 2008 had historically high turnout of young, black and Democrat voters, while Republican turnout was suppressed, it is not difficult to see how many polling organizations can have algorithms which are biased to a greater or lesser degree in favor of Democrats. However the initial sample bias isn't why those polls might be inaccurate, the algorithm bias is the real culprit.
But there is a more critical reason I distrust polls more and more every year. And that is because polls have become self-sampling exercises, where the vast majority of "randomly sampled" polling targets simply opt out of the polling process. While I can't find any scientific basis to "prove" that conservatives are more likely to "opt out" than liberals, I am pretty confident that is true.
So between the algorithm bias and the self-sampling bias, I believe that most polls give Democrats a slight edge, maybe a point or two. But that's likely all it amounts to.
In the end the proof is in the pudding. We will see tomorrow if the polls have any predictive value at all.
I suspect that my "gut feeling" of a 1% - 2% bias in favor of Democrats will prove pretty close and that will mean that Obama's "slight edge" might very well vanish. However, I don't feel like it will become a Romney edge.
I think the end result is likely to be one of the closest elections in history. Maybe even closer than 2000 if more than one state ends up in recount.
Comments (2)