Replies

I’ll keep an eye out for your posts. How do you adjust for toplines that are impacted by biased sampling? It could come in the margin of error calculation you do perhaps. My problem with Nate Silver is the garbage in garbage out problem. If you input the toplines of 80% of polls that are oversampling Dem leaners just a little bit in every ID, like slightly too many women, slightly too many non-white ID, slightly too little white ID, slightly too many college degrees/post-grads, etc. and they don’t balance those out with comparable oversamples of GOP-leaning bias in other categories, won’t your model end up leaning too Dem? Maybe you take a cross section of polls where you have a even number of GOP-leaning and Dem-leaning polls.

Yes, you are correct that my model will be imbalanced, too. As I said, I'm not trying to correct the balance, I'm telling what the polls are saying but hopefully better than just the RCP average. I add the probability of winning that the MOE brings to the polls. Just because the poll is "within the margin of error" doesn't mean it's a "statistical tie" as the LAAP-dog media likes to say.

My hope was that, over the entire breadth of the polls it would balance out. In 2020, Emerson's polls were the most balanced while Quinnipiac's were the most tilted to Democrats. If I see an egregious error in the sample, I will exclude that poll from the average, but that's as far as I go to try to "correct" the polls.

I had hoped that peer pressure and social stigma from wildly incorrect polls would self-police the polling firms, based on their wanting to remain credible from a business model perspective.

-PJ