Fox News poll: Jones leads Moore by 10 points

[karynnj]

78. I was not attacking scientific polls -- I am not sure which ones you even think I question!

What I said is that polling on this is likely to be tricky because - for BOTH candidates - there are people who would not openly proudly say who they will vote for.

My background is having worked for 25 years doing statistics, operation research and mathematical modeling for AT&T, mostly in Bell Labs. I even had a turn at teaching a sampling lecture developed by the experts in the group I was then in. I never did polling or opinion research, but I did attend conferences where they were discussed. While many samples designed in our group were intricate and had complicated variance estimates, the things we were estimating were based on measurable qualities - that could always be measured for the sample chosen.

Polling has become progressively harder to do. Gone are the days where you could simply take a random sample of registered voters and actually get a large percent of them to answer. When the "non response rate" was low, the assumption that the non responders were similar to the responders in their demographic cell was a necessary and reasonable assumption that had to be made. As the people willing to even answer the phone plummets, it is harder to justify that assumption.

In addition, the pollsters cannot just put out an estimate based on registered voters, they need a model of likely voters. It is probably the model of likely voters that creates the biggest difference between the various polls. Estimating which voters are more likely to vote is as much an art as a science. Where the sampling portion is scientific - even if there are caveats due to non response, the likely voter models are more heuristic. The first hint of how well the likely voter models worked will come tomorrow when reports come in that show how many people are voting in various places.

Here, the Fox poll is an outlier compared to most other polls. It can be that it is one that got it right, while many were extremely off.

Please do not think that because of this I do not look at the polls -- I do. However, even though statisticians have not gotten worse - they actually are likely better with a wealth of specialized software that makes their job easier - there have been far more surprises in both directions. Consider that no one predicted the big margin in the Virginia race and the massive turnover in their legislature.