"I didn't think that ladies and negroes would vote!!"
http://www.dickmorris.com/why-i-was-wrong/
Morris starts off by saying "Well, I was wrong, polls were right, whataya know" and then immediately continues by offering gut feeling / qualitative analyses of how Sandy halted Romney's momentum and Christie helped Obama. His arguments seem intuitive enough, but are they true? It'd be nice if there was some kind of quantitative measure to figure out when exactly "Romney's momentum" stopped and how, if at all, Christie "helped" Obama. Maybe if some kind of company polled American citizens at different points in time so we could see how the dynamics of support in the race shifted.
Jonah Goldberg said:
Silver, a statistician, has been predicting a decisive Obama victory for a very long time, based on his very complicated statistical model, which very, very few of his fans or detractors understand.
"very complicated statistical model"
- Take the most recent poll in a state from every polling agency
- Decay old polls with time in case an agency stops polling
- Weight polls based on their historical error, weight down internet and other methodologically questionable polls.
- Introduce per-state fudge factors based on state demographics and the economic climate which help fill early gaps in the data, weight these down as the time to the election decreases. Silver has repeatedly commented on what these include on a state by state basis. In 2008 they included proportion of population that's evangelical, unemployment rate, proportion of population that's African-American, and a few others. We are made to understand that these factors have essentially no weight in the final stretch of the race.
- Calculate the current standing of the race and the margin of error for that standing
- Do this for every state
- Given the figures, run 10,000+ runs of the race with the vote in each state distributed randomly (evenly?) over the margin of error
- Report results as a histogram
VERY COMPLICATED STATISTICAL MODEL
I DON'T UNDERSTAND
