Friday, September 2, 2016

Trump v. Clinton, State and National polls 9/2

The purpose of measuring the state and national polls is to see if they align and how the states will turn out if the national number proves to be accurate on election day. I've refined me methodology in an attempt to get the state and national polls to align. The more I thought about it there wasn't a lot of information that could be gleaned from having the state polls as Clinton +6.4% and national polls as Clinton +4.5%. All you got out of it was that Trump was probably doing better than the state polls. What you didn't know was how much better in each state.

I'm using state polls that were conducted no later than August 22. There's been a shift toward Donald Trump and I want to capture that. In some cases I'm only using the IPSOS poll that was conducted ending August 25. In normal circumstances I wouldn't only use one poll, but I'm doing so for two reasons. First, older polls are more skewed toward Clinton and the idea is to see how the states will play out based on where the election is now. Second, I'm trying to find out how the states will play out if the national number is accurate. Thus, the most important thing is to get the state margin and national margin to align, not to worry about whether states might be a little off. I'm trying to show what the state number would be with this national margin, not making a definitive prediction.

Fortunately, there is at least one current poll in 43 states. The states without a current poll are all small with 3-4 electoral votes. For these states I used FiveThirtyEight's election day estimates. Even if these are off the number of votes in each state don't really impact the national percentage. I made adjustments to two large states, California and Texas. The Ipsos poll has Hillary Clinton winning California by 36%. Barack Obama won California by 23%. An increase that dramatic doesn't align with the other state and national polls and really skews the state result. I gave Clinton a 17% win. Ipsos had Trump winning Texas by 13%. I bumped that to 18%. Without these adjustments Clinton wins by 6.2%. With them she wins by 4.0%.

Some of you might cry foul. How can I change two polls but keep the rest? This exercise is to see how the states will play out based on the current national percentage. If I used the unusual California and Texas numbers the state total would be out of whack. I don't feel adjusting all states by 2.2% to the 4.0% would get more representative numbers.

The new spreadsheet is here. Clinton wins the electoral college by 338-200. The good news for Trump is that Clinton might not have the electoral college advantage people think she has. Clinton leads in Nevada, Florida, North Carolina, Virginia, and Wisconsin by 2 points or less. These states are worth 73 electoral votes, enough to put Trump over the top. Thus, it's possible that Trump could lose the national vote by 1.0-1.5% and still win the electoral college.

A Clinton backer might believe that this batch of polls is off and that Clinton leads by more in those battleground states. She might but not if she leads nationally by 4.0%. Remember that the way you get the national number is by adding up all the state votes.

No comments:

Post a Comment