I am analysing the recent US election(s) and conducting some thought experiments to see what might happen this November. I'll post some results here. Please feel free to make suggestions for analyses and thought experiments. 

There are similarities with UK elections in that the election is 'indirect' with the President (in the UK the Government) chosen via Electoral Votes/Colleges. But there are big differences, most obviously in there being realistically only two parties. My gut feeling is that this makes it harder for voters to change allegiance than in the UK. (E.g. in the UK if one is dissatisfied with the Conservatives but unable to stomach voting Labour then there are many intermediate parties to vote for.) This and the supposed appeal of Trump to a certain type of person leads to my first experiment.

The first results I'd like to share are in the above graph. Here I've assumed the very simple scenario that people vote exactly as they did in 2012 but that some of those who did not vote in 2012 come out in favour of Trump. The horizontal axis is the percentage of non voters who will vote for Trump.

Key numbers are:

1. Trump needs 5.5% of previous non voters in order to get 50% of the popular vote. This does not make him President however. Although this also coincides with him getting a majority of states (26 out of 50 + DC) he only gets 253 Electoral Votes. To be President requires a majority of the 538 available.

2. With 8% of previous non voters he gets 50.85% of the popular vote, 28 states, and 286 Electoral Votes and thus becomes President. Going by swings in turnout from election to election this 8% is not easy but possible.

The nonlinearity in the problem, that becoming President is not simply a function of number of votes, is similar to the UK. In the UK the current significant bias in favour of the Labour Party is (will be?) easily corrected by boundary changes. Removing any bias is obviously not so simple in the US.

P

Continuing with the same experiment, the above shows which states are Trump/Clinton as a function of non voters moving towards Trump.

P

Interesting model.

Why do you think non-voters will vote for Trump?

At this point any women non-voters aren't going to vote for Trump and any Hispanic non-voters aren't going to vote for Trump -- that's 59% of the non-voters that won't vote for him. I'd have thought that Trump's voters (cranky Caucasians) would have voted against Obama in the last election.

I'm not saying all non voters will now vote for Trump!! Just that some % will, and that's my horizontal axis. Your 59% may well be correct. But I've only been looking at up to 10% of non voters now voting, so no inconsistency there.

There are over 200 million voters to model potentially! So one has to start simple, and add complexity one step at a time.

My thinking is:

1. Ignore independents. Treat them as non voters.

2. How many people who voted for Obama last time will vote for Trump? Not many!

3. How many who voted Rep. last time will vote for Clinton now? Interesting. A lot of intellectuals are saying they will. That's a small number of very noisy (and annoying) people. And in the UK these people turned many people towards Brexit. So I'm assuming not many.

4. How many people who voted last time will just not vote? No idea. I can see voters on both sides not liking their own candidate. But if the % is the same for Rep. and Dem. then it doesn't affect anything. (Just means scaling the NV -> Trump %.)

5. What will previous non voters do? Vote Clinton? No. She's not exactly inspirational enough to persuade non voters out of their La-Z-Boys. Vote Trump? Yes. That's sort of the whole point of Trump.

So this is the first step in forecasting.

P

So there is  no recent poll info in this model? 

The swings in the polls might say something about the non-firm voters?

Good question. I was going to use that info, like I did in the UK election. But the Trump % of voters in the polls is too far away from the % he needs in my model for victory. So it seemed pointless.

I tried doing the same trick with a weighted average of elections going back 30 years. Makes no difference. Maybe if you took a Reagan election as the starting point he would be in with a chance, but not with recent elections.

I think most of the swinging is statistical errors(?)

P

Yes, I think there is alway a bias between what people think they will do, and what they end up doing.

it would also be interesting to see if the current poll volatility is higher/lower than previous elections. The relative difference might say something about "swing capacity" or voters having a firm opinion.

If the vol was sample noise then it would look more like white noise, not slow waves, so I do think the poll dynamics reveal info.

It reveals a lot about pollsters and the media. Large polls that don't change don't make the news. Everybody loses!

P

Yes, interesting.
Last week someone said that the goal of the media is to make it end at 50-50 pre-election, that's a smart way to look at it!

At 50-50 the "delta" for each vote is maximum! (I did some unpublished work on this with Yury Rojek) That should encourage people to vote. But sometimes (Brexit yet again!) people want to vote one way but are afraid of winning!

P

Hmm...

From this side of the pond, it looks like a fair number of "republicans" will be voting for Clinton but it's unclear if any "democrats" will be voting for Trump.  Trump's negatives are much worse than Clinton's although Clinton's negatives are high, too.

Clearly, Trump does have a small loyal core who will vote for him.  But the larger republican base isn't behind him and may become non-voters.

All v likely but that just adds to my base case: Clinton looks almost certain to win.

Do you have any stats on Rep -> Dem?

(Although I would add that some negatives are not seen as such by many people.)

P

there's a sports statistician Nate Silver runs fivethirtyeight.com (a reference to the electoral college)
I think he's just using polling combined with the electoral college makeup

He's made the point that Gary Johnson is (or was) polling at 24% in his home state of NM, so there's an outside chance that Gary Johnson could win NM and subsequently that no candidate gets a majority in the electoral college

Assange has announced that he's going to announce something big on Thursday. He not going to do a balcony scene this time because of fear of getting assassinated.