Monday, 7 November 2016

The Demise of Donald J Trump

A bold prediction? I don't think so. The media has a bad habit of calling every election "too close to call" in the final days, regardless of whether or not that's technically true. To wring any hint of a nail-biter out of this contest one has to look at the polling data so superficially it takes genuine effort.

But before I get to this analysis, a few quick observations: despite being November the weather is looking not at all terrible in most of the US. In a non-compulsory voting nation, this means more of the less-enthused voters will turn out to vote (or more accurately, rain will cause these voters to not vote). Based on pure energy levels, I think the unenthused voter demographic is broadly not Trump's demographic. The main areas of rain seem to focus on the states around Texas, and those around Washington State, reasonably safe seats for the Republicans and Democrats respectively where weather will have a less drastic impact than a more marginal state.

The other thing I'd like to point out is that whoever gets elected will be unpopular. This means the usual mid-term trend of votes away from the President's party is likely to be particularly pronounced, and either candidate may find themselves unable to achieve very much at all after their first two years.

And now...

The Unpredictable Candidate

Most conventional wisdom in this campaign has gone out the window. On conventional wisdom, Trump should have lost right at the start of the primaries. Since that point there have been 100 more that should have ended his campaign. Trump defies most understanding about American politics. As such, traditional methods of prediction are likely to prove unreliable. Trump has driven more than a few Republicans to the Democratic side this election, and enthused demographics that have not traditionally had the faith, patience or trust in politics to vote before. Historical data about state voting patterns has been used in this post, but only in a minor way when no other solution presented itself.

For the most part this post attempts to predict the results of this irregular election without recourse to historical data. Instead, we're entirely using this year's polling data. A brief glance at the national polls gives a result of 46.6 for Clinton and 44.8 for Trump: a mere 1.8 percentage point gap between the candidates. But to call this election close ignores the realities of the Electoral College.

Each state has two senators and at least one member of the House, making a minimum of three representatives. (Washington DC has no senators and thus only has one elected representative, but for the purposes of the electoral college counts as having three). These equate to seats in the electoral college with several states having the bare minimum three, while more populous states like California sail well into the double digits. These seats are filled by votes during the election and, in turn, elect the president. Every state except Nebraska and Main operate on a winner-takes-all system. Therefore the national polling is irrelevant. A candidate can win election with just 270 seats in the college, or the 11 largest states combined. We need a state-by-state approach.

Here is a table of all the state polls found at this source:

Unfortunately there is no polling data from Mississippi, Hawaii and Washington D.C., so these states (and federal district) are dealt with at the end.

Now, to calculate each state's natural lean and project this on the national polling we need to work out how each poll diverges from the national norm. For consistency of data, I've averaged the state polls on each day rather than the national polls which may not align with the state polling dates. This does raise some issues where a lot of polling in red or blue states in a cluster might skew the average, however.

To smooth out outliers and deal with days that have few polls conducted, I've also produced a 5-day average that averages the polls of each day, the two proceeding days and the two following days. (Note that this is different to the average of the relevant daily averages as some days have more polls than others).

This data allows us to measure the degree to which each state poll diverges from an average that (hopefully) follows the rise and fall of candidate popularity throughout the campaign.

Next we need to refine this bulk of data into an assessment of each state's bias. The two measures I'll be using are each state's most recent divergence, and each state's average divergence:

 These divergences (which in truth are measured in percentage points, not percentages) can then be applied to national polling data (currently 51% once normalised so that Trump + Clinton = 100%) to emulate the state's natural lean and allow us to count the electoral college votes (EC) that each candidate can count on.

(Strictly speaking, Maine and Nebraska will divide their EC votes between the candidates based on their respective results on election day. However, I'd expect Maine to divide 2/2 and Nebraska 3/2 in favor of Trump on these numbers. This comes to Trump 5, Clinton 4 which is the same as just giving all of Maine to Clinton and Nebraska to Trump)

All I have done is award the states that a candidate wins on all four measures to them. Yet already Clinton has 266 of the required 270. Since the smallest state has 3 votes, Trump draws on the loss of any undecided states at this point, and Clinton can win outright with any one of these states except North Dakota or Alaska. Realistically Clinton will get one of these and win.

Increased Confidence

Now, this data all but gives Clinton the presidency off the bat, but it'd be foolhardy to claim a win for Clinton on this untested method alone. Interestingly, though, ALL of these predictions agree with the latest state polling except New Hampshire, worth a mere four votes. Everything I look at makes me view these allocated seats as the "safe bets", and however you cut the cake Clinton is over 260/270. Trump, by comparison, can't break 180 on these declared seats and needs to pretty well sweep every tossup to win.

Let's test his chances in each of the seats in doubt:

Alaska (3): With 3 votes, Alaska isn't crucial for Trump. And by not crucial, I mean he can still draw if he loses it, or if he somehow wins New Hampshire.
Historically Alaska should be a republican win, but Trump has alienated enough Alaskans, it seems, for Clinton to be ahead in the latest polls. Still, he wins 2/3 of the non-even predictions I've made, so I'll tentatively give him this state. [REP]

District of Columbia (3): Again, 3 votes is something Trump can draw without, but he can't really afford a draw. If he loses DC (and he will, as you will see) he needs to win New Hampshire to have the slightest chance at winning.

Why am i certain Clinton wins DC when we have no district poling data? History. Yes, I know this is an atypical election, but LOOK AT THOSE NUMBERS! There are world dictators with worse reported support than that! The Democrats' worst result in recent history is 85.2% to the Republicans' high of 9.3%.

Assuming my call on New Hampshire stands up, DC puts Clinton at a draw before any other state on this list is called. [DEM]

Florida (29): At 29 votes Florida is an auto-win for Clinton if she gets it. Florida is always an important seat, due to having the equal-third most EC votes and having a mixed voting record (unlike #1 California and equal #3 New York, which are reliably Democrat, and #2 Texas which is firmly Republican). This is the kind of seat, therefore, where history is of no use to us. Instead, latest polling has Clinton ahead in Florida, and she wins 3/4 of our metrics. I'm calling Florida democrat this year, and Clinton president. [DEM]

Georgia (16): 16 votes. Auto-win for Clinton. The flipside of Florida, Georgia is 3/4 pro-Trump in my calculations and Trump leads the latest poll. Unlike Florida, Georgia's history is more solid -- pro-Republican. Georgia goes to Trump in my view. [REP]

Hawaii (4): If my predictions are correct and Clinton wins Hawaii as well, she reaches 270 and claims the presidency. Trump can stave this off by winning New Hampshire and every other state on this list. Otherwise Hawaii is a must-win.

There's no state data to fall back on, but Hawaii has a strong Democrat history, and the lowest white population in the US which is where Trump gets most of his support. Hawaii for Clinton, putting all Trump's hopes on New Hampshire, but realistically, that won's save him. [DEM]

Mississippi (6): If Trump loses Mississippi's 6 votes he needs to win New Hampshire and every other state on this list. If he can pull that off, he wins by a single vote.

Without any state data, we're forced to rely on history. Still, Mississippi is a reliable Republican seat. I don't have too many doubts about handing it to Trump at this point. [REP]

New Hampshire (4): We've assumed Clinton wins this one based on our data, but this is the only state where our system disagrees with the latest polling. I'm sticking to my guns, partly because I need to pick a side and want to back my method, but partly because that latest poll looks like it may be an aberration. If Trump can win this seat, however, it gives him breathing space to lose Alaska, Hawaii or North Dakota as well as his inevitable loss in DC. [DEM]

North Carolina (15): at 15 EC votes, Clinton wins if she carries North Carolina.

Trump claims 2/3 of our calculations above. Clinton is ahead in the latest polling, but NC is historically GOP. Under the old system where I'd get 5% of the predictions as tossups, this'd be one of them (rounding up to 3 tossups I'd also claim New Hampshire and Alaska). But since we don't keep tally any more I should at least take a punt. Eeny Meeny Miny Trump. [REP]

North Dakota (3): North Dakota is small. If Trump loses this and doesn't win New Hampshire, it's a tie at best. If he loses NH or any other state on this list, he's done for. History and the latest polls suggest I should break this tied data in Trump's favour, and I have no better indicator. [REP]

Ohio (18): Ohio's 18 votes alone will give Clinton the presidency if she wins it. 50:50 on out calculations, and an equally mixed history make this an edge seat. Clinton leads the latest polls, and that's all I have to go on. [DEM]

That leaves us with this as a prediction:

TL; DR: Clinton wins with 320 votes, 50 more than required, and thus a victory with 59.5% of the EC vote. So not a landslide, but for Trump to win this requires one of a handful of very specific and unlikely situations. DC is a guaranteed loss, meaning he has to win everything else on the tossup list (Alaska, DC, Florida, Georgia, Hawaii, Mississippi, New Hampshire, North Carolina, North Dakota and Ohio) to win.

No comments:

Post a Comment