Why and How Clinton will Defeat Donald Trump on Election Day

Part 1: Modeling Assumptions and the National Vote

It seems like this cycle, the new vogue is up to date, breathless models giving you real time odds of political results. Each day, a new poll or two can throw around a model’s real time odds exceptionally, and every day it seems I have to read about people panicking about Nate’s model this of that.  This kind of real-time modeling is amusing, but also distorting and inaccurate in my opinion, and this is why: just because you get new data, doesn’t mean things have changed or its better than your old data. Political modeling really needs to emphasize longitudinal analysis of polling, demographics, and clear assumptions of fundamentals in politics. The reason for this is quite simply, individual polls are almost useless, even more so in recent elections than in the past (particularly when trying to reach many demographics; spanish-speaking voters, young voters, and so on). The only thing that is modestly useful is the polling aggregate, but even this can be impacted in real-time models by movement that is pure noise. In particularly, response bias and lack of voting enthusiasm seem to tack closely to whatever scandal/media attack cycle dogs Clinton over whatever Trumped up story they are running with at that moment. In general, I would say polling aggregates, especially from about mid-August until mid-October, are noisier than the electorate in reality is, and much of what we see is response rate bias and shaky attempts at modeling likely voters by pollsters themselves.

So here are the core elements I do to make actual predictions (not just betting odds) on various states and the popular vote: I look at polling aggregates over 3 different points: February 3rd, July 1st, September 1st, and now (October 2nd). I also look at how these dates varied for Obama in 2012. As far as pollsters go, I don’t include internet polls aside from YouGov which has an extensive track record and has been decently accurate, I don’t include Rasmussen or Gravis because of their noted accuracy issues and methodology problems. In predicting what the electorate will look like, I worked from the Roper Center’s data from 1988-2012, as seen in the following table.

As you can see, aside from the outlier of 1992 (which could reflect faulty sampling, given how consistent the trend was (and it also runs counter to the trend between 1980-1988), there is an extremely consistent growth in the nonwhite electorate that has purely to do with demographics. Obama’s impact is negligible; this is no special Obama magic like the media assumed in 2012 (pollsters too). Indeed, in 2008 Obama had less impact in growing the nonwhite electorate than Kerry did in 2004, and his 2012 performance likewise saw a growth in the electorate equivalent to the 2000 Presidential election. Because of this, I have estimated a 69% white electorate nationally for 2016, the first time its ever been below 70%, and this is a hair lower than a regression plotting would suggest, but I am making an assumption based on preliminary reports from LatinoDecisions polling and Hispanic GOTV groups of record interest in this election and a massive spike in requests for naturalization and voter registration so far, that Hispanic turnout which has typically lagged other demographics, will go up modestly this year (and I’m mean modestly, maybe a 5-10% increase in Hispanic turnout, which would still leave it well below that of white and AA turnout).

I am projecting an electorate that is 12% black, 12% Hispanic, 4% Asian, and 3% Other (mixed race, indigenous). Speaking purely of demographic expectations, prior to actually looking at polling and making a real topline projection, that breaks down to Trump getting 2% of the black voter (this is actually being generous to Trump based on how well he’s been polling thus far with black voters), 15% of Hispanic voters, 12% of Asian voters, and 29% of Other, as well as 59% of white voters. In raw demographics, this equals 44.1% of the vote for Trump.

How does that match up with polling from the 4 dates I selected?

There is noted stability there, especially for Trump’s numbers. Trump has hardly ever broken 42% in the aggregate polling since he announced for President almost a year and a half ago. The big variable is how many people commit to Clinton. This is also consistent with the fact that opinion polls of Trump have found that man has been reviled by over 60% of Americans going back to the year 2000. Trump has always been deeply unpopular and well-known as a celebrity and businessman. Hillary Clinton though, and this is why the media and the jokes at ‘hahaha, look how shitty our choices are this year, look how much Americans hate all these candidates’ irritate me is that Hillary Clinton has in fact regularly been a very popular political figure in America. As First Lady her approval ratings ranged from 44% when she was pursuing health care reform (i.e. taking a leadership role outside the traditional domestic image of the First Lady), to 67% right after the affair, and back down to the low 40s a few months after she announced her campaign for Senate. Once in the Senate a few years though, Hillary’s national approvals were back up into the upper 50s. Until the crashed back into the low 40s when she ran for President. Then as Obama’s Secretary of State, she was both one of his administrations most prominent and well-known members and also its most popular, even during 2010 she routinely got approval ratings in her job of over 62% and this continue through the end of 2012 and the real beginning of the endless Benghazi investigations and intentional attempts to weaken Hillary Clinton’s presidential candidacy, and really collapsed back to the low 40s once she was into her campaign for President. Hillary still routinely polls significantly higher than Donald Trump in aggregate favorables, and in a host of important benchmarks such as how has the better temperament and better qualifications for the job. The point being, Hillary the politician/leader is usually very popular with voters, the Media Hillary/Campaign Hillary is always deeply unpopular, and it can’t be understated how much this related to traditional forms of sexism that affect women who aggressively seek promotions in every walk of life.

Trump’s numbers are horrid. The undecideds are high, which reflects a 5-6% total I expect for third parties. Here there are several important points to understand as well. One, Johnson is fading as I expected. I think he will only get around 4-4.5% of the vote nationally, with another 1-2% for other minor candidates. What’s bad news for Trump is that Second-choice Hillary Johnson-voters are much softer for Johnson than are Second-choice Trump Johnson-voters. This means that when Johnson inevitably underperforms his polling average once voters are in the booths and firmly faced with the binary nature of the electoral system, the soft-Johnson voters are strongly second-choice Hillary. True undecideds also seem to lean overwhelmingly Democratic, as is not surprising. In several swing states I’ve seen multiple polls giving Obama 65%+ approval ratings with undecided voters. These are mostly Democratic voters, and the bulk of them will still end up backing Hillary Clinton in the end.

The 2016 election has been remarkably stable. In the immediate aftermath of his own convention, Trump managed to tighten the race to about 3 points, with Hillary at 42.7% and him at just 39.7%, but other than that he has never actually led in this race a single time in aggregates. Not once. 3 points is the closest he comes. In terms of the polling aggregate he’s never broken 42%. His range has been 39%-42%; Hillary’s has been a much wider 42% to almost 49%. If we look at the 2012 General election aggregate (sans the awful Gallup tracker and online polls other than YouGov), you see a similar story that Obama led the entire cycle except for a week in mid-October post-debates were Romney had a super-narrow lead in the polling aggregate. Obama’s final margin though was 5 million votes and 4 percentage points. It wasn’t particularly close in any metric for a modern election and was actually healthier than Bush’s 2004 reelection which was painted by the media as a devestating defeat for Democrats with long-term implications for their ability to be able to win national elections and the Senate. The media tried to present 2012 as this coin-flip race, but the Obama team’s internal numbers were consistent at around +4 from mid-summer on, and the only changes they ever saw were response bias. The two campaigns had relatively equal resources between them and allies, though outside Romney groups vastly outspent outside Obama groups, and both sides ran competent campaigns, though Romney’s data program was nowhere near as sophisticated as Obama’s and he didn’t have the same volunteer intensity or GOTV effort. The Romney campaign has mentioned in various media stories that this really hit them hard in Florida, which they expected to win based on their internal polling and public polls, but that Obama’s GOTV effort was so massive and so effective that it single-handedly turned a narrow Romney win into a narrow Obama in the state (and this is a story brought up by veterans of both campaigns).

Essentially, and this may sound like a bold assumption though I think its not, given that General Election dynamics are far different from primary election ones, there is a reason why campaigns do certain things in certain ways. There’s a reason why Trump’s campaign being a dumpster fire is going to kill his GOTV efforts and already is in early voting. Trump is not taking contact information or voter information from the many people that attend his rallies. The campaign barely has any semblance of a volunteer database, and until mid-August had no campaign offices set up in Florida or Ohio. The campaign is still vastly behind in the number of official campaign offices, staffers, and organizational apparatus it has in place everywhere, and is essentially relying on state parties and the RNC to do all that for them, which those groups simply can’t do alone. The RNC's database has gone largely unused this election, and Trump’s data is non-existent, while Hillary has taken Obama’s 2012 data team core and made it even larger and more sophisticated this time, also honing it over the primaries, when they collected massive amounts of data and simulations around everything from most-effective use of advertising dollars per delegate, to what time is the best time to contact voters for GOTV and what sentences work best. The answer to whether campaigns matter is that they don’t except on the margins; they matter 6-10 percentage points of turnout and perhaps 3-4% in margins (there are several interesting political science papers I have read on this subject working from 2012).  But they do certain things because those things work. When one campaign is completely crushed in both media spending and organization, the disparate impact will show itself, an impact that is normally cushioned (save in close states like Florida 2012), by the fact that the disparity is usually marginally, i.e. Romney 2012 had a much larger and better run campaign apparatus than Trump does.

This matters for a second reason as well; Trump is relying on blue collar white voters without a college education to make up a disproportionate amount of the electorate. This is one of the weakest turnout demographics in the white community. It’s also one that’s been shrinking for 4 decades as a result of simple demographic change. He’s doing quite well with it, but Romney already won 66% of the vote in this demographic and lost by 4 nationally. And this is with Romney doing quite well with college educated white voters, a demographic universally shown by polls (and also what states Trump has been weakest in compared to Romney), to be very weak for Trump. Simply put, Trump can peel off all the soft-partisans left in blue collar white voters, he can pull upward to 75% of the vote there and still make none of the over-all gains he needs to in the white electorate as he loses ground with college educated white voters, and he needs to do better with white voters, substantially better, than Romney did in order to win. His lack of a campaign infrastructure and any kind of sophisticated contact and turnout modeling apparatus are big issues here. Simply put, there is no evidence in the primary that Trump got non-voting white voters to suddenly turn up and vote for him. Most preliminary analysis suggests that one, not only did Trump’s core base in the primaries lean older (reliable voters), and wealthier (over 70,000 year incomes, also more reliable voters in aggregate), but I haven’t seen any analysis that offers any evidence for Trump turning out non-voters. Regular general election voters voting in a Republican primary is no real threat to a Democratic win, a threat would be unusually high turnout from demographics that have favored Trump and Republicans, namely lower middle-class to poor blue collar white voters/white voters without a college education. This is not a high turnout group though, and I think any campaign/group of supporters that is banking their performance on a low-turnout group based purely on the magic of “my super unpopular candidate is going to inspire them to vote” is I think no different than polling unskewerers of 2012. The fundamentals of the election are baked in, demographic trends in the nonwhite voting population are baked in. Even the over all Democratic top-line with white voters is fairly consistent for that last 28 years, varying between 39-45 of the vote, which is not a big a variance. Simply put, Democrats are not going to get much less than 30% of the blue collar white vote; I don’t think that there are that many soft-partisans left for Democrats in this group, nor that 100% of remaining soft-partisans will abandon the party in one election, nor do I think “Trump won the primaries so toss-out everything you think you know about politics” especially since I consistently thought Trump would win the primary going from October of 2015, and I thought so because of what I knew about politics and how the dynamics were working in Trump’s favor in a severely fractured field of Republican candidates (including my projection that Trump would be strongly favored if the last conservative alternative standing was Ted Cruz).

The polling margin about 2 and half weeks after the Republican convention proved very accurate for Obama’s final margin (4.1%). In fact after mid-Sept the polling was very noisy and in the end less accurate than it was during the weeks after both conventions concluded, which ended being the time period that gave us the clearest picture of how the fundamental divide in the electorate was. The comparative point this time, is mid-August, when Clinton had a 45.1% to 38.6% lead in the aggregate polling, or a 6.5% margin. Those are my topline projection for this election is that Clinton wins by a 7.5% margin, as I am making the previous, clearly defined assumption that campaign organization does matter in outperforming polling fundamentals by a small amount. Right now, the toplines I envision are something like 51.4% Clinton, 43.9% Trump, 4.7% third party. What’s more is that I have an extremely high-level of confidence in this projection, with a range of basically 51-53% Clinton, and 41-44% Trump in virtually all scenarios I can model.

Polling chart for reference:

Clinton versus Trump

Part 2: Electoral College

My projection is also remarkably consistent with how the Clinton campaign is acting. Simply put and I’ve said this many times, don’t rely on noisy models testing out new assumptions, and don’t rely on public polling. Public polling has been unusually infrequent in this election, but much of it has also been very chaotic, noisy, and suffering severely from response bias and increasing difficulty to reach the large chunk of voters that are cellphone only and ignore unsolicited calls/hang up on pollsters. What the Clinton campaign does is far more expensive and more sophisticated than anything any media pollster is doing. That’s just something that anyone following this election needs to understand. It would be true of Trump too if Trump were running any semblance of a data campaign, which it isn’t (Trump’s campaign even said it didn’t poll for debate reactions, which is important data for formulating future debate strategy). First off what the Clinton campaign does is large-volume polling based on an extensive analysis of the DNC’s data bank, amassed over the past 4-5 elections. From there, they are able to model a whole range of likely electorates and their odds over an entire range of results while also using data-intensive statistics to correct for things like response bias.

Even in the third week of September, when many on the Left were in full-panic mode over a few days of bad polling after a negative news story that would have obvious impact on response bias, one, Hillary kept an aggregate lead nationally almost as large as the largest lead Obama had in the final six months of the 2012 election, and two, HfA did not change its campaign strategy in the slightest. Hillary for America and the associated Super Pac, are making only modest buys for her core 270 electoral vote firewall states like Colorado, Pennsylvania and Virginia (where her remaining election ad buys are comically small to what Obama spent in 2012), and focusing heavily still on Florida, North Carolina, and Ohio. Her campaign schedule is also focusing heavily on North Carolina, a state Obama triaged in late September of 2012, giving it up as unwinnable in order to focus more on Virginia, Ohio, and Florida. Furthermore, Hillary is running huge national ad campaigns, some of which are hitting more voters in various states than Trump’s total campaign spending, and this includes states the campaign initially tested the water out in but declined to make larger ad buys in (probably because polling didn’t show the initial buys as efficient + effective as they wanted), states like Georgia and Arizona. HfA still however, is keeping staffers, offices and GOTV efforts open in Arizona and Georgia, and has also opened offices in Dallas County and Harris county Texas. One internal poll, plus comments from Texas Democratic operatives, have said Hillary’s opened up a double-digit lead in Harris county (one of the 10 most populous counties in the nation, and one Obama won by just 900 votes in 2012; this is Houston and its immediate suburbs). She has an office, staffer, and GOTV program up and running mostly to help Democrats pick up a swingish state house seat, and sweep some county-wide offices and County Judge positions in order to build a long-term infrastructure to turn Harris hard blue, sort of like what happened to traditionally Republican Dallas county after 2006.

Simply put, if the very costly and comprehensive data analytics, being run by veterans of Obama’s 2012 campaign and Terry McAuliffe’s 2013 campaign for Virginia governor, showed substantial tightening and need to worry, Clinton’s campaign would not be wasting staffers and GOTV money to ensure downballot wins in Texas. You would also see more money and also more campaign time going away from states like North Carolina, and into the upper midwest, like Michigan, Minnesota, and Wisconsin, were the Hillary campaign has essentially made no specially earmarked ad buys nor focused on campaigning much in. This is also consistent with what David Plouffe has said, and he’s said straight out that Ohio and Iowa are tougher than North Carolina, but that ultimately, the Clinton campaign is very confident about winning all three, and that Florida is not as close internally as public polls suggest, oh and also that the most favorable electoral modeling the campaign can do in Pennsylvania can’t get Trump closer to winning than a 3 point loss. So given that 7.5 point national win, what does my national electoral college map look like? Like this:

Image may be NSFW.
Clik here to view.

Trump flips Iowa, Hillary flips NE-03 and North Carolina, and carries Ohio extremely narrowly, while losing Iowa narrowly. I don’t think that Hillary will lose either of Maine’s electoral votes.

I am very confident in this projection; at this point the only change I foresee happening could be in Hillary’s favor if Donald Trump continues to have a major implosion going into October and screws up the second debate even worse than the first one. Hillary’s ceiling is probably around 53%, but Trump’s floor is as low as 39% in my opinion. Right now, it looks like we could see another campaign shake up and chaos grip his campaign apparatus at the start of October. Trump is both extremely unpopular, and he has a very poor to non-existent GOTV effort, and even with state parties and the RNC taking up the slack, Republicans are deep in the whole in most states, and in some states, it is the Koch brothers and the State Republican party both working in tandem not for Trump, but for a downballot candidate, in the case of Ohio, Rob Portman, which means they aren’t working to help Trump and will even GOTV anti-Trump voters as long as they are Portman voters.

My instincts and understanding of politics and political campaigns both scream to me that Trump’s performance is going to be good in a lot of blue collar white areas, but not great and that turnout is not going to be that good for him in those areas, while there is a big drop off in typical Republican turnout in many suburban areas. I know a couple in Louisiana who both used to be registered Republicans, who have changed to independent because of Trump and are dedicated to backing Johnson, even though they both voted for Romney in 2012. These are two college educated, Church youth group, middle-class, middle-aged white ex-Republicans who live in a suburban community who find Trump so shameful they changed their life-long partisan affiliation. They aren’t angry like Trump is angry. They have secure lives and jobs and aren’t threatened like Trump and many of his core supporters feel threatened. These kind of white, white-collar educated suburbanites have been one of the GOP’s core constituencies, and its been an extremely reliable, easy to turn out group that is growing as a percentage of the electorate, and Trump has the GOP abandoning it for an unreliable, shrinking demographic that matches his mix of rage and prejudice.  This is going to have a big impact as GOP gains among blue collar white voters are neutralized by substantial losses among college educated suburban white voters, both to Johnson, Hillary Clinton, and in many cases, just not voting. The pattern I think that will be present is a slight underperformance by Hillary in many lower-educated, blue collar white heavy area from 2012 Obama, more drastic in some places than others (eastern Iowa, Appalachia, Eastern Ohio, southwestern Pennsylvania, parts of Massachusetts, Rhode Island, southern Illinois, and Northern Maine), and elsewhere have Hillary exceeding Obama’s 2008 and 2012 high water marks in many suburban areas (think Chicago suburbs, Philadelphia suburbs, ares of Texas like Fort Bend, Harris, and Tarrant, SoCal, Western Connecticut, South Florida, NoVa, and around Denver).

Closer to election day I’ll also make final projections on the margins in each swing state, and revisit my projection for the national margin. My impression remains the race really isn’t that close, or that unstable/unpredictable. There was nothing unfathomable to me about the way Trump ate up all the oxygen in the Republican primary and how he won there. I think the actual trendlines are pretty stable nationally, and notice that the projection of Trump’s final vote percentage based on his aggregate vote % at different points in the race is almost exactly where my demographics only estimation puts Trump. To me, Trump’s loss is going to have been largely preordained and predictable from long-term, decades-ongoing demographic patterns, and his own long-standing issues with educated voters and wealthier suburbanites.

Anyway, thanks for reading. Please vote in the poll too and let me know what you think.