2016 has become the year that jumped the shark a long time ago. It's a post-factual, anything goes, not required to obey logic, reason, or science year. But, usually, the newer world of data geeks taking polls and coming up with probabilities has taken some of the pundit nonsense out of the process. Election year polls that you can't make heads nor tails of have given way to polling averages and complicated formulas for predicting how various states will vote based on demographics and likely turnout models. A few outlets are even experimenting with predicting results early based on turnout and whether candidates can hit precinct-level benchmarks. (I welcome it, with reservation, but many are not so optimistic and I could do a whole post just on that topic alone.) Long story short is that in years past forecasting models have been fairly competitive with each other for accuracy in determining election results. A few reach near 100% precision. But those of you with a science background also know accuracy and precision are two different things. Yes, you can hit the target this time. But can you hit it over and over. It's that last one that has become a sore spot.
This year's forecasts are all over the place. As I write this, they range from giving Hillary a 60% chance of winning (538) to 86% (Princeton Election Consortium). Not only am I going to argue that we should throw out the high and low forecast like we do in sports judging, but I'm also going to argue that the New York Times' The Upshot is probably the best so far this season. It hasn't gone high. It hasn't gone too low. And it--at the moment--seems to reflect reality on the ground the best. It's at 74% Hillary at the moment. But, even more, it has a better breakdown of the state-by-state distribution.
We can argue all day about it. Do I think 538 or Sam Wang are "wrong?" Not necessarily. But I would say that I've had frustrations with both for being a little too "horse race" this campaign in giving too much weight to more recent polling rather than using the average. And, I think, a few forecasts are denying some basic facts on the ground. This can be seen, if you want to get super nerdy, by reading the recent analysis of the Electoral Vote distribution graphs between several forecasters. 538 is flat like a skyline. PEC is more clustered. But in terms of statistics, the tails at the end of the curve remain very unlikely even if you include them in your plot. Could Hillary win Georgia? Sure. Could Trump win Florida? That one is more difficult if you also pause to consider field operations and ground game. And 538 and NYT split. 538 currently lists Florida as 54% Trump while NYT lists Florida as 59% Clinton. Does this make it a true tossup?
538 earlier today published on Twitter a piece on how as the race tightens Hillary can't rely on the Electoral College helping her or being in her favor. The assumption is that a tightening race tightens all over the country. But this isn't exactly true. States like Wisconsin, Virginia, Pennsylvania remain longshots out of reach to Trump. But Ohio is perfectly winnable for him. So their claim that her map is worse than Obama's isn't entirely painting the same picture I would. It's a a very 2012 race, but she has the ability to pick up states Obama lost (NC) and do better than he did. It depends on turnout and how a few of these states swing. But, ultimately, she's probably LESS likely to lose the overall Presidency than he was in 2012. Which makes no sense until you get into some of the nuts and bolts of how Colorado is very likely blue while Iowa is maybe less likely to go Hillary's way than it was to go Obama's way. It's complicated and I find it disheartening that the very people who have elevated the discussion about data have this year perhaps gone against their own advice of "don't pay attention to individual polls," etc..
Moral of the story? 538 has Hillary with 284 Electoral Votes. Upshot has Hillary with 323. PEC has 306. I very easily get Hillary to 272. She wins. But the results in FL, OH, NC, IA, NV are very much in doubt. Hillary could very easily sweep all those states. That gives her more EV than Obama in 2012. But you can see where the doubt is. (I strongly suspect IA and OH go to Trump. FL/NC are truly close.)
This year's forecasts are all over the place. As I write this, they range from giving Hillary a 60% chance of winning (538) to 86% (Princeton Election Consortium). Not only am I going to argue that we should throw out the high and low forecast like we do in sports judging, but I'm also going to argue that the New York Times' The Upshot is probably the best so far this season. It hasn't gone high. It hasn't gone too low. And it--at the moment--seems to reflect reality on the ground the best. It's at 74% Hillary at the moment. But, even more, it has a better breakdown of the state-by-state distribution.
We can argue all day about it. Do I think 538 or Sam Wang are "wrong?" Not necessarily. But I would say that I've had frustrations with both for being a little too "horse race" this campaign in giving too much weight to more recent polling rather than using the average. And, I think, a few forecasts are denying some basic facts on the ground. This can be seen, if you want to get super nerdy, by reading the recent analysis of the Electoral Vote distribution graphs between several forecasters. 538 is flat like a skyline. PEC is more clustered. But in terms of statistics, the tails at the end of the curve remain very unlikely even if you include them in your plot. Could Hillary win Georgia? Sure. Could Trump win Florida? That one is more difficult if you also pause to consider field operations and ground game. And 538 and NYT split. 538 currently lists Florida as 54% Trump while NYT lists Florida as 59% Clinton. Does this make it a true tossup?
538 earlier today published on Twitter a piece on how as the race tightens Hillary can't rely on the Electoral College helping her or being in her favor. The assumption is that a tightening race tightens all over the country. But this isn't exactly true. States like Wisconsin, Virginia, Pennsylvania remain longshots out of reach to Trump. But Ohio is perfectly winnable for him. So their claim that her map is worse than Obama's isn't entirely painting the same picture I would. It's a a very 2012 race, but she has the ability to pick up states Obama lost (NC) and do better than he did. It depends on turnout and how a few of these states swing. But, ultimately, she's probably LESS likely to lose the overall Presidency than he was in 2012. Which makes no sense until you get into some of the nuts and bolts of how Colorado is very likely blue while Iowa is maybe less likely to go Hillary's way than it was to go Obama's way. It's complicated and I find it disheartening that the very people who have elevated the discussion about data have this year perhaps gone against their own advice of "don't pay attention to individual polls," etc..
Moral of the story? 538 has Hillary with 284 Electoral Votes. Upshot has Hillary with 323. PEC has 306. I very easily get Hillary to 272. She wins. But the results in FL, OH, NC, IA, NV are very much in doubt. Hillary could very easily sweep all those states. That gives her more EV than Obama in 2012. But you can see where the doubt is. (I strongly suspect IA and OH go to Trump. FL/NC are truly close.)