Increasingly, I'm finding the electoral vote logic compelling[1]. There's a nice, clear presentation of that logic in a recent Washington Post piece: "If Clinton wins Florida and carries the 19 states (plus D.C.) that have voted for the Democratic presidential nominee in each of the last six elections, she will be the 45th president. It's that simple"[2]. Well, I wouldn't want to use the word "simple" in the same breadth as "presidential election," but, yeah, that would do the trick. Two questions arise here: (i) Are any of the six-time D states likely to defect? and (ii) how probable is a D win in Florida?
Let's start with (i). The most likely defectors are Wisconsin and Pennsylvania. The Cook Political Report tells us that both of these states have a "lean D" status[3]. Federal legislative positions that are rated as lean D end up going D 87.7% of the time[4]. (Presidential elections are too rare to supply useful date, I suppose.) Note that the WaPo arithmetic doesn't presuppose that the Ds will win either Colorado or Virginia, both of which are also listed as lean D.
What about (ii)? How likely is it that Florida will be blue this fall? CPR also lists it as lean D. The WaPo piece points to a poll referred to in Politico article, which has HRC up 13 percentage points on Mr. Trump. But once you run it down, you see that the poll was based on 604 voters, not enough to get very excited about[5].
More important, perhaps, are the demographics of the state: Registered Democrats = 40%, registered Republicans = 35%, and registered for neither = 25%. Hispanics make up about 20% of the state's population[6], and Mr. Trump's problems with hispanics don't deserve a footnote. All of this information, however, has to be taken with a grain of salt: Hispanics skew younger, and many younger folks can't vote (under 18) or don't vote (enter your favorite joke about millennials here). Interestingly, 15% of Florida are vets or serving military, and Mr. Trump is reported to poll well with this group[7][8].
Oversimplifying (a lot), I'd like to say that Mr. Trump can't win without Florida, so his odds of winning the presidency can't be higher than his odds of winning Florida. And at the moment, I don't think his odds of winning Florida are higher than 30% (though that number will likely change a lot as more data becomes available.) Furthermore, while winning Florida is a necessary condition, it is not a sufficient condition of victory. He could, for example, win Florida but lose Ohio (where he also lost the primary contest). Ergo, Mr. Trump's odds of winning should be *less* than 30%. Let's call it 29% just to be safe.
I initially had an estimate somewhat like this - but then i did a simple MC simulation based on the cook ratings, and if you were to re-assign the "Lean D" states "to Toss-up", that brings prob of democratic victory down to about 75%. However, that would be far from my Base Case, and assigning only a 75% probability for D win for the Lean D states instead of 50% puts you at 95%+ Clinton. Assigning the historically observed probabilities (as you note based on the legislative elections) would give you c.99% Dem victory.
zettel314, as it so happens I too did a MC simulation or two a few days ago with the CPR data, but using Cook's own information about reliability[1]. My results were overwhelming favorable to the Ds, just as were yours. And this isn't the only line reasoning that favors the Ds. Sam Wang has an interesting little model based on current polling[2] that I discussed a little at some point[3].
But I'm not ready to go all in the Ds since other evidence points in other directions. I haven't tried to develop a taxonomy of the kinds of evidence that are salient here (too early), but one category will surely involve historical trends. The short take is the it's quite rare for a party to win the White House after it has held it for 2 contiguous terms. I think there's something to that, and I've discussed it elsewhere[4]. The main problem is that the historical data set is *tiny* and, therefore, it's probative value is questionable. It's also possible that there is something like the gambler's fallacy at work here. I don't mean to say that presidential elections are independent of one another; obviously, they're not. Yet when there's a new candidate at the top of the ticket, as there is this year, then maybe there's less dependency than the historical argument assumes. At any rate, if this historical argument were as strong as some are making it out, then we'd expect to see something similar in gubernatorial races too. I'd love to know whether that's true, but I don't quite have the time to look into it. Hmmm...I'm beginning to talk myself out of taking the historical line of evidence as seriously as I have been. But, of course, in such cases, the difficulty is: to stop ;-)
On the historical point, I just took a quick look at this and I make out there have been 31 presidential elections where incumbent party was in power for 8+ years. In 17 of those elections, the incumbent won.
Restricting to 20th Century, there have been 17 elections of which incumbent won 7. This seems not dispositive either way. Now, post-WW2 there have been 9 such elections of which incumbent has won only 2, but that seems like a very small sample size on which to draw a conclusion, and I don't really know how much to weigh different historical periods.
This feels like a bit of an analytical rabbit hole to me for information of questionable predictive value.
Thanks, zettel314. This is what I said earlier bout the historical data (in the comments somewhere or other):
"I think it’s possible to overplay the importance of the historical data. Presidential elections happen so rarely that the actual sample size to work with is rather small. It’s true that since 1900 the Ds have had the White House for 3 terms in a row only 1 time. But there have been very few opportunities: 1920, 1940, 1968, and 2000. What does 1 victory in 4 tries tell us? Not much. Moreover, the data can be reframed quite naturally in ways that favor the Ds cause. Suppose we ask how many times since 1900 the Ds have won the White House after holding it for at least 2 terms. The answer is that they failed in 1920, 1968, and 2000 but succeeded in 1940, 1944, and 1948. To be clear, I’m not saying the historical data is unimportant; I’m just saying it’s not as important as I originally thought."
I still stand by that. The reason I'm not all in for the Ds is largely because I think that at this point *everything* is of questionable predictive value. The CRP data is more predictive, as is the polling data, than the historical data but not so much so (right now) that I'd give the latter a weight of 0. There's simply not much that I can learn right now that would make me go all in; I want to see how the game unfolds before doing that, even after Mr. Trump says things that must make fiscal conservatives want to vote for HRC[1].
To be frank, I have another reason for holding back: I really want the Ds to win. And I think that predisposes me to overvalue evidence that suggests they will. So I'm self-consciously tapping the break from time to time. If you don't suffer from my psychological oddities (here's hoping!), then that might not be necessary for you.
this has been very helpful. i have the same bias, but at the same time i do think that name recognition of both candidates is higher than normal at this stage, which makes me more inclined to place more credence on polling data relative to the average race. This conversation has persuaded me to come down a bit, but if nothing substantial were to change post-Conventions I would be back up to close to 90% (worth noting that the point in time which the CPR accuracy was tested was from ratings between July and the end of August [1]).
Good analysis. http://www.realclearpolitics.com/epolls/2016/president/fl/florida_trump_vs_clinton-5635.html It's important to note that the polls changed drastically in the past 2 months. We still have 6 more months, so there are a lot of possibilities to change. Trump has already started to shift his platform to woo Sanders supporters. I wouldn't be surprised if he ends up being more liberal than Clinton by the general election. Both Trump and Clinton go with the public opinion and have no real platforms they personally believe in. Difference is that Trump dares to take more risks, so he changes his positions faster than Clinton and he doesn't lose as much when he say politically incorrect things.
Increasingly, I'm finding the electoral vote logic compelling[1]. There's a nice, clear presentation of that logic in a recent Washington Post piece: "If Clinton wins Florida and carries the 19 states (plus D.C.) that have voted for the Democratic presidential nominee in each of the last six elections, she will be the 45th president. It's that simple"[2]. Well, I wouldn't want to use the word "simple" in the same breadth as "presidential election," but, yeah, that would do the trick. Two questions arise here: (i) Are any of the six-time D states likely to defect? and (ii) how probable is a D win in Florida?
Let's start with (i). The most likely defectors are Wisconsin and Pennsylvania. The Cook Political Report tells us that both of these states have a "lean D" status[3]. Federal legislative positions that are rated as lean D end up going D 87.7% of the time[4]. (Presidential elections are too rare to supply useful date, I suppose.) Note that the WaPo arithmetic doesn't presuppose that the Ds will win either Colorado or Virginia, both of which are also listed as lean D.
What about (ii)? How likely is it that Florida will be blue this fall? CPR also lists it as lean D. The WaPo piece points to a poll referred to in Politico article, which has HRC up 13 percentage points on Mr. Trump. But once you run it down, you see that the poll was based on 604 voters, not enough to get very excited about[5].
More important, perhaps, are the demographics of the state: Registered Democrats = 40%, registered Republicans = 35%, and registered for neither = 25%. Hispanics make up about 20% of the state's population[6], and Mr. Trump's problems with hispanics don't deserve a footnote. All of this information, however, has to be taken with a grain of salt: Hispanics skew younger, and many younger folks can't vote (under 18) or don't vote (enter your favorite joke about millennials here). Interestingly, 15% of Florida are vets or serving military, and Mr. Trump is reported to poll well with this group[7][8].
Oversimplifying (a lot), I'd like to say that Mr. Trump can't win without Florida, so his odds of winning the presidency can't be higher than his odds of winning Florida. And at the moment, I don't think his odds of winning Florida are higher than 30% (though that number will likely change a lot as more data becomes available.) Furthermore, while winning Florida is a necessary condition, it is not a sufficient condition of victory. He could, for example, win Florida but lose Ohio (where he also lost the primary contest). Ergo, Mr. Trump's odds of winning should be *less* than 30%. Let's call it 29% just to be safe.
--
[1] https://www.gjopen.com/comments/comments/229516
[2] https://www.washingtonpost.com/news/the-fix/wp/2016/05/02/republicans-have-a-massive-electoral-map-problem-that-has-nothing-to-do-with-donald-trump/
[3] http://cookpolitical.com/presidential/charts/scorecard
[4] http://cookpolitical.com/about/accuracy
[5] http://static.politico.com/00/fd/3b66ba454039b21f6797f039096e/april28-lge-600-final-slides.pdf - But it is worth noting that the source of the poll was Associated Industries of Florida, which appears to be a business friendly body not known for a bias toward the Ds. Politico reports, however, that AIF "teamed with Democratic strategist Steve Vancore for the poll." See http://www.capitalnewyork.com/article/florida/2016/05/8598048/florida-poll-republican-brand-damage-bolsters-clinton
[6] https://en.wikipedia.org/wiki/Demographics_of_Florida
[7] http://www.washingtonexaminer.com/why-troops-and-vets-are-voting-for-trump/article/2584171
[8] http://www.militarytimes.com/story/military/election/2016/03/14/military-times-election-survey-donald-trump-bernie-sanders/81767560/
Nicely done, very informative. Thank you.
I initially had an estimate somewhat like this - but then i did a simple MC simulation based on the cook ratings, and if you were to re-assign the "Lean D" states "to Toss-up", that brings prob of democratic victory down to about 75%. However, that would be far from my Base Case, and assigning only a 75% probability for D win for the Lean D states instead of 50% puts you at 95%+ Clinton. Assigning the historically observed probabilities (as you note based on the legislative elections) would give you c.99% Dem victory.
Thanks, zettel314 and sevedb.
zettel314, as it so happens I too did a MC simulation or two a few days ago with the CPR data, but using Cook's own information about reliability[1]. My results were overwhelming favorable to the Ds, just as were yours. And this isn't the only line reasoning that favors the Ds. Sam Wang has an interesting little model based on current polling[2] that I discussed a little at some point[3].
But I'm not ready to go all in the Ds since other evidence points in other directions. I haven't tried to develop a taxonomy of the kinds of evidence that are salient here (too early), but one category will surely involve historical trends. The short take is the it's quite rare for a party to win the White House after it has held it for 2 contiguous terms. I think there's something to that, and I've discussed it elsewhere[4]. The main problem is that the historical data set is *tiny* and, therefore, it's probative value is questionable. It's also possible that there is something like the gambler's fallacy at work here. I don't mean to say that presidential elections are independent of one another; obviously, they're not. Yet when there's a new candidate at the top of the ticket, as there is this year, then maybe there's less dependency than the historical argument assumes. At any rate, if this historical argument were as strong as some are making it out, then we'd expect to see something similar in gubernatorial races too. I'd love to know whether that's true, but I don't quite have the time to look into it. Hmmm...I'm beginning to talk myself out of taking the historical line of evidence as seriously as I have been. But, of course, in such cases, the difficulty is: to stop ;-)
--
[1] https://www.gjopen.com/comments/comments/229516
[2] http://election.princeton.edu/2016/05/01/what-do-head-to-head-general-election-polls-tell-us-about-november/
[3] https://www.gjopen.com/comments/comments/223874
[4] Somewhere, can't find it at the moment.
Many thanks for your response @pythagoras
On the historical point, I just took a quick look at this and I make out there have been 31 presidential elections where incumbent party was in power for 8+ years. In 17 of those elections, the incumbent won.
Restricting to 20th Century, there have been 17 elections of which incumbent won 7. This seems not dispositive either way. Now, post-WW2 there have been 9 such elections of which incumbent has won only 2, but that seems like a very small sample size on which to draw a conclusion, and I don't really know how much to weigh different historical periods.
This feels like a bit of an analytical rabbit hole to me for information of questionable predictive value.
Thanks, zettel314. This is what I said earlier bout the historical data (in the comments somewhere or other):
"I think it’s possible to overplay the importance of the historical data. Presidential elections happen so rarely that the actual sample size to work with is rather small. It’s true that since 1900 the Ds have had the White House for 3 terms in a row only 1 time. But there have been very few opportunities: 1920, 1940, 1968, and 2000. What does 1 victory in 4 tries tell us? Not much. Moreover, the data can be reframed quite naturally in ways that favor the Ds cause. Suppose we ask how many times since 1900 the Ds have won the White House after holding it for at least 2 terms. The answer is that they failed in 1920, 1968, and 2000 but succeeded in 1940, 1944, and 1948. To be clear, I’m not saying the historical data is unimportant; I’m just saying it’s not as important as I originally thought."
I still stand by that. The reason I'm not all in for the Ds is largely because I think that at this point *everything* is of questionable predictive value. The CRP data is more predictive, as is the polling data, than the historical data but not so much so (right now) that I'd give the latter a weight of 0. There's simply not much that I can learn right now that would make me go all in; I want to see how the game unfolds before doing that, even after Mr. Trump says things that must make fiscal conservatives want to vote for HRC[1].
To be frank, I have another reason for holding back: I really want the Ds to win. And I think that predisposes me to overvalue evidence that suggests they will. So I'm self-consciously tapping the break from time to time. If you don't suffer from my psychological oddities (here's hoping!), then that might not be necessary for you.
--
[1] http://www.vox.com/2016/5/9/11639292/donald-trump-default-print-money
this has been very helpful. i have the same bias, but at the same time i do think that name recognition of both candidates is higher than normal at this stage, which makes me more inclined to place more credence on polling data relative to the average race. This conversation has persuaded me to come down a bit, but if nothing substantial were to change post-Conventions I would be back up to close to 90% (worth noting that the point in time which the CPR accuracy was tested was from ratings between July and the end of August [1]).
[1] http://cookpolitical.com/about/accuracy
Good analysis. http://www.realclearpolitics.com/epolls/2016/president/fl/florida_trump_vs_clinton-5635.html
It's important to note that the polls changed drastically in the past 2 months. We still have 6 more months, so there are a lot of possibilities to change. Trump has already started to shift his platform to woo Sanders supporters. I wouldn't be surprised if he ends up being more liberal than Clinton by the general election. Both Trump and Clinton go with the public opinion and have no real platforms they personally believe in. Difference is that Trump dares to take more risks, so he changes his positions faster than Clinton and he doesn't lose as much when he say politically incorrect things.
Another consideration is this: http://blog.dilbert.com/post/131552504961/trumps-third-act-part-of-the-trump-persuasion
http://blog.dilbert.com/post/144105138301/the-apology-gambit
Trump is acting out a story, not acting like a typical politician. I think it's very possible that he can reverse his stance on the Wall to promote a general sense of heroic change and humility after he secures the Republican nomination. He's already starting to appeal to Latinos and Bernie supporters.
Thanks, hyperspace!