Can Biden beat Trump? Political science, consistency, and me
This question keeps coming up. So, I suppose I should write this post.
After the 2016 election, I made the following pledge: I will never question the political science forecasting models, or more specifically, the "Time For A Change" model, again.
Here's what happened, as a refresher. Contrary to popular misconceptions, political science actually got 2016 right. A generic R was supposed to beat a generic D, based on underlying political fundamentals. The model that I have consistently cited as my favorite-- Alan Abramowitz's "Time For A Change" model-- was one of the many models that worked. It forecast an R victory. Why? It had three variables: GDP growth in the second quarter of the election year, the incumbent's approval rating (even though he wasn't running), and a two-term penalty for the incumbent party because one party almost never wins a third term. That final variable was the thing that really worked against the Democrats in 2016, according to the math from Abramowitz. That model predicted an R victory.
Of course, Alan himself, like most political scientists, threw out that prediction and said that Hillary Clinton would win. Why? Well, first there were the polls, and then there was... Trump. You know, that guy who tells you that you should take cleaning products intravenously, forcing a press release from the makers of Lysol about the improper use of their products. Let that sink in. Metaphorically!
Anyway, I was one of the people who disregarded the political science forecasting models in 2016, and predicted a Hillary Clinton victory. In the aftermath, I made a promise. Never again would I disregard those models. Yup. From then on, I'd just go with the forecasting models. What, am I some kind of fool? You know, the kind who would guzzle bleach to cure a virus?
So here I am drinking my coffee on a Saturday morning, hiding in my house in a shelter-in-place order, which is totally different from how I would otherwise spend a Saturday morning... um... how? Anyway, let's go back to those three variables from Alan Abramowitz's model: GDP growth in the second quarter, the incumbent's approval rating, and the two-term penalty.
As the 2020 election approaches, let's think about imputing some values here, and ask what will happen.
Trump won't face a two-term penalty. His approval rating? It has never really crossed the 50% threshold, for rather obvious reasons. RealClearPolitics currently has him at a 46.1% average approval rating. That's... less than stellar, with a 51.4% average disapproval rating. Net negative. That'll work against him, and for the cardboard cut-out of a Democrat who will be propped up on a debate stage.
And then there's Q2 GDP. That'll be bad. Very, very bad. Look, folks, this is a recession. A bad one. The reason Trump is pushing to end shelter-in-place orders and mandatory closures is this. We are in Q2 right now. It runs April through June. Even if businesses start opening up in May, we won't be out of the woods for a long, long time. You may be reading speculation on whether the economic recovery, when it eventually happens, will be a V-shaped recovery, or an L-shaped recovery, and you can picture what that means, but the point is that right now, we are either around the trough of the V, or the angle of the L. Either way, impute a number that can be plugged into the Abramowitz model, and there's a pretty high likelihood that it will say, "D beats R."
So, if I'm true to my word, which you can't go back and quote to me because it was on The Unmutual Political Blog, and that is no more, but I'm an intellectually honest person, so I'm putting this up anyway, but... where was I? Oh, yeah. If I am true to my word, I ignore everything weird about the world, plug numbers into the Abramowitz model, the Erikson & Wlezian model, and everything else that gets printed in the October 2020 issue of PS: Political Science & Politics, compute the weighted average that I compute every election, and just say, "here's what's going to happen. Ignore the noise. Here's the signal. Watch me be true to my word."
Or, I could get squirrelly. Watch my tail twitch. Why? Because the coolest superhero of all is obviously Squirrel Girl.
Anyway, remember that thing I posted last week about how the social sciences are different from the physical sciences? The world changes, people are contextual critters, and all of that?
So, when was the last time you went into a) work, physically, or b) the classroom, physically (assuming they are different for you)?
Yeah, that's what I thought. The gist of that post last week was that we are living in one of those moments during which things change. And rules change. Economic structures can change, political structures can change, and everything social scientists know about one system can cease to apply when the world goes through a dramatic transition. And the thing is, you don't necessarily know what will continue to apply, and what won't.
[Tail twitching...]
There is a high likelihood that political science forecasting models will, when numbers become available, predict a Democratic victory based on the economic collapse we are experiencing. Is that, at this point, definitive? No, but we have unemployment figures that are hard to reconcile with any kind of Q2 GDP that wouldn't put an incumbent below reelection, particularly when that incumbent has never been above water in approval ratings.
So, why is my tail twitching?
1) Let's start with the voting process. We have no idea what the state of coronavirus will be in November. Will people be willing to go to the polling places, physically? We have no idea, and there is a high likelihood of another wave of infections by November. This isn't a "one candidate is just really weird" observation. This is a "can democracy even work, mechanically?" observation.
2) If there is another wave of infections by November, either nobody votes, or voting requires widespread shifts to vote-by-mail. That latter possibility would depend on state-level action. What do states do? We're back to the mechanics of democracy, with state-level strategic actors making decisions from a partisan perspective. Trump has already said, point-blank, that he opposes vote-by-mail because he thinks it would help Democrats, and yet without it, combined with a second wave of infections in November, the mechanics of elections break down completely. Forecasting that? You want me to trust old models in that circumstance? I don't get to reconsider?
3) Will elections be canceled? You have probably read various analyses about the legality of the alternative possibilities for election cancelations in November. Ask me how much I factor "it's blatantly illegal and unconstitutional" into my prediction of whether or not something happens anymore. Go ahead and ask me.
Remember how we used to have a system in which Congress appropriated money to the projects it chose to fund? That was Congress's power? Yeah, not so much anymore.
If I start down this road, I'll never stop, but the point is that the legality or constitutionality of various "cancel the election" plans will not factor into whether or not they happen. The practical consequences will. If governors are considering state-level decisions to cancel their elections and they decide that it would result in blood in the streets, and they'd rather face a Biden presidency than violence, then say hello to President Probably-Rapes-Fewer-Women-But-Hey-What-Are-You-Going-To-Do? My point is that we no longer have a constitutional order. Practical and strategic concerns will continue to bind political actors, and those will include expectations of "democratic" procedures, but laws? Constitutional processes? Those don't exist anymore. Political science forecasting models amid this? Really?
4) Election interference. In 2016, I wrote off Russia's interference as a minor factor at most, but this year, we will see more foreign interference by orders of magnitude, because it is being invited by the President, openly. Congressional Republicans gave him permission, and he has publicly stated that he intends to collaborate with foreign governments during the campaign. This will be like nothing you have ever seen before. I can't even begin to predict what this will do to elections. I consider this a lesser factor than 1-3, but worth stating.
5) Post-election court challenges. In 2016, one of the things that was clear was that Trump was not going to concede. It wasn't clear that he would win, but it was clear that he was not going to concede. The question is, what happens if a sitting president approaches the election with a known refusal to concede, possession being 9/10ths. The bigger point, beyond the possession observation, though, is what has happened to the court system. Even if Biden wins, we know, with 100% certainty, that Trump will refuse to concede, and challenge the results in various courts. One of the biggest differences between now and then is that Trump has appointed a lot of those judges, and 2 of the Associate Justices of the Supreme Court.
There is more to say here, but I think this is enough to justify my unwillingness to defer as mechanically as I promised I would to the great, Delphic oracle of Emory, Alan Abramowitz. As I wrote last week, I must grudgingly admit that political science really is different from the physical sciences because people are contextual, and right now, context is different from any time in modern history. This is a moment in which society is shifting. I don't know what rules will continue to apply.
Do I keep my post-2016 promise? Um... I'd really like to, but right now, I'm not sure how much of political science I want to throw into the dustbin of intellectual history, and frankly, forecasting models were never more than a sideline hobby for anyone. They are fodder for blogs, and fun as far as they go, but at this point, social science needs to worry about whether or not the big stuff is about to get tossed overboard.
After the 2016 election, I made the following pledge: I will never question the political science forecasting models, or more specifically, the "Time For A Change" model, again.
Here's what happened, as a refresher. Contrary to popular misconceptions, political science actually got 2016 right. A generic R was supposed to beat a generic D, based on underlying political fundamentals. The model that I have consistently cited as my favorite-- Alan Abramowitz's "Time For A Change" model-- was one of the many models that worked. It forecast an R victory. Why? It had three variables: GDP growth in the second quarter of the election year, the incumbent's approval rating (even though he wasn't running), and a two-term penalty for the incumbent party because one party almost never wins a third term. That final variable was the thing that really worked against the Democrats in 2016, according to the math from Abramowitz. That model predicted an R victory.
Of course, Alan himself, like most political scientists, threw out that prediction and said that Hillary Clinton would win. Why? Well, first there were the polls, and then there was... Trump. You know, that guy who tells you that you should take cleaning products intravenously, forcing a press release from the makers of Lysol about the improper use of their products. Let that sink in. Metaphorically!
Anyway, I was one of the people who disregarded the political science forecasting models in 2016, and predicted a Hillary Clinton victory. In the aftermath, I made a promise. Never again would I disregard those models. Yup. From then on, I'd just go with the forecasting models. What, am I some kind of fool? You know, the kind who would guzzle bleach to cure a virus?
So here I am drinking my coffee on a Saturday morning, hiding in my house in a shelter-in-place order, which is totally different from how I would otherwise spend a Saturday morning... um... how? Anyway, let's go back to those three variables from Alan Abramowitz's model: GDP growth in the second quarter, the incumbent's approval rating, and the two-term penalty.
As the 2020 election approaches, let's think about imputing some values here, and ask what will happen.
Trump won't face a two-term penalty. His approval rating? It has never really crossed the 50% threshold, for rather obvious reasons. RealClearPolitics currently has him at a 46.1% average approval rating. That's... less than stellar, with a 51.4% average disapproval rating. Net negative. That'll work against him, and for the cardboard cut-out of a Democrat who will be propped up on a debate stage.
And then there's Q2 GDP. That'll be bad. Very, very bad. Look, folks, this is a recession. A bad one. The reason Trump is pushing to end shelter-in-place orders and mandatory closures is this. We are in Q2 right now. It runs April through June. Even if businesses start opening up in May, we won't be out of the woods for a long, long time. You may be reading speculation on whether the economic recovery, when it eventually happens, will be a V-shaped recovery, or an L-shaped recovery, and you can picture what that means, but the point is that right now, we are either around the trough of the V, or the angle of the L. Either way, impute a number that can be plugged into the Abramowitz model, and there's a pretty high likelihood that it will say, "D beats R."
So, if I'm true to my word, which you can't go back and quote to me because it was on The Unmutual Political Blog, and that is no more, but I'm an intellectually honest person, so I'm putting this up anyway, but... where was I? Oh, yeah. If I am true to my word, I ignore everything weird about the world, plug numbers into the Abramowitz model, the Erikson & Wlezian model, and everything else that gets printed in the October 2020 issue of PS: Political Science & Politics, compute the weighted average that I compute every election, and just say, "here's what's going to happen. Ignore the noise. Here's the signal. Watch me be true to my word."
Or, I could get squirrelly. Watch my tail twitch. Why? Because the coolest superhero of all is obviously Squirrel Girl.
Anyway, remember that thing I posted last week about how the social sciences are different from the physical sciences? The world changes, people are contextual critters, and all of that?
So, when was the last time you went into a) work, physically, or b) the classroom, physically (assuming they are different for you)?
Yeah, that's what I thought. The gist of that post last week was that we are living in one of those moments during which things change. And rules change. Economic structures can change, political structures can change, and everything social scientists know about one system can cease to apply when the world goes through a dramatic transition. And the thing is, you don't necessarily know what will continue to apply, and what won't.
[Tail twitching...]
There is a high likelihood that political science forecasting models will, when numbers become available, predict a Democratic victory based on the economic collapse we are experiencing. Is that, at this point, definitive? No, but we have unemployment figures that are hard to reconcile with any kind of Q2 GDP that wouldn't put an incumbent below reelection, particularly when that incumbent has never been above water in approval ratings.
So, why is my tail twitching?
1) Let's start with the voting process. We have no idea what the state of coronavirus will be in November. Will people be willing to go to the polling places, physically? We have no idea, and there is a high likelihood of another wave of infections by November. This isn't a "one candidate is just really weird" observation. This is a "can democracy even work, mechanically?" observation.
2) If there is another wave of infections by November, either nobody votes, or voting requires widespread shifts to vote-by-mail. That latter possibility would depend on state-level action. What do states do? We're back to the mechanics of democracy, with state-level strategic actors making decisions from a partisan perspective. Trump has already said, point-blank, that he opposes vote-by-mail because he thinks it would help Democrats, and yet without it, combined with a second wave of infections in November, the mechanics of elections break down completely. Forecasting that? You want me to trust old models in that circumstance? I don't get to reconsider?
3) Will elections be canceled? You have probably read various analyses about the legality of the alternative possibilities for election cancelations in November. Ask me how much I factor "it's blatantly illegal and unconstitutional" into my prediction of whether or not something happens anymore. Go ahead and ask me.
Remember how we used to have a system in which Congress appropriated money to the projects it chose to fund? That was Congress's power? Yeah, not so much anymore.
If I start down this road, I'll never stop, but the point is that the legality or constitutionality of various "cancel the election" plans will not factor into whether or not they happen. The practical consequences will. If governors are considering state-level decisions to cancel their elections and they decide that it would result in blood in the streets, and they'd rather face a Biden presidency than violence, then say hello to President Probably-Rapes-Fewer-Women-But-Hey-What-Are-You-Going-To-Do? My point is that we no longer have a constitutional order. Practical and strategic concerns will continue to bind political actors, and those will include expectations of "democratic" procedures, but laws? Constitutional processes? Those don't exist anymore. Political science forecasting models amid this? Really?
4) Election interference. In 2016, I wrote off Russia's interference as a minor factor at most, but this year, we will see more foreign interference by orders of magnitude, because it is being invited by the President, openly. Congressional Republicans gave him permission, and he has publicly stated that he intends to collaborate with foreign governments during the campaign. This will be like nothing you have ever seen before. I can't even begin to predict what this will do to elections. I consider this a lesser factor than 1-3, but worth stating.
5) Post-election court challenges. In 2016, one of the things that was clear was that Trump was not going to concede. It wasn't clear that he would win, but it was clear that he was not going to concede. The question is, what happens if a sitting president approaches the election with a known refusal to concede, possession being 9/10ths. The bigger point, beyond the possession observation, though, is what has happened to the court system. Even if Biden wins, we know, with 100% certainty, that Trump will refuse to concede, and challenge the results in various courts. One of the biggest differences between now and then is that Trump has appointed a lot of those judges, and 2 of the Associate Justices of the Supreme Court.
There is more to say here, but I think this is enough to justify my unwillingness to defer as mechanically as I promised I would to the great, Delphic oracle of Emory, Alan Abramowitz. As I wrote last week, I must grudgingly admit that political science really is different from the physical sciences because people are contextual, and right now, context is different from any time in modern history. This is a moment in which society is shifting. I don't know what rules will continue to apply.
Do I keep my post-2016 promise? Um... I'd really like to, but right now, I'm not sure how much of political science I want to throw into the dustbin of intellectual history, and frankly, forecasting models were never more than a sideline hobby for anyone. They are fodder for blogs, and fun as far as they go, but at this point, social science needs to worry about whether or not the big stuff is about to get tossed overboard.
Why toss away so much data if the only prediction is "win" or "lose?" Most of the models predict some kind of ratio-level variable--2-party vote share, EC vote share, vote share, etc.
ReplyDeleteAbramowitz's model predicted about a 3-point win for Trump, and he got a 3-point loss. To my mind, the model WAS wrong in 2016. But, it HAD been right for a long time.
Norpoth's model was even further off: a 5-point win. Fair predicted a 10-point win. The fact that the direction of their predictions was in the direction that a terrible distribution of voters (plus Comey) ended up producing is dumb luck, not accuracy. Erikson & Wleizen come off relatively good, but as the Abramowitz model shows, it's really possible that models can outlive their predictive usefulness and we won't know when. Abramowitz himself, for all his model's parsimony, keeps messing with it to get it to produce the result he wants. At the end of the day, I think you and I trust Abramowitz's core model more than he does.
I actually still believe in the core model: call it Hibbs' Bread and Peace or whatever. Then, we just have to add in a factor for candidates still mattering. Abramowitz has a fairly parsimonious way of doing that: presidential approval, which gets us robust data, but only really on something less than 1/2 of the data (1/2 for the parties, minus a factor because Al Gore isn't Bill Clinton, etc.) For most of modern history, the parties nominated normal people, so candidate effects largely washed out, making approval a great proxy. Trump and Fox have basically blown the underlying assumptions of that model out of the water.
That's kinda what you're saying, but it's different. I don't think the constitutional stuff really factors in for the predictive purposes. We can still predict vote shares. We just know that there's a non-zero chance that those will be fixed or ignored. We were never really in the business of forecasting winners; we were always forecasting votes. And I think those are just as predictable as they used to be; it just may take different variables to predict them. The context may have changed, but our units of analysis have not: voters.
I'm not sure how to handle a "wrong" prediction regarding a popular vote/EC split when, as I often have to remind people, there's no such thing as "the popular vote." It doesn't exist. It is an illusion because that's not the contest. Ultimately, it is about predicting a winner, and Abramowitz's model predicted the right winner.
DeleteRegardless, the biggest modification Abramowitz ever did was the modification for polarization, and in retrospect, I agree with that one. As far as economic measures, I prefer Hibbs, but unfortunately, he didn't put an entry into the 2016 PS issue, so... whatever. Regardless, you are right that you and I trust "Time For A Change" more than Abramowitz. After all, he said himself that he distrusted it in 2016.
As far as the constitutional/court stuff goes, here's the problem. It's all about counting. Court challenges are going to be about which votes get thrown out, creating the final official tally. That was what the Florida challenges were in 2000. The final official tally was a Bush margin of 537 votes. What would it have been under various alternative standards? Depends on the standard. Gore only wanted a recount in three counties, but a full statewide recount? Eh... So, what happens if Trump loses, and we see court challenges across the country changing the tallies? We get new counts. You and I can create models to forecast votes, but we can't forecast court challenges that selectively throw out some votes if one set of judges doesn't like the result.