Impeachment, prediction and modes of thinking
I have psychic powers! You doubt me?! Observe. You want to ask me whether or not Donald Trump will be impeached!
Let's distinguish between the impeachment of Donald Trump and the conviction of Donald Trump in the Senate. How likely is it that the House of Representatives will pass articles of impeachment? I don't know. When we say that a president "is impeached," that is what it means, technically. The House has passed articles of impeachment. Andrew Johnson was impeached. Bill Clinton was impeached. Neither were removed from office. Richard Nixon was not impeached. He resigned based on the fear that he would not only be impeached, but convicted by the Senate. Will Trump actually "be impeached?" I have no idea. With respect to investigations of Russia and obstruction of justice related to Russia, it was always clear to me that he wouldn't be impeached because Nancy Pelosi knew that a) he wouldn't be convicted by the Senate, and b) the electoral blowback in 2020 would hurt the Democrats. (That's... still clear, but other things are afoot.)
Ukraine looks dicier. Nancy Pelosi is being dragged, kicking and screaming, into impeachment proceedings because she sees the politics the same way, but more members of her caucus see things differently.
A few reminders on terminology. First, "negative agenda control." Party leaders in the House control the chamber, not usually by twisting arms, but by preventing proposals from getting a vote if a) the vote would pass, and b) leadership doesn't want it to pass. Second, "conditional party government." Party leaders are permitted to exercise power when the party is unified. Party leadership loses power as the party starts to divide itself. Right now, that's happening, so Nancy Pelosi is losing her ability to exercise negative agenda control. So, I have no idea whether or not she can prevent a vote on articles of impeachment. However, should those articles pass, what is the likelihood that the Senate will convict Donald Trump?
Zero. There's a little electronic shock device in my brain, installed by the probability theorists and econometricians who conferred a degree upon me, lo-those-many-years-ago, and that electronic shock device will punish me if I tell you that the probability of Trump's conviction is absolute zero, or that the probability of anything is absolute zero, but I call it functional zero. Let's call it epsilon, which is the Greek character we use to represent a mathematical value arbitrarily close to, but not precisely equal to zero. There, are you happy now, gizmo-in-my-brain?
Now, let's have a little discussion about Bayesian updating. Here's what has happened. My assessment of the probability of Trump's impeachment, meaning, the probability of passage of articles of impeachment in the House, has gone up. It has gone up from some really low number (for the sake of inputting a value, let's call it 5%) to throwing up...
... my hands in uncertainty. For the sake of typing a number now, let's call that 45%. I lean towards no, but it is close enough to a coin toss that I have no confidence in my "no" assessment. Betting this morning at PredictIt calls it a coin toss. If someone gave me an estimate of 50-50, I wouldn't quibble. Either way, my estimated probability of impeachment went way up.
What happened? I got new information. The Ukraine phone call, along with Trump's decision to put military aid to Ukraine on hold, shortly before the call, along with the transcript of the call, Giuliani's role, and so on. More importantly than the information itself, though, I know how Democrats are going to react to it. I see new information, and my "prior" probabilistic assessment of the likelihood of impeachment goes up. This is how "Bayesian updating" goes. Probabilistic assessments are essentially what you do in a state of uncertainty, and probabilities are just mathematical representations of the quality of information you possess at any given time. As you get more information, you change your probabilistic assessments to reflect the new information you possess.
That's Bayesian statistical theory, in a non-math-y form.
Notice, though, that I have not "updated" my probabilistic assessment of the likelihood that Trump will be convicted in the Senate. What's the deal?
The short version is that I'm not using statistical reasoning to get to my assessment of Trump's "likelihood" of conviction. So, no Bayesian updating necessary. In the same way that I don't need new information to re-consider whether or not 2+2=4, I don't need new information to re-consider whether or not Senate Republicans will turn on Trump. They won't. This isn't a probabilistic statement. It is an axiomatically derived statement. I approach the question in a completely different way.
This is about epistemology.
Democrats. How do I make predictions about what they will do? In this particular case, there are several issues. The most important of those issues is that they are divided, primarily along lines of strategy. Are any of them truly stupid enough to think that the Senate will convict Trump, and force him from office? Not many.
The Democrats who support impeachment-- those who are forcing Pelosi's unwilling hand, lest she go the way of Joe Cannon and John Boehner-- mostly know that the result of Trump's formal impeachment will inevitably be acquittal in the Senate. So... why do it?
You can go read the arguments elsewhere, but they fall along two general lines. There is the moral line, and the indirect effect line of reasoning.
The moral line of argument is that Trump just should be impeached because he did wrong, so even if he'll be acquitted in the Senate, House Democrats just should impeach him. No prosecutor in any other context follows this line of reasoning, but it seems to be prominent. It is the tilt at a windmill, dream the impossible dream approach to politics, which also seems to be taking over other debates within the Democratic Party. So... do they like windmills, or want to bang away at them with swords? I'm confused...
Then, there's the bank-shot thinking. It won't remove Trump, but maybe impeaching him will have some other effect, like deterrence, or helping Democrats in 2020. Of course, I'm not sure how a guaranteed acquittal deters anything, and the only modern example of a failed impeachment is Clinton, which also gave us one of only two modern midterm elections in which the president's party gained seats. The other was the post-9/11 election of 2002. That makes the electoral theory ahistorical, at best.
There's more, but I'm just painting with a broad brush in order to demonstrate the wide range of ideas that don't exactly follow from, well... game theory. Remember last week's post on "backwards induction?" I'm coming back to that. The basic point is that there is a wide range of thought within the Democratic Party on what impeachment may or may not do.
That makes the Democrats hard to predict. They aren't thinking about this in consistent, historically-grounded, game-theoretic terms. That leaves me with Bayesian updating. So, new information. Probability of impeachment goes up. Will it backfire if they do? The only modern precedent of a failed impeachment says, yes.
So, what about the Republicans? Why haven't I been doing Bayesian updating with them? Because I'm doing game theory with them.
The only way the Republicans lose is if they decide to lose. This is the old Silent Cal problem, which is a reference I have been making for a while now. There is an old, and probably apocryphal story you have probably heard about Calvin Coolidge, the notoriously taciturn President. A woman bet him that she could get him to say more than two words at a dinner party. His response: "you lose."
The problem with the apocryphal woman's bet was that she was not in control of whether or not she would win. Cal was. Completely. All he had to do to win was choose to win. The only way he would lose was if he chose to lose. OK, there's something to be said for intelligence, focus and willpower, and all of that jazz, but when you strip that all away and focus on the structure of the game... it wasn't even a "game," in the technical sense of "game theory." It was "decision theory," in microeconomic terms. Either Silent Cal would choose to win, or choose to lose. He chose to win.
Why? Because why would he choose to lose?! Stupid bet.
The threshold for conviction in the Senate is 2/3. The only way Trump could be convicted-- even if there were an impeachment after Democrats picked up some hypothetically plausible majority in 2020, with Trump retaining office-- would be if a lot of Republicans chose to convict.
That choice would destroy, not just Trump, but the Republican Party. Watergate was a disaster, not just for Nixon, but for the entire GOP. How bad? It actually gave us a great story about... partisan gerrymandering! The way it works is that the party trying to gain maximum advantage gives itself narrow majorities in as many districts as possible, but there's a risk to that. What if opinions of the two parties shift? A small shift can cause you to lose a lot of seats. That... happened.
Republicans tried a partisan gerrymander in New York. Right before Watergate. Oopsies! All of those narrow Republican districts they gave themselves? Yeah, they went Dem when the scandal broke. (The political science term for this is: "dummymander.") Then... Carter. Watergate was a disaster for the GOP.
And to keep my Thomas Schelling references going, congressional Republicans have tied themselves to Trump far more inextricably than any Republican ever tied themselves to Nixon. The effect is to add cost were they to turn on him. They can't anymore. In 1974, the Republican Party wasn't first and foremost, "The Nixon Party," but today, the GOP is "The Trump Party." That means if Senate Republicans turned on him and convicted him, the consequences for the party would be far worse than Watergate. That's why they can't. That's the point. They're locked in. The cost of turning on him is higher than the cost of continuing to defend him.
So here it is. If the Senate GOP convicts Trump, it's a party meltdown. If they stand in lockstep behind Trump, the best historical reference is 1998. The rational choice is the lockstep choice.
And there is a greater point here-- the rational choice for the Republican Party is the same choice as the psychological choice. They are psychologically predisposed to side with Trump, and that's their best move. For the Democrats? It isn't at all clear what the rational choice is. An impeachment is highly likely to backfire, but at the same time, it isn't clear what their options are if not impeachment, and the psychological choice-- impeachment-- is at odds with the probability of backfiring.
The Democrats' basic problem is that they can't make Silent Cal talk. Not even with the empty threat of contempt charges, when Jerry Nadler hauls him into a hearing.
Republicans? I can just do the game theory here. The structure of the politics won't change. They need to keep marching in lockstep. So, they will. They have no psychological pressure to do otherwise, and in fact, they have psychological reinforcement for that rational choice, so I'm not doing any Bayesian updating. I'm just doing game theory.
The Democrats have an extraordinarily difficult position, though. I don't know what they'll do. I'll keep watching and keep Bayes-ing. Watch me "verb" a name!
Let's distinguish between the impeachment of Donald Trump and the conviction of Donald Trump in the Senate. How likely is it that the House of Representatives will pass articles of impeachment? I don't know. When we say that a president "is impeached," that is what it means, technically. The House has passed articles of impeachment. Andrew Johnson was impeached. Bill Clinton was impeached. Neither were removed from office. Richard Nixon was not impeached. He resigned based on the fear that he would not only be impeached, but convicted by the Senate. Will Trump actually "be impeached?" I have no idea. With respect to investigations of Russia and obstruction of justice related to Russia, it was always clear to me that he wouldn't be impeached because Nancy Pelosi knew that a) he wouldn't be convicted by the Senate, and b) the electoral blowback in 2020 would hurt the Democrats. (That's... still clear, but other things are afoot.)
Ukraine looks dicier. Nancy Pelosi is being dragged, kicking and screaming, into impeachment proceedings because she sees the politics the same way, but more members of her caucus see things differently.
A few reminders on terminology. First, "negative agenda control." Party leaders in the House control the chamber, not usually by twisting arms, but by preventing proposals from getting a vote if a) the vote would pass, and b) leadership doesn't want it to pass. Second, "conditional party government." Party leaders are permitted to exercise power when the party is unified. Party leadership loses power as the party starts to divide itself. Right now, that's happening, so Nancy Pelosi is losing her ability to exercise negative agenda control. So, I have no idea whether or not she can prevent a vote on articles of impeachment. However, should those articles pass, what is the likelihood that the Senate will convict Donald Trump?
Zero. There's a little electronic shock device in my brain, installed by the probability theorists and econometricians who conferred a degree upon me, lo-those-many-years-ago, and that electronic shock device will punish me if I tell you that the probability of Trump's conviction is absolute zero, or that the probability of anything is absolute zero, but I call it functional zero. Let's call it epsilon, which is the Greek character we use to represent a mathematical value arbitrarily close to, but not precisely equal to zero. There, are you happy now, gizmo-in-my-brain?
Now, let's have a little discussion about Bayesian updating. Here's what has happened. My assessment of the probability of Trump's impeachment, meaning, the probability of passage of articles of impeachment in the House, has gone up. It has gone up from some really low number (for the sake of inputting a value, let's call it 5%) to throwing up...
... my hands in uncertainty. For the sake of typing a number now, let's call that 45%. I lean towards no, but it is close enough to a coin toss that I have no confidence in my "no" assessment. Betting this morning at PredictIt calls it a coin toss. If someone gave me an estimate of 50-50, I wouldn't quibble. Either way, my estimated probability of impeachment went way up.
What happened? I got new information. The Ukraine phone call, along with Trump's decision to put military aid to Ukraine on hold, shortly before the call, along with the transcript of the call, Giuliani's role, and so on. More importantly than the information itself, though, I know how Democrats are going to react to it. I see new information, and my "prior" probabilistic assessment of the likelihood of impeachment goes up. This is how "Bayesian updating" goes. Probabilistic assessments are essentially what you do in a state of uncertainty, and probabilities are just mathematical representations of the quality of information you possess at any given time. As you get more information, you change your probabilistic assessments to reflect the new information you possess.
That's Bayesian statistical theory, in a non-math-y form.
Notice, though, that I have not "updated" my probabilistic assessment of the likelihood that Trump will be convicted in the Senate. What's the deal?
The short version is that I'm not using statistical reasoning to get to my assessment of Trump's "likelihood" of conviction. So, no Bayesian updating necessary. In the same way that I don't need new information to re-consider whether or not 2+2=4, I don't need new information to re-consider whether or not Senate Republicans will turn on Trump. They won't. This isn't a probabilistic statement. It is an axiomatically derived statement. I approach the question in a completely different way.
This is about epistemology.
Democrats. How do I make predictions about what they will do? In this particular case, there are several issues. The most important of those issues is that they are divided, primarily along lines of strategy. Are any of them truly stupid enough to think that the Senate will convict Trump, and force him from office? Not many.
The Democrats who support impeachment-- those who are forcing Pelosi's unwilling hand, lest she go the way of Joe Cannon and John Boehner-- mostly know that the result of Trump's formal impeachment will inevitably be acquittal in the Senate. So... why do it?
You can go read the arguments elsewhere, but they fall along two general lines. There is the moral line, and the indirect effect line of reasoning.
The moral line of argument is that Trump just should be impeached because he did wrong, so even if he'll be acquitted in the Senate, House Democrats just should impeach him. No prosecutor in any other context follows this line of reasoning, but it seems to be prominent. It is the tilt at a windmill, dream the impossible dream approach to politics, which also seems to be taking over other debates within the Democratic Party. So... do they like windmills, or want to bang away at them with swords? I'm confused...
Then, there's the bank-shot thinking. It won't remove Trump, but maybe impeaching him will have some other effect, like deterrence, or helping Democrats in 2020. Of course, I'm not sure how a guaranteed acquittal deters anything, and the only modern example of a failed impeachment is Clinton, which also gave us one of only two modern midterm elections in which the president's party gained seats. The other was the post-9/11 election of 2002. That makes the electoral theory ahistorical, at best.
There's more, but I'm just painting with a broad brush in order to demonstrate the wide range of ideas that don't exactly follow from, well... game theory. Remember last week's post on "backwards induction?" I'm coming back to that. The basic point is that there is a wide range of thought within the Democratic Party on what impeachment may or may not do.
That makes the Democrats hard to predict. They aren't thinking about this in consistent, historically-grounded, game-theoretic terms. That leaves me with Bayesian updating. So, new information. Probability of impeachment goes up. Will it backfire if they do? The only modern precedent of a failed impeachment says, yes.
So, what about the Republicans? Why haven't I been doing Bayesian updating with them? Because I'm doing game theory with them.
The only way the Republicans lose is if they decide to lose. This is the old Silent Cal problem, which is a reference I have been making for a while now. There is an old, and probably apocryphal story you have probably heard about Calvin Coolidge, the notoriously taciturn President. A woman bet him that she could get him to say more than two words at a dinner party. His response: "you lose."
The problem with the apocryphal woman's bet was that she was not in control of whether or not she would win. Cal was. Completely. All he had to do to win was choose to win. The only way he would lose was if he chose to lose. OK, there's something to be said for intelligence, focus and willpower, and all of that jazz, but when you strip that all away and focus on the structure of the game... it wasn't even a "game," in the technical sense of "game theory." It was "decision theory," in microeconomic terms. Either Silent Cal would choose to win, or choose to lose. He chose to win.
Why? Because why would he choose to lose?! Stupid bet.
The threshold for conviction in the Senate is 2/3. The only way Trump could be convicted-- even if there were an impeachment after Democrats picked up some hypothetically plausible majority in 2020, with Trump retaining office-- would be if a lot of Republicans chose to convict.
That choice would destroy, not just Trump, but the Republican Party. Watergate was a disaster, not just for Nixon, but for the entire GOP. How bad? It actually gave us a great story about... partisan gerrymandering! The way it works is that the party trying to gain maximum advantage gives itself narrow majorities in as many districts as possible, but there's a risk to that. What if opinions of the two parties shift? A small shift can cause you to lose a lot of seats. That... happened.
Republicans tried a partisan gerrymander in New York. Right before Watergate. Oopsies! All of those narrow Republican districts they gave themselves? Yeah, they went Dem when the scandal broke. (The political science term for this is: "dummymander.") Then... Carter. Watergate was a disaster for the GOP.
And to keep my Thomas Schelling references going, congressional Republicans have tied themselves to Trump far more inextricably than any Republican ever tied themselves to Nixon. The effect is to add cost were they to turn on him. They can't anymore. In 1974, the Republican Party wasn't first and foremost, "The Nixon Party," but today, the GOP is "The Trump Party." That means if Senate Republicans turned on him and convicted him, the consequences for the party would be far worse than Watergate. That's why they can't. That's the point. They're locked in. The cost of turning on him is higher than the cost of continuing to defend him.
So here it is. If the Senate GOP convicts Trump, it's a party meltdown. If they stand in lockstep behind Trump, the best historical reference is 1998. The rational choice is the lockstep choice.
And there is a greater point here-- the rational choice for the Republican Party is the same choice as the psychological choice. They are psychologically predisposed to side with Trump, and that's their best move. For the Democrats? It isn't at all clear what the rational choice is. An impeachment is highly likely to backfire, but at the same time, it isn't clear what their options are if not impeachment, and the psychological choice-- impeachment-- is at odds with the probability of backfiring.
The Democrats' basic problem is that they can't make Silent Cal talk. Not even with the empty threat of contempt charges, when Jerry Nadler hauls him into a hearing.
Republicans? I can just do the game theory here. The structure of the politics won't change. They need to keep marching in lockstep. So, they will. They have no psychological pressure to do otherwise, and in fact, they have psychological reinforcement for that rational choice, so I'm not doing any Bayesian updating. I'm just doing game theory.
The Democrats have an extraordinarily difficult position, though. I don't know what they'll do. I'll keep watching and keep Bayes-ing. Watch me "verb" a name!
Comments
Post a Comment