Kamala Harris and Philip Tetlock

You should know both of those names, but you probably know at least one of them.  Anyway... fine.  News of the week.  Besides, I did a very, very early veepstakes post here, and that brings me to the topic of prediction, which brings me to Tetlock.

Last fall, I wrote a veepstakes post.  Long before Biden was the nominee, and long before he made a promise to select a woman or suggest an African-American woman, I predicted that no nominee would select a white male, and that the process would fall, to a significant extent, along demographic lines.  That was easy.

Of course, I didn't get the name.  I got the process, but not the name.  Let's examine the process of prediction, because this presents some interesting lessons about the nature of wrongness.

Hi!  I'm an expert in wrongness.

When I started musing about VP picks, Elizabeth Warren was leading the Democratic field.  So, I wrote about who Warren might pick, and who the other contenders might pick if Warren fell.  My call for Biden's most likely pick was... Warren.  Y...eah, not so much.

Reference time.  Philip Tetlock, Expert Political Judgment.  Whether or not I can be called an "expert," on politics rather than wrongness is a matter of some debate, but I'z got's me letters after my name and either you're reading this for some glimmer of insight, or to play some stupid game of gotcha like a jerk.  Of course, whether or not I have insight is a debatable matter.  So... ya' got me.  Today's topic:  why you got me.

There are several basic elements of Tetlock's book that are relevant here, and that I keep referencing.  First, a "prediction" is not as simple as a 1 or 0 when the forecaster is actually assigning probabilities.  This is the most difficult thing for a layperson to grasp.  If I say, "there are three possibilities:  A, B and C.  A has a .4 probability of being true, B has a .3 probability of being true, and C has a .3 probability of being true," and C happens, was my statement wrong?  The layperson is going to say yes.  I "predicted" A.  But... that's not actually what I did.  I assigned A the highest probability of occurring, among the set of all possibilities.  However, I also said there was a 60% chance A would not occur.  So, to say that I "predicted" A when I said there was a 60% chance it wouldn't occur... doesn't make mathematical or grammatical sense.  This is where we get to the gap between how statisticians make "predictions," and how laypeople talk about them.

And even if a forecaster says, "X has a 60% chance of occurring," if X does not occur, does that mean the probability assessment was wrong?  No.  You can't actually test that claim without repeated trials to compare the proportion of the time X occurred to .6 with repeated trials.  Once ain't-a-gonna do it, scientifically speaking.  'Cuz scientists talk that way.  Again, this is the gap between how statisticians discuss forecasts and how laypeople do.

This is just a linguistic thing, though.  Tetlock made important distinctions between forecasts and the reasoning behind forecasts.  Here, I'm going to pick on a very famous but very inept person named Allan Lichtman.  You'll see this... individual trotted out by media figures who have no training in the social sciences 'round about election time because he is supposedly such a brilliant prognosticator.  You see, he predicts winners!

If enough monkeys bang away on enough typewriters, eventually, one of them will write Hamlet.

So let's shut Lichtman down, with math.  He has a bunch of "keys" that he uses to predict elections, and he has a set of historical elections that his "keys" predict.  13 keys.  He uses them to predict elections.  He's been right since 1984!  Check your math here.  That's 9 elections-- 9 equations-- and 13 variables.

No.  No.  Bad Allan!  No treat.

You wanna go back before 1984?  If we're in straight-up algebra territory, '68 puts you in 13 equations/13 variables territory, but scientifically, this is still a joke.

After all, his model works by having it be a majority of keys.  That puts you in combinatorics territory, so it isn't really 13.  He wants to pretend it's binary-- whichever candidate has 7 out of 13-- but against what does he test that?

Nothing.  That'd be science.  Introducing the notion of an interaction effect puts us in combinatorics territory and requires even more cases, which he doesn't have.

This is not how math works.  This is an unsolvable system.  It's not a system at all.  That's before we get into the fact that there is no coherent social scientific theory behind his "keys."  It's just a bunch of garbage he pulled out of... somewhere, which makes no mathematical sense.  He doesn't understand the problem because he can't do math, and has no understanding of social science.

And as for his supposedly amazing record of prediction (monkeys and typewriters...)...

You know who else predicts elections?  Alan Abramowitz.  He can go back further into history.  With three variables.  GDP in the second quarter of the election year, the incumbent's approval rating, and whether or not the incumbent party has won two terms in a row.  With those three variables, he can tell you not just which party will win, but the margin.

Why is Lichtman the one who gets trotted out by the press?  Affinity bias.  Abramowitz uses a parsimonious, social scientific, mathematical model.  Lichtman is a bloviating bullshitter who can't do math and hasn't studied the subject upon which he pontificates.

Abramowitz is your guy.  (Realistically, the way to go is to take the forecasting models together, when they are printed in the October issue of PS: Political Science & Politics, and Abramowitz will publish his paper along with several other forecasters.  Point being, ignore Lichtman.)

I honestly don't know what historians think of Lichtman, but to political scientists, he's a joke, and his "model" isn't a joke of a model, because to be a joke of a model would require being... a model.  What does any of this have to do with Tetlock?  There is such a thing as making a correct prediction for the wrong reason.  There is also such a thing as making a wrong forecast for the right reason.

Point being, you need to examine not just a forecast itself, but the reasoning behind it.  Lichtman has made correct predictions.  His reasoning is less unscientific than an astrologer's, but not by much.  And by the numbers, if you go looking, you can find plenty of astrologers who have gotten it right every year since 1984 too.  That doesn't mean you have found an authentic prognosticator.

So.

Why'd I say that Warren would be a likely pick for Biden, as of last Fall?  And similarly, why did I discount Harris?  In order for Biden to take down Warren, it would have required a bruising contest, creating a rift between Biden and the left-wing faction of the party.  A Warren pick would be a conventional way to heal that rift.  Why did I discount Harris?  She wouldn't give Biden a state he wouldn't win anyway, and her debate performance after her Biden zinger fell flat, which took away her biggest appeal as a nominee-- crushing Pence like a bug during the VP debate.

At the time last Fall, Warren was the rising star of the field.  The Democratic Party has been going hard left for years.  Warren, opportunist that she is, tried to ride that wave to a presidential nomination.  Given that, it wasn't out of line to suggest Warren as the likely pick if Biden managed to beat her.  As for Harris, at the time, it wasn't clear what she added, so the reasoning at the time holds up, I think.

And what clearly held up was my demographic argument.  I was way ahead pointing out that no nominee could pick a white male.  I made that forecast long before Biden got the nomination, before he made it explicit that he would nominate a woman, and before the politics of the process moved attention towards women of color.  So, my framework was largely correct.  Biden first narrowed the search to women, then to women of color, and then made a decision within that subset.  Last Fall, I got the right process, but came to the wrong name.

Tetlock has a lot to say about this.

The importance of external events.  The political world, unlike Newtonian mechanics, is probabilistic.  One of the primary reasons is that exogenous events, which cannot be predicted, intervene to affect the basic framework within which we are working.  Two things happened between my veepstakes post and the Harris pick.  One potentially big, and one very big.  First, Rep. James Clyburn (D-SC) endorsed Biden.  The importance of endorsements is... a complicated matter.  I've picked on The Party Decides more than almost any other book I've ever read, because I think it's a bad book.  (Not as bad as what Lichtman writes, but it wasn't a good book.)  Basically, the authors say that party elites control the nominating process through mechanisms such as endorsements.  That's... not quite this.  However, one could make the case that Clyburn turned the election.  One could make the case that Clyburn swung South Carolina for Biden, which changed the dynamic.  I'm going to want to see a lot more data on that-- more than I have seen so far, and in particular, the rolling cross-sectional data from Annenberg with bunches-'o-analysis-- but let's say it's true for the sake of argument.  It means that Biden doesn't necessarily pick Warren.  It means Biden's nomination is inextricably linked to his support from the African-American community.

Add to that the really big thing-- the death of George Floyd and the rise of the Black Lives Matter movement.  Before that, in principle, Biden could have picked a white woman, although never a white male, but after George Floyd, the politics of the Democratic Party required an African-American woman.  Absolute, 100%.  He was locked in, at that point.  The party would not have tolerated a different choice.  George Floyd's death and the rise of the Black Lives Matter movement took Warren out of it completely.

This is one of those exogenous events that could lead an intellectual extremist towards what Tetlock calls radical skepticism, in reference to a philosophical/epistemological approach, but which those of us in the social sciences simply accept as a limitation on our ability to make deterministic forecasts of discrete events.  In other words, it's why we mostly predict patterns, and leave stuff like veepstakes forecasts to blogs and other meaningless outlets for weekend ramblin'.  From my perspective last Fall, Warren looked like a reasonable pick, should Biden defeat her, not being able to predict something like the rise of the Black Lives Matter movement in the wake of George Floyd.  After that, Warren was off the table completely.

Of course, as of last Fall, I was writing off Harris, and looking at Stacey Abrams, who had a lot of buzz.  Um... not so much.  What happened?

Well, have there been any other exogenous events that might have made Harris more appealing?  Um... is there anything going on in the world right now?  Anything that might make you crave a sort of no-nonsense competence over a base-mobilizer?

My assessment of Abrams last year was that she would help Biden heal some rifts with the left wing.  The thing is... she is not ready for the national stage.  She hasn't held national office.  Compare her, on basic qualifications to Harris and Rice, who became the top two.  Now, what's goin' on right now?

COVID.  The world was turned upside-down, or perhaps Up-and-Under (as I read Seanan McGuire's Middlegame), and we've seen what happens when you hand the keys to an amateur.

Whatever your ideology is, take that out of the picture.  Give the keys to Mike DeWine right now, and you'd kind of relax, wouldn't you?  Why?  Because he's competent, and experienced, and even-keeled, and that's what the country needs.  Take ideological disputes out of the picture, and DeWine would say relax, I got this.  Same thing with Harris or Rice.

Put Harris in the White House in a crisis, and we're fine.  She's got it.  Relax.

Abrams may have been appealing, based on the premise that she could put Georgia in play as it turns purple, and she could heal Biden's rift with the left.  From the perspective of 2019, with Warren in the lead and no COVID, an unconventional pick like Abrams might have made some sense, but a lot of that was buzz creating buzz.

Right now, in this environment, it's just not realistic to go with anyone other than a nationally-tested, conventionally-credentialed choice.  That meant either Harris or Rice, and really, that meant Harris.  In a different political environment, Biden could have gone with someone else, but the combination of factors that have turned the world Up-And-Under since this Spring took my basic framework and narrowed the list effectively to Kamala Harris.  Thanks, Phil.

The thing is, this process was probably hurt by Biden's statement that he was going to nominate a woman, and the movement to nominate, specifically, a woman of color because it affects how people respond to Harris.  Take away all of that.  Take away the affirmative action stuff, the Bass/Castro stuff, and everything else.

Imagine that Biden just said, "we have a crisis.  I'm looking for the person with the best qualifications and the best temperament to take over if necessary at a moment's notice because that's what's most important."  Look at the field of senators and governors, think about who'd be cool in a crisis completely separately from anything demographic, and it's not a stretch to say you land on Harris anyway.

Sort of changes the dynamic, though, doesn't it?