Thread by @Moshe_Hoffman

Recent editorial 👇 (by Kahneman & Renshon) offers a readable summary of many interesting ‘judgment and decision making’ findings pertinent to trigger-happy leaders.

But imo ALSO exemplifies a key limitations of jdm-style explanations.

(Thread.)

foreignpolicy.com/2009/10/13/why…

So the article talks about -overconfidence
-reactive devaluation
-sunk cost fallacy
-the fundamental attribution error

etc.

(I’ll explain.)

2/

That is,

We are prone to go to war b/c
-we over-estimate our likelihood of victory
-we presume our adversaries concessions are not meaningful
-we throw good troops after bad. (Think “the surge.” Or basically every stage in Vietnam.)

3/

And

-we are likely to see our own hostile acts as being “forced”, but not realize when our adversaries were likewise pushed into a corner.

All true.

All well documented phenomena. All valuable insights. All worth keep in mind when interpreting current events.

4/

But.

K&R seem to be missing the key underlying cause.

And instead ascribing causality to a hodgepodge of phenomena that imo are merely proximal.

Allow me to explain what I mean. And why i think this matters.

5/

The first thing I’ll note:

Whenever you see ten causal arrows all pointing to the same endpoint, you gotta wonder if there is an ultimate cause. Something pushing so many arrows in that direction.

6/

If you saw a map of the ancient Mediterranean, and noticed that all roads lead to Rome. Hopefully you wouldn’t conclude that so much commerce and culture and armies are trafficked through Rome *because* all roads lead that way.

7/

Imo this is a common failing in jdm.

Seeing a bunch of psychological phenomena converging on behavior x. And then listing these phenomena (or worse: debating which among them!) as explanations of x.

8/

(For instance, what causes us to be scope insensitive when giving to public goods. Here’s ten heuristics, emotions, or other cognitive limitations that lead to scope insensitivity.)

9/

Another clue something is amiss:

If each of the enumerated phenomena have on and off switches. And all just *happen* to be on in this instance.

10/

Like overconfidence. Are we *always overconfident*?

Not when we want to convince others we ain’t a threat or are in need of help. For instance.

11/

Like if you are playing a board game. And are in the lead. But don’t want everyone else to gang up against you. You act as if you are doing worse than you are. That’s under-confidence.

12/

If you are homeless, do you act over-confident about your ability to feed yourself when begging for money?

13/

When that guy in “the office” is flirty w/ the assistant, and her burly boyfriend shows up, does he spread his arms and show how buff and formidable he is. Or does he, umm, cower in the corner and make clear he isn’t a threat?

14/

And when a president is going to war, yeah, he may very well want to sell to his country how likely the war is to “spread democracy to the Middle East” and be “greater as liberators.”

15/

But, like, what about when he needs to rally troops or get women to take men’s place at the factories. Consume less butter. And buy war bonds. Does he say the war is gonna be a walk in the park? Or does he say your country *needs* you in order to win?

16/

Also, overconfidence can go both ways. The president can be overconfident about likelihood of success in war OR success at peace.

Like he could think appeasement is especially liable to succeed. Peace especially liable to be stable and prosperous.

17/

Why does k&r think the overconfidence swift, which very much is a switch not a fixed feature, is on for war? And off for peace?

18/

The same could be said about the other phenomena they list.

Yeah we send good troops after bad.

Sunk cost fallacy.

But always?

19/

What about when we have a new president come in from a different party.

Trump didn’t invest MORE in the nuclear treaty w/ Iran.

He tore it up. Cause his predecessor did it.

An inverse sunk cost, if you will.

20/

That’s commonly the case. When, instead of needing to show your past decisions were valid, you need to show your predecessors weren’t, you switch boats mid-stream.

21/

(Seems to me this has more to do w/ justifying than with some cognitive ailment. Likewise re overconfidence: has more to do with persuading than w/ a fixed cognitive bias. And the phenomena of overconfidence and sunk cost fallacy are just *means* at our disposal to achieve such.)

What about thinking you are stuck in a corner and forced to fight but your opponent is fighting cause he’s amoral?

Again, you should be asking: Do we *have* to think this way? Do we always think this way?

22/

Of course not. If the adversary was instead an ally, we wouldn’t.

Which goes to show it’s not a fixed part of our psyche. It’s a convenient tool our psyche has at its disposal. For disparaging our adversaries. And exonerating ourselves.

23/

Let’s talk about “devaluing concessions offered by your adversary.”

Cause that illustrates yet another red flag.

24/

Imo there is a fairly obvious explanation for this: in a bargaining setting you benefit from arguing your counterparts offers aren’t as significant than they are. And yours are more significant than they are. Cause then you can offer less and make them offer more.

25/

Which imo is very much the same fundamental underlying cause as the other phenomena mentioned.

They are all ways to justify.

Justify going to war. Justify having gone to war. Justify demanding more. Justifying.

26/

And when all the phenomena listed are presented as distinct. But in fact have an underlying commonality. You should wonder. Are they really distinct. Are they really best understood as a list of separable phenomena.

27/

(I guess we haven’t yet addressed why so many phenomena have a tendency toward presidential war lingering. My guess: presidents have perverse incentives to go to war. As the founding fathers anticipated. Which is why they put the power of declaring war in the hands of congress.)

(Perverse incentives like they have more resources and contracts to dish out. They can take the glory for successes. And hide the failures. And benefit from increased nationalism.)

So to reiterate... there are a bunch of psychological phenomena that tend to point presidents in the direction of war mongering. (For presidents who have perverse incentives to be war mongerers.

28/

All of these phenomena come with on off switches. That happen to be on. When the president needs to justify going to war, negotiate better offers during war, or hide his losses from war.

29/

But instead of recognizing the ultimate causal role of perverse incentives toward warfare, and the need to justify, and the fact that the cognitive switches can be turned on or off as the incentives dictate...

30/

K&R think of the cognitive switches as if they are always on and the cause.

31/

And that’s where I think jdm *often* goes wrong.

Not always.

Some of the switches are in fact invariably on. Like just a fixed constraint on cognition. (But imo they should more often be asking whether this is the case, instead of *presuming* it’s the case).

32/

And sometimes there isn’t an underlying common feature, an ultimate cause. (Just a coincidence of follies.)

But imo, when many roads lead to Rome, they should wonder.

Cause ow they risk miscontruing the underlying causality.

33/

Who cares? What’s the big deal.

The phenomena are still valid. They still have a causal impact. As documented in internally valid, controlled lab experiments.

34/

Who cares if those experiments are done in an environment where the causal forces are clipped? Where the larger broader causal network, which includes the many roads, the on off switches, and the upstream common forces pushing those switches in coordination, are severed?

35/

Hopefully some reasons we should care are already apparent.

If you have the wrong causal model, you will misconstrue the significance of certain arrows, focus too much effort on activating or deactivating those.
...

28/

Miss out on the significance of the upstream causes. And interesting causal pathways involving these.

Leading you to miss out on potentially more fruitful interventions. And potentially make erroneous predictions.

29/

That is, you are liable to over-estimate the importance of tampering with the cognitive phenomena.

Not realizing that upstream causes are liable to counteract or sidestep interventions therein.

30/

And you are liable to under appreciate the significance of the underlying incentive structures. Or other more ‘ultimate’ causes.

On the outcomes of interest (war mongering).

As well as the psych phenomena. (Overconfidence).

31/

And likely to miss out on, perhaps the most fruitful interventions. Ones that mess with these more ultimate causes. Perhaps. Like addressing the perverSe incentives for presidential war mongering. And related deceptions.

32/

Like:

-Not letting the power of war-making slide into the executive branch.
-Requiring acts of war be accompanied by proof of casus belli and full reports on costs+casualties+contracts etc
-Introducing penalties for fabricating or withholding evidence and arguments wrt such

33/

Obv just ideas. But my point is: these are ideas Kahneman missed. Along with the key role of perverse incentives. The endogeneity of the psychological phenomena he lists. And perhaps the ineffectiveness of intervening directly on those phenomena.

Eom

Addendum: for those interested, let me explain a bit more bout the causal model I have in mind.

(Thanks to @asbear91 for the key idea here.)

A1/

The causal model I have in mind is one with feedback loops.

Whereby the outcome variable, like warmongering, when not at the level that’s incentivized, sends messages upstream, that causes the proximate mechanisms, like overconfidence, to ramp up.

A2/

Like a thermostat regulated heater. The temperature is the degree of aggressiveness. Overconfidence is like the heating mechanism. The number input into the control pad on the thermostat is like the incentives to go to war.

A3/

The feedback loop means this causal model isn’t a DAG, as is often presumed when modeling causality.

A necessary feature for capturing the role of incentives, or other ultimate drivers, when paired with proximate psychological mediators.

A4/

This feedback loop is also what causes lab experiments—which nicely document a causal effect of the proximate psychological phenomena—to be so misleading.

A5/

Since the lab experiments either put the subject in a context stripped of the feedback loop, or don’t give time for the system to stabilize and the feedback loop to come into play.

A6/

Returning to the thermostat analogy.

The heating apparatus is in some sense the cause of the temperature.

You can show a causal link between the two. Cause if you tweek the heating element, eg by adding a leak to make it less effective, the temperature will fall.

A7/

But that’s misleading. Cause the temperature only falls in the short term. Provided the thermostat regulator is still online.

A7/

In the long term, the best way to reduce the temperature is to change the number input on the control pad.

And it would be rather stupid to naive, imo, the control pad, and thermoregulation when trying to understand how the room gets heated.

Eom

*should say “would be rather naive, imo, to ignore...” oops.