Potpourri
Once again the browser becomes turgid with tabs:
* Jim Manzi has been having a good discussion/argument with Karl Smith, on the confidence economists should have in their models. If you go to this post, you can get up to speed. It’s an interesting discussion because if you read either of them, you totally “get” what he is saying, but they are fundamentally disagreeing with each other.
* Seeing that the CIA has taken no hostile action against Julian Assange, Robert Wenzel comes out of hiding and does a podcast with Lew Rockwell.
* Speaking of Wenzel, I think he often bites below the belt, but anyone can appreciate this (from his review of Matt Taibbi’s new book): “In thinking about the book, the only people I think could find the book useful would be the miners in Chile who were trapped for 69 days, but only if they would have been trapped for 6.9 years, instead. It would have brought them up-to-date in an almost picture book like way.”
* I don’t share his flippant dismissal of Arrow’s Theorem, but nonetheless Silas Barta makes some good points in this post, especially his response to a famous result by Amartya Sen.
* This is a YouTube of a new mother who is given the runaround because she doesn’t want her breast milk to go through the x-ray machine. She actually printed out the TSA’s own policies, but she couldn’t show the printout to the agents because her luggage was already screened and she was still in the hot zone. I think some of the people go to the airports looking for a fight, but this one (no sound, unfortunately) looks like they were really just screwing with her to teach her a lesson.
* Here’s my podcast from a couple of weeks ago on “Live Free Austin” (John Bush and Jason Rink). I don’t even remember what we talked about. I’m sure it was not favorable to Bernanke.
Silas writes: “Rather, [irrelevant alternatives] give evidence about the relative _strengths_ of preferences and therefore SHOULD affect the aggregated preference ordering!”
This looks like a muddle to me. How can the relative strength of A > B > C mean that, when we remove C, we get B > A.
@Gene_Callahan: That’s explained in the Black Belt Bayesian post:
Because people know how the voting algorithm works, then a strategic (i.e. maimially informative) preference ordering presented by a voter can be chosen so that it will force the aggregate ordering to better reflect the individual’s true preferences, even as their submitted preferences don’t match their true ones. This difference leads to the kind of flipping that IIA prohibits, and it does so only by “capping” how much information people can convey through their preference ordering.
Yeah, I read Black Belt Bayesian, and that seemed a muddle as well. If we want our voting mechanism to reflect strength of preferences, than it should just do so, without the need for strategic voting. The fact that some systems lead to non-sincere voting is a problem illustrating Arrow’s theorem, not a counter to it!
Or, in other words, it may be the best we can do is have these “irrelevant preferences” included to allow strategic voting. Arrow probably would agree. But, ideally, we’d like to avoid that — and Arrow shows we can’t, and fulfill the other goals at the same time.
That’s one way to interpret the result. A more fruitful way, in my opinion, is that constraining voters to express their preferences purely through an ordinal ranking forces them to throw out useful information. Alternately, that limiting the information you can extract from a set necessarily limits your ability to concisely describe it. The problem, either way, is with the voting systems you limit yourself to, not with seeking a faithful aggregate preference ranking.
I’m not trying to “counter” Arrow’s Theorem, except in the sense of disputing its significance and applicability to practical situations. For example, I agree that Arrow show’s how a system with the preferences in my example would have to throw out determinism, so in that respect the Theorem is valid. But I dispute that having to use randomized tie-breakers in symmetric edge cases is some kind of real-world problem, despite its violation of (the inutitively appealing desideratum of) determinism.
Like I said in the comment sections, you can indeed find rigid mathematical theorems with practical significance — but this isn’t one of them.
Bob sent this to me by email, saying he couldn’t post, so I’ll give it a try. What follows is Bob’s words, not mine.
***
I haven’t read BBB, so I might be misunderstanding your guys’ discussion. But anyway, I think as originally conceived, social choice theorists were starting from the POV of an omniscient central planner. I.e. if we knew everybody’s preferences, how would we map them to a social welfare ordering.
(Now it’s true, maybe we could come up with the rule in theory, but then it would break down in practice because people would lie about their preferences. Yet Arrow showed that it wasn’t merely a problem of incentives.)
As far as ordinal rankings, this is something that economists seem to like, while people from other disciplines think is dumb. Anyway, originally many economists really did believe in cardinal utility. Believe it or not, even Bohm-Bawerk’s examples involve intensity of preferences. But a lot of people objected to this on philosophical grounds.
A really important result was Pareto showing (I forget exactly when, but either late 1800s or early 1900s) that you could “do consumer price theory” (my words) without assuming cardinal utility. Specifically, if you picture the standard indifference curve map, the original conception was that it was a topographical map showing the contour lines going up the mountain of utility. So each indifference curve connected all points of equal height on the mountain, i.e. points of equal utility. And the slope of the indifference curve was the ratio of the marginal utilities of good x and y at that point (combination of x,y).
But Pareto showed that we didn’t need to think like that. Instead, all you had to assume was that each consumer could rank any two combinations of goods x,y such that one bundle was strictly preferred to the other, or the consumer was indifferent. Then you plotted indifference curves by connecting bundles of indifference. The slope was no longer the ratio of marginal utilities, but was instead the “marginal rate of substitution.”
Maybe it was the sheer drive to get the same results with as few assumptions as possible, but for whatever reason, a lot of economists (not just Austrians) felt it was important that preferences are ordinal in nature. That is still how standard micro is taught at the graduate level, even though probably 75% of new PhDs don’t realize it.
Now, to reply to Bob’s comment, posted under my name:
To a voter, it certainly matters whether they find A to be much better or just a little better than B. If you force them to express their preferences in a way that “deletes” this information, then it can’t be represented in the final solution either. But this is a deliberate choice to deem certain information irrelevant and no longer reflects the lay intuitions about what desiderata a voting system should meet, especially if it only considers candidates and no other goods.
I think what this boils down to is what it means for A to be “strongly” preferred” to B vs “just a little” preferred to B. From the ordinal perspective, this is simply the difference between whether you need to give that person a lot or a little with B to move that set to the same indifference bundle as A. This all assumes that there are goods not yet accounted for in a particular ranking. But perhaps, as you mention, we’re considering the case where all goods-bundles really are ranked. Does *that* situation make it meaningless to speak of preference strength?
I would agree that it might make it meaningless if we were, say, only concerned with mapping preferences to actions. But if we’re considering social welfare, we know, from introspection, that our rankings can place consecutive items “close” or “far apart”; A can be “much better” or “just a little” better than B. So for questions of social welfare, you want this information and an ordinal ranking indeed discards it.
(And I don’t know what BBB is. Did you mean EvBB?)