IUCs and the Law of Large Numbers

I've been keeping silent through most of the recent exchange over IUCs (which has been fascinating), though you can probably guess where I stand based on my earlier comments. Let me first say that I heartly endorse Patri's most recent post below, and aim merely to supplement it.

First, it's interesting to me that while Glen says he agrees with Brian, he doesn't stick to the strictly formal objections Brian makes. And with good reason, I think -- this is a situation where one has to choose between formal tractibility and empirical accuracy. The standard contention that utility is strictly ordinal is tautologically true within the confines of an economic model, but here in the real world we can and do say more than just "I'd like an apple more than an orange" -- we frequently make statements like "I'd like an apple way more than an orange." To a consistent ordinalist this must sound like "baa baa buff" -- complete gibberish, since preferences can't have magnitudes, you see, and hey wait where are you going? Come back!

Ahem. So yes, we really do feel likes and dislikes with magnitudes, not just orderings, and if we're interested in describing real human beings we may want to take account of this. Glen understands this, so he falls back to a different line of defense. When Glen says that "we've got different brains, and an endorphin in my brain just ain't the same as an endorphin in yours," he's making an empirical claim which is perfectly true so far as it goes -- everyone's brain is a little bit different. But for many purposes this is a point without teeth, so long as we can expect the differences in people's brains to be approximately normally distributed with non-dramatic variance.

On page 70 of Stumbling On Happiness, Dan Gilbert uses the following analogy:

If the workers at the factory that makes all the world's tape measures, rulers and yardsticks got sloshed at a holiday party and started turning out millions of slightly different-sized measuring instruments, we would not feel confident that a dinosaur was larger than a turnip if you measured one and I measured the other. After all, we may have used pickled rulers. But if hundreds of people with hundreds of rulers stepped up to one of these objects and took its measurements, we could average those measurements and feel reasonably confident that a tyrannosaurus is indeed bigger than a root vegetable. After all, what are the odds that all the people who measured the dinosaur just so happened to use stretched rulers, and that all the people who measured the turnip just so happened to use squashed rulers? Yes, it is possible, and the odds can be calculated quite prescisely, but I will spare you the math and promise you that they are so slender that writing them down would endanger the world's supply of zeroes.

Patri's example of "funniness" is apt, though I can still imagine someone taking the consistent skeptic position that no, we really can't compare humour between two seperate persons. I submit, however, that the skeptic's stance retains plausibility only because of the number two in that statement. What if we show a random sample of one thousand people the same two video clips of stand-up comedians, then ask them to rate which was funnier? If 90% of those people rate Comic A as funnier than Comic B, would we not have grounds to say that Comic A generates more subjective humour?

The implausibility of hardline agnosticism in this case increases with N, and thus it is with the measurement of any subjective feeling. Incidentally, this also answers Arnold Kling's question when he asks "if whether a Ramirez walk-off home run makes you happy or not depends on which team you were rooting for, can we be confident about generalizations that we make about whether being a father makes you happy ... ? I don't see how." The law of large numbers, that's how. Or, as J.V. Cunningham put it much more elegantly in his poem Meditation On Statistical Method:

Plato, despair!
We prove by norms
How numbers bear
Empiric forms,

How random wrong
Will average right
If time be long
And error slight,

But in our hearts
Curves and departs
To infinity.

Error is boundless.
Nor hope nor doubt,
Though both be groundless,
Will average out.

Related posts:

Why IUCs?
Cardinal Schmardinal, Ordinal Schmordinal
Encoding Happiness
No Soul Suggests IUCs
Love and Intrapersonal Utility Comparison
What color does a submarine weigh? (True or False?)
Exploding IUCs on the roadside
Interpersonal Utility Comparisons
Pareto Efficiency and Justice
Can the Paradox of the Non-Comparability of Interpersonal Utility be Resolved?

Share this

So yes, we really do feel

So yes, we really do feel likes and dislikes with magnitudes

Sure we do, and likes and dislikes are not choices. Talk all you want about things that have magnitudes, just don't keep being so sloppy about it. Don't keep confusing different things. There's more room in the universe for other things aside from preference. You don't have to reduce everything in the whole world to preference. Next are you going to tell us that the chemical formula for water should be expressed in terms of preference?

Constant, What would you say


What would you say the difference is between "preference" and "like"?

What would you say the

What would you say the difference is between “preference” and “like"?

The very concept of like involves magnitude. It's firmly embedded in our language. You can like something a little or a lot. Preference is about choice. If I choose A over B, then I don't choose A "a lot" over B, I simply choose A over B.

The concepts are just different. But maybe you're asking about the underlying reality. Well, let's consider music. Music makes you feel good things. Some music makes you feel good things more intensely than other music. That intensity might be measurable. It might be reducible to the amount of some chemical released or to the amount of firing of some neurons. Same with food. You might be able to tell whether someone gets more pleasure out of some dish than out of music.

But preference is entirely a different thing, because it concerns the choices that a person makes. And a person may, or may not, choose what he likes over what he dislikes. Specifically, a person may dislike certain vegetables but force himself to eat them anyway, and he my like ice cream but force himself to stay away from it.

Now, you might say, "but in that case he likes the vegetables more than the ice cream in a larger sense - in a sense of liking their health effects on him." But if you say that then I think you are now using "like" as a synonym for choice. The ice cream that he rejects *still* produces the cascade of neural impulses or chemical releases that are associated with pleasure of some intensity, and the vegetables *still* produce the cascade associated with displeasure of some intensity. In *that* sense, the sense that likely admits of some absolute magnitude which it is easy to believe we can tie to the release of chemicals (or whatever physiology underlies it), he likes ice cream to a certain degree and dislikes certain veggies to a certain degree.

Preference, in the economic sense, is about choice, and we choose things for a variety of reasons. At our simplest, yes, we do often choose what we like more over what we like less, we do often choose the greater pleasure over the lesser pleasure. But degree of pleasure is not the only basis of the choices we make. An intellectual understanding of the ill effects of certain foods and a decision to maximize our own lifespan to the extent we can is not necessarily tied to a corresponding degree of pleasure that can in any canonical way be compared to the sensual pleasures of eating.

So, what is the point of all

So, what is the point of all this? Even if we could make valid (i.e., quantifiable) IUCs that could reliably account not only for present but also future consesequences of certain actions, what should we do with that information? Redistribute income and wealth? Such an outcome assumes not only that we can reliably quantify IUCs (about which all I've read in these various posts is theoretical blather), it assumes implicitly that there is a "social welfare (or utility) function" that can be optimized, given reliably quantified IUCs.

Well, you know who defines the "social welfare function": the same people who tell me that it is my privilege to pay taxes in order to finance bike trails for yuppies, luxurious government edifices, welfare for druggies, muscle-building equipment for thugs in prison, etc., etc., etc. Somehow such things actually diminish my utility. And I don't care if such things increase the utility of 90% of the population. They still diminish my utility. Or are we to bow to the law of large numbers? Seems anti-libertarian to me.

Taking money from X and giving it to Y may make Y happier, but it only makes X unhappy (unless he's a rich liberal). X's happiness and Y's happiness are not additive, unless you're into "cosmic consciousness." Let's stick with Pareto -- except when it comes to de-funding government employees and their various parasitical constituencies, of course.

Constant, Good answer! I


Good answer! I actually wouldn't disagree with any of that: revealed preference and subjective pleasure don't perfectly overlap (nor do either of them perfectly overlap with what's actually good). If we want to define utility strictly in behaviouristic terms and nothing else, then my post, Glen's and Patri's (and Scott's and Brandon's...) are all moot. However, once we've taken the step of opening up the hood (so to speak) and talking about what motivates choices, as Glen apparently has (though not Brian), everything Patri and I have said kicks in. I'm mainly arguing in this way because Glen is a professed utilitarian who presumably buys all of the premises assumed in the above post.


Bad answer! I haven't breathed so much as a word about income redistribution, which is a bad idea even under charitable assumptions, but that's a whole other post...

So, what is the point of all

So, what is the point of all this?

In order of decreasing confidence on my part:

1. What I and Matt (who does it better than I) and everyone else are arguing is true, or at least stands a chance of moving us closer to truth, which is a noble pursuit that ought to be pursued.

2. Pragmatically, it doesn't bode well for my side when they make bad arguments.

3. I suspect that without the notion of comparable utility, the idea of economic efficiency becomes hollow (or it puts a great deal more weight on a theory of property rights than it will support) and I'd like to see that idea preserved, since it's useful.

Are we done with this yet?

Are we done with this yet? I'm surprised to learn that many people whose opinion I respect don't believe there's any ordinal ranking of choices without a cardinal system to back it up, but I can accept that it seems to be the case. I'm even more surprised that folks are arguing against "you can't compare ordinal rankings" with the non-sequitor "I can measure some things cardinally".

I think we all agree that there are dozens of things about people that can be measured and compared, some of which are usable to various degrees as predictors of future choices.

Whether to call any of these measures "happiness" or "utility" or "X" is just not that interesting to me. Heck, I'm not even sad to lose "utility" as the word to describe ordinal preferences behind an action. I've long preferred "preference", but even that doesn't describe it well, so I'm open to suggestions.

If y'all want to sloppily confuse whatever you're measuring (endorphin level, tension in a smile muscle, response to a survey, etc.) with the ordinal ranking of choices that are often somewhat correlated to your measure, I can't stop you. If you want to believe it's the same thing, I think you're glossing over the massive complexity of human software, but again you're welcome to do so.

So move on to Tom's question: so what? Say you can measure a person's X-ness (letting X be defined by the measure you choose) with a magic device. What do you propose to do with such information?

I'll be making another post

I'll be making another post on this subject, so here I'll just respond to one point:

"[B]ut here in the real world we can and do say more than just “I’d like an apple more than an orange” – we frequently make statements like “I’d like an apple way more than an orange.” To a consistent ordinalist this must sound like “baa baa buff” – complete gibberish..."

It's not gibberish to an ordinalist, it's just an imprecise way of speaking. Assuming that "like" is an all-things-considered notion (so that it includes not just tastiness but also future health benefits or whatever else the speaker thinks relevant), the ordinalist interpretation is that the amount of other goods and services the speaker would sacrifice for an apple is relatively large compared to the amount of other goods and services the speaker would sacrifice for an orange.

The Stumbling quote is pure

The Stumbling quote is pure comic gold.

So, what is the point of all

So, what is the point of all this?

In order of decreasing confidence on my part:

I don't think you answered the question, at least not the way I read it. The point of measuring temperature is manifold. For example, it's useful to know the temperature of food as you cook it so that you know when to take it out of the oven. That sort of thing.

Analogously, we can ask, what is the point of the particular kinds of measurement (specifically, cross-measurement, comparison) you're arguing exists. The *obvious* point of it is for a government to decide who to rob and of how much and who to give the loot to (to put it bluntly). Since *some* (not all) disavow any such intention, the question remains, what's the point of such a measurement.

I can answer the question for subjective preferences. The point of knowing a person's preferences is to know what he's going to do. What, then, is the point of knowing whether one person prefers X to Y more than another person prefers Z to W? It is not to know what they will do, because utility monsters will, by assumption, act exactly the same way as normal people.

Constant, That's silly. I


That's silly. I never argued that we should make such comparisons (much less for redistribution), only that they are possible.

If I argued that my house is flammable, does it imply I want it burnt down?

Sometimes the things I believe are true can be used against me, sometimes for me---nonetheless, I like knowing the truth. Do I really have to defend that preference? To Foucault, maybe, but to you?

Come on!

That’s silly. I never

That’s silly. I never argued that we should make such comparisons (much less for redistribution), only that they are possible.

Scott, I didn't say that you argued we should make such comparisons.

I am suggesting that we apply a pragmatic test of reality. Weight, height, length, temperature, ordinal utility, all easily pass this test. The kind of utility that allows IUCs has not, so far, passed the test that I have seen in this long discussion.

Pragmatism and Occam's Razor converge. From Wikipedia on pragmatism:

"Most of the thinkers who describe themselves as pragmatists point to some connection with practical consequences or real effects as vital components of both meaning and truth."

I'm asking for you to point to some connection with practical consequences or real effects. If someone prefers X to Y, then a real effect of that preference is that, if given the choice, he will choose X to Y. But if Bob prefers X to Y more than Bill prefers Z to W, what is a real effect of that?

Occam's Razor:

"entities should not be multiplied beyond necessity."

How are IUCs necessary? I've asked you what they are good for.

Economic utility DOES HAVE

Economic utility DOES HAVE relative magniture. Von Neumann-Morgenstern shows this. What it doesn't allow is absolute maginutes (one scale over another) or interpersonal comparisons.

You can read about here in the subchapter "Von Neumann Utility" of David Friedman's Price Theory, Chapter 13.

Economic utility DOES HAVE

Economic utility DOES HAVE relative magniture. Von Neumann-Morgenstern shows this.

David Friedman's discussion mentions lotteries, which is what I used earlier today to show that you can answer the question:

"Suppose someone prefers A to B to C to D and you want to know whether he prefers A to B more than he prefers C to D."

You can answer that by giving him a choice of two lotteries. Does that strike you as somewhat along the lines of what David is talking about? Is that (part of) what von Neumann and Morgenstern were talking about?