Intellectual Garbage Pickup: Donald Luskin

Yes, it's time for our once-every-three months websurf over to Donald Luskin. Why? As a public service: somebody needs to lay down a marker that he simply does not know what he is talking about, and that anyone who believes anything he writes without very careful verification is asking for big trouble.

And it is unbelievable. You don't have to read a dozen paragraphs before you run across something so bats--- ignorant that it should cause every National Review editor and writer to resign in shame, move to Rwanda, and take up a life of anonymous service to others.

But I don't have the heart to surf over and read any more. So let me pull out a true turkey from the archives: the time Luskin denied the existence of the entire discipline of statistics:

The Conspiracy to Keep You Poor and Stupid: The Times and the Journal cite many authoritative-sounding studies.... But to get an accurate picture [of inequality]... you'd have to track hundreds of millions of individuals.... [N]one of this is reliable... the Panel Study of Income Dynamics... tracks only 8,000 families out of a U.S. population of 295 million individuals.

The whole science of statistics exists because Luskin is wrong. As long as you can take a random sample of your population, you can find out an enormous amount about the population from a relatively small number of observations. You can find out what proportion of rich people had poor paretns, or what proportion of twenty year olds think they will graduate from college, or pretty much any other average proportion that you want.

Now the "random sample" part of this is very important. But if your sample is random--if the fact that the yes-no pattern of observations so far makes it no more (or less) likely that you next observation will be a "yes"--then the law of large numbers tells us that the sample average you compute will converge to the true population average at a frighteningly rapid speed.

The standard demonstration of this is to repeatedly flip a coin and count the excess proportion of heads over tails. We know that--with a coin flipped and caught in the air by a human being at least--the population average taking all coins that have ever been flipped of the excess proportion of heads is zero. How many observations do we have to take--how many coin flips--before the sample average converges to this population average of 0% excess heads?

Let's see. Here's one run of 1,000 "flips" from Excel's internal random number generator:

Here are ten more:

Here are ten more:

Impressive, no?

Try some yourself.

You could have a population of 295 million flipped coins. Yet you don't need to look at "hundreds of millions" of them to determine what is going on. Looking at a couple of thousand will do.

This is the principal insight of the science of statistics. it is an important insight. It is a powerful insight. It is also not an obvious insight--that's what makes it powerful and important. Yet it is one that Donald Luskin has never managed to grasp.

That's really sad.

Comments

[Matt]

But who wants to flip all those coins every time we do statistics?

[Rob]

obvious side issue: Why are we testing every student in "no child left behind" rather than statistically testing children in classes from time to time in some true random fashion. If you are going to test every kid, of course teacher will teach to it. Urged of course by school boards and administrators.

[Rob]

obvious side issue: Why are we testing every student in "no child left behind" rather than statistically testing children in classes from time to time in some true random fashion. If you are going to test every kid, of course teacher will teach to it. Urged of course by school boards and administrators.

[Rob]

obvious side issue: Why are we testing every student in "no child left behind" rather than statistically testing children in classes from time to time in some true random fashion. If you are going to test every kid, of course teacher will teach to it. Urged of course by school boards and administrators.

[Roy]

We count on you for fresh Donald Luskin lowlights ... a rerun, even if it's a classic, is not quite the same.

[But that would be just too painful...]

[PeterE]

Suppose that you took a random sample of 8000 Americans. I wonder how many believe (a) in the devil (b) that we are in the Final Days (the Rapture, etc.) (c) in evolution (d) the law of large numbers or any central limit theorem (assuming they understand the statement of them). In one of his essays H.L. Mencken wrote that if you throw a banana peel out of a railway carriage and it hits someone it's likely to be a fundamentalist. That's not true in the Bay Area-- but may still be true elsewhere.

[Julian Elson]

Well, getting an accurate broad picture of income inequality with 8,000 random observations is fine, but given the fractal nature of contemporary inequality, it may not be enough looking at the very top. If the top 0.001% is pulling way ahead of the top 0.01%, 8,000 random observations from across the whole distribution of income won't cut it.

[vt codger]

***obvious side issue: Why are we testing every student in "no child left behind" rather than statistically testing children in classes from time to time in some true random fashion.*** Because NCLB is a festival of lunacy created by the fringes of America's right and left. ***If you are going to test every kid, of course teacher will teach to it. Urged of course by school boards and administrators.*** Just as they will if you only sample. The problem is the use of dubious metrics and methods, not which dubious metrics and methods are used. ===== The real problem with NCLB testing is that assumption that the tests are strongly correlated with actually knowing anything. And the problem with NCLB overall is the bizarre premise that everybody can continuously improve constantly in every respect and that the results can be determined with infinite precision. There is no such thing as a random variation or trade off in NCLB land.

[save_the_rustbelt]

Beating On Luskin is like bullying the small, fat kid in the class, it is too easy.

(The dumbest conversation on the planet occur when Kudlow is interviewing Luskin.)

There may be a related issue that is legitimate, do social scientists draw overly broad conclusions from their empirical research? Luskin couldn't understand that I doubt.

[save_the_rustbelt]

Beating On Luskin is like bullying the small, fat kid in the class, it is too easy.

(The dumbest conversation on the planet occur when Kudlow is interviewing Luskin.)

There may be a related issue that is legitimate, do social scientists draw overly broad conclusions from their empirical research? Luskin couldn't understand that I doubt.

[stewart]

Now can you do the same for Ben Stein's latest NY Times piece? As I read it, he's saying 'I rely on history, anything more than 10 years old is prehistoric'.

[neutrino]

Why even bother to surf on over to Luskin's site? The law of large numbers applied to previous samples has already shown what is to be expected. If you still should doubt, however, check out Luskin's post in late May 2005 about peak oil.

neutrino

[neutrino]

Why even bother to surf on over to Luskin's site? The law of large numbers applied to previous samples has already shown what is to be expected. If you still should doubt, however, check out Luskin's post in late May 2005 about peak oil.

neutrino

[jerry]

Sadly, this sort of innumeracy is not the domain of Luskin, or the rightwing, but pervades society and can often be seen in the true believers of many movements, including many of our so-called reality based left wing.

It may have been a long time since High School, but I would have sworn we learned about stats there. I would have hoped that, statistically speaking, holding a high school diploma would mean this problem would have been solved.

(Are you folks with the double posts using typepad? I'm trying to figure out how a system using unique "captchas" can allow a double post. What sort of luskin was integrating the database and software?)

[paul]

He keeps using that phrase [The Conspiracy to Keep You Poor and Stupid]. I do not think it means what he thinks it means.

He's part of it, not a member of the valiant resistance. Is this how the Enlightentment ends, not with a bang, but with a krgh!! mfsghjkft! ?

[vt codger]

***(Are you folks with the double posts using typepad? I'm trying to figure out how a system using unique "captchas" can allow a double post.)*** The problem does indeed seem to be something in Typepad. It does that multiple post thing sometimes if you try to back up to the main message stream using the Browser's back icon. I've never taken the trouble to figure out the details. Neither has Typepad apparently (surely, they must be aware of it). The lack of new lines in my posts are also a Typepad thing. They are there in my browser. They are there in the Preview. But they are missing in the final post. I keep on hoping that Typepad will fix that someday also. Let's hear it for modern technology.

[DILBERT DOGBERT]

Jerry,
Using Firefox I have to avoid the dialog box that appears when I apply the captchas and scroll to the top and click "main" or one of the links on either side of "main". If I enter anything in the dialog box I double post. The solution to this is above my pay grade.

[Steve Jung]

Professor DeLong,

I don't think your eplanation for why polling/sampling works is accurate. The math behind polling is that we can graph a distribution of every possible sample that could be taken from a population, and then construct a bell curve as to how all of those samples vary from the true population. The larger the sample drawn, the more compressed the distibution of all possible samples becomes in relation to the true population, and the greater the likelihood that the sample drawn if truly random will approach the actual population within a percentage point or two either direction. The size of the sample increases the number of possible combinations that are relatively close to the true sample, so it becomes more likely that your sample drawn is reasonably likely to be close to reflecting reality.

[save_the_rustbelt]

Firefox does not like DeLong.

Sorry

[Gerard MacDonell]

Steve, your point plus diminishing returns to increased sample size equals Delong's point.

But the real action is in what we choose as the null hypothesis. Hume's problem has not been solved.

Separately, I have a general question that I would like to pose to any readers here. The country is run by elites who are chosen by some process that accords some weight to the preferences of non-elites. Given this, I think it is important to understand how efforts at educating the decision makers should be directed. Should guys like Delong aim at convincing the current and prospective elites to go this way or that? Or should he aim at the broader public?

My guess is the former. I think the broader public is quite clever when it comes to its own peculiar economic interests. But I really doubt it is ever convinced by abstract arguments about how the general welfare is constructed or how it hits on a personal level. Rather, the public's reasoning is largely adaptive. They try this and if it works, they stick with it; if it fails, they switch. And all the while, they filter the information based on some primal bieases.

But nobody outside the elite is going to have their opinions affected by a reasoned argument from Delong. When was the last time the public switched its opinion on the basis of speculative reasoning, as opposed to being punched hard and repeatedly in the face?

The reason conservatism is headed for a fall is that the median voter is falling behind economically and is embarassed by America's reduced standing in the world. They need not know why they are poorer or why the world has the lower opinion, but they see the facts and react by shifting their votes leftward.

It ain't about Delong or his usually-airtight arguments. So his job is to make sure the liberal elite does not screw it up when they get the power back, which will be soon. The public, for the most part, is hopeless at speculative reasoning.

And this raises another issue. The only salvation of this Republic is if we continue to luck into negative, stabilizing feedback loops. If the system ever switches from sink to source, we are royally screwed. The public really has no ability to look forward, except as relates to their own circumstance, which is not enough.

[wood turtle]

As wealth is concentrated more in the upper regions, I think a stratified random sample would be more accurate.

[j]

Should H.L. Mencken throw a banana peel out of a railway carriage and it hits someone it's most likely to be one who believes that Donald Luskin is right and DeLong wrong.

[j]

Metaphorically, of course.

[albatross]

Gerrard:

Are elites any more likely to be swayed by logical arguments than other smart, educated people? I'll grant you, it's hard to make a statistics-based argument with someone who doesn't know much about statistics, but I've seen little evidence that the powerful decisionmakers in our society are mostly smarter or less emotionally driven than I am.

Spreading good information and explanations of the world is important, because those who can use them, will.