While you may find appeals to arguments from the regression to the mean to be “horrendously bad”, I can only report that, so far as I have been able to make out, the logical legitimacy of such arguments is pretty much taken for granted among the disputants on both sides of the IQ nature/nurture controversy.
The first link you point to, which seems most directly to address the issue of regression to the mean, in turn points to papers which were written about 30 years ago or more, without, it seems, anyone in the dispute taking them seriously.
Don’t you think that that would suggest that there’s something deficient in the argument that use of regression to the mean in this context is a logical fallacy?
Here’s the basic problem with claiming that regression to the mean in the context of, say, human traits is simply some mathematical artifact: it does nothing to explain WHY there should be a regression to the mean.
Yes, not only do the average IQs (or heights) of children regress to the mean from the average IQs of their parents; the opposite is also true—the average IQs (or heights) of parents regress to the mean from the average IQs of their children. Does that mean that there is no causal relation established by regression to the mean effects? No, absolutely not. It only establishes that the direction of a causal arrow can’t be determined from the fact of regression to the mean alone. But we know the direction of that arrow, if the cause is genetic (or environmental, presumably): it goes from parents to children, not the other way around. When we understand this, we can also explain why we see regression to the mean in the other direction as well; the same underlying set of causes are working, though, again, the direction of the causal arrow is opposite.
The fact of regression to the mean strongly argues that there is SOME underlying causal mechanism (be it genetic or environmental or a combination) that explains that fact. Why is it that the children of high IQ parents regress partly to the mean, but not all the way?
Regression to the mean in traits in both directions, from children to parents and vice versa, can be explained by luck—those parents or children who have greater IQs or greater heights are, on average, luckier than average; they are, in particular, luckier than their own children or parents, respectively. But what are they luckier AT? What have they received more of? If one says, genes that increase the trait in question, then a perfectly coherent explanation emerges. One might say that they’ve received a better environment—but that becomes a very difficult explanation in the case of IQ, since typically quite the opposite seems to be true (parents with high IQs have on average greater incomes and generally should establish a better environment for their children than they themselves experienced.)
In short, the existence of regression to the mean in the expression of traits across generations presents an important fact—one that one might not a priori expect. Something must explain that fact. Do you seriously think that that explanatory problem simply goes away by declaring that appeals to regression to the mean constitute a “logical fallacy”?
While you may find appeals to arguments from the regression to the mean to be “horrendously bad”, I can only report that, so far as I have been able to make out, the logical legitimacy of such arguments is pretty much taken for granted among the disputants on both sides of the IQ nature/nurture controversy.
The first link you point to, which seems most directly to address the issue of regression to the mean, in turn points to papers which were written about 30 years ago or more, without, it seems, anyone in the dispute taking them seriously.
I pointed out these papers because among the literature I’ve read on the topic, they present the best discussions and explanations of this issue. They are definitely not the last thing that’s ever been written on the subject. And while Mackenzie’s paper is indeed (yet undeservedly!) forgotten and obscure, Furby’s has been cited widely throughout the last four decades (just google for its title).
Furthermore, the logical validity of the regression argument is by no means “taken for granted” on both sides. I recommend that you read James Flynn’s 1980 book Race, IQ, and Jensen (dated, but still well worth reading), which presents a refutation of it by a prominent participant in the controversy. (It’s on pages 64-67 -- you might be able to find it on Google Books preview.) Another refutation, written by Nathan Brody, can be found in the 2003 volume The Scientific Study of General Intelligence: a Tribute to Arthur Jensen, edited by Helmut Nyborg (pages 404-407). The regression argument has also been dismissed as invalid in numerous books and papers by Richard Nisbett and many others, with refutations of varying detail and quality.
Also, an interesting critical discussion of the quality of Jensen’s statistics in general, which also addressed the regression arguments, was featured in the fall 2001 issue of the journal Chance. (Jensen himself also contributed.)
On the whole, unfortunately, a rather stupid situation has persisted since the seventies on this issue. Jensen and the other hereditarians stubbornly keep insisting on the same decades-old regression arguments, and their critics reply with more or less the same refutations. Neither side has made any further advance. However, while the anti-hereditarians can be blamed only for not coming up with more readable, clear, and in-depth counter-arguments, the hereditarians are, in my view, much more to blame because they keep bringing up the same invalid argument over and over.
(I have to add that on the whole, I have a lot of respect for Jensen as an intellectual figure, and I’m puzzled by his behavior when it comes to this particular issue. I should also stress that here I’m stating my opinion only on the specific issue of regression-based arguments, not about any other disputes that are relevant for this controversy.)
Here’s the basic problem with claiming that regression to the mean in the context of, say, human traits is simply some mathematical artifact: it does nothing to explain WHY there should be a regression to the mean. [...] In short, the existence of regression to the mean in the expression of traits across generations presents an important fact—one that one might not a priori expect. Something must explain that fact. Do you seriously think that that explanatory problem simply goes away by declaring that appeals to regression to the mean constitute a “logical fallacy”?
Honestly, with all due respect, I think you lack the necessary knowledge of statistics to reason about this issue correctly. Regression to the mean is not some unusual phenomenon that calls for a special explanation when observed. On the contrary, it is a mathematical necessity that happens whenever you have two imperfectly correlated variables (under some very generous mathematical assumptions, to be precise). For a rudimentary intuitive view, see the already discussed article by Neuroskeptic, and for detailed explanations, check out the above cited references.
In your post, you take the hopelessly muddled argument from Rushton and Jensen’s 2005 paper—which is, incidentally, restated in their 2009 rebuttal of Nisbett’s subsequent criticism of it, thus completing another round of the decades long non-debate I described above. You then proceed to make an even bigger muddle out of it. If you insist, I can post a more detailed criticism, but if you intend to debate these topics publicly, I would advise you to acquire a greater familiarity with the relevant literature and the pertinent topics in statistics. Reading through the above listed references should give you an idea of where the problems with your argument are.
While you may find appeals to arguments from the regression to the mean to be “horrendously bad”, I can only report that, so far as I have been able to make out, the logical legitimacy of such arguments is pretty much taken for granted among the disputants on both sides of the IQ nature/nurture controversy.
The first link you point to, which seems most directly to address the issue of regression to the mean, in turn points to papers which were written about 30 years ago or more, without, it seems, anyone in the dispute taking them seriously.
Don’t you think that that would suggest that there’s something deficient in the argument that use of regression to the mean in this context is a logical fallacy?
Here’s the basic problem with claiming that regression to the mean in the context of, say, human traits is simply some mathematical artifact: it does nothing to explain WHY there should be a regression to the mean.
Yes, not only do the average IQs (or heights) of children regress to the mean from the average IQs of their parents; the opposite is also true—the average IQs (or heights) of parents regress to the mean from the average IQs of their children. Does that mean that there is no causal relation established by regression to the mean effects? No, absolutely not. It only establishes that the direction of a causal arrow can’t be determined from the fact of regression to the mean alone. But we know the direction of that arrow, if the cause is genetic (or environmental, presumably): it goes from parents to children, not the other way around. When we understand this, we can also explain why we see regression to the mean in the other direction as well; the same underlying set of causes are working, though, again, the direction of the causal arrow is opposite.
The fact of regression to the mean strongly argues that there is SOME underlying causal mechanism (be it genetic or environmental or a combination) that explains that fact. Why is it that the children of high IQ parents regress partly to the mean, but not all the way?
Regression to the mean in traits in both directions, from children to parents and vice versa, can be explained by luck—those parents or children who have greater IQs or greater heights are, on average, luckier than average; they are, in particular, luckier than their own children or parents, respectively. But what are they luckier AT? What have they received more of? If one says, genes that increase the trait in question, then a perfectly coherent explanation emerges. One might say that they’ve received a better environment—but that becomes a very difficult explanation in the case of IQ, since typically quite the opposite seems to be true (parents with high IQs have on average greater incomes and generally should establish a better environment for their children than they themselves experienced.)
In short, the existence of regression to the mean in the expression of traits across generations presents an important fact—one that one might not a priori expect. Something must explain that fact. Do you seriously think that that explanatory problem simply goes away by declaring that appeals to regression to the mean constitute a “logical fallacy”?
liberalbiorealist:
I pointed out these papers because among the literature I’ve read on the topic, they present the best discussions and explanations of this issue. They are definitely not the last thing that’s ever been written on the subject. And while Mackenzie’s paper is indeed (yet undeservedly!) forgotten and obscure, Furby’s has been cited widely throughout the last four decades (just google for its title).
Furthermore, the logical validity of the regression argument is by no means “taken for granted” on both sides. I recommend that you read James Flynn’s 1980 book Race, IQ, and Jensen (dated, but still well worth reading), which presents a refutation of it by a prominent participant in the controversy. (It’s on pages 64-67 -- you might be able to find it on Google Books preview.) Another refutation, written by Nathan Brody, can be found in the 2003 volume The Scientific Study of General Intelligence: a Tribute to Arthur Jensen, edited by Helmut Nyborg (pages 404-407). The regression argument has also been dismissed as invalid in numerous books and papers by Richard Nisbett and many others, with refutations of varying detail and quality.
Also, an interesting critical discussion of the quality of Jensen’s statistics in general, which also addressed the regression arguments, was featured in the fall 2001 issue of the journal Chance. (Jensen himself also contributed.)
On the whole, unfortunately, a rather stupid situation has persisted since the seventies on this issue. Jensen and the other hereditarians stubbornly keep insisting on the same decades-old regression arguments, and their critics reply with more or less the same refutations. Neither side has made any further advance. However, while the anti-hereditarians can be blamed only for not coming up with more readable, clear, and in-depth counter-arguments, the hereditarians are, in my view, much more to blame because they keep bringing up the same invalid argument over and over.
(I have to add that on the whole, I have a lot of respect for Jensen as an intellectual figure, and I’m puzzled by his behavior when it comes to this particular issue. I should also stress that here I’m stating my opinion only on the specific issue of regression-based arguments, not about any other disputes that are relevant for this controversy.)
Honestly, with all due respect, I think you lack the necessary knowledge of statistics to reason about this issue correctly. Regression to the mean is not some unusual phenomenon that calls for a special explanation when observed. On the contrary, it is a mathematical necessity that happens whenever you have two imperfectly correlated variables (under some very generous mathematical assumptions, to be precise). For a rudimentary intuitive view, see the already discussed article by Neuroskeptic, and for detailed explanations, check out the above cited references.
In your post, you take the hopelessly muddled argument from Rushton and Jensen’s 2005 paper—which is, incidentally, restated in their 2009 rebuttal of Nisbett’s subsequent criticism of it, thus completing another round of the decades long non-debate I described above. You then proceed to make an even bigger muddle out of it. If you insist, I can post a more detailed criticism, but if you intend to debate these topics publicly, I would advise you to acquire a greater familiarity with the relevant literature and the pertinent topics in statistics. Reading through the above listed references should give you an idea of where the problems with your argument are.
As a fence-sitter on this topic, I’d put in a vote for such a discussion, when your schedule allows.