“Andrew Sullivan”:http://andrewsullivan.theatlantic.com/the_daily_dish/2007/10/genetics-and-ra.html links briefly to the post below on whether _g_ is a statistical myth, describing it as another expression of the “conventional left-liberal view,” and defending again his decision as editor of _The New Republic_ to publish extracts from _The Bell Curve._ I would have much preferred to have seen a substantive response to the “essay”:http://cscs.umich.edu/~crshalizi/weblog/523.html by Cosma Shalizi that the post linked to and summarized. I don’t see anything in Cosma’s essay that requires subscription to left-liberal views, conventional or otherwise – instead, I see a (to me entirely convincing) methodological critique of the basis for statistical claims that _g_, the purported general factor for intelligence, exists. To quote Cosma again:
If, after looking at your watch, you say that it’s 12 o’clock, and I point out that your watch has stopped at 12, I am not saying that it’s not 12 o’clock, just that your watch doesn’t actually give you any evidence about the time. Similarly, pointing out that factor analysis and related techniques are unreliable guides to causal structure does not establish the non-existence of a one-dimensional latent variable driving the success of almost all human mental performance. It’s possible that there is such a thing. But the major supposed evidence for it is irrelevant, and it accords very badly with what we actually know about the functioning of the brain and the mind.
If Andrew would like to take issue with something, these are the claims that he needs to be taking issue with. And there’s nothing stopping him, if he has even a moderate grasp of statistical reasoning (Shalizi’s arguments are quite comprehensible to someone with a basic minimum of statistical training, as evidenced by the fact that a gawp like me can reasonably claim to understand them). What Cosma is saying is that the entire body of research on _g_ is demonstrably based on bad statistical reasoning. Nor is it only Cosma who says this. Nor is this a product of political druthers – it clearly flows from a set of methodological claims that are widely accepted among statisticians, and that have many applications outside this particular and highly heated debate. If Andrew wants to show how Cosma’s methodological critique is fundamentally flawed in some way because of left-liberal preconceptions, he really should do so. If not, then all of his claims about “conventional left-liberal view”s and “going to challenge many assumptions of right-thinking liberalism” are by-the-by – they don’t count for anything unless they are actually backed up by, like, methodologically sound science.
{ 2 trackbacks }
{ 154 comments }
John Quiggin 10.21.07 at 3:34 am
It’s not only the failure to respond on g that’s striking. A perfectly conventional statement of the need to be vigilant about the way in which technology is applied, is turned by Sullivan into “vigilance against science“.
Of course, as we’ve discussed before the standard US rightwing position is pro-technology, anti-science and largely unaware of the difference between the two (the archetypal example is provided by creationist engineers like Henry Morris). I’d hope for something a bit better from Sullivan, now he has had his eyes opened to other aspects of the way in which his former comrades treat truth on any issue where it collides with politics.
Neil 10.21.07 at 4:15 am
By the by, surely?
Aulus Gellius 10.21.07 at 4:17 am
I don’t want to defend Sullivan, and I don’t think it’s particularly surprising that he’s misrepresenting his opponent’s point and portraying himself as the only seeker of truth in a world of mindless political drones; as far as I can tell, that’s his usual method. But I will say that I think you may be setting the “basic minimum of statistical training” too high. I don’t know what Sullivan’s background is, but I’m a well-educated person (I admit, mostly in the humanities), and Shalizi lost me pretty early. I got some of the arguments, and the others SOUNDED good, but some of them were definitely over my head. (This applies mainly to this last post; I clicked the links to earlier ones, and found them easier to understand.)
rilkefan 10.21.07 at 4:22 am
Isn’t “myth” a strong overstatement of the post’s argument? Shalizi says he doesn’t like g, and that the techniques and data cited for it aren’t good evidence, but I missed where the last step to the apparent thrust comes from.
I’d also be interested in a cite to the response in the literature of the researchers who are arguing the other side, if there is one.
Henry 10.21.07 at 4:27 am
John – I thought so too, but decided to concentrate my post on the methodological issues.
Neil – thanks – fixed.
Aulus – it seemed pretty straightforward to me, albeit I had to read carefully and slowly. And when I say that I’m not a sophisticated stats person, I’m not being modest. I have the usual basic pol-sci training – e.g. multiple regression etc, plus bits and pieces of categorical data analysis, factor analysis etc. I can thread my way through an argument, spot obvious fallacies and perform basic analysis myself, but that is about it.
Henry 10.21.07 at 4:28 am
rilkefan – when I use the term ‘myth’ I am appropriating it from Cosma.
Bruce Baugh 10.21.07 at 4:34 am
“Myth” seems like just the right word: an explanation of observed phenomena’s cause that is not itself resting on observations but rather held as a matter of belief. It’s a story about why stuff happens. It may turn out to be a true story, but it has at least not yet been shown to be true, not in the way that the things it’s supposed to explain are tested and accepted or rejected.
bi 10.21.07 at 4:39 am
rilkefan:
The last part of the thrust is I think this: “What the modern _g_-mongers do, however, is try to use exploratory factor analysis to uncover hidden causal structures.”
And I think I understand that: it’s like if you were to measure the speed at which this blog serves up blog entries (call it _y_), and you manage it to express it as a function of various observables such as the number of characters on the blog entry (_x_[1]), the version number of WordPress (_x_[2]), the size of the blogroll (_x_[3]), etc. So we have
_y_ = _f_(_x_[1], _x_[2], …)
Then someone comes along and says that _y_ is actually The Latent Factor (_g_) that controls the expression of _x_[1], _x_[2], etc.
bi 10.21.07 at 4:43 am
(Ugh, underlining for italicization doesn’t always work.)
rilkefan 10.21.07 at 5:22 am
6 – sure, I don’t understand why (except perhaps for politicalish reasons) he’s saying “myth” instead of something more neutral – even “artifact” might be too much. If you say “I measure x” and I respond, “Your test isn’t good enough to determine x because blah and what’s more here’s my Monte Carlo where your technique can’t show if x is in there or not”, I can’t therefore say “x is a statistical myth”. I’d have to go see what those people are claiming they are doing to really understand the argument – it doesn’t seem on the face of it the sort of thing one could determine given the tests in question – but at the moment it seems to me the post is overreaching.
bi 10.21.07 at 5:34 am
rilkefan: The problem, again, is that the people who claim to be measuring _g_ are in fact measuring something else, and that something else hasn’t been shown to be _g_. An observed statistic isn’t a latent factor.
Bruce Baugh 10.21.07 at 5:52 am
The politics of it are precisely the point. Assertions about g are part of an extremely important argument about both morality and policy. The legal and social status of large populations will be set in part by what policy makers and those who influence them come to believe. And if it turns out that almost all of the existing debate is no more founded than the guy with the stopped watch’s conviction that it’s noon, that’s really important, and urgent, and worth communicating in terms that signal strongly.
rilkefan 10.21.07 at 6:40 am
The probability that a stopped watch is right is say 1/(12*60) – what it reports is not what I’d normally call a “myth”. And isn’t the correct analogy to say that time is a myth because I can show that your watch can’t measure it? I can’t even begin to consider how likely it is that there’s a g given the data because I don’t see how one would extract it, but I don’t see how to get from there to saying it’s intrinsically unmeasurable or to implying there’s no such thing – which is what “myth” suggests to me.
That is, unless I’m missing something, the word “myth” is an absence of evidence/evidence of absence mistake, which brings in politics (though I don’t actually understand why one side or the other of the question should be liberal or conservative).
JFTR, Wikipedia very unhelpfully says: ‘Most psychometricians still recognize and employ “g” as a valid and coherent evaluation of human mental ability[citation needed].’
albert 10.21.07 at 7:58 am
The idea of “myth” is also, and more accurately, used to refer to an event or process that is believed in by a group of people who attach significance to the content of the idea. Myths in this sense are generally neither provable or disprovable, but analysis of them may reveal interesting things about the group that believes in the myth (contrasted with others who don’t believe).
Steven Poole 10.21.07 at 9:06 am
Shalizi’s essay is excellent, thanks for repeating the link. I liked the note in passing that “The psychologists make up some tests where a high score seems, to intuition, to go with a high degree of the quality.” (emphasis added). There’s an accessible history of IQ testing called IQ: The Brilliant Idea that Failed, by Stephen Murdoch, which emphasizes how unscientific IQ tests have always been and still are.
By the by, it seems plain to me that if you are worrying about whether Africans are on the whole dumber than Europeans, instead of, say, worrying about how to stop Africans dying from Aids or malaria etc, then you are ipso facto operating under a set of racist priorities.
dsquared 10.21.07 at 9:47 am
in general, anyone talking about “Africans” at all as if they were a homogeneous group and as if it obviously made sense to make claims about Benin that were also meant to apply to Zambia, is probably talking out their arse.
dsquared 10.21.07 at 9:49 am
(actually thinking about it that’s a bit too strong – there are plenty of generalisations that can reasonably be made about African history or politics. But I will defend it as a weaker claim that generalising about African people is a lot less sensible than generalising about “Americans” or “Europeans”.
bi 10.21.07 at 10:15 am
I have no idea why rilkefan keeps reiterating and reiterating that something must be provably untrue in order to qualify as a “myth”. WordNet says that a “myth” is “a traditional story accepted as history; serves to explain the world view of a people”. Drislane and Parkinson’s Online Dictionary of the Social Sciences even goes so far as to explicitly say that the word “myth” is “Often used incorrectly to refer to a claim considered to be untrue” (hah!).
And the absence of evidence — and more importantly, of methodology — is precisely the whole point. And this whole absence of methodology is, itself, a harsh truth which the _g_-mongers better have the courage and strength to bear.
Hidari 10.21.07 at 10:26 am
‘in general, anyone talking about “Africans†at all as if they were a homogeneous group and as if it obviously made sense to make claims about Benin that were also meant to apply to Zambia, is probably talking out their arse.’
It’s worse than that. People still talk about “black people” as if they are a homogeneous group. But (I read somewhere…seems quite plausible but might not be 100% true!) the indigenous inhabitants of Australia, due to the fact that they arrived in Australia so early, and were isolated from then on, are as genetically distinct from modern day Africans as it is possible to be (whilst still being human, presumably!).
Literally, the only reason to classify these peoples together is on the basis of the completely trivial fact of skin colour (and you might as well start classifying people on the grounds of the colour of their eyes or hair if you go down that road).
Essentially, if you talk about ‘black people’ and ‘white people’ and ‘yellow people’ as if these are scientific categorisations, you are, as you say, talking out of your arse.
John Quiggin 10.21.07 at 11:21 am
For people who find the stats difficult, it’s worth stressing a point made by Cosma Shalizi. The central objections to g are not new and have been repeated many times (Stephen Jay Gould said much the same in The Mismeasure of Man and Cosma notes much earlier versions). There is, as far as I know, no standard rebuttal – advocates of g just ignore the central point that they are confusing correlation with causality.
Tim Worstall 10.21.07 at 1:16 pm
Hmm, well:
“By the by, it seems plain to me that if you are worrying about whether Africans are on the whole dumber than Europeans, instead of, say, worrying about how to stop Africans dying from Aids or malaria etc, then you are ipso facto operating under a set of racist priorities.”
Aren’t we supposed to take things in context? From what I’ve seen Watson was originally talking about what solutions might be applied to the problems in Africa, then going on to say that we might not be able to use the ones we ourselves used because of that difference in intelligence.
Unless we’re going to ascribe that last to malnutitrion in childhood (a well known cause of some stunting of development and yes, also known to happen in Africa) then I don’t believe it either: partly because the genetic variability in Africa is, at least so I’ve been told, larger than it is on any other continent. That they’re all genetically dumb just doesn’t seem sensible (and of course what I find sensible has ne bearing on any scientific question but there we go).
So however wrong Watson’s argument was (insert any other description of it you like here)he was, at least from what I have heard, starting from exactly the point you want him to be: how do we solve the problems in Africa?
Brett Bellmore 10.21.07 at 1:23 pm
I would say that, while Shalizi makes a really good case that the existence of “g” can’t be proven by the usual statistical tools, that’s more an indictment of those tools, than a reason to believe “g” doesn’t exist.
Whatever the mechanisms that underlie all the various aspects of intelligence, they all are implemented by the same physical brain, and rely on many of the same functional features of that brain.
Cell level learning, propagation speeds, neurotransmitter supply, the number of aspects of the brain which will cause ‘intelligence’ to improve or degrade globally, rather than just for specific tasks, are legion. And they all vary from one person to another, from both genetic and environmental causes.
“G”, then, would simply be a function of those global determinants of intellectual capacity, and how they stack up in a specific individual. And from the perspective of neurobiology, rather than statistics, it’s existence is pretty much indisputable, though the question of how much of the variation in performance on any given task is due to these global influences, and how much is task specific, is an open question.
PJ 10.21.07 at 1:31 pm
Brett, surely not ‘indisputable’, since you are already positing a multi-factorial causation (“Cell level learning, propagation speeds, neurotransmitter supply” etc) which is exactly Shalizi’s point about the statistical fiction that g represents a single causal factor.
Steve LaBonne 10.21.07 at 1:46 pm
This is kind of bass-ackwards, seeing as the whole point is that the way those tools are being used by the g-believers gives absolutely no real warrant for thinking it DOES exist. Extraordinary claims, etc.
Lack of any credible evidence that such a thing exists IS in fact a very good reason for not believing in it.
Brett Bellmore 10.21.07 at 1:54 pm
“G”, as I understand it, is supposed to be a common factor across a variety of intellectual tasks. But I don’t think this was ever meant to imply that it was a single causal factor, like maybe the concentration of a particular polypeptide. Clearly, to the extent that it exists, it was going to be a very polygenetic trait.
Brett Bellmore 10.21.07 at 1:58 pm
Steve, I just gave you a real warrant for thinking it does exist.
We see every day evidence that a laundry list of causal factors can influence all intellectual capacities across the board, en mass. Just because the statistical tools are inadequate to analyze this doesn’t mean we have to ignore it.
Henry 10.21.07 at 2:36 pm
Shorter Brett Bellmore: Teh awesomeness of my concepts is too ginormous to be captured by your puny statistics.
Does anything more really need to be said in reply??
Steve LaBonne 10.21.07 at 2:43 pm
No, you didn’t. Your #25 is a meaningless form of words, offering no warrant at all for believing even in your common factor that isn’t really a common factor, or at least not a causal one (what, exactly, is a non-causal factor?), or whatever logic-torturing point you thought you were trying to make. It’s a typical example of ideology-driven rationalization.
Really, John Quiggin already gave in the last sentence of his #20 the only comment that this whole g business deserves. It’s nothing but a fancily gussied-up version of the original statistical sin, confusing correlation and causation. You can put all the fancy statistical lipstick you like on a pig, but it’s still a pig.
Cryptic Ned 10.21.07 at 2:53 pm
I believe Mr. Bellmore is right in his apparent point that g has not been conclusively demonstrated not to exist. I say this without any idea of what g is supposed to be, since a claim that something has not been proven not to exist is not a very strong claim.
rilkefan 10.21.07 at 3:14 pm
“There is, as far as I know, no standard rebuttal – advocates of g just ignore the central point that they are confusing correlation with causality.”
I had the vague idea that the psychometricians claim that g is correlated with hard-science stuff from brain scans or brain wave measurements – if so that’s worth noting. Anyway, I wonder if any psychometricians blog so one could ask.
Brett Bellmore 10.21.07 at 3:16 pm
“G” isn’t a specific thing, it’s just shorthand for “whatever factors influence intellectual performance across all tasks”, or something like that.
And I’d say that some of the components of g are very conclusively proven to exist.
Barry 10.21.07 at 3:21 pm
Brett: “I would say that, while Shalizi makes a really good case that the existence of “g†can’t be proven by the usual statistical tools, that’s more an indictment of those tools, than a reason to believe “g†doesn’t exist.”
This is a good summary of why a lot of people have given up on any good faith being assumed a priori for people asserting racial/gender differences in intelligence. We keep saying (a) there is a large and provable set of confounding factors at the population level, (b) there a many, many experiments on the individual level which demonstrate powerful effects of confounders and prejudices (e.g., blind interviews for orchestras, name changing experiements on papers and resumes, matched-pair interview studies), (c) you don’t have proof of anything worth a d*mn in light of those confounders and (d) your claims are suspiciously pointed in the same direction (e.g., genetic differences in racial intelligence are used to claim that blacks have lower IQ’s than whites).
And the Bell Curvists keep restating variations on the same old garbage – when they’re feeling active; much of the time they don’t even do variations.
fred lapides 10.21.07 at 3:22 pm
this is all beyond me, I admit up front, and the Watson thing seems absurd, since he badmouths any and all darkskinned people if they live in Africa (and perhaps elsewhere?), and that seems silly. But it was the practise to deny genetic superiority to Kenyon runners and assume training gave them the records, over and over. Not so today. But as far as dark skin: Kenyons are great distance runnrs while black in America are lousy at long distance but champs, over and over , as sprinters. go figure.
Steve LaBonne 10.21.07 at 3:31 pm
This is either confused or disingenuous. (Have you actually read Shalizi’s eassy? Really?) The claim of the g-believers is precisely that those factorS, plural, are correlated because they are all manifestations of some kind of SINGLE “general” (that’s what “g” stands for) “intelligence” that underlies all of them and is supposedly some real property of the brain and not just a convenient statistical summary of a bunch of data. There has never been any valid reason for believing that assertion.
Frank 10.21.07 at 3:36 pm
‘..advocates of g just ignore the central point that they are confusing correlation with causality.’
The proponents of psi behave Similary. No matter how many times the concept gets consigned to the dust-bin of history, its fans find a way of re-cycling it. We should have a generic name for such people; psi & g, if scrambled together, can spell pigs.
novakant 10.21.07 at 4:19 pm
Whatever the mechanisms that underlie all the various aspects of intelligence
“Gâ€, then, would simply be a function of those global determinants of intellectual capacity
a common factor across a variety of intellectual tasks
g is correlated with hard-science stuff from brain scans or brain wave measurements
I think this kind of vagueness is a good indication that I had a point last time we discussed this, when I said that the only way to get a grip on this problem, if it is at all possible, is to give neuro-science another few decades to figure out how all this “stuff” actually works.
engels 10.21.07 at 4:33 pm
Because obviously in “another few decades” we will be able to explain all mental phenomena in physical terms.
novakant 10.21.07 at 5:01 pm
Who knows? I don’t. We’ll see. Maybe the Churchland’s are right or maybe neuro-science will have to surrender to the complexity of it all at a certain point. It’s a worthy undertaking either way and I don’t understand the hostility it generates with some people.
bi 10.21.07 at 6:43 pm
Cryptic Ned, Steve LaBonne:
Here’s the proof that Brett Bellmore is absolutely and ginormously right: simply consider the vector comprising the 3 billion or so base pairs of a human’s DNA. Call this vector _g_; then obviously _g_ completely determines a person’s innate “intelligence”.
And you know what the nicest thing is? The 3 billion base pairs can be accurately summed up in one single integer — by writing out the base pairs as ones and zeros and interpreting them as a binary numeral! Quod erat demonstratum.
Moral of the story: Brett Bellmore just repaired a falsifiable — and unproven — claim by turning it into an unfalsifiable (or rather, trivially useless) claim. (Of course, the above ‘proof’ isn’t something which the _g_-mongers will want to hear: what they want is a single latent factor which induces a total ordering _and_ also controls the expression of all “intelligent” behaviours.)
Drm 10.21.07 at 7:37 pm
If you really believe the following things (all dubious in my opinion):
-IQ is a valid measure of general intelligence (if there is such a thing)
-racial groups differ significantly in IQ
-IQ is highly heritable
-achieving maximum IQ in the human population is an important objective
Then you probably should also want to do your utmost to promote interracial marriage. Because intelligence (general or specific) is almost certainly polygenic, it is unlikely that any one group ended up with the best alleles at all loci. Hence, it would be best to mix things up a bit if you want to maximize the potential for intelligence within the human population.
(Needless to say, you don’t need to believe those things to promote interracial marriage on other grounds.)
christian h. 10.22.07 at 2:14 am
Steve has already nailed it – it’s really correlation vs. causality again – but may I engage in a little bit of credentialism and say that as a mathematician, my reaction to Cosma’s essay was “duh!” It makes the most important point well: purely for reasons of mathematical statistics, it is possible to “find” a dominant factor even in a completely random data set. In other words, using “factor analysis” to “find” a dominant factor does precisely nothing.
David Kane 10.22.07 at 2:33 am
Why don’t Henry and John organize a seminar on the topic of Cosma’s essay at Crooked Timber? Invite discussions from, say, some of the smart folks at Gene Expression, allow Cosma to reply, and then open the floor to discussion. Wouldn’t that be the academic way to handle the discussion?
Alas, if past performance is any good, the authors at Crooked Timber prefer that there pages be unsullied by such disquieting viewpoints . . . And that’s a shame.
adam 10.22.07 at 3:27 am
I read that blog entry. Reading it awoke my dormant love of mathematics.
But I’m just not good enough. Abstract algebra was the death of me. Poor working memory? Who knows!
Bill Gardner 10.22.07 at 3:40 am
Christian H at #42:
I think you are reading just a bit too much into Cosma’s essay. Using EXPLORATORY factor analysis to find a single dominant factor won’t do, and of course a correlation matrix based on completely random data will always have a largest eigenvalue. However, a completely uncorrelated set of variables is unlikely to produce a single LARGE eigenvalue (given a large sample, etc.), and you can use ‘confirmatory’ factor procedures to look at this. He also showed that there are structures of non-zero correlations among variables other than a g-structure that will produce a large eigenvalue. Cosma’s right: this won’t be sorted out until we can do it from neuroscience. But factor analysis isn’t quite as useless as you suggest.
bi 10.22.07 at 3:43 am
David Kane employs the “blow lots of smoke in the name of promoting ‘reasoned discussion’ but don’t put in any actual effort to start this reasoned discussion and in fact try to avoid it” antipattern.
Bill Gardner 10.22.07 at 4:02 am
Fred @ #33:
“Kenyons are great distance runnrs”
The Kenyon Lords own NCAA Division III swimming. They are not distinguished runners.
Carter 10.22.07 at 6:06 am
“Shalizi’s arguments are quite comprehensible to someone with a basic minimum of statistical training”
And yet Shalizi’s arguments are completely incomprehensible to anyone with a low IQ. Strange, isn’t it?
Tim Worstall 10.22.07 at 7:38 am
Fred @ 33.
You’ve hit on one of the reasons why Watson’s claim of “Africans” as a group is nonsense. The Kenyan long distance runners (and the Ethiopians etc) are quite different genetically from the West Africans who provide the sprinters in the US (and the Caribbean etc). Simply put, the term “African” as a genetic grouping simply doesn’t exist: quite independently from whether “g” means anything or not.
zdenek v 10.22.07 at 8:18 am
In a paper to which Cosma links, Clark Glymour adds some additional force to the point about why establishing causation ( as opposed to correlation )is so elusive, by methods used by Hernnstein and Murray.
The difficulty can be brought out in another way : they rely on the following principle :
If some characteristic is largely genetic and there is an observed difference in that characteristic between two groups, then there is very likely a genetic difference between the two groups that goes in the same direction as the observed difference.
Applying this to Intelligence, we get : given the heritability of IQ, if East Asians are superior in measured IQ, then, they are highly likely to be genetically superior; and if Blacks are inferior in measured IQ, then they are highly likely to be genetically inferior in IQ.
Is this right ? The problem is that ‘genetic’ can mean ‘heritable’ or it can mean ‘genetically determined’ and for their argument to work they need to show not just heritability ( which they do seem to show ) . They need to show that IQ is genetic in the sense of being genetically determined , but they have not done that.
PJ 10.22.07 at 9:07 am
There is another problem – to establish heritability you will look within a narrow population with similar environmental factors. But when generalising to the cause of differences between groups you are looking at populations with different environmental factors. So even if the high heritability you’ve found is driven by genetics there is no reason to believe that the differences between the groups is due to genetics because the genetic effects and heritability are specific to a given environment. My favourite example of schizophrenia in African Americans and British African Caribbeans:
Alex 10.22.07 at 9:25 am
To put it in really simple terms, if you break down any measure into more than one factors, which are positively correlated with your measurement, there will always be one which is the biggest. (If they are all the same, then they are meaningless; you’ve just renamed the original measure)
That there is one that is bigger than the others doesn’t tell you anything in itself about causation, truth, or explanatory power. It may be that this residual value is simply another way of describing the one you started with.
This is what happened here; all we have learned is that people who do well on IQ tests still have a high score when you strip out other factors, which is both obvious and trivial. Just juggling the figures doesn’t answer the question of whether or not the IQ test itself is useful.
John Quiggin 10.22.07 at 9:28 am
DK at #43. Last time you raised this, I said “I’ve said my piece for now, but I’d be interested to read a GNXP view of the main developments in recent decades, taking account of the Flynn effect. If someone at GNXP wants to write something like this, I’ll certainly make some comments and may respond at greater length.” AFAIK, that’s where it stands.
I’ll take the chance to restate what I said about g that time “I don’t have a firm view one way or the other on the validity of the g/general intelligence construct – I haven’t looked hard at it. The expositions I’ve seen relied on a rather naive reification of the outputs of factor analysis – maybe you can point me to something more sophisticated.”
I think it’s pretty clear from the discussion on these threads that no such sophisticated defence exists – all we have is a bunch of people who impute magical powers to a change of bases in vector spaces (this is, in effect, what factor analysis consists of).
But all that said, I agree with you. If someone from GNXP wants to write a rebuttal of Shalizi, showing some actual evidence that g is more than a statistical construct, I’d certainly be in favor of CT providing a venue for disussion.
Alex 10.22.07 at 9:28 am
Mind you, the presence of David Kane on this thread does tend to indicate that different forms of right-wing scienciness are strongly correlated; if you’ve got Lancet report, you’ve probably also got g, and probably climate change denial too.
We could define a residual factor called c for crap; but as Cosma has so wisely pointed out, that would be unscientific.
zdenek v 10.22.07 at 9:54 am
Also, does not the so called ‘Flynn Effect’ on its own , show that IQ cannot be genetically determined ( though not that it is heritable ) ?
I mean as is well known ,IQ has been rising about 3 points every 10 years worldwide. Since 1945, IQ in many countries has gone up 15 points, about the same as the gap separating Blacks and Whites in this country. For example, average IQ in Holland rose 21 points between 1952 and 1982.
If having 2 arms ( genetically determined ) varied in this way with changes in environment, would we think still that having two arms is genetically determined ? So again heritable yes but genetically determined no.
soru 10.22.07 at 10:00 am
I follow this stuff well enough to understand it, but not enough to apply it safely to new and novel situations.
So if somone asked me ‘doesn’t that argument imply that climate change is also a statistical construct, produced by aggregating different local measurements on a complex system?’, how would I best respond?
I can see a couple of possible reasons why that could be invalid, but I can’t see one that would obviously refute the idea ‘climate change is a myth: you could replace climate change, ‘c’, with a system of correlated variables c1 to cn with no visible effect on any experimental results’.
PJ 10.22.07 at 10:21 am
“So if somone asked me ‘doesn’t that argument imply that climate change is also a statistical construct, produced by aggregating different local measurements on a complex system?’, how would I best respond?”
Without getting into statistical mechanics, changes in global temperature levels are real physical measures with causal effects. g is a statistical construct that does not, as far as we know, have any causal effect other than that mediated by the factors it correlates with.
“Also, does not the so called ‘Flynn Effect’ on its own , show that IQ cannot be genetically determined”
No, it shows that there is a large environmental influence on IQ, and one that seems to be changing with the generations, but it does not tell us whether, if we fixed these environmental factors constant, genetics would account for much of the variation in IQ, or whether, even given the large environmental factors, genetics does in fact contribute to a lot of the variation.
PJ 10.22.07 at 10:30 am
Ah, soru, I see what you mean now, you mean that global temperature is just a statistical construct that supervenes on local temperature, which is true, but if you think of it as such it doesn’t alter any of the conclusions regarding climate change. As I initimated, temperature is simply a summary statistic of the actions of individual particles. But we can usefully think of it as causal at a certain level because we understand how it works.
With g we have a sort of summary of IQ test performance (or rather the correlations between subscales), but it doesn’t represent anything deeper than that, any predictive value it might have is correlatory rather than as a result of being a summary statistic. And, of course, it doesn’t represent or summarise anything physical that we know of, unlike temperature.
John Quiggin 10.22.07 at 10:49 am
To follow up on pj, the various silly alternatives to the arithmetic mean temperature suggested by McIntyre and McKitrick are indeed meaningless statistical aggregates. There is a well-developed, and well-verified, physical theory explaining why arithmetic mean temperature is meaningful.
Matt McIrvin 10.22.07 at 11:17 am
Anyway, Shalizi at one point links to a paper finding no evidence for a genetic component to the dominance of East African distance runners.
Brett Bellmore 10.22.07 at 11:23 am
“Also, does not the so called ‘Flynn Effect’ on its own , show that IQ cannot be genetically determined”
No. It shows that it can’t be 100% genetically determined. The Flynn Effect is perfectly consistent with the heritability of IQ rising with time, as a matter of fact, if it represents a trend towards the environmental influences on IQ being optimized.
I find it remarkable the hostility you can generate here, by observations which, in the end, don’t amount to anything more controversial than noticing that you can’t do any intellectual task well if your brain cells aren’t working.
zdenek v 10.22.07 at 12:07 pm
“No. It shows that it can’t be 100% genetically determined. The Flynn Effect is perfectly consistent with the heritability”
It is consistent with heritability of IQ ( that is what I say ) but not with the hypothesis that IQ is genetically determined.
In other words you seem to be confusing ‘heritable ‘ with ‘genetically determined’ . Hernnstein and Murray do the same thing.
PJ 10.22.07 at 12:55 pm
“It is consistent with heritability of IQ ( that is what I say ) but not with the hypothesis that IQ is genetically determined.”
‘Genetically determined’ and ‘heritability’ can be a bit meaningless if you don’t specify what you mean.
As you pointed out above, number of arms is genetically determined to a large extent (we mostly have 2, and that’s because of our genes), but variance in the number of arms is mostly environmental. Heritability is talking about variance, so number of arms is not very heritable, but is genetically determined.
Something can have a high heritability estimate in a given situation because it is confounded with an environmental factor, but also, since heritability estimates only attribute variation to differences in genotype compared environmental variation in the context they’re measured, you could have a high heritability in a given study population while still having large environmental factors influencing phenotype between generations.
So the Flynn effect could be compatible with high genetically determined heritability within a generation, but with large environmental effects between generations.
Of course the presence of the Flynn effect also suggests the likelihood of causal environmental factors that vary within the generation – but does not establish them.
Barry 10.22.07 at 1:24 pm
Brett:
“No. It shows that it can’t be 100% genetically determined. The Flynn Effect is perfectly consistent with the heritability of IQ rising with time, as a matter of fact, if it represents a trend towards the environmental influences on IQ being optimized.”
That last sentence is gibberish, but it does illustrate a major (if not *the* major) objection to Bell Curvism – the tendency to look at massive environmental effects, and to ignore their effects.
“I find it remarkable the hostility you can generate here, by observations which, in the end, don’t amount to anything more controversial than noticing that you can’t do any intellectual task well if your brain cells aren’t working.”
The hostility you are seeing is due to fraudulent, racist junk science.
PJ 10.22.07 at 1:31 pm
“That last sentence is gibberish”
Strictly speaking it isn’t. In theory, if we were progressing in some Whiggish fashion towards some environmentally perfect world without malnutrition or poor education then we would expect the genetic heritability of a trait to increase as the effects of the environment become uniform across the population leaving only genetics (and other random factors) to contribute to variance.
christian h. 10.22.07 at 2:02 pm
Bill (#45), yes, I wasn’t sufficiently precise in what I wrote – of course a completely random set of data might not give a dominant factor that’s very dominant. What I should have said is: exploratory factor analysis cannot, by its very nature, do anything but show that a certain linear map (Cosma also remarks that the linearity assumption is questionable) factors, up to a small error term, through some vector space of low dimension (preferrably, 1). It can never tell you why.
Let’s do a thought experiment: say there are two uncorrelated mental abilities, A and B. And say we measure two outcomes, C and D. Now if C and D happen to be expressable in A and B with pretty much the same coefficients, then factor analysis will produce a single factor G “explaining” the correlation – but this G will simply be an artefact of the mathematics.
Brett Bellmore 10.22.07 at 3:23 pm
“In theory, if we were progressing in some Whiggish fashion towards some environmentally perfect world without malnutrition or poor education…”
And we are. We’re no longer using lead in gasoline, for instance; That’s got to be having a measurable impact on IQ tests by itself, given the known effects of lead on developing brains.
Frankly, our environment is a long way from being optimal for intellectual development, and a longer way from being uniform. While I’m opposed to ruling out a priori any genetic contribution to group IQ differences, environment is the sensible place to expect most of those differences to be explained.
Bill Gardner 10.22.07 at 3:36 pm
Christian @ #66:
I think we agree completely. BTW, my impression is that the technical issues in non-linear factor analysis are pretty hard. As you would expect. It is hard enough to identify a latent structure when you constrain the function space to be linear.
cheers
Bill
rb 10.22.07 at 3:56 pm
They need to show that IQ is genetic in the sense of being genetically determined
In addition, they would need to show that the “races” are genetically distinct.
But they’re not distinct.
Ballgame.
The Raven 10.22.07 at 4:12 pm
Anybody who does IQ testing for the state, for SSI, for the VA, etc., can tell you that the Stanford Binet and WISC are fairly accurate and reliable. That is, two different testers at different times will arrive at a score that has only a small margin of error. And the tests do, in fact, give you a measurement of what they purport to measure.
Question: Is this child here mentally retarded? (score below 60)
Question: Is this child here bright? (score over 120)
In no circumstances would either child obtain the other’s score. If our general academic conception of g is valid, this is the expected outcome. People arguing that there is no such thing as g, that there are “multiple intelligences” and such, are arguing that a very bright student might obtain a low IQ score as well as the reverse. That never happens.
jayackroyd 10.22.07 at 4:22 pm
rilkefan:
. And isn’t the correct analogy to say that time is a myth because I can show that your watch can’t measure it?
No. The original post very clearly says that your watch state is irrelevant to the claim. Using a method that fails to show that g exists doesn’t prove g doesn’t exist. Good practice has one assume that a phenomenon doesn’t exist absent proof to the contrary, but it is careless to say that something is false.
On the post itself, Sullivan’s problem is typical of people who write columns and magazines. He starts from a political position–a left or a right wing point of view on the question of g–and then assumes that researchers are proceeding from such a point of view. This seems to be an especially serious problem among convervatives….
liberal 10.22.07 at 5:27 pm
brett bellmore wrote,
What is “cell level learning” in the context of actual neuroscience (as opposed to, say, a connectionist model)? I’m talking about a construct that’s actually used by real neuroscientists.
H 10.22.07 at 5:33 pm
“Race” (skin, hair, eye color) is determined by genetics. “Intelligence” or that quality psychometricians can measure may have more to do with epigenetics.
PJ 10.22.07 at 5:35 pm
“What is “cell level learning†in the context of actual neuroscience (as opposed to, say, a connectionist model)?”
LTP/LTD? Although granted they are model systems.
Bloix 10.22.07 at 5:44 pm
I’m in the market for a new car right now, and I’m reading a lot of reviews that discuss all kinds of factors relating to performance, safety, comfort and reliability. For each of these there’s a whole range of considerations. Take performance: horsepower, torque, steering, braking, shifting, tires, etc. At the end of these reviews, often there’s a number rating: this car gets an 8.6, another one a 6.9, a third a 6.2.
I accept that the number rating is a useful shorthand expression of someone’s overall opinion of the car based on an evaluation of the various factors. But it’s obvious that the factors are more-or-less independent variables which are themselves composed of many variables that are more or less independent.
Now, why would anyone suppose that a human brain is a less complicated object than a sedan?
Sebastian Holsclaw 10.22.07 at 6:05 pm
The strange thing about the argument politically is that many of the same people who want to assert a superior ‘g’ also want to cite superior culture as a large part of the reason for the success of ‘the West’. That seems rather greedy. One would think superior culture would be explanatory enough, if true.
Henry 10.22.07 at 6:07 pm
sebastian – did you get the PDF I sent you? I didn’t hear back from you and wanted to know whether it made it (it was a large file, so may have been bounced).
Tom Fuller 10.22.07 at 6:16 pm
Sense check:
One large group of people does not score well on a test designed by members of another large group of people, a group that lives a very different lifestyle and essentially belongs to anothe culture with another vocabulary designed to describe it.
Those in the first group have limited, if any, access to the culture that inspired the language of the test. They also are poorer, less-educated, often less-well cared for. Their diet is often poorer. They are more likely to be exposed to chemicals (such as lead) that affect mental performance. Their upbringing is dramatically different than members of the second group.
They do less well on the test. It must mean they are genetically inferior and should have remained slaves. Right?
Why do IQ differences disappear when measuring between ‘races’ but within income or class groups? Just asking…
Will Roberts 10.22.07 at 6:20 pm
Brett @ 67:
If all we were doing were learning more about the effects of our actions while not actually performing any new actions, you’d have a point. Since we are not and never will fulfill this condition, the Whiggish notion of progress is nonsense. We will continue to find new and spectacular ways to mess up our environment in ways that impact human flourishing. It’s entropy, baby.
Sebastian Holsclaw 10.22.07 at 6:39 pm
I forgot to check my public mail to look until you reminded me. I just got it. Thanks.
bi 10.22.07 at 8:02 pm
Jeez, The Raven (#70) can’t tell the difference between a convenient summary bogometric and a latent factor that controls everything else? I mean, can he even grasp the idea that a statement such as “this child is bright” doesn’t necessarily mean there must be a unitary latent factor called “brightness” which is the sole cause of all bright behaviours?
“People arguing that there is no such thing as g, that there are ‘multiple intelligences’ and such, are arguing that a very bright student might obtain a low IQ score as well as the reverse.”
Well, I think we just found a perfect counter-example — assuming that The Raven has a high IQ, that is.
soru 10.22.07 at 9:29 pm
Now, why would anyone suppose that a human brain is a less complicated object than a sedan?
Actually, the brain would be _more_ complex if, in addition to all those trade-off factors, it _also_ had a more or less linear scale marked ‘better/worse’. For example, if instead of just looking at currently-marketed sedans, mix in cars from previous decades. Current models would presumably beat those from the 1970s cars on most counts, and those from the 1950s or 1930s on almost all.
So ‘year of manufacture’ is a simple linear factor that coexists with a complex pattern of partially correlated factors. Just as ‘C02 concentration’ coexists with complex patterns seasonal weather.
At some scale, something like this must happen: a human brain is simply smarter, not just differently smart, than that of other apes. The list of brain-tasks an average chimp can do better than an average human is about similar in size to the list of transportation tasks a 1920s Rolls Royce can do better than a 2007 Toyota.
I think the key point is, that as #58 implies, we simply don’t understand the mathematics of how the brain processes information well enough to be able to say whether such a factor exists or not. The brain might, surprisingly, be a simple cpu with a clock speed printed on it.
More likely, it is a university, where no matter what statistics you gather, any claims of simple better/worse not directly tied into a plan of improvement are most likely just an excuse to start a fight.
Walt 10.22.07 at 10:17 pm
I would like to object to John Quiggin’s claim that there’s nothing magical about changing bases of vector spaces. Since I’m really good at changing bases of vector spaces, I think we can safely conclude that this constitutes the sine qua non of magic.
Cian 10.22.07 at 10:20 pm
#70 – no IQ tests measure how good one is at IQ tests. How strongly this correlates to intelligence, assuming such a thing can be well defined (I suspect not) is a different question. And while they may be robust tests within their own parameters, it is perfectly possible to get significant improvements on IQ scores through practice, experience of tests/IQ style puzzles. In general, kids from poorer socio-economic backgrounds are more likely to be intimidated by such things, which might explain why they do less well on IQ tests (and respond so well to early interventions designed to raise IQ performance). Other factors have have a significant affect include motivation, something for which there is a fairly strong correlation with socio-economic background.
“People arguing that there is no such thing as g, that there are “multiple intelligences†and such, are arguing that a very bright student might obtain a low IQ score as well as the reverse. That never happens.”
And what is your measure of very bright? An IQ test? Are you beginning to see why there might be a problem here?
Brett Bellmore 10.22.07 at 10:34 pm
“People arguing that there is no such thing as g, that there are “multiple intelligences†and such, are arguing that a very bright student might obtain a low IQ score as well as the reverse.”
No, they’re arguing that, depending on what intellectual tasks you measure performance on, and how you weight them, people will end up ordered differently. Which is quite true. And that there’s no objective way to determine how to weight them. Also true.
Unfortunately, sometimes they’re arguing that there’s no such thing as smart and stupid, which is rather less true.
JBL 10.23.07 at 1:10 am
I thought pj’s answers (at #57 and #58) to soru’s question (at #56) were interesting, partly because I would have answered in an entirely different way: the claim that climate change is a statistical construct is absurd because we understand much more about the climate than we do about intelligence. In particular, climate models are not designed to apply statistical methods to a bunch of historical data. Instead, they are physical models, i.e. their predictions are based not on past climate patterns but rather on the underlying physical and chemical laws. If we could model intelligence based on the physical and chemical interactions in the brain in the same way, we wouldn’t need discussions like this one.
Bloix 10.23.07 at 1:15 am
#82- All 2008 model sedans are qualitatively different from all 1950 model sedans – no disagreement there. And I will grant you that pretty much everyone will agree that BMW 5-series is a better car than a Honda Fit. My point is that if you measure one parameter of a car – say horsepower – you are not necessarily going to learn anything about the car’s braking, or its sound system, or the number of passengers it will carry.
ScottS 10.23.07 at 1:28 am
To the advocates of multiple intelligences:
if IQ tests designed to measure “g” are flawed assessments bound by the social contexts that created and received them, and at best only create correlations that should not be used as a basis for asserting causation, then how can you be sure that any of the individual components of multiple intelligences aren’t similarly compromised?
Now of course it is common sense that some people are more naturally adept at picking up social cues, that language ability and mathematical ability can run on rather different tracks, and that some obviously very smart people can’t read a map to save their life. That does not mean that each of these separate intelligences is based on hard-wired biology, right?
Yet, most educators-in-training in the US are force-fed an oversimplification that posits that there are 7 distinct intelligences, and everyone has their strengths and weaknesses, and this is all based on settled, scientifically impeccable brain research, so there is nothing to discuss here; now, let’s move on the implications of these findings.
When a few naive teachers-to-be asked too many questions of the naive P.Hd candidate in Psych that was tasked to our section, he finally admitted that a lot of serious people in his field thought Gardner et. al. were full of it on the deep down science, even though there were many interesting hypotheses with interesting implications that improved the ability of instruction to make a difference in the education of a great many individuals.
I really don’t care whether g exists, or not. Pretending that only one side comes to this discussion with political blinders on is pretty shabby.
Dress it up in all the statistical analysis you wish; if God were to come to earth and say “not only does “g” exist I gave a bit more of it out in the North, sorry about that” I’m sure a few people with very high IQs would simply say, “God, I’m sorry, but since you are a social construct, the potential for your existence is irrelevant. Good day.”
Brett Bellmore 10.23.07 at 2:08 am
Actually, in the real world if you know a car’s horsepower, it WILL tell you a great deal about it’s braking… Since they’re designed so the brakes can bring the car to a stop with the engine running flat out, as a safety feature. ;)
bi 10.23.07 at 5:02 am
“how can you be sure that any of the individual components of multiple intelligences aren’t similarly compromised?”
We aren’t. Next?
“Pretending that only one side comes to this discussion with political blinders on is pretty shabby.”
Political blinders which justify discriminating against a group of people who have historically been subject to the manacles of outright slavery, ah yes. I don’t need to subscribe to some theory of 7 latent factors of intelligence (or is it 200 factors?… I forgot) to know that using the troubles brewing on the African continent to say bad things about black employees _in America_, as Watson did, is a totally dumb idea.
bi 10.23.07 at 5:03 am
Wait, was that another instance of the “Clinton did it too!” antipattern?
Bloix 10.23.07 at 5:44 am
Well, hell. I thought I was being clever and now that I’ve bothered to read Cosma Shalizi’s discussion I see that he used the very same car analogy that I did. So I’m not stupid, just lazy.
SG 10.23.07 at 6:31 am
If I may indulge in the magic of vector spaces for a moment, Bloix:
While the BMW 5-series may have a very high score on the factor “performance” (which has a very good correlation with the observed variables horsepower, braking etc.), in my non-linear factor model it has an infinite loading on the obliquely rotated factor “tastelessness”. As a consequence, it is inevitably a shittier car than every other car on the market (except the BMW 4WD, which is so shitty that no statistical model will fit it).
In fact I would go so far as to say that this is a causal model: 99% of the variability in people who bought one of these cars can be explained by this factor.
PJ 10.23.07 at 9:37 am
That is sort of what I was trying to say, but was rather hamstrung by the obvious fact that temperature is a statistical construct. A meaningful one, but still a construct. Which makes talking about the causality of temperature difficult. But as JQ and you say, it is a statistical construct that summarises something physical with well studied laws governing it which is why we can usefully think of it as causal.
mugwump 10.23.07 at 9:52 am
So it seems the arguments against “g” are all of the “correlation is not causality” variety. Fair enough.
There’s a very infamous group that have also consistently argued the “correlation is not causality” line: the tobacco companies.
So tell me, what is the difference between the majority of commenters here and the tobacco companies?
Ray 10.23.07 at 10:05 am
Most commenters here are smart enough not to use stupid analogies.
PJ 10.23.07 at 10:05 am
I think the argument here is that correlation doesn’t imply existence. Things that don’t exist obviously can’t be causal, but that’s a secondary point.
As far as tobacco goes, I refer you to the medical literature.
JJ 10.23.07 at 10:14 am
Even if there was a g factor, even if “race” actually existed genetically, even if the g factor was genetically inheritable, even if it was shown that g factor was a part of the genetic racial definition, and even if the intelligence “caused” by the genetics of the g factor is measured by IQ tests: the range of IQs found in one race (or gender more importantly), and the observed lack of correlation of that IQ to the success or failure of a life lived – all preclude us from making any decisions, or taking any actions towards any individual human being based upon membership in a genetic group, or an IQ category. It is patently reprehensible and monstrous to even think that if it were so it would then be perfectly ok to begin stereotyping individuals and treating them as faceless members of a few defined groups. One size never fits all – or even a majority.
mugwump 10.23.07 at 10:22 am
But at least one of them is apparently not smart enough to explain why the analogy is stupid.
Ok, but the “g-spotters” don’t need g to exist to make their arguments (although apparently they don’t know that).
The analogy with tobacco companies is strong: the medical profession never postulated a magic “tobacco-g” that was doing all the causal work – they just pointed to the strong correlations and said “there has to be causality”.
Likewise,”IQ-g” is not necessary for the arguments of the IQ-g-ers to have some legs: the strong correlations between IQ and certain groups suggests there must be causality (they don’t need the causality to be mediated by some mythical “IQ-g”)
At the very least, if you want to argue that the IQ-g-ers are wrong, you need to reconcile that with your position on tobacco.
zdenek v 10.23.07 at 10:51 am
“I think the argument here is that correlation doesn’t imply existence. Things that don’t exist obviously can’t be causal, but that’s a secondary point.”
Actually this is not accurate , because corelation also involves existence : if E1 and E2 are corelated then the relationship is between them is real real. The key thing is dependence : if E1 and E2 are causally related such that E1 causes E2 then E2 depens on E1; this is why this sort of relationship supports counterfactuals : if E1 did not happen then E2 would not happen .
PJ 10.23.07 at 11:02 am
mugwump,
everyone else here is discussing whether g, the general factor of intelligence is a real thing or not – which was the point of the Shalizi post. Shalizi is arguing that fans of IQ argue that g is a real causal factor because it comes out in factor analysis – Shalizi points out that is an incorrect conclusion. g or IQ may correlate with other things, but that is not the focus of this argument and so nobody needs to justify their position on the “correlation doesn’t equal causation” point with regards to tobacco because g is not like tobacco because there is no evidence, other than factor analysis, that it does actually exist. You can’t even get into arguments about whether the correlation between two things implies one causes the other if you don’t have any reason to believe that one of them actually exists at all.
With regards to tobacco and its ill effects (lung cancer, COPD etc) the weight of evidence is rather strong not just from correlations over time but through case control and cohort studies, the increased risk in smokers is huge, and related to dose. The question of whether causality can be inferred from epidemiology (correlations) is rather well known in medicine, for instance the Bradford Hill criteria are widely discussed. You cannot just take a “correlation doesn’t equal causation” position on everything, it is more complex and nuanced than that.
PJ 10.23.07 at 11:06 am
“Actually this is not accurate , because corelation also involves existence”
In what sense? The argument here is that g exists in a statistical sense only. Correlation coefficients do not have causal effects in my non-Platonic world!
Ray 10.23.07 at 12:23 pm
mugwump, in the case of tobacco, the situation was that a lot of people smoked, and a lot of people got cancer, and some medical researchers thought “Hey, perhaps these two observed phenomena are related!”
In this case of intelligence, there are a bunch of IQ test results, and, well, that’s it. Every now and then, some people who are keen on the idea that there are important differences between races, say “Hey, perhaps these IQ test results vary according to some mysterious underlying factor (and maybe that underlying factor is race but SSHH!! this is our big surprise in part 2)”
As I understand it, Shalizi’s point is that there are a whole load of real, observable factors that cause variations in the results of IQ tests – so many that there is no way to know (and no reason to believe) that there is a g, let alone whether or not g varies with (anything that could be sensibly described as) race.
But, you know, if you’re having fun with the stupid analogies…
zdenek v 10.23.07 at 12:27 pm
We need to distinguish between two different claims , one ok, one not ok. :
1) The ok claim about g is that given that tools ( factor analysis )used by Herrnstein and Murray do not licence causal claims , they are not entitled to conclude that there is g .
2) The not ok claim is that given that Herrnstein and Murray use factor analysis which does not licence causal claims g does not exist.
The general point is that we are not entitled to make this stronger claim ( g does not exist ) because g may exist and maybe discovered with better analytic tools. ( of course Herrnstein and Murray are not entitled to say this without such tools ).
mugwump 10.23.07 at 12:35 pm
pj, I am not talking about causation from IQ-g to IQ, I am talking about causation from racial genetic factors to IQ to socioeconomic status (SES).
I understand that the argument here is about IQ-g, but reading between the lines, it seems the point of debunking IQ-g is to debunk the race->IQ->SES chain.
My point is that the race->IQ->SES chain doesn’t need IQ-g and causation; correlations are enough.
Just as when arguing for the tobacco-lung cancer link, you don’t need to prove the existence of an underlying tobacco-g that causes lung cancer, the evidence from correlations is overwhelmingly supportive of a causal link (although, of course, discovering all the tobacco-g’s and hence the exact causal pathways from tobacco to lung cancer would put paid to any residual objections to the link).
mugwump 10.23.07 at 12:44 pm
ray, there’s plenty of correlation between IQ and “race”. Just like there’s plenty of correlation between smoking and lung cancer. If you want to argue against correlation implying causality in the former case, but for it in the latter, you have to delve into the details.
Repeatedly labeling the analogy as “stupid” is not what I would call “delving into the details”.
PJ 10.23.07 at 12:52 pm
But mugwump, in that case you don’t have evidence of a similar strength to the tobacco/cancer link that racial genetic factors cause differences in IQ, which was the point of Shalizi’s last post (or at least he was talking about the heritability of IQ, whether racial differences are due to genetic differences is another step).
It is more like links between pancreatitis and smoking, where there is also a great big confounding effect of all the smokers being big drinkers of alcohol. You need to tease out the causal factors in that case. In this case there are rather big social and economic differences between racial groups, so you can’t just assert that IQ differences are due to genetic differences.
Ray 10.23.07 at 1:10 pm
It’s funny how short a time it takes to get from “I’m interested in finding out if there is an underlying g” to “I’m interested in arguing that black people are poor because they’re stupid”.
Have you actually read any of Shalizi’s posts?
zdenek v 10.23.07 at 1:34 pm
pj : The argument here is that g exists in a statistical sense only. Correlation coefficients do not have causal effects”
This is too quick surely ( Cosma concedes this point too ). Strictly speaking we do not know whether g exists and hence we do not know whether g causes anything. Some sort of verificationism is required to establish the principle that you seem to be relying on : O exists if it is perceived or if it is analytical truth that ‘ O exists ‘ . But this seems counterintuitive.
mugwump 10.23.07 at 1:36 pm
Yes, ray, I read Shalizi’s post linked to in the article. Spends an awful long time pointing out the bleeding obvious to a statistician, but I guess he’s up against it with social science: give them a pointy tool and they’re guaranteed to injure themselves with it.
I will read the post linked to by pj above before I comment (another long one).
PJ 10.23.07 at 1:44 pm
“This is too quick surely ( Cosma concedes this point too ). Strictly speaking we do not know whether g exists and hence we do not know whether g causes anything.”
I’m not claiming that g definitely doesn’t exist, just that, contrary to your claim in #99, that a factor can be derived from the correlations between scores does not mean that this factor represents anything physically real or plays any causal role in those scores – i.e. it is a statistic, nothing more.
Therefore the argument is about more than whether correlation equals causation, because we’re not talking about two real things that correlate but essentially a correlation coefficient. And correlation coefficients are statistics, not real things.
zdenek v 10.23.07 at 2:03 pm
“that a factor can be derived from the correlations between scores does not mean that this factor represents anything physically real or plays any causal role in those scores – i.e. it is a statistic, nothing more.”
Agree, but do you not want to say also that ‘g is a statistical construct only’ meaning that g does not exist ? The problem is that that conclusion does not follow from the fact that the tools Herrnstein and Murray use do not get them the result they want.
Or am I missing something ?
Cranky Observer 10.23.07 at 2:12 pm
> Actually, in the real world if you know a car’s
> horsepower, it WILL tell you a great deal about
> it’s braking… Since they’re designed so the brakes
> can bring the car to a stop with the engine
> running flat out, as a safety feature. ;)
After 1980, yes: as a result of (here it comes) social pressure on automobile designers. Had you studied the correlation between engine power and braking power between 1885 and 1980 there would have been at best a very weak correlation; after 1980 a strong one. But nothing intrinsic to the design of automobiles or the laws of physics or mechanical engineering would explain the change.
Cranky
Bill Gardner 10.23.07 at 2:25 pm
“There’s a very infamous group that have also consistently argued the “correlation is not causality†line: the tobacco companies.”
1. From very early in tobacco research, there was also experimental evidence in animal studies showing that cigarette smoking caused cancer. The idea that the evidence was just observational was one of many distortions of the scientific record by the tobacco companies. (To appreciate the deliberate and systematic quality of the this distortion, see Allen Brandt’s The Cigarette Century.)
2. Even in the observational data, the health effects of cigarette smoking are MUCH larger than the associations between ability tests and demographic factors.
3. There are concerns about correlation and causality in IQ research, but I don’t think that is CS’s main point.
4. If you are really interested in this issue, may I recommend the work of my friend Eric Turkheimer?
Eric Turkheimer, Andreana Haley, Mary Waldron, Brian D’Onofrio, Irving I. Gottesman (2003). Socioeconomic status modifies heritability of iq in young children. Psychological Science 14 (6), 623–628.
Turkheimer E, Waldron M. (2000) Nonshared environment: a theoretical, methodological, and quantitative review. Psychological Bulletin;126(1):78-108.
Turkheimer, E. (1998). Heredity and biological explanation. Psychological Review, 105 4 782-91.
Steve LaBonne 10.23.07 at 2:59 pm
Strictly speaking you can say the same thing about Santa Claus. What’s lacking in both cases is evidence that the claimed entity DOES exist. Until some is provided neither is worth discussing by grownups.
Alex 10.23.07 at 3:20 pm
The difference between g and inferring that smoking causes cancer is that Richard Doll had an empirical referent; he could count smokers and count the cancerous. Here, however, we’ve got half the picture; we’re comparing one empirical result (the IQ test) with something we got out of analysing the IQ tests, so there is no external check on its validity.
Cigarettes would be an analogy if rather than counting the smokers who got cancer, Doll had calculated a general factor of smokiness – call it s – and concluded that some sort of inferior group prone to lung cancer were genetically predisposed to high s and hence smoking.
zdenek v 10.23.07 at 3:45 pm
“What’s lacking in both cases is evidence that the claimed entity DOES exist. Until some is provided neither is worth discussing by grownups.”
Not always true. We talk about gravitons , and singularities etc. even though there is no evidence for their existence and they are theoretical entities and pure creatures of our current theories only.( i.e they are taken seriously unlike, Santa Claus, without any empirical evidence ; notice that for some of these theoretical entities there could not be direct empirical evidence ; think of singularity ).
Secondly it is never a bad idea , surely, to be clear about what has been shown and what has not been shown in the debate about the g factor.
My point is that to say that g factor does not exist, because we have shown that factor analysis cannot establish causality, is misleading : if you mean ‘there is no good argument for g’s existence ‘ then yes , but if you mean ‘there is good argument for g’s non existence ‘ then no.
LizardBreath 10.23.07 at 4:09 pm
Doll had calculated a general factor of smokiness – call it s – and concluded that some sort of inferior group prone to lung cancer were genetically predisposed to high s and hence smoking.
I’d have to google a bit to remember who it was, but this theory was put forth seriously at one time.
Alex 10.23.07 at 4:11 pm
Well, that’s one for the Global Stupidity Monument…
Steve LaBonne 10.23.07 at 4:42 pm
Gravitons and singularities are entities predicted by highly elaborated theories that have been very well-verified with respect to other observations. Not so g or Santa Claus.
Cian 10.23.07 at 5:22 pm
Another problem with the cancer analogy, is that we don’t actually have a measure of intelligence (we have IQ tests, but that’s rather different), whereas there’s a fairly easy way to directly verify if somebody died of cancer.
Actually a better analogy would be schizophrenia, and what do you know, there’s a similar controversy there. Science, eh.
mq 10.23.07 at 6:09 pm
No, they’re arguing that, depending on what intellectual tasks you measure performance on, and how you weight them, people will end up ordered differently. Which is quite true. And that there’s no objective way to determine how to weight them. Also true. Unfortunately, sometimes they’re arguing that there’s no such thing as smart and stupid, which is rather less true.
This was a sensible comment by Brett. But he doesn’t seem to have noticed that the first two true claims *do* rather imply that there is no such single quality as “smart”, full stop, or “stupid”, full stop. Or at least that there is no objective way to define such a quality. People will tend to be be smart in some areas, stupid in others. Claims that one is generally across-the-board smart can be taken as a rhetorical attempt to cover up one’s areas of stupidity and deny others who are different their forms of intelligence.
Muhammad Ali had a tested IQ of below 80, but I have no doubt at all that he was in certain ways much smarter than I am, and I do well on IQ tests. And I’m not talking about athletic skill, but verbal skill and decisionmaking ability. This strikes me as so obvious as to be hardly worth remarking on, but a follower of “g” would presumably say it is false.
ScottS 10.23.07 at 7:38 pm
“People will tend to be be smart in some areas, stupid in others. Claims that one is generally across-the-board smart can be taken as a rhetorical attempt to cover up one’s areas of stupidity and deny others who are different their forms of intelligence.”
They could be taken that way, and maybe, sometimes, they were meant that way. More often than not, claims that someone’s intelligence has been overlooked has more to do with an adult’s unwillingness to face the truth about their child or student and his or her general lack of academic ability. If you can’t remember things so well and /or have a lack of curiosity, you are at an incredible disadvantage (g or no g), and the teaching profession tying itself in knots trying to hit all the modalities of low-performing students “intelligences” results in a pretty phenomenally poor return on the investment and a lot of frustration because they desperately want a pedagogical solution to a social problem. Having others folks in the mix trying to prop up a biological explanation, even at the margins, is an awfully convenient strawman.
Brett Bellmore 10.23.07 at 8:58 pm
I think there’s something positively bizarre about the absolute determination on some people’s parts to reject any biological explanation for differences in performance of a biological system, the human brain.
Bill Gardner 10.23.07 at 9:34 pm
Brett @ #123:
“I think there’s something positively bizarre about the absolute determination on some people’s parts to reject any biological explanation for differences in performance of a biological system, the human brain.”
Is that the issue here? What CS said, sensibly, is that the dimensionality of variation in human performance will be determined by neuroscience, not psychometrics.
mugwump 10.23.07 at 9:45 pm
RE #107: I started to read the post you referred to, but quickly encountered such an elementary statistical error there seemed little point in continuing. FTA:
Wrong. The deviation of IQ from the mean mu is |X – mu|. The expected value of |X – mu| for normally distributed X with standard deviation s is (2/pi)^1/2^*s, not s.
In other words, the expected deviation from average of the IQ of a randomly selected person is (2/pi)^1/2^ *15, not 15.
bza 10.23.07 at 10:14 pm
Mugwump, it’s pretty obvious that in that remark Cosma is casually and not using “expect” in a technical sense. You’re really reaching here if that’s the best objection to his post you can come up with.
mugwump 10.23.07 at 10:30 pm
That’s extraordinarily charitable, bza.
If he knew the expected absolute deviation is not the same as the standard deviation, at the very least he would have said something like “Although standard deviation is not absolute deviation, roughly speaking, we expect the IQ of a random person to be around 15 points from the population average”. The use of “That is,…” indicates he probably doesn’t understand the difference.
The point is not trivial: the absolute deviation is considerably smaller (80%) than the standard deviation. An average person is a lot closer to average (12 IQ points) than the standard deviation (15 IQ points).
cosma 10.23.07 at 11:37 pm
Mugwump, I would like to appeal to your sense of altruism. Carnegie Mellon is, sadly, a very expensive school, and the people who foot the bill for our students rightly demand much more than basic competence from the faculty. if you really think my use of “expect” and “that is” implies that I don’t know the difference between the mean absolute deviation and the standard deviation, then given my position I really think you ought to share your proof with my department, before they let me teach again.
(Why yes, I have had a few too many beers.)
Vincent 10.23.07 at 11:42 pm
For much of the 19C. there was a strong correlation between areas of high rainfall and areas of high employment in Britain. For much of the late-20C. there was a strong correlation between areas of high rainfall and areas of high unemployment in Britain (this may or may not actually be so, but that is by the by). Neither of these correlations would allow us to conclude that rainfall causes either high employment rates or unemployment rates; nor could we conclude that employment or the lack of it causes rain to fall in buckets.
The correlation might prompt some researchers to go out and look for meaningful relations, some may or may not be found, but, clearly, correlation does not equal causality here.
Now forget the weather, look to geographical variations in employment and say, ‘hmm, there must be a reason why people in one area tend to be employed and not in others, it must be because of ‘e’, a genetically determined quality of employability.’ After all, jobs are tasks, as are IQ tests, that require brains (though those brains tend to be in to people, who tend to be in cultures).
mq 10.23.07 at 11:59 pm
I think there’s something positively bizarre about the absolute determination on some people’s parts to reject any biological explanation for differences in performance of a biological system, the human brain.
total straw man. There was nothing in Cosma’s post that rejected biological explanations for the human brain. I didn’t see anything in this discussion that did so either, although with 129 comments perhaps I missed something.
Anyway, I’m really tired of this kind of junk. It’s like seeing a phrenologist say “it is bizarre that people deny the obvious importance of the brain, housed in the skull, for human thought and personality”. It’s not like IQ is in any way rooted in brain science or neuroscience either.
mq 10.23.07 at 11:59 pm
oh, my post should have said “rejected biological explanations for differences in performance of the human brain”.
mugwump 10.24.07 at 12:42 am
Cosma, your background suggests you do understand the difference between the mean absolute deviation and the standard deviation, which makes the error so much more surprising. Why didn’t you just say “we expect the IQ of the average person to differ by 12”?
What do you do when you encounter elementary errors in blog posts written by people whose work you’re not familiar with and who you know nothing about? I almost always stop reading, because at that point I know I am going to have to check carefully every calculation and claim, and there’s just too much good stuff to read without going to that kind of effort.
Anyway, I stand corrected; the rest of your post is probably worth reading and I will do so.
BTW, since we’re being tools about it, I once turned down a position from your institution.
Brett Bellmore 10.24.07 at 1:11 am
“total straw man. There was nothing in Cosma’s post that rejected biological explanations for the human brain.”
So, what makes you think I included Cosma among those people? Cosma, by contrast to a lot of commenters here, is rationally agnostic about the implications of her point: Lack of proof not being proof of lack, and all that.
mq 10.24.07 at 2:56 am
Brett, what makes you think commenters here don’t accept the clear connection between biology and intelligence? The argument here isn’t between people who acknowledge biology and people who don’t, it’s between people who want to use “hey, it’s biological” as an excuse for crude reductionism and people who show more respect to the complexity of the phenomenon.
Look, thinking uses the brain. Same as, say, playing basketball uses the body. Obviously basketball skills are related to physical abilities that are purely genetic. But there are a lot of different skills that let you be a good basketball player, and it’s perfectly possible for very physically different people to be equally good at the game, although in a different way. If someone came along and claimed to have identified a “basketball-g” that measured ability in the game, said it was genetic, and said only black people had enough of it to be good basketball players, they’d effectively be racist toward white players. They’d also be wrong. White European players regularly beat the best black Americans now. Figuring out a single summary physical metric for definining a good basketball player based on genetics alone is in fact extraordinarily difficult, to the point that no one has done it. What seemed to be obvious genetic and hereditary correlations (like the black American one) have fallen apart in the space of a decade or two. How physical / genetic skills interact with environmental factors like training, team strategy, and the rules of the game is extremely complex (small differences between the NBA and international game have big effects on how much various physical skills help you).
Well, news for ya — thinking is a lot more complicated than basketball, and my guess would be that there’s a lot more ways to be good at it and a lot more abilities in play than there are in basketball.
Harold 10.24.07 at 4:22 am
The race is not to the swift nor the battle to the strong.
Alex 10.24.07 at 9:59 am
….nor bread to the wise, nor riches to men of understanding, which is more to the point.
cosma 10.24.07 at 4:49 pm
Mugwump: I considered explaining the difference between measuring the average distance in the mean absolute sense and a root mean square sense; that the variance gives us the standard deviation, and so the RMS sense; that there is no fixed proportionality between the two, unless you make distributional assumptions which are not part of the framework of heritability as such; and that if you are willing to make those assumptions, they shrink the whole population and the genetically-identical sub-population by the same factor. Then I decided that readers who’d appreciate, and grasp, that stuff would be able to supply it on their own anyway, and that being explicit about it would just create another obstacle for everyone else.
I am perfectly prepared to believe that this was a mistake.
LizardBreath 10.24.07 at 5:34 pm
Muhammad Ali had a tested IQ of below 80, but I have no doubt at all that he was in certain ways much smarter than I am, and I do well on IQ tests.
Heh. I remember this from a Newsweek article on the Bell Curve back when it came out, and thinking that there was an indictment of the accuracy of IQ tests — anyone who’d ever listened to an interview with Ali from back before the Parkinson’s hit could tell that he was an awfully bright guy, in a very conventional sense of bright.
Cian 10.24.07 at 9:10 pm
IQ tests from when Ali was a lad were heavily biased in their cultural assumptions. When they fixed the more obvious biases, there was a sharp jump in black IQs. They have improved, allegedly.
mugwump 10.24.07 at 11:53 pm
cosma,
The problem is that this is two precise statements of a common misconception: that the expected deviation from the mean of a random sample is equal to the standard deviation. Among other things, standard deviation is used because variances of independent random variables add, not because it gives you the expected deviation from the mean.
As for distributional assumptions, while I agree that they don’t matter to the discussion, IQs are conventionally assumed to be normal, and no reader would object to that assumption being unstated.
Anyway, more important to me is the thrust of the rest of the post, now that I have read the whole thing.
Your argument essentially boils down to “there’s too many confunding factors to be able to reliably estimate heritability of IQ”. But your arguments apply equally well to estimating heritability of almost any human trait. Which to me means there has to be a better way to get to the bottom of this.
When I watch NFL football and see that nearly every running back is black, I don’t think “those poor white guys growing up in athletically disadvantaged environments”. Likewise when the entire lineup for the finals of the 100m sprint at every major international sports meet is almost always black.
Nor, when yet another Nobel prize is won by an Ashkenazi jew, do I think “those poor protestants, growing up in intellectually disadvantaged households”.
When overwhelming talent percolates out the top end, there’s got to be more going on than just environment.
Harold 10.25.07 at 3:58 am
I think it may be the same genetic factor that makes Koreans have vegetable stores and Pakistanis newsstands.
Kevin Donoghue 10.25.07 at 9:57 am
Nor, when yet another Nobel prize is won by an Ashkenazi jew, do I think “those poor protestants, growing up in intellectually disadvantaged householdsâ€.
Nor should you. The House of Lords is overwhelmingly Protestant. Membership is a highly heritable trait, though not as much so as it used to be.
Knecht Ruprecht 10.25.07 at 12:29 pm
if he has even a moderate grasp of statistical reasoning
Assumes facts not in evidence
Ray 10.25.07 at 1:13 pm
Forget running backs and Nobel Prize winners – how come all the best dancers are black and all the bankers are Jewish?
harold 10.25.07 at 4:19 pm
Beethoven was black
http://www.straightdope.com/columns/050527.html
and so were Colette & Pushkin & possibly St Augustine
harold 10.25.07 at 4:25 pm
You may also well ask, why did Ashkenazi Jews only start winning Nobel prizes (or the equivalent) when they moved to America?
Didn’t Wagner write (anonymously) that Jews never could be really good composers — or even achieve anything original at all, since they were cosmopolitan, etc etc.
We also heard that women couldn’t write, or drive cars or jet airplanes, announce the news, and on and on.
Japanese people were not supposed to be able to make anything of high quality. (it is to laugh)
I even remember hearing in my childhood that only white musicians could really play jazz!!
engels 10.25.07 at 4:39 pm
Which to me means there has to be a better way to get to the bottom of this.
ie. reeling off half-baked anecdotes and prejudices as in the remainder of your comment above…
mugwump 10.26.07 at 12:08 am
Apparent one man’s fact is another man’s prejudice when that fact doesn’t suit his political philosophy. Running backs and top male sprinters are almost universally black. Ashkenazi Jews do win way more than their fair share of Nobel prizes.
I picked on these examples for a reason: my gut feeling (details to be worked out) is that you may be able to say something more statistically reliable about heritability vs environment on the basis of outliers than you can based on the performance of the general population.
Harold 10.26.07 at 12:26 am
Yes, but it seems that only (or by far the majority) Ashkenazi Jews born in the US win these prizes.
Why didn’t they win equivalent distinctions during over a thousand years residence in Poland and Germany? How come they were “inferior” until the last 100 years?
mugwump 10.26.07 at 1:13 am
Einstein did the work for which he was awarded the Nobel prize in Switzerland.
And were the Ashkenazi Jews that won earlier last century born in the US or immigrants?
One of the principal reasons the US developed the bomb before Germany in WWII was because the US had all the genius Jewish physicists expelled by Germany working on the Manhatten project.
[Full disclosure: I am not an Ashkenazi or any other flavor of Jew].
Another reason: the US wins more than its fair share of Nobel prizes period, because it has a better environment for research. It may be that the Ashkenazi Jews rose to such great Nobel-winning prominence in the US for the same reason blacks only rise to great running-back prominence in the US: there’s not a lot of NFL played in Africa.
Harold 10.26.07 at 3:25 am
I looked on a website that listed Jews who had won Nobel prizes and it said that the great majority of them were US citizens.
Yes, the environment — that of pre-WWI Germany in encouraging people to pursue scholarship no matter what their class background — and Einstein did have the benefit of a very good education — may indeed have contributed to bringing out the latent abilities of the citizenry.
Many working class Brits and Americans by the way, went to Germany in that period to get an education, from which they otherwise would have been debarred in their native countries for class reasons, and many became world-famous scientists.
The moral to be drawn is that one never knows where great abilities may be found and only fools would exclude people from opportunities pre-emptively on the grounds of race, religion, or sex.
mugwump 10.26.07 at 5:40 am
Of course. But that’s not the question being considered here.
Harold 10.26.07 at 2:18 pm
Actually, it is the question. But history doesn’t confirm that because Great Britain produced a Shakespearean age once, another one is going to turn up in Great Britain any time sooon. Or because Athens, Florence, and Rome .. etc., etc. Experience tends to show the opposite is the case. The locus for these spots is ever shifting.
Though perhaps we could learn something about the conditions that nurture talent. Shouldn’t thatreally be the question? Not the essentialist one of who belongs to the chosen clan.
cosma 10.26.07 at 4:50 pm
mugwump: It seems that what’s bugging you is my use of the word “expect”. If you would care to suggest another term, which would get the idea of root mean square distance across to a non-statistician without offending your sensibilities, I’d be obliged to you, and cheerfully update the post.
“Your argument essentially boils down to ‘there’s too many confunding factors to be able to reliably estimate heritability of IQ’.”
I think it’s stronger than that. There is a single coefficient of heritability when the right model is (trait) = (genes) + (environment). If the right model is non-additive, the coefficient of heritability just isn’t well-defined. You might still hope that the deviations from this model are small enough that it holds as a reasonable approximation, but that needs to be supported with evidence, and results like Turkheimer’s show that this is going to be hard to do. Once those obstacles are cleared, then you have to confront confounding.
“But your arguments apply equally well to estimating heritability of almost any human trait”
Yes.
“Which to me means there has to be a better way to get to the bottom of this.”
I agree: it’s experimental biology and experimental psychology.
As for your examples, they leave me puzzled. You evidently care greatly for precision and rigor, and I presume are not one to strain out the gnat and swallow the camel. Yet in all the cases you instance, systematic genetic differences accompany systematic differences in culture and in position in social networks, so ugly words like “confounding” and “multicollinearity” spring insistently to mind. They do not go away when I tell them that you want to look at the extremes rather than the averages. In fact, they insist that this path has even more obstacles blocking it than the twin/adoption route. But no doubt all this has already occurred to you, and it’s superfluous for me to point out that a solution to these difficulties would revolutionize the whole field of causal inference. Beside that accomplishment, uncovering the genetic reasons why the Boston police force used to be so heavily Irish — and no doubt the genetic differences dividing the Irish emigrants to Massachusetts from the Irish emigrants to England — will be as nothing. I won’t ask you to risk being scooped by revealing your secret in a comments thread, so I will just have to keep a watchful eye on the journals.
Comments on this entry are closed.