The Bell Curve

[_Chap]

If I may observe briefly - a major question relevant to the discussion on this thread is precisely whether or not 'intelligence' is a 'thing', whether abstract or not, or whether it is mostly just a word that people use when they are articulating certain political or social positions in the guise of stating facts about human beings.

I don't claim to have a definitive answer to that question.

There are three major theories of intelligence in the milieu of psychology with some other ancillary theories here and there. The work Analytics is interacting with relies on factor analysis in an effort to prove that there exists general intelligence that explains relative performance a wide variety of different cognitive tasks. It's a respectable position, and is only worth challenging in this context if someone declares or makes an argument that relies on the idea that g factor is settled science.

I had noticed that, yes. In fact I had to study this stuff in detail (quite) a while ago. However, on the assumption that we are not a board made of of specialists in psychological testing, I think my remark was apposite.

I could've course simply have said, in the Joad mode "It all depends what you mean by 'intelligence' ." Because it certainly does. In non-expert discussion of such matters it is often falsely assumed that if someone constructs a test that produces numbers of some kind, then we may assume that there exists some separate and objective thing that is being measured by the said test. This is sometimes far from being the case. Given the large political dimensions of any discussion of intelligence testing, this should be kept in mind.

[Guest]

Here is my actual comment addressing this, prior to that last sentence you quoted:

So if individual factors, NOT captured by the statistics, are the "driving determinants" of success, then the only way that IQ could still have "the most predictive power" would be if those individual factors WERE correlated with IQ. But the main argument of these outliers, if I am reading you correctly, was that these individual factors were NOT correlated with IQ. So, which is it? IQ predicts? Or individual factors predict? Or is it that the authors are carefully asserting no individual is being talked about here, but their analysis still holds for groups and averages, hence EA's "exaggerated relationship" comment?

I only have a minute, but I'll try to answer this. The idea is that IQ has the most predictive power of the variables we have been able to study. Individuals do better or worse than what the prediction based solely on IQ predicts. On average however, IQ does a good job of predicting.

Why do some people do better than the prediction and others do worse? Presumably there is a reason. Or maybe it's just luck. Or both? The whole concept of predictive analytics and big data is to expand the scope of analysis, bring in more factors, look at other model forms, and make better predictions. The fact that the model has room for improvement doesn't invalidate the statistical relationships that it does explain.

Whether it "exaggerates" the relationship is a subjective claim, one that I'm not sure I agree with and don't know how to address.

Does that help?

Who is it that you think is asking for help? That's twice now. Just weird. I'm not asking for help, I am registering some serious reservations regarding your the assessment of the statistical work you are reviewing.

You say "presumably there is a reason" the model fails for some, but you don't seem to understand that if the statistical analysis is not set up correctly by the humans it will fail--not because the principles of statistical analysis are weak, but because the set up has errors!

And your italicized part "of the variables we have been able to study," does not make sense. There are a myriad of variables and data available, and in your example, the authors chose TWO, found a MODEST relationship, with an extremely large number of unexplained data points, and then made sweeping, unsupported conclusions.

The fact that the model has room for improvement doesn't invalidate the statistical relationships that it does explain.

Actually, yes it does. If the model is not properly capturing the relationship, then the incorrect statistical relationships it IS producing are indeed invalidated.

Whether it "exaggerates" the relationship is a subjective claim, one that I'm not sure I agree with and don't know how to address.

No, the way you've presented the stats it's an objective issue that CAN be addressed by evaluating the robustness of the inputs into the statistical model compared to using other, apparently unevaluated independent variables, and the relative strength of the resulting statistical relationships.

What I'm seeing from your summary and excerpts is that the authors have engaged in some extremely weak and unreliable statistical analysis, which they are then using to unjustifiably support their conclusions. I don't have the confidence you do that they are accurately portraying the situation; in my opinion you are taking their work at face value without adequately considering and understanding the statistical issues.

[_Morley]

From what I've read so far, I'm getting the distinct impression that many of this book's critics are often attacking strawmen.

Maybe you'll list these, so they can be addressed.

[_Morley]

Of course not. I've been studying this stuff for less than a week. My main argument on this thread has been about what the book in question actually says. I think I'm qualified to judge the statistical arguments, but I don't have first-hand knowledge of the research it is all based on.

Then perhaps I'm missing something. You seem to arguing not "what the book in question actually says." Instead, I see you suggesting that what the book says is credible. I can go back and find examples of this, if you wish.

You're also ascribing validity to research that's two decades old. Even in that, you're not addressing the consensus of opinion in the field. I can make a case from research that climate change is not occurring. That won't make it the consensus of opinion in that field.

[_Morley]

The fact that the model has room for improvement doesn't invalidate the statistical relationships that it does explain.

Actually, yes it does. If the model is not properly capturing the relationship, then the incorrect statistical relationships it IS producing are indeed invalidated.

[_Morley]

Which is it?

This:

Sam Harris: People don’t want to hear that intelligence is a real thing.

Sam Harris: Some people have more of it than others.

Sam Harris: IQ tests really measure it.

Sam Harris: Differences in IQ matter because they are highly predictive of differential success in life.

Sam Harris: People don’t want to hear that Intelligence is due in large measure to his or her genes.

Sam Harris: There is very little we can do environmentally to increase a person’s intelligence, even in childhood. It’s not that environment doesn’t matter, just that it appears that genes are 50% to 80% of the story.

Sam Harris: Average IQ differs across racial and ethnic groups.

At this point, it is clear to me that Sam Harris was right. The things that Sam Harris said are facts really are facts.

(Emphasis mine.)


or this:

Clearly different people have different opinions and "intelligence" is an incredibly abstract thing that can be thought about in different ways. That's what I think.

(Emphasis mine.)

[Guest]

edit: deleted for now

[_The CCC]

At the risk of over-generalization if we put Einstein in a closet at birth. We'd still end up with a smart guy with plenty of hangups.

[_Analytics]

Who is it that you think is asking for help? That's twice now. Just weird. I'm not asking for help, I am registering some serious reservations regarding your the assessment of the statistical work you are reviewing.

Sorry for offering "help." I really wasn't trying to be patronizing. It just seems that the way you judge the validity of a model and the way statisticians do is quite different. In terms of this kind of model, saying it "fails for some" doesn't make any sense. You equate a model "failing" with a model not explaining everything. Statistics doesn't work that way. The model doesn't claim that there are no outliers. The model succeeds if the residuals are independent of each, normally distributed, and have a constant variance. The magnitude of the variance doesn't matter.

You could look that up in a statistics text book.

Maybe I'm misunderstanding what you are saying. If you think I'm being patronizing in what I'm saying and that you are qualified to judge this in terms of statistics and what the book actually says, we can roll up our sleeves and rigorously perform an Analysis of Variance and engage it that way.

[_Analytics]

Which is it?

This:

Sam Harris: People don’t want to hear that intelligence is a real thing.

Sam Harris: Some people have more of it than others.

Sam Harris: IQ tests really measure it.

Sam Harris: Differences in IQ matter because they are highly predictive of differential success in life.

Sam Harris: People don’t want to hear that Intelligence is due in large measure to his or her genes.

Sam Harris: There is very little we can do environmentally to increase a person’s intelligence, even in childhood. It’s not that environment doesn’t matter, just that it appears that genes are 50% to 80% of the story.

Sam Harris: Average IQ differs across racial and ethnic groups.

At this point, it is clear to me that Sam Harris was right. The things that Sam Harris said are facts really are facts.

(Emphasis mine.)


or this:

Clearly different people have different opinions and "intelligence" is an incredibly abstract thing that can be thought about in different ways. That's what I think.

(Emphasis mine.)

I think it is both. There are things that have been objectively measured, and there are models we construct to explain it.