Wednesday 13 December 2017

Common error - 'regression to the mean' of intelligence

People say that the offspring of high IQ individuals will regress to the mean (average) of their 'population' - and they calculate this as if it were a mathematical law...

But intelligence is a measure of a biological variable, and 'regression towards the mean' happens, if-and-when it does, for biological reasons - it is not a mathematical law.

When a high IQ individual is a descendant of high IQ parents, grandparents etc - there is no regression to the mean.

(Except for the trivial reason that test-takers who score highly because they 'have a good day' will re-test at lower scores. This can be somewhat dealt with by having several measurements of IQ - although this also increases the chance of 'having a bad day' maybe from non-random illness, and falsely dragging down the average. In practice there is no substitute for high quality data and increased numbers/ averageing does not help. This means excluding from the data any people who are suffering from acute, test-score suppressing illness or any other systematic cause for false measurement. In biology; smaller higher quality studies are always better than larger, poorer quality studies.)

In other words, to the extent that a high IQ individual comes from a genetically-relatively-intelligence-inbreeding caste or class; there is no regression to the mean.

And, in fact this is a very common situation - at least to the extent that regression to the mean is insignificant in amongst other factors. 

The point to hold in mind is that no variation/ distribution is really random; randomness is just an assumption, a model, which may be expedient for specific purposes - but is not a general truth; indeed randomness is usually a false model when it comes to biology.

In sum, human behaviour and ability cannot be explained by mathematical rules - at most such rules summarise a specific data set - which must then be evaluated in terms of scientific quality. We cannot explain unless or until we know something of causes.  

https://charltonteaching.blogspot.co.uk/2010/10/scope-and-nature-of-epidemiology.html
https://charltonteaching.blogspot.co.uk/2008/05/social-class-iq-differences-and.html 


11 comments:

Wm Jas Tychonievich said...

Ti the extent that a high IQ individual comes from a genetically-relatively-intelligence-inbreeding caste or class, regression is to the mean of that high-IQ caste, not to the mean of the general population.

Bruce Charlton said...

@Wm - Strictly, the caste or class is irrelevant except insofar as they are actual ancestors. My point is that there are biological, not statistical, reasons for intelligence. The (sometimes) statistical regularities presumably have causes unknown or only partly known - the question is whether these causes are relevant in a specific situation under consideration.

Wm Jas Tychonievich said...

Of course all statistical phenomena have physical causes, and statistical regularities are not a useful tool for understanding specific individual situations.

Wm Jas Tychonievich said...

There's nothing special about ancestors. My mother or father's genes correlate no more closely with my own than do those of my brother or sister -- and correlation, not causation, is what is statistically relevant.

Bruce Charlton said...

@Wm - The qualitative distinction between statistics and biology is exactly what I am emphasising here. The one is a summary - the other a science.

Anonymous said...

Do you have any speculations about the underlying biological reasons why regression to the mean occurs?

Wm Jas Tychonievich said...

So are you challenging the view that regression to the mean is a valid statistical generalization when it comes to the heredity of intelligence (and height and other such things)? Or are you accepting regression but stressing that a statistical generalization is not a cause and that all causes are physical rather than mathematical?

Regression to the mean certainly does occur, as a statistical rule, but it is important to understand its limits -- particularly, the fact that it's a one-time thing. We would expect regression in a child born to high-IQ parents -- but not in one whose grandparents, aunts, and uncles were all high-IQ as well.

I suppose the biological cause of regression to the mean intelligence is something like this: (1) high intelligence is the result of highly complex combinations of genes and (2) many quite different such combinations can yield high intelligence. As a rough analogy, if you take two very strong poker hands, combine them, shuffle the 10 cards, and deal them out again, the resulting hands will likely be somewhat stronger than average but not as strong as the original two hands.

Chiu ChunLing said...

I think it is more appropriate to say that the mean to which the children of unusually intelligent parents regress is the mean of their entire ancestry, not the mean of the population at large. In other words, I should expect my children to be about average for my family history.

Which is true whether the mean of their ancestry is above or below that of the general population. In other words, children of parents who are unusually intelligent themselves, but come from an ancestry which was below the global mean, will tend to regress back towards that lower mean, just as children of intelligent parents who have an ancestry with a mean higher than the general population will only regress back to that higher mean.

It is important to note that this is not invariably the case with unusually low intelligence, which can be the result of significant (and inheritable) genetic damage in the last few generations. As uncouth as it is to say it, high intelligence requires relatively undamaged genetics, with few mutations. The purely Darwinistic idea of slight improvements occurring through random mutation and then being amplified by natural selection is false. The genetics that produce something as refined and complex as a highly functional intellect cannot be random, the changes are not 'slight improvements', they are invariably harmful.

This damage can occur to children of parents with superlative intellects, and it largely erases the expectation that the grandchildren will have high function. Such damage occurs fairly frequently, which is why I'm not fond of appealing to distinguished distant ancestors. If it's been more than three generations since any of your ancestors were geniuses, then probably one of the intervening generations had some damage that basically eliminated that potential in their descendants.

Bruce Charlton said...

My point - which I think the discussion tends towards confirming - is that regression to the mean is a kind of fake knowedge, an approximate/ overall statsitcal regularity seen in some mixed populations... but by regarding it as a kind of causal-theory or universal phenomenon - it makes people think they know something when they don't.

(Note: As in Life - fake knowledge is only dangerous when it contains some true aspects - as does the regression to the mean idea. Nobody would take any notice of the idea of it didn't contain something correct. I am concerned by the *overall* tendency of the idea; which is malign.)

So regression to the mena doesn't have A cause; but it could be caused By the intelligence in persons being a product of large numbers of genes of small and similar-magnitude effect - which seems a reasonable approximation.

But intelligence doesn't have A cause, neither does height - and there may be single genes with larger effects. Indeed, the conceptualisation of general intelligence is poor, indeed, I would say that intelligence is much Less understood at present than it was 50 years ago, because of the decline in quality (and, especially, *honesty*) of scientists and of society - and the historical fact that IQ research has been dominated by educationalists and statisticians - not biologists.

(Plus, modern 'biologists' are narrowly educated genetic technicians, almost wholly.)

In such a situation, fake knowledge is especially harmful.

Paul Bonneau said...

"When a high IQ individual is a descendant of high IQ parents, grandparents etc - there is no regression to the mean."

I have to question this statement.

Case A: IQ 140 parents have a son with an IQ of 130.

Case B: IQ 140 parents have a son with an IQ of 150.

It doesn't make sense to say (if that's what you are saying), that regression to the mean happened in Case A but it didn't in Case B. Regression is not measured by individual cases. Instead, what it means (I think - not quite sure) is that if those parents had 100 kids, the average IQ of those kids would most likely be somewhat less than 140. There is always regression to the mean, no matter what a particular child's IQ happens to be, because it is a statistical statement depending on a population, not on an individual case.

Bruce Charlton said...

@PB - My point is that there aree specific causes to do with the inheritance of specifci genes why "Case A: IQ 140 parents have a son with an IQ of 130. Case B: IQ 140 parents have a son with an IQ of 150." - there is not a general (statistical) cause which explains it.

Statistical statements about populations are not science, they are not 'theories'. They are, instead, measurements. And these measurements are IQ tests - not actual intelligence. Therefore "There is always regression to the mean," is Not true. There could be populations where, for various reasons, regression does not happen.

For example, for much of the 20th century, children's raw score IQs inflated compared with their parents (i.e. Flynn Effect). Or, by another measure - simple reaction times, there was a decline in the underlying 'genetic' intelligence (g) with every generation - that affected below average IQs as well as above-average.

Furthermore, there are different reasons - different causes - that *might* explain why low scores increased and high scores decreased. For example, some low scorers are ill and underperforming in a specific testing session - their IQ will be underestimated. Some high scorers will just have made lucky guesses - their IQ will have been overestimated. We would expect to see 'regression to the mean'.

But some high scorers will be underestimated - because they were ill on the day, or have motor or percetual problems. They would not show rttm - but their offspring would have even-higher IQs than the parents.

Therefore whether there is rttm or the opposite would depend on whether the specifci population being tested on that day contained (for example) a lot of ill people (maybe there was a flu epidemic?) or few/ no such people.

My point is that rttm is a kind of pseudo-understanding, a pseudo science, based on no more than some observed statistical regularities - it doesn't *explain* anything.