Yes, I was thinking along the same lines. When you see a lot of large families, in a population which is mildly heterogeneous for skin color (as I do), you see a lot of families in which there is significant variation in skin color among siblings. Having two light-skinned children out of five when the parents are both darker is not absurdly improbable.

this is especially true when parents are from recently admixed populations, as the likelihood for homozygosity at any locus is probably pretty low since the populations haven't had a long time to fix at any given locus.

There are lots of situations in the world where there is more variation within many nuclear families than within entire racial groups. E.g., many nuclear families in Brazil have more variation in skin color than all the Bantus or all the Nordics put together.

The genetic regulation of skin colour is pretty complex. First of all, there are two main pigments (carotene and melanin) and subtypes of each (for example eumelanin (black/brown) and pheomelanin (red/yellow)). Secondly, the synthesis of eumelanin itself (only one pigment of the total visible colour) is complex, and is affected by so many proteins: a mutation in any of these could affect skin colour.
A lot of proteins involved in this are very general, and active in multiple pathways, but they still affect skin pigmentation too.
I guess what I'm trying to say is: not only are there more than 6 loci, changes in one part of the pathway (through, for example, an inherited mutation) can also be circumvented in different ways.

As the genetics of skin colour are not fully understood, the probability that two 'black' (African-American or Afro-Caribbean) parents will have a child lighter than either of themselves can at present only be determined by observation. False paternity would of course have to be excluded. I can only go by the casual observation that the children are *usually* intermediate between their parents, though sometimes much closer to one than the other.

In the Cosby case, I wouldn't say that the colour distribution of the offspring is 'absurdly improbable', just rather unlikely - less than, say, 1 in 20. One point to note is that Bill Cosby is very dark, so it is fair to assume that he has nearly a full-house of 'dark' skin genes. *All* of his offspring would therefore get around half their genes 'dark', even if they happened to get only 'light' genes from their mother.

Here is a discussion of eye color genetics.

Rereading my comments, I think I have been implictly assuming that the parents themselves are significantly different in skin colour. In this case I would expect the children *usually* to be intermediate. But I would acept that if the parents are quite similar in colour - say, both mid-brown - then their ofspring will often be somewhat lighter or darker, due to random recombination.

But I would acept that if the parents are quite similar in colour - say, both mid-brown - then their ofspring will often be somewhat lighter or darker, due to random recombination.

This is true even if parents are quite different, as long as they are not at extreme opposite ends of the spectrum.

OK, so long as the number of genes involved is not very large. If it is, the variance in the genetic contribution of each parent separately may be small compared with the difference between the parents, in which case the offspring would nearly always be intermediate. Suppose, to be fanciful, there are 1000 loci involved, and one parent has 'black' and 'white' genes in the ratio 70:30, while the other parent has them in the ratio 30:70. I haven't worked out the sd's, but I guess that the offspring would nearly always have them in an intermediate ratio, e.g. 60:40. Presumably the number of genes involved in skin colour is much less than this, though the overall number of genes involved in racial differences may not be.

david,

but you are assuming additive independence. some loci might have strong interactional/regulatory effects globally....

razib: yes, I was assuming additive independence. But, off-the-cuff, I don't see that introducing dominance or epistasis would produce offspring more extreme than either of the parents.