He's saying that if you consider the race results as a control of sorts, and you operate under the assumption that there is nothing inherent in race that would create the results shown, it is obvious that the variable being measured is not well controlled, and there is plenty of noise in the results.
> The reason i'm speaking in probabilities is because that is what the facts are, that we have more variance within races than between.
thank you, exactly.
> africans are genetically predisposed to be taller.
"african" is not a category that has any meaningful genetic definition.
> genes play a predictive role in attributes like height and IQ through twin studies, but we have already begun to find genes that can predict physical attributes like height.
none of which means that "race" has any meaningful existence.
>Category theory is made up, does that make it not real? Something that has no value whatsoever?
it has value in organizing information. another system of organizing information might be harmful. suppose you had a system of organizing information that weighted irrelevant data and disregarded relevant data. would it have positive value or negative value?
> you seem to believe race is subjective
race is intersubjective, as is language and culture.
> to the point it has 0 predictive value
what are you trying to predict? either the trait that you select for when you delineate a sub-population is connected to the variable you predict, in that case you define the category by the specific trait, not some giant bag the size of a continent; or the trait you select for when you delineate the sub-population IS NOT connected to the variable you predict, so you're spouting nonsense. there is no third option.
> we can say that what people see as race almost perfectly corresponds to what a computer would sort them into based on purely genetics.
the computer is just doing what the programmer told it to do based on a bunch of genetic categories that are artificially constructed. you get a guy's genes. what makes you say that guy is asian? oh thats right, a hand wavy mix of phenotypes and culture. thats entirely objective, dont know what I was thinking.
> We can say that a self proclaimed hippie is more likely to smoke weed than a self proclaimed Muslim or Christian. Do you believe we cannot do this?
you're going to classify him as a hippie because he smokes weed and predict that he smokes weed based on the fact that he is a hippie. and you dont see the circularity?
> Race isn't some wavy concept where you have asians thinking of themselves as africans
actually it is much more fluid than you seem to think. parts of africa and asia are geographically close, and have mingled for millennia.
> Sure you can point to exceptions but they are just that, not even 1% of the people in these studies when using more than just a dozen SNPs
thats more an artifact of the limitations of your studies, if you know anything at all about this subject you should know just how genetically diverse the human species is.
> there is no evidence for the existence of a race that does not rely on referencing the idea of a race. the idea is self-referential. to understand this, you must understand philosophy.
Huh? Race is defined by a combination of genetics in the form of phenotypes and culture. I don't see the circular paradox here.
> another way of saying this is that assigning a person to a race doesn't give us any more information than the information we used to classify him in the first place.
But we can? We can say that Chinese are more likely to be shorter than africans. That asians are more likely to have black hair than blonde hair or that africans are more likely to have higher testosterone than european whites or indians.
Ask any doctor or pharmacist and they can list you a few ways they are taught about how to treat people of different races differently because of biological differences.
We can even go further and say for instance that ashkenazi jews have a higher average IQ compared to Africans due to a large part their genetic differences and not purely environmental reasons.
The bell curve which was published a few decades ago proves many of these kind of claims so I don't know where you get this idea that we cannot determine anything useful from race.
> ok, I'll bite: what defines a race? what test do I apply to a specimen in order to sort him/her into a race category?
I gave a definition previously which is pretty accepted, for how you test for a race the answer is that's pretty easy, 99% of the time you can just ask someone and they will tell you.
We can prove this through cluster analysis, where you take a sample set of people and first ask them what race they are from a set number of selections. They then can read SNPs from each of the participants and hand that to a computer without the self reported data the people provided.
What you see is that even with 100 random SNP's you have a near perfect correlation between what people say their race is and the computer correctly grouping them into the race they chose.
> Race, or some hidden variable correlated to it, is highly predictive even after accounting for income and similar things.
Since you're convinced of this position, please provide a biological definition of each "race" you think shows this effect, and we'll see if the data actually agrees with your claims.
I'm trying to counter the claim that race doesn't predict anything. More than nothing is something. How much? I haven't quantified it but neither has the OP.
>I should probably also mention that one interesting and politically relevant result of these population studies is that the concept of race doesn't really have a solid statistical base, because we have found the variation inside the group to be so large compared to the variation across groups, that there is no unambiguous way to cluster the human population.
If you concentrate on a large pool of genes, yes. But then you might come to the same findings when studying dogs or cats and conclude that breeds don't have a solid statistical base.
With statistics everything is provable, given carefully selection of samples and traits.
The danger here is of assuming the conclusion, in this case, that race doesn't correlate with anything meaningful. This is still a hotly disputed point.
Ultimately, transparency and reproducibility are the key criteria I think, like it is in science.
It just seems lazy, using race as a proxy variable without positing a hypothesis for what it means. It doesn't tell us, as readers, anything about why some kids didn't perform as well as others.
> What specifically is the definition of race, and how does that definition help with mapping population movement?
Quoting from OED, 2nd Ed., a race is "a group of persons, animals, or plants, connected by common descent or origin".
> they exist on a spectrum.
Possibly, but that doesn't tell you anything useful about whether there are useful categories or not. It depends significantly on the shape of the distribution whether useful categories arise in the first place, how many and so on. Fixating on the connectivity of the kernel would be ditching an awful lot of information. That this connectivity exists is a trivial consequence of the fact that we're not talking about different species.
> For example, check out this plot of genetic differences in Europe. The differences smoothly smear out over the geography.
Yes, and that tells you a lot about the population of Europe. So clearly this is useful.
> If races look different, then the possibility of other metrics being different is a logical induction.
This doesn't follow. It becomes more clear why if you put it in more abstract terms. Take a bunch of balls, half red, half blue. Now assume some half of these balls are heavy (say 10lbs) and some half are light (say 2 lbs). Assume the balls have a bunch of other characteristics this way. Now what you're saying is that if we know a ball is e.g. Red, we can conclude something about the probability that it's e.g. heavy. Clearly that's not the case, why would it be? Unless you just assume a priori that the characteristics are distributed in a way that's correlated with ball color, there's no reason we should think there's an association between color and e.g. weight.
Wading into a minefield here, but is this even remotely true?
My understanding of DNA testing is that it can only tell you what genes a person has. I wasn't aware of any reliable way to cluster genes into racial groups. (The 23 & me stuff is all probabilities, wide percentage ranges, etc.)
Further, I thought it was the case that science has basically abandoned the concept of race specifically because it couldn't be reliably measured, even by DNA analysis. Is this not the case?
What am I missing?
"Race" as in "the color of the skin allows statistically significant predictions about physical and mental aptitude"? No, that's a stupid categorization, because it's far too wide and imprecise.
However, I do believe there are (small) genetic differences between subpopulations from different geographic origins. For example, it's hard to deny that Kenyans have an unmatched edge in long distance running, one which can't be explained using non-genetic factors.
>There really shouldn't be a debate that there are genetic differences amongst 'races'. It shouldn't be something that is "taboo" to discuss or talk about, since it is a factually correct statement to make.
I disagree. Who defines what a "race" is?
Divide a population into two groups arbitrarily, and you'll find statistical differences.
Pick any metric, and you'll get clusters.
There are so many data scientists here on HN - so I hope many do understand how sensitive clustering is to the metric, algorithm, and other parameters!
And for everyone else: race is like colors in a rainbow. Yes, the rainbow is not monochromatic. And yet there is no division in the continuous spectrum in nature.
> races are a social construct, not a biological reality
I do agree with you, but then what should anyone care about the distribution of ethnic groups in a medical trial if those ethnic groups have absolutely no impact on the result of the trial.
Either it makes sense to differentiate the vaccine according to $variable and testing groups as well as certifications need to reflect that. Or it makes no biological sense to care about $variable, then we should have no requirements about making sure that $variable is correctly sampled in the test groups.
In this case, ethnicity probably has 0 impact on the effectiveness of the vaccine, therefore why does anyone care about the distribution of ethnic groups in the sample ?
> We can say that Chinese are more likely to be shorter than africans.
whats a chinese? whats an african? and what is it about those categories that makes it necessary for you to speak in probabilities and not certainties? if those categories accurately described features of reality, and were not mere artifacts that stem from your worldview, you would be able to rigorously tell me a relationship between gene(s) and features.
> Race is defined by a combination of genetics in the form of phenotypes and culture.
> for how you test for a race the answer is that's pretty easy, 99% of the time you can just ask someone and they will tell you.
exactly, its a made up category, i.e. not a real thing.
> We can prove this through cluster analysis, where you take a sample set of people and first ask them what race they are from a set number of selections. They then can read SNPs from each of the participants and hand that to a computer without the self reported data the people provided.
you can do the same thing for things like "hippie", "biker", "nerd" etc. it doesn't make those categories leave the intersubjective realm.
>"The trouble with "race" is that it's pretty meaningless biologically;"
One of the things that shocked me about taking my first Bioinformatics class was how casually the professor talked about race. It seems like the concept of race is well accepted in the field, at least as it denotes large haplogroups of the human species. If you gave me the mitochondrial DNA from an Indian, then I am pretty sure that I could tell you it came from an Indian.
Also, I think well-designed statistical studies would easily meet all your objections.
As to whether or not such studies would be useful - I think they would. A common public policy problem is the disparate performance of students from different racial groups in school. As long as public policy sees racial groups as meaningful, it is probably useful to study them.
reply