If you are going to "debunk" race with gene testing, please stop.

Humans do not have discrete races. Racial groups in humans do not have reproductive boundaries. Genetic variation in humans is clinal. Allele frequencies of most human genes do not vary by much between populations.

I have to put all that up front, because an editorial in the NY Times yesterday seems unaware of any of it (via Dienekes).

This point [i.e. mixed ancestry] came through with resounding clarity recently at Pennsylvania State University, where about 90 students took complex genetic screening tests that compared their samples with those of four regional groups. Many of these students thought of themselves as "100 percent" white or black or something else, but only a tiny fraction of them, as it turned out, actually fell into that category. Most learned instead that they shared genetic markers with people of different skin colors.

Really? Who knew? People share genetic markers with others who live on different continents?

This part is the worst:

One "white" student learned that 14 percent of his DNA came from Africa - and 6 percent from East Asia. The student told The Daily Collegian, the student newspaper, earlier this year: "When I got my results I was like, there's no way they were mine. I thought it was just an example of what the test was supposed to look like. Then I was like, Oh my God, that's me."

So it's all about making people find out they aren't as "white" as they thought they were? I guess so; the column goes on to describe Samuel Richards' aims: "showing students they aren't what they think they are", causing them to question "rigid and received notions about the biological basis of identity", and showing that "race and ethnicity are more fluid and complex than most of us think".

My question: do the DNA tests actually do any of this? My answer: no.

The class

The Pop Culture blog has posted an archived version of an earlier Times article about the class, with facts and quotes.

Samuel M. Richards, who teaches Sociology 119, Race and Ethnic Relations, to 500 students each semester, said the DNA tests, which were conducted last year for the first time, were very popular with the class. "Everyone wants to take the test, even students who think they are 100 percent one race or another, and almost every one of them wants to discover something, that they're 1 percent Asian or something. It's a badge in this multicultural world," he said.
About half of the 100 students tested this semester were white, he said, "And every one of them said, 'Oh man, I hope I'm part black,' because it would upset their parents. "That's this generation," he said. "People want to identify with this pop multiracial culture. They don't want to live next to it, but they want to be part of it. It's cool."

Every one? Personally, I think if you are going to tell students that the results of some genetic test challenge their identity, you should be providing counseling.

Another article on the class testing from American Renaissance News says Richards wants to test 1,000 students next year. But this article reveals a slight problem with the tests themselves:

The case also illustrates the limits of DNA testing, says Mark Shriver, a professor of anthropology at Penn State and a consultant for DNAPrint. Native Americans are believed to have immigrated from central Asia thousands of years ago. These same central Asians also migrated into eastern Europe, meaning that her "native American" DNA could have come from there, he says. Greeks and Ashkenazi Jews also may show significant percentages of "native American" ancestry for the same reason. Eventually, a more sophisticated test will be able to sort out these differences, Dr. Shriver adds.

So in other words, the test doesn't even show what it purports to show! No wonder "20 percent were 'very surprised' to find out they had a mixed heritage"!

The test comes from DNAPrint Genomics. You, too, can pay for this service, called AncestrybyDNA, for only $219.00. Richards' students got the test for free, under the condition that their results go into Shriver's database.

The test

If you are like me, your idea of "accurate" is very different from that employed in these genetic tests. A determination of your ancestry cannot actually be made from genes alone: it always requires additional information about your origin, and still may be off by more than 10 percent. Ten percent one way or the other may not sound like much between forty and fifty, but the difference between 10 percent and zero is a lot. Depending on your purposes, this kind of accuracy may be of interest, or it may be completely meaningless. Applied to a class of several hundred students who may or may not know anything about their ancestry, a high percentage of students will have results that misinform rather than enlighten.

Usually, reputable companies post exactly what they are assaying, and what statistical confidence the results come with. DNAPrint is no exception, and its website is very informative.

The reported accuracy of the test is rather better than 10 percent. But these figures are based on computer simulations that use false assumptions about human populations. Feel free to read through some of the reasons AncestrybyDNA is supposed to work. Notice how every example assumes that the four race groups (European, African, Asian, and Native American) are real? Notice how all the estimates of standard errors assume that each of the races is a single gene pool with no spatial heterogeneity? All these assumptions make it easier to classify; all violations of these assumptions make the results less accurate.

And the errors are not only potential errors -- they are plain in the data that have already been collected. These errors are too large to go unnoticed, and some of them are characterized by the company. Consider this quote from the Ethnicities page:

For example, a European with 5% Native American affiliation can conclude from our tables in Accuracy and Precision that the result indicates Native American affiliation with less than 95% certainty (it would be about 80% or so).
However that doesn't mean that the 5% reading is not useful, and in fact, if you were Greek you would expect a result in this ballpark and most of your Greek friends would show similar results, such that the score for the average Greek was about 5% Native American. That is to say, the 5% Native American affiliation type, on a European background, is systematically obtained for Greeks, meaning it is seen for most individuals of this group. Other affiliation types on other backgrounds are characteristic signatures for other ethnic groups. This section of the website is designed to give you a feel for these signatures.

Consider that 6 percent admixture is close to one ancestor in the great-great grandparents' generation. What in most contexts would look like the boundary of statistical significance, in the context of genealogy is highly significant. If you are wanting a test to find out whether an ancestor four generations ago was Native American, these genes are of little use: their variation within continents is too great to accurately indicate relationships on that level. Indeed, the site says:

the limit of detection for our test is such that a person could have a great grandparent that is 50% Native American and still obtain a 0% Native American score. Therefore a negative result with our test does not prove that there is no Native American or Amerind genetic heritage - it only suggests that if it exists, it is likely to be relatively low.

The discussion of different ethnicities on this site appears to espouse the assumption of pure races completely. How can it be that Mediterranean populations show alleles common in Native Americans? According to the site, these alleles may be traced to a maternal Native American contribution to these Mediterranean populations (!).

Please understand if you are reading this, that what actually explains the results is not the movement of Native Americans into southern Europe. Instead, some alleles that are common in New World populations are more common in southern Europe and less common in northern Europe. When "Europe" is taken as a single pool, these southern European alleles cause those groups to look more like Native Americans than do other Europeans. Do these allele frequencies actually reflect a common ancestry -- for example, did some ancient Siberians move to Beringia and send their cousins to Greece? Who knows? At the time scale that is relevant here, nobody has any idea why some alleles are more common in some places than others. Could be ancient relations, could be something else. But none of this is about recent mixing between "Europeans" and "Native Americans" -- the kind of answer that someone looking for their own genealogy would want.

I've quoted from the Native American admixture examples, because these are discussed in the most detail. But looking at the numbers, some other populations are a complete mess. The results for South Asian populations appear incomprehensible to me. How does it have any meaning to assign Indian people to some proportion of European (which sometimes appears intended to equal "Caucasian") ancestry and some proportion of East Asian ancestry? Other populations are not considered at all. What if your ancestors come from East Africa? Madagascar? Hawaii?

The bottom line about the testing: most populations give results that are potentially misleading. And worse, all the analyses are based on a false assumption: that people are a mixture of one or more "pure" races.

Searching for the pure races

To me, that is the worst of all of this. An instructor is "challenging rigid notions of race" by assuming that every person is some mixture of races that were once pure. A test is telling people about their ancestry by giving them a proportion of admixture from different races.

For some people, these are the answers they want to hear. If you were going to have a test that told you what race you are, what kind of answer would you expect?

On the other hand, we already have a test to tell us what race we are. It's called society. And society doesn't care what our genes say.

It is certainly relevant to the concept of race that genetic ancestry and social definitions do not match. That is what the testing of the class is purporting to demonstrate.

Instead, the test is telling students (and readers of the NY Times) that science can determine what their racial ancestry is. And science is giving them a result. The lesson the test teaches is clear; if it wasn't clear, then it wouldn't be telling Indian students that they are part European and part East Asian. Or Greek students that they are part Native American.

It's worse than meaningless. It means the wrong thing entirely.