More Testing Follies and Other News

More 23andMe fails and other news.

As background, read this.  Also read this.

Prepare for an unexpected shock – Sallis is proven right once again.

Over the last year or two, companies such as 23andMe have been updating their customers’ ancestry results; in almost all cases that has been as a direct result of expanding their parental (reference) population sample database with all sorts of non-European samples. They do this (concentrating on the expansion of samples from outside Europe) even though they have grossly insufficient coverage from various parts of Europe (particularly the South and East) and even though most of their customers are of European descent. 

In the months since I posted the above linked criticisms, I’ve been studying online forums in which customers discuss their results, including the more recent updates, as well as looking at statements by the companies themselves, and also material forwarded to me by correspondents.  

The problems accompanying these updates, combined with the pre-existing problems of the tests, essentially completely confirm my previous criticisms and interpretations of these ancestry tests, particularly with respect to the issue of “parental privilege.”

In these updates, in general, the changes in ancestral proportions perfectly mirror the additions of parental population samples that are likely inappropriate for the customers in question (based on their actual, proven genealogical ancestry). Thus, customers who have poor parental population coverage of their actual ancestry exhibit increased ancestral proportions for precisely those (genealogically non-ancestral) parental reference populations that had their numbers increased in companies' databases.

Therefore, and 100% consistent with my past criticisms, the results are completely dependent upon the choice of parental populations, and the degree to which particular parental populations are represented in the databases. More of a certain parental population shifts ancestral proportions precisely in that direction, causing customer results to fluctuate wildly dependent on the parental population choices.

In addition, with these updates, the "unassigned" percentages for the conservative estimate (90% confidence) markedly increased for these same customers (in these cases specifically 23andMe, which provides confidence levels and "unassigned" portions of the genome - other companies do not generally do so), clearly demonstrating that the updated results are less accurate than the preceding. 

As a model of this, look at the first example here. Consider a scenario in which  the testing company refuses to add (more, if they have any) “green” parental population samples, but significantly increases the representation of “yellow” (but not “blue”) samples in their database. What happens? “Green” individuals are suddenly shown to exhibit a much greater percentage of “yellow” ancestry – which is purely a consequence of the shifting representation of different groups in the parental population database.  What if the number of “yellow” was decreased, and “blue” increased?  Then the “greens” would be more “blue.” But, here’s the rub – if significant numbers of “green” were introduced, then the “blue-yellow”” “greens” would – presto! – be represented as mostly “green.”

Again, my criticisms have been 100% confirmed as legitimate by the direct correspondence between expansion of certain parental populations in the databases and the increased ancestral proportions for those same populations among customers who lack proper parental population representation.

An equally valid conformation of my criticisms is that for many of these customers, the updated ancestral proportions have been accompanied by an ever-increasing “unassigned ancestry” percentage when considering results at the (more proper) “conservative” (“90% confidence”- which itself is a bit too low) estimate levels – often increasing to ludicrous levels. If “updates” markedly increase the amount of unknown ancestry at reasonable confidence levels, then this is strong evidence that the updates are providing ancestry estimates that are less accurate than those preceding.  How could it be otherwise? By introducing parental populations that are more distant from customers’ actual ancestral backgrounds, in the context of refusing to increase the appropriate parental population representation for those customers, of course the results will be less accurate, with less of the genome being reliably assigned at higher levels of confidence. The more the parental populations are unrepresentative of the customer, the less likely they will fit the data at the highest confidence level – hence, “unassigned ancestry.”

Anyone getting over 20-25% “unassigned” at the 90% level should view their results with extreme skepticism.  What if it is over 40%?  That is in my opinion essentially useless.  And what about levels exceeding 50% (!) – and some of them (believe it or not) do?  That is in my opinion a tragicomic embarrassment. That’s what one could expect if one tried to represent, say, Russians using some English reference samples and an increasing Japanese reference database. That the company actually releases data with such high “unassigned” levels is shocking.  If person A has an “unassigned” (at 90% confidence) of, say, 5-15% (or less) and someone else has 40-55% (or more) – how can you possibly equate the validity of those two sets of data?  In some cases, the differences are at the level of an order of magnitude.  

Note to testing companies: More references samples from Europe. Many, many more, covering ALL areas.  Most of your customers are of European origins.  You need high level coverage from throughout Europe, all of Europe, before you do your SJW sampling of other areas to satisfy the diversity-mongers.  Get all of Europe covered before you handle those Egyptians, Tibetans, Nepalese, Martians, Neptunians, or whatever. Your customers are your customers, not SJWs screeching about “diversity” in reference populations. You want “diversity?”  First start with Europe.

ALL of your customers should have “unassigned” in the low range at 90% confidence - not just those with “parental privilege.” And even for those latter customers, who are much better off than the others, the results are still suboptimal.  Consider Derbyshire’s data, which is not fully matching his actual ethnic ancestry; however, at least Northwest Europeans fall within the correct sub-region, even if national-ethnic affiliations are not always on target. The swarthoids and slavoids often do not get even that.

For now, 23andMe may be useful for the raw data (that can in theory be used for kinship analysis, which is biopolitically relevant) as well as the health data. The ancestry testing is laughable.  And, by the way, the “timeline” feature is a bad joke, based as it is on the flawed “chromosome painting” and consequent ancestry estimates. Note to company geniuses: Just because you model someone’s ancestry with your limited and inappropriate reference parental samples, does NOT mean their actual ancestry derives from those sources, so that you can “time” when that non-existent ancestry entered their ancestral line (shown to be ludicrously - and objectively mistaken - recently).

Going back to the Russian (23andMe) customer scenario, let's model it differently for the sake of illustration. In one scenario, there are no Russian reference (parental) samples, only Germans and Central Asians.  At 50% confidence, the Russian would likely be represented as mostly German but with a significant Central Asian ancestral component.  At the low level of 50% (!) confidence, some chromosome fragments would seem slightly more Central Asian than German and would be assigned thus - it's only at the coin-flip level of confidence, remember.  At 90% confidence, likely 40-50+% of the chromosome fragments, and hence the ancestry, would be "unassigned" - since at that more reasonable level of confidence, many of the chromosome fragments do not at all match either German or Central Asian. Of the remainder, most would be German, with a small minority of Central Asian. What if the Central Asian reference population was suddenly increased with more samples - increasing the chances that at 50% confidence a match was more likely with some new Central Asian sample than with the original German parental samples?  The Central Asian proportion of the Russian customer's "results" would be increased at 50% confidence, and the "unassigned" would increase at 90% confidence - the latter occurring because these new results are actually less accurate than the preceding. Thus, at 90% confidence, the chromosomal fragments are not matching these new Central Asian samples. What if the parental populations were Sardinian and Central Asian? Likely the Central Asian component would be larger at 50% confidence than with the German and Central Asian parentals, since Russians are more genetically distant from Sardinians than they are to Germans. And here, with Sardinian parentals, the "unassigned" at 90% confidence would be even larger than with the German parentals.

Now, let's do another scenario.  Here, there is a large and very comprehensive Russian parental population - many reference samples from ethnic Russians from all parts of Russia. What happens then? This same Russian customer - the same individual with the same genome - is now represented as being overwhelmingly Russian (and since Russian would be considered "European" by the company labeling, the customer would be so labeled), with only smaller amounts of other ancestries (since the customer may not be an exact fit to the co-ethnic reference samples). Note that the results from the two scenarios would be completely, utterly different. Also, in the latter scenario, at 90% confidence, the "unassigned" percentage would be low, since there would be a good fit between the Russian customer's chromosome fragments and a large and comprehensive Russian reference population.

Consider another scenario.  Imagine if "German" was defined only by samples from North Germany. A Bavarian at 50% confidence might be mostly German but with a strong minority of other ancestries, with a hefty "unassigned" at 90% confidence. If "German" was subsequently redefined to also include many South German/Bavarian samples, then the Bavarian would see his German results greatly increase and his "unassigned" decrease.  

This isn't rocket science or nuclear physics.  When you identify ancestral components by comparison to reference samples, then the composition of those references will of course determine the outcome of the ancestry determination. The accuracy of that determination can be ascertained by how much of the ancestry is "unassigned" at higher levels of confidence.

Can you believe this petty, puerile, and utterly childish attack on Spencer, coming from – surprise! – the obsessives of Counter-Currents.  That’s an embarrassment.  I suppose though it is a useful distraction from the real criticisms of Spencer and of the Alt Right that would hit too close to home to those currently attacking Spencer for talking about hamburgers.  After all, those screaming “Kek” (figuratively and/or literally) three years ago would like to pretend it never happened.

Comments on the comments: 

Ivan White
Posted November 26, 2019 at 11:17 am | Permalink

Spencer might not be a great leader…

Better: “Spencer might not be a leader…”

…but he is certainly a brave man that has risked his money and personal safety for Our People. He deserves some recognition for that.

I agree.  I have never questioned his physical courage. As far as money goes, I don’t know his personal circumstances – indications are that he is from a very wealthy family, but who knows what he has personally.

I do not know the writer of this piece…

All you have to do is click on his name on the side-bar.  Is that so difficult?

…but I get the feeling he has a personal grudge against Spencer.

Welcome to Counter-Currents.

Maybe the writer is a Groyper…

Groper maybe…for all those young lads in “rainbow thongs?”

…which is ironic considering Greg Johnson is one of the few people in the movement that uses the term “White Nationalist”. Isn’t that bad optics?

Ah…he’s a “white advocate” now.

John Wilkinson
Posted November 26, 2019 at 3:45 am | Permalink

I’ll have a Faustian burger with a side of pan-European fried potatoes, please.

No, I’ll have a Traditionalist burger with a side of ethnonationalist fried potatoes, please – in Hungary, even though I’m not Hungarian.  Chow down! (Morgan salivates)

Samuel Nock
Posted November 25, 2019 at 11:43 pm | Permalink

It appears to have been flushed down the Internet memory hole, but a few years back there was a quite funny meme consisting of the below-linked photograph with thought bubbles above each person as follows:
Brimelow thought bubble: image of Ronald Reagan
Derbyshire thought bubble: image of a young Asian woman
Taylor thought bubble: ‘they look Hwhite to me’
Sam Dickson thought bubble: image of a frosted donut
Richard Spencer thought bubble: image picturing himself as Bond in tuxedo with gun

That’s about right…but Brimelow would be more appropriately thinking about depositing money into a Happy Penguins account.

See this.  Emphasis added:

…universities threw open the gates, admitting Jews, women, blacks, other races, cripples, and sexual deviants. Since these people do not belong at university…He goes out of his way to bemoan the fact that as women become professors, the profession becomes devalued. He notes that there’s a general pattern that as women are pushed into jobs previously done by men, the market devalues those same jobs – almost as if the market were correcting for falling standards.

But, but, but…Joan of Arc!

Real professors ascend through a hierarchy and are eligible for tenure, which means job security for life. 

Actually, real tenure is becoming increasingly scarce in academia.  It may exist in some tangible form at some of the top universities, no doubt, as well as at some lower level institutions desperate to attract and retain anyone, but, in general, it’s becoming more and more an empty honorific, and in those cases certainly does not mean “job security for life.”

See this.  But it is not only about adjuncts. Full time faculty either are being not offered tenure track or the “tenure” offered is the equivalent of “job well done, here’s another title for you” but has zero practical meaning.

Once again: Wrong, wrong, they’re always wrong.