The Unz Review - Mobile
A Collection of Interesting, Important, and Controversial Perspectives Largely Excluded from the American Mainstream Media
 BlogviewJames Thompson Archive
A Piffer Pause?
🔊 Listen RSS
Email This Page to Someone

 Remember My Information



=>

Bookmark Toggle AllToCAdd to LibraryRemove from Library • BShow CommentNext New CommentNext New Reply
Search Text Case Sensitive  Exact Words  Include Comments
List of Bookmarks

publications in library

We are facing a French dilemma: Piffer has an approach to the genetics of racial differences in intelligence which seems to work in practice, but should not work in theory. His technique appears to run against the general trend of genetic research, in that he appears to be getting good results predicting group differences in intelligence on the basis of just 18 SNPs, while genetics researchers are getting only reasonable results in predicting individual intelligence on the basis of lots of SNPs.

For example, people skilled in these matters tell me that they did an out of sample prediction in an independent but European population, and they got 4.8% of the variance, using all SNPs. That is the upper limit of prediction in a non-European population using all SNPs. Pfiffer used just 18 SNPs in non-European populations and his correlation is huge, which does not make sense.

In any case, I think that most genetics researchers will probably wait till Piffer publishes his updated paper before commenting. Of course, there is a difference between predicting IQ within populations using GWAS and applying GWAS results to between-population differences, which is what Piffer does.

He explains:

These SNPs that explain variance within populations are markers of polygenic selection. They do not have to explain a lot of variance between populations, or even within populations. The polygenic evolution model predicts that a few SNPs will have frequencies correlated to frequencies of countless other SNPs. I just need to know the few most important SNPs to gather a signal and infer to the distribution of the other unknown SNPs.

If selection pressure acted on these 9 SNPs by driving their frequencies up in population A compared to B, then it has also done the same to other SNPs. We don’t need to know what these other SNPs are because theory predicts that they will have similar distribution.

So, the official response to Piffer is: “publish, and then we will give you our comments in reply.” This will take time, but it is the traditional way of doing things.

The unofficial response is to encourage more criticism right now, because if the finding is the result of a simple error, it should be exposed and corrected as soon as possible.

It is open season on Piffer’s methods. Recruit critics and get them conduct their peer reviews right now.

 
Hide 9 CommentsLeave a Comment
9 Comments to "A Piffer Pause?"
Commenters to Ignore...to FollowEndorsed Only
    []
  1. Why shouldn’t it work in theory? Not a rhetorical question. I’m not taking a stand on the method (haven’t looked at it closely enough), but I don’t see the theoretical contradiction between a tiny r2 for predicting individual IQ from 8 SNPs and a large r2 for predicting group orderings, given Piffer’s hypothesis.

    Read More
    • Replies: @utu
    I don’t see the theoretical contradiction

    Correct. But genome is not theoretical. It is what it is. SNPs have particular distributions and structures within populations and human genome in general.

    If ps (polygenic scores of 9 SNPs) have large variance within population then Piffer formula must also work (to some extent) within populations to be valid on population averages. If it doesn't work within population, Piffer's result is invalid.

    If however ps has very narrow variance within populations then Piffer's formula will not work within population but this fact will not preclude a possibility of it being correct on population averages. So, what is variance Var(sp) for various populations? This should be estimated from available databases.

    While I do not know for sure I do not think that Var(ps) is small. So if Piffer's formula does not work within population then its exceptionally good fit (r=0.91) to population averages must be spurious.

    Note 1: Theoretically Var(ps) can assume wide range of values. For frequencies from Table 8 one can estimate min and max of variance. In terms of iq the approx range is (50,300). Keep in mind that Var(iq)=225. The lowest would be realized in the most "egalitarian" population where everybody has a polygenic score equal to its population average (±0.5 SNP). The highest when SNP's are distributed randomly independently of each other.

    Note 2: The Piffer's formula =24.5+131.1* exactly predicted IQ of Craig Venter but overestimated the IQ of James Watson by 21 pnts. Both Venter and Watson allegedly have 8 out of 9 SNPs.
    ReplyAgree/Disagree/Etc.
    AgreeDisagreeLOLTroll
    These buttons register your public Agreement, Disagreement, Troll, or LOL with the selected comment. They are ONLY available to recent, frequent commenters who have saved their Name+Email using the 'Remember My Information' checkbox, and may also ONLY be used once per hour.
    Ignore Commenter Follow Commenter
    Sharing Comment via Twitter
    /jthompson/a-piffer-pause/#comment-1902225
    More... This Commenter This Thread Hide Thread Display All Comments
  2. I agree with you. However, I haven’t looked at the method enough to be sure that I haven’t missed some artefact, or that the Monte Carlo method does not fully control for chance findings. To me, Piffer has taken a very intriguing first step with a method which can be tested again and again as each new SNP for intelligence is found.
    However, the signals I am getting from genetic researchers is that it is so against the general trend of the literature that they rate the finding as unlikely. The results are so good that it raises the fear that they are too good to be true. Not that Piffer has invented them in any way (he has set out his workings in full) but that there is problem with the general method and assumptions.
    I know I won’t be able to take this further, and also admit I am not going to take time out to learn R. I was used to SPSS and then Statistica, and only had very passing knowledge of genetic stats programs, only by talking to researchers in our Psychiatry dept who did the early work on the genetics of schizophrenia, and nothing on the much more powerful later stuff.
    Ideally, more geneticists with all these skills at their fingertips would pile in right now, and knock holes in the technique. However, I think we will have to sit out the usual publication delays, and hope they take it up again later.

    Read More
    • Replies: @res

    that it is so against the general trend of the literature that they rate the finding as unlikely.
     
    Under normal circumstances I would tend to share this view. However, given the strength of the zeitgeist (aka Narrative) against Piffer's results I don't think much of this as a reason in these circumstances.

    Speaking of which, more on the recent Vox/Charles Murray controversy:
    http://quillette.com/2017/06/11/no-voice-vox-sense-nonsense-discussing-iq-race/
    And from one of the authors (Paige Harden) of the Vox article: http://www.geneticshumanagency.org/ff/the-science-and-ethics-of-group-differences-in-intelligence-part-1/
    I am getting the impression that the people arguing for the Vox article seem to be forgetting that the article asserted zero genetic effect and are instead arguing for non-zero environmental effect. That seems to qualify for a number of fallacies (e.g. Motte and Bailey).

    It is also a nice touch that the Vox headline screams "junk science" but much of the argument revolves around ethics.

    I am not going to take time out to learn R.
     
    That's understandable. If you ever change your mind, I thought this was a good starter course: https://www.coursera.org/learn/r-programming
    R in combination with RStudio provides a powerful (and free!) working environment. If you work with others who are fluent with R it is possible to run and extend their data and analysis files (if supplied) fairly easily. I was able to replicate Piffer's RPubs work with the data files he provided. Recreating the RMD (analysis, stands for "R Markdown") file was the hardest (more time consuming than hard, it might help replication efforts if Davide is willing to make his own analysis file available, GitHub provides nice facilities for sharing entire R projects) part, but once that is done rerunning it and creating HTML output (as on RPubs) is trivial as is actually sharing it on RPubs. There are a number of other nice features, such as the creation of PDF and Microsoft Word documents for sharing tables, figures, etc.
    ReplyAgree/Disagree/Etc. More... This Commenter This Thread Hide Thread Display All Comments
  3. res says:
    @James Thompson
    I agree with you. However, I haven't looked at the method enough to be sure that I haven't missed some artefact, or that the Monte Carlo method does not fully control for chance findings. To me, Piffer has taken a very intriguing first step with a method which can be tested again and again as each new SNP for intelligence is found.
    However, the signals I am getting from genetic researchers is that it is so against the general trend of the literature that they rate the finding as unlikely. The results are so good that it raises the fear that they are too good to be true. Not that Piffer has invented them in any way (he has set out his workings in full) but that there is problem with the general method and assumptions.
    I know I won't be able to take this further, and also admit I am not going to take time out to learn R. I was used to SPSS and then Statistica, and only had very passing knowledge of genetic stats programs, only by talking to researchers in our Psychiatry dept who did the early work on the genetics of schizophrenia, and nothing on the much more powerful later stuff.
    Ideally, more geneticists with all these skills at their fingertips would pile in right now, and knock holes in the technique. However, I think we will have to sit out the usual publication delays, and hope they take it up again later.

    that it is so against the general trend of the literature that they rate the finding as unlikely.

    Under normal circumstances I would tend to share this view. However, given the strength of the zeitgeist (aka Narrative) against Piffer’s results I don’t think much of this as a reason in these circumstances.

    Speaking of which, more on the recent Vox/Charles Murray controversy:

    http://quillette.com/2017/06/11/no-voice-vox-sense-nonsense-discussing-iq-race/

    And from one of the authors (Paige Harden) of the Vox article: http://www.geneticshumanagency.org/ff/the-science-and-ethics-of-group-differences-in-intelligence-part-1/
    I am getting the impression that the people arguing for the Vox article seem to be forgetting that the article asserted zero genetic effect and are instead arguing for non-zero environmental effect. That seems to qualify for a number of fallacies (e.g. Motte and Bailey).

    It is also a nice touch that the Vox headline screams “junk science” but much of the argument revolves around ethics.

    I am not going to take time out to learn R.

    That’s understandable. If you ever change your mind, I thought this was a good starter course: https://www.coursera.org/learn/r-programming
    R in combination with RStudio provides a powerful (and free!) working environment. If you work with others who are fluent with R it is possible to run and extend their data and analysis files (if supplied) fairly easily. I was able to replicate Piffer’s RPubs work with the data files he provided. Recreating the RMD (analysis, stands for “R Markdown”) file was the hardest (more time consuming than hard, it might help replication efforts if Davide is willing to make his own analysis file available, GitHub provides nice facilities for sharing entire R projects) part, but once that is done rerunning it and creating HTML output (as on RPubs) is trivial as is actually sharing it on RPubs. There are a number of other nice features, such as the creation of PDF and Microsoft Word documents for sharing tables, figures, etc.

    Read More
    • Replies: @James Thompson
    Thank you for your comments, and the link to R. I have clicked on it, out of gratitude, but make no promises.
    ReplyAgree/Disagree/Etc. More... This Commenter This Thread Hide Thread Display All Comments
  4. utu says:
    @Charles Murray
    Why shouldn't it work in theory? Not a rhetorical question. I'm not taking a stand on the method (haven't looked at it closely enough), but I don't see the theoretical contradiction between a tiny r2 for predicting individual IQ from 8 SNPs and a large r2 for predicting group orderings, given Piffer's hypothesis.

    I don’t see the theoretical contradiction

    Correct. But genome is not theoretical. It is what it is. SNPs have particular distributions and structures within populations and human genome in general.

    If ps (polygenic scores of 9 SNPs) have large variance within population then Piffer formula must also work (to some extent) within populations to be valid on population averages. If it doesn’t work within population, Piffer’s result is invalid.

    If however ps has very narrow variance within populations then Piffer’s formula will not work within population but this fact will not preclude a possibility of it being correct on population averages. So, what is variance Var(sp) for various populations? This should be estimated from available databases.

    While I do not know for sure I do not think that Var(ps) is small. So if Piffer’s formula does not work within population then its exceptionally good fit (r=0.91) to population averages must be spurious.

    Note 1: Theoretically Var(ps) can assume wide range of values. For frequencies from Table 8 one can estimate min and max of variance. In terms of iq the approx range is (50,300). Keep in mind that Var(iq)=225. The lowest would be realized in the most “egalitarian” population where everybody has a polygenic score equal to its population average (±0.5 SNP). The highest when SNP’s are distributed randomly independently of each other.

    Note 2: The Piffer’s formula =24.5+131.1* exactly predicted IQ of Craig Venter but overestimated the IQ of James Watson by 21 pnts. Both Venter and Watson allegedly have 8 out of 9 SNPs.

    Read More
    • Replies: @Davide Piffer
    Excuse me, but how did you infer that my formula overestimated Watson's IQ? I pooled together the alleles across Watson and Venter's genomes because each individual's count was too small. Watson in fact had a lower allele count.
    And the SNPs are 9, not 8. Originally I used 8 but then I found also the 9th.
    ReplyAgree/Disagree/Etc. More... This Commenter This Thread Hide Thread Display All Comments
  5. @utu
    I don’t see the theoretical contradiction

    Correct. But genome is not theoretical. It is what it is. SNPs have particular distributions and structures within populations and human genome in general.

    If ps (polygenic scores of 9 SNPs) have large variance within population then Piffer formula must also work (to some extent) within populations to be valid on population averages. If it doesn't work within population, Piffer's result is invalid.

    If however ps has very narrow variance within populations then Piffer's formula will not work within population but this fact will not preclude a possibility of it being correct on population averages. So, what is variance Var(sp) for various populations? This should be estimated from available databases.

    While I do not know for sure I do not think that Var(ps) is small. So if Piffer's formula does not work within population then its exceptionally good fit (r=0.91) to population averages must be spurious.

    Note 1: Theoretically Var(ps) can assume wide range of values. For frequencies from Table 8 one can estimate min and max of variance. In terms of iq the approx range is (50,300). Keep in mind that Var(iq)=225. The lowest would be realized in the most "egalitarian" population where everybody has a polygenic score equal to its population average (±0.5 SNP). The highest when SNP's are distributed randomly independently of each other.

    Note 2: The Piffer's formula =24.5+131.1* exactly predicted IQ of Craig Venter but overestimated the IQ of James Watson by 21 pnts. Both Venter and Watson allegedly have 8 out of 9 SNPs.

    Excuse me, but how did you infer that my formula overestimated Watson’s IQ? I pooled together the alleles across Watson and Venter’s genomes because each individual’s count was too small. Watson in fact had a lower allele count.
    And the SNPs are 9, not 8. Originally I used 8 but then I found also the 9th.

    Read More
    • Replies: @utu
    From your email to D. Thompson:

    http://www.unz.com/jthompson/piffer-replies-to-prof-posthuma/#comment-1901748
    From what you wrote to Dr. Thompson http://www.unz.com/jthompson/the-dna-of-genius-n2/

    “One of those SNPs was missing from Watson and Venter “

    we know that Watson and Venter members of population Cloud 8 that has expected IQ of 141 according to the formula you have established. And lo and behold

    Watson IQ=120 (https://www.simonsfoundation.org/science_lives_video/james-d-watson/)
    Venter IQ=141 (https://www.bbvaopenmind.com/en/craig-venter-the-man-who-knew-himself/)
     
    But if Watson have all 9 SNPs then the estimate is way off. Your formula IQ=24.5+131.1*PS gives IQ=155.6 for PS=9/9=1 which is 35 IQ pnts too high for Watson.
    ReplyAgree/Disagree/Etc. More... This Commenter This Thread Hide Thread Display All Comments
  6. @res

    that it is so against the general trend of the literature that they rate the finding as unlikely.
     
    Under normal circumstances I would tend to share this view. However, given the strength of the zeitgeist (aka Narrative) against Piffer's results I don't think much of this as a reason in these circumstances.

    Speaking of which, more on the recent Vox/Charles Murray controversy:
    http://quillette.com/2017/06/11/no-voice-vox-sense-nonsense-discussing-iq-race/
    And from one of the authors (Paige Harden) of the Vox article: http://www.geneticshumanagency.org/ff/the-science-and-ethics-of-group-differences-in-intelligence-part-1/
    I am getting the impression that the people arguing for the Vox article seem to be forgetting that the article asserted zero genetic effect and are instead arguing for non-zero environmental effect. That seems to qualify for a number of fallacies (e.g. Motte and Bailey).

    It is also a nice touch that the Vox headline screams "junk science" but much of the argument revolves around ethics.

    I am not going to take time out to learn R.
     
    That's understandable. If you ever change your mind, I thought this was a good starter course: https://www.coursera.org/learn/r-programming
    R in combination with RStudio provides a powerful (and free!) working environment. If you work with others who are fluent with R it is possible to run and extend their data and analysis files (if supplied) fairly easily. I was able to replicate Piffer's RPubs work with the data files he provided. Recreating the RMD (analysis, stands for "R Markdown") file was the hardest (more time consuming than hard, it might help replication efforts if Davide is willing to make his own analysis file available, GitHub provides nice facilities for sharing entire R projects) part, but once that is done rerunning it and creating HTML output (as on RPubs) is trivial as is actually sharing it on RPubs. There are a number of other nice features, such as the creation of PDF and Microsoft Word documents for sharing tables, figures, etc.

    Thank you for your comments, and the link to R. I have clicked on it, out of gratitude, but make no promises.

    Read More
    ReplyAgree/Disagree/Etc. More... This Commenter This Thread Hide Thread Display All Comments
  7. utu says:
    @Davide Piffer
    Excuse me, but how did you infer that my formula overestimated Watson's IQ? I pooled together the alleles across Watson and Venter's genomes because each individual's count was too small. Watson in fact had a lower allele count.
    And the SNPs are 9, not 8. Originally I used 8 but then I found also the 9th.

    From your email to D. Thompson:

    http://www.unz.com/jthompson/piffer-replies-to-prof-posthuma/#comment-1901748
    From what you wrote to Dr. Thompson http://www.unz.com/jthompson/the-dna-of-genius-n2/

    “One of those SNPs was missing from Watson and Venter “

    we know that Watson and Venter members of population Cloud 8 that has expected IQ of 141 according to the formula you have established. And lo and behold

    Watson IQ=120 (https://www.simonsfoundation.org/science_lives_video/james-d-watson/)
    Venter IQ=141 (https://www.bbvaopenmind.com/en/craig-venter-the-man-who-knew-himself/)

    But if Watson have all 9 SNPs then the estimate is way off. Your formula IQ=24.5+131.1*PS gives IQ=155.6 for PS=9/9=1 which is 35 IQ pnts too high for Watson.

    Read More
    ReplyAgree/Disagree/Etc. More... This Commenter This Thread Hide Thread Display All Comments
  8. EH says:

    I think the way in which a few SNPs can predict groups’ relative intelligence is going to turn out to be one of those obvious-in-retrospect things that mark a really big discovery. I think it may be something like how accuracy rises as the square root of many independent measurements with Gaussian error. Evolution has conducted many “measurements” on genomes’ fitness, the fitter segments are correlated with each other by the process of evolution, as well as with certain neutral marker sequences in the same way that that a sequence of measurements correlates with the quantities being measured. The relative frequencies of the most fitness-correlated sequences in different populations reflects the outcome of an iterated, ecologically co-evolving fitness-measuring process which includes not only all the individuals in the population but also the populations of their ancestors back to the time when the two populations last separated.

    Looking at whole populations, some individuals are lucky or unlucky, but they generally cancel each other out in large groups so the evolutionary fitness-measuring/correlating process is what governs the genetic makeup of groups and not the random chance one sees in individual cases, so one would expect the population frequencies of some sequences to have high correlations with whatever fitness-affecting trait one were to study. Some of these variants will date back to before the populations separated and they will have undergone selection in different environments with different demands/rewards for the trait under consideration, and also subject to uncorrelated random effects, so the populations will differ in the frequencies of these certain sequences in a way that is highly correlated with the trait, even though the sequences generally don’t cause the trait, they’re just correlated with all the bits that do affect the trait. Or something like that. Genetic simulations should show this correlation effect emerging naturally if the hypothesis is correct.

    Read More
    ReplyAgree/Disagree/Etc. More... This Commenter Display All Comments
  9. populations will differ in the frequencies of these certain sequences in a way that is highly correlated with the trait, even though the sequences generally don’t cause the trait, they’re just correlated with all the bits that do affect the trait. Or something like that.

    This is what I would like geneticists to comment on. I will be posting sometime on another way in which this general technique can be used.

    Read More
    ReplyAgree/Disagree/Etc. More... This Commenter Display All Comments
Current Commenter says:

Leave a Reply - Comments on articles more than two weeks old will be judged much more strictly on quality and tone


 Remember My InformationWhy?
 Email Replies to my Comment
Submitted comments become the property of The Unz Review and may be republished elsewhere at the sole discretion of the latter
Subscribe to This Comment Thread via RSS Subscribe to All James Thompson Comments via RSS