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

 Remember My Information



=>

Bookmark Toggle AllToCAdd to LibraryRemove from Library • BShow CommentNext New CommentNext New ReplyRead More
ReplyAgree/Disagree/Etc. More... This Commenter This Thread Hide Thread Display All Comments
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
Search Text Case Sensitive  Exact Words  Include Comments
List of Bookmarks

sex diffs cartoon

Will sex differences never end? Sometimes they seem to go one way, sometimes the other, with the gaps closing or staying resolutely the same, but this is the March of Science, as different schools contend, and as new results are added to the old. We should be glad that researchers delve into these matters, particularly in times when sex differences are considered somehow dangerous, and apt to cause offence, public disorder, and a rending of the very fabric of society, as when a diaphanous blouse is torn from a lady’s breast by a brutish leering lout. I digress. I may not have got the hang of current sensibilities.

Anyway, back to “tilt”. I find this a relatively new concept, but perhaps it is more succinct than the former description “verbal-performance discrepancy”. Which way do the fair sex incline: to matters verbal or mathematical? Verbal, it would seem, and all the more so as you go up the ability spectrum.

Sex differences in ability tilt in the right tail of cognitive abilities: A 35-year examination Jonathan Wai, Jaret Hodges, Matthew C. Makel. https://doi.org/10.1016/j.intell.2018.02.003 Intelligence, Volume 67, March–April 2018, Pages 76–83

https://drive.google.com/file/d/1Y0kQvdjiuinNsPlIHdXS65PjmHsxLvV_/view?usp=sharing

The authors highlight the following findings:

Sex differences in math-verbal ability tilt in the right tail were examined across 35 years.
Sample included >2 million gifted adolescents across multiple measures in the U.S. and India.
Ability tilt favored males for math > verbal and females for verbal > math.
Sex differences in ability tilt remained fairly stable over time and replicated across measures.
Trends should be monitored given their potential to impact future workforce trends.

Abstract
Sex differences in cognitive ability level and cognitive ability pattern or tilt (e.g., math > verbal) have been linked to educational and occupational outcomes in STEM and other fields. The present study examines cognitive ability tilt across the last 35 years in 2,053,265 academically talented students in the U.S. (SAT, ACT, EXPLORE) and 7119 students in India (ASSET) who were in the top 5% of cognitive ability, populations that largely feed high level STEM and other occupations. Across all measures and samples, sex differences in ability tilt were uncovered, favoring males for math > verbal and favoring females for verbal > math. As ability tilt increased, sex differences in ability tilt appeared to increase. Additionally, sex differences in tilt increased as ability selectivity increased. Broadly, sex differences in ability tilt remained fairly stable over time, were consistent across most measures, and replicated across the U.S. and India. Such trends should be carefully monitored given their potential to impact future workforce trends.

Before going further, I should digress to sing the praises of Gerd Gigerenzer. Whereas Kahneman and Tversky concentrated on puzzles which many have found contrived, Gigerenzer concentrated on actual, everyday problems with interpreting statistics. Here is an example. We can all guess those people in the top 1% of intelligence are 1 in 100 intellects, but what does 0.01% mean? It “feels like” 1 in a 1000, but is in fact 1 in 10,000. As Gigerenzer says, don’t mix decimals with percentages. They are different tribes. Confusing. Instead, use natural frequencies. In a town of 10,000 people there will be 100 people who are in the top 1% of the population, but only one person who is 1 in 10,000. That is the level that Benbow and Lubinski studied. Eminent minds, Galton called them.

So, skipping a thousand words, here is the pictorial summary, which shows that sex differences increase as ability tilt increases:

sex diff violin plot of tilt Wai

To my eye, starting from the bottom for all students, these violin plots show the following: women are almost perfectly balanced between verbal and mathematical ability, but men incline towards being better at maths than at verbal tasks. Men are more likely to calculate. This relatively slight difference might be the source of contrary imaginations, and some exasperated arguments.

At the higher intellectual level of the top 1 in a 100 of the population both men and women incline more to mathematical thinking, but men predominate more.

At the eminent level of the top 1 in 10,000 of the population, men outnumber women by about 2.5.

The authors caution:

It should be noted it is likely that the magnitude of the selectivity level moderator (top 5%, top 1%, top 0.01%) is underestimated. In the analysis, the ratio of male to females was skewed toward males. This skew increased as selectivity increased. This unbalanced design can lead to the true effect being greater than what is reported. In other words, it is possible that the magnitude of tilt is greater than what is reported.

Of course, the top mathematicians will be even brighter, say 1 in 1,000,000. They will be mostly men.

The paper is rightly restrained in its conclusions, but this stable result is a causal component in the observed sex differences in STEM studies and subsequent occupations. Sex differences in preferences for occupations are mentioned, but are not part of this study. India, which is far less favourable to women’s careers, shows the same trend. Probably, men take to maths more than do women, and at the brightest levels that general trend is accentuated. The current study may under-estimate the male advantage.

 
• Category: Science • Tags: Gender 
Hide 147 CommentsLeave a Comment
Commenters to Ignore...to FollowEndorsed Only
Trim Comments?
    []
  1. Larry Summers was fired as president of Harvard for making a similar observation about how and why men predominate in STEM fields. What’s less often noted about that episode is what Summers had to say about where else men predominate: in prisons and homeless shelters. His point being not that men are more intelligent, but that from the top to the bottom of societal achievement, women are both less likely to populate the peaks as well as the troughs. Whereas men predominate both high and low. Presumably if the scope of this study were expanded to include the whole range of intellectual ability, we would also find that men have an outsized presence among the dumbest 1 in 10,0000 grouping.

    • Replies: @dearieme
    , @bjondo
    , @hyperbola
  2. That is the level that Benbow and Lubinski studied. Eminent minds, Galton called them.

    Elitism…

    • Replies: @Anon
  3. Anon • Disclaimer says:

    Of course, the top mathematicians will be even brighter, say 1 in 1,000,000. They will be mostly men.

    every genius and/or spiritually deep mind will be male as well.

    No need for papers, research, and “scientific proof” to what anyone resistant enough to societal cultural programming learns to be true over the years. Observing and connecting the dots is enough.

    a title=”https://illimitablemen.com/2015/09/27/educated-women-vapidity/>This is a place where what all of us who still have the children who knew the Emperor to be naked inside us know is put in an elegant, truth-respecting way.

    • Replies: @AndrewR
    , @Paw
  4. Anon • Disclaimer says:
    @Santoculto

    Envy (insecure ego wounded).

  5. dearieme says:

    Personally I’d settle for an IQ test consisting of one question and a follow-up. The question: what does “impact” mean?

    The follow up: then why do you dolts keep misusing it – is it because you have too small a vocabulary, or that you think it a posh use, or are you just mutton-headed?

    I shouldn’t be in the least surprised if the fair sex did better on that test.

    • Replies: @Pericles
    , @res
    , @andrenovartis
  6. dearieme says:
    @Habakkuk Mucklewrath

    “Summers was fired as president of Harvard for …”: Summers was fired because many academics at Harvard had come to loathe him (with good reason). His observation gave them a desirable pretext and useful allies. No martyr he.

  7. dearieme says:

    “The current study may under-estimate the male advantage.” That’s my suspicion based on spending much time STEMing in the company of STEMers.

    P.S. I’m a fan if Gigerenzer too. I’ve read only one of his books but it really is splendid.

    https://www.amazon.co.uk/Reckoning-Risk-Learning-Live-Uncertainty/dp/0140297863

    • Replies: @Dieter Kief
  8. Factorize says:

    Dr. Thompson, thank you very much for this one!
    I am preparing to write a post to ask the help of the blog’s group intelligence to help with my assignment.

    I making room for a whole new section in my thesis for this.
    The role of women is particularly highlighted in human development research, so this should fit in well. The top panel of the figure especially caught my eye. In the two other panels of the figure the tilts for men and women were almost symmetric. Yet, when you look at the top ( most extreme selection) a bimodal distribution emerges for women but not men.

    This is very very interesting and possibly of substantial importance. When genetic enhancement of IQ begins a distribution such as that seen in the top panel might emerge, possibly even more extreme. There would be high achieving men and women quants that could program and do calculations together and there would also be high performing women comms without an obvious gender matchup. Enhancing IQ might create a nearly unbridgeble cognitive divide between the genders.

  9. j2 says:

    Two questions:

    There are ethnic groups which have stronger verbal than spatial IQ. How does it reflect in SAT, that is, do they have higher verbal than mathematical SAT score?

    If so, what if anything it implies to the main observation of the post that if ability goes up, the tilt to mathematics and away from verbal becomes stronger?

  10. Pericles says:
    @dearieme

    A good candidate for one of those All Souls exams.

    • Replies: @DFH
  11. “I may not have got the hang of current sensibilities.”

    Or you might not be writing here, but thank you for it.

    It’s all good fun to take a look at Uni undergrad populations by discipline – I don’t know if UCL does that, but Bristol have stats for all their undergrads.

    http://www.bristol.ac.uk/ssio/statistics/

    sarc We all know that historically men have been discouraged from being doctors, so it’s no surprise that the ratio for undergrad medics is 2/1 female to male /sarc.

    But that holds for biochemistry, biological science, physiology, neuroscience, vet, cellular/molecular med, pharmacology and anatomy too.

    Maths is pretty male – 550/280 – but applied science – engineering maths, mechanical, civil and aero engineering – is where the boys really dominate – anything from 3/1 to 7/1.

    Only Art History, with it’s 8/1 female/male ratio, skews so much to one sex. Must be nice if you’re a straight male Art History student.

    (I was originally going to look at the Institute of Actuaries exam results to see what the m/f ratio was in this most nerdishly numeric of fields, but the number of subcontinental/far eastern names defeated a quick analysis – I did see more females than I expected and that not a single Smith (one Smyth) or Jones passed for Fellowship – the sort of thing that Gregory Clark might note.)

    https://www.actuaries.org.uk/studying/exam-results/fellowship

  12. res says:

    Interesting paper. Thanks! I think the SAT tilt for the top 0.01% plot in Figure 1 above merits some additional discussion.

    First, I think a big part of what we are seeing in the shape of that curve is due to the separate M/V cutoffs. I believe this selects for large tilts as demonstrated by the bimodality. This is most clearly seen for women (which I think is an important fact itself, why is the male V tilt side of the distribution bumpy?)) where the two modes are roughly at +-150 tilts. A similar but less obvious effect appears in the 1% group where dual modes at roughly +-50 fits well with what we see.

    One thing that is worth emphasizing is that the violin plots are probability densities, not population. This obscures the male/female difference in the top 0.01%. The respective sample sizes for each SAT group are:
    5% female = 673,756, male = 670,134, total = 1,343,890
    1% female = 411,978, male = 448,787, total = 860,765
    0.01% female = 1472, male = 3451, total = 3923

    One oddity is that the 1% and 0.01% sample sizes are 225x different in size rather than 100x. I’m not sure if they were more selective than intended, this is an artifact of the different waves, or something else. The 5% and 1% sample sizes have a similar discrepancy at 1.6x vs. 5x.

    By my eye males are overrepresented on the high M side by about 4-5x (2.5x sample size, 1.5-2x density). This is less than I would have expected at the 0.01% level. Perhaps I need to recalibrate. Any thoughts?

    An interesting alternative would be to look at the tilts in groups selected by a single combined threshold. I don’t know if they have the data to do that. I think it would be a better approach for looking at the tilt in the top 0.01% since the original study selects disproportionately for high tilts in both directions (hence the bimodality). An additional problem with the dual thresholds is that it appears the math threshold is easier (in “percent of population which qualifies” terms) than the verbal. I suspect the difference in +/- tilt populations influences the conclusions, but I have not thought this all the way through yet.

  13. res says:
    @dearieme

    But what would be the impact of your test? ; )

    I think the battle has been lost given that at least some current dictionaries have both meanings.

    It is interesting to compare American and English definitions at https://dictionary.cambridge.org/us/dictionary/english/impact
    and also to compare the proficiency levels given with the English definitions: https://dictionary.cambridge.org/help/

    I expect you will be scandalized by the definition “the force or action of one object hitting another” being noted as the highest level of proficiency (C2) while the other two are B2 and C1.

    • Replies: @dearieme
  14. Factorize says:
    @res

    Res, I am disappointed that the raw data was not presented. For example with the tilt figures, all they are showing is the relative comparison against THEMSELVES! For someone to be at the far right of the figures one needs to be relatively weak in verbal ability (while still being in the top 01%). I would really love to see the MRI slices at different levels of absolute achievement. What would be the m/f quant ratio be at the extremes?

    I am also interested if this male tilt to the right as ability levels increases also apply when moving upwards from very low ability.

    • Replies: @res
  15. The tilt graph’s caption bothers me:

    Tilt is the student’s SAT math score minus their SAT verbal score.

    This makes investigations of tilt at the “high end” worthless, because the SAT’s math ceiling is much lower than the SAT’s verbal ceiling, even though both are scored at 800 “points”.

    When I took the GRE, I scored 780 on the math and 750 on the verbal. That would show up as a tilt toward math by this methodology, despite the fact that 750 verbal is a higher score than 780 math is.

    And sure enough, the graphs show that males and females both tilt toward math at the high end, which they’re required to do by the study’s methodology.

    • Replies: @res
    , @James Thompson
  16. res says:
    @Michael Watts

    Keep in mind that these people took the SAT at 13. That means there is much more ceiling than for those taking the test at the usual time. Still, your different difficulties for a given M/V score point is relevant as I also mentioned in comment 12.

    As noted in section 3.3 they transformed scores to the post-1995 SAT scale (losing some of the ceiling for earlier scores). Their score thresholds were:
    Top 1% – SAT-M 430+ or SAT-V 450+
    Top 0.01% – SAT-M 700+ or SAT-V 700+

  17. res says:
    @Factorize

    I always like seeing more data ; ), but keep in mind that the Duke TIP is a long ongoing study with an extensive publication record: https://tip.duke.edu/resources/research/publications
    You might want to explore that and see if the data you want is available. I don’t know if their data was used for a MRI study, but it was used for a case-control GWAS last year:
    Zabaneh, D., Krapohl, E., Gaspar, H. A., Curtis, C., Lee, S. H., Patel, H., … & Breen, G. (2017). A genome-wide association study for extremely high intelligence. Molecular Psychiatry, 1-7. https://www.nature.com/mp/journal/vaop/ncurrent/pdf/mp2017121a.pdf

    There was a preprint of this paper last year: https://psyarxiv.com/rwgpy/
    which lacks the violin plots and presents the results as numerical tables. For example see Appendix C on page 28 for a look at the top 0.01% tilts. I like the violin plots better, but the tables present population numbers rather than distributions which gives a better idea of the relative group sizes. One thing that tripped me up at first is the bins are inclusive (e.g. >= 100 includes >= 200 and 300). You can see this by summing the <= -100, -100 < x = 100 bins for males and seeing 326 + 675 + 2450 = 3451 (i.e. the total 0.01% male sample).

    I am also interested if this male tilt to the right as ability levels increases also apply when moving upwards from very low ability.

    Me too. ETS has data (all SATs) that would be good for exploring this at most levels, but not at the very lowest. I wonder if Dr. Thompson knows of any research like this.

    Regarding the tilt, remember the thresholds they used:
    Top 1% – SAT-M 430+ or SAT-V 450+
    Top 0.01% – SAT-M 700+ or SAT-V 700+
    So a +250 tilt might correspond to top 0.01% M and top 1% V. A big difference, but hard to call the lower score weak.

    BTW, the Duke TIP India program was new to me. Does anyone know much about its history?

    • Replies: @Factorize
    , @James Thompson
  18. Factorize says:
    @res

    res, sorry I was using MRI slices more as a figure of speech.

    I was thinking about the ability distributions that you would find as you sliced through the
    multivariate space. It is too bad that we still have not moved to a time when science would
    naturally dump its datasets somewhere for people to poke around.

    It was not completely clear to me what might happen at the extreme low end. With certain developmental disabilities, you are no longer talking about merely being at the low end of the distribution for common variants, but more about being on an qualitatively different distribution.

    This idea about the entire top end of the ability curve shifting to a mathematical orientation is quite startling. I have never encountered it before. The figures suggest that it is substantial. I would love to see the animation starting at the bottom and working up, using perhaps 1% frames and watching the curve shift right. Also of interest to see how this might apply to Asian populations who are thought to be naturally relatively math tilted.

  19. Yan Shen says:

    Speaking of uh math/verbal tilt, I have an article coming up on Unz that should published next Monday that explores precisely the importance of the math/verbal split.

    To my eye, starting from the bottom for all students, these violin plots show the following: women are almost perfectly balanced between verbal and mathematical ability, but men incline towards being better at maths than at verbal tasks.

    This kind of characterizes the difference between East Asians relative to Europeans, but more to uh come on Monday…

    • Replies: @res
  20. I call this bullshit. Females just talk, talk, talk…. but their verbal ingenuity is lesser than male. Just look at significant writers in any field after 1960 (so, no that ogre of “patriarchy”).

    Take simply imaginative literature. How many female novelists, poets, essayists, memoirists, dramatists, short story writers are worth reading, in comparison with males of the same profession?

    Yeah, I know …

    • Replies: @dearieme
  21. @res

    Thanks. It is very good to have someone comment in detail on matters that I noted, worried about, and then decided to put aside in my brief account of the paper. As far as I can see, you are right that there are a number of problems with the depiction of the results, probably caused by cut-offs, and not having the plots of the raw data to tease out what may be happening. I think the actual male/female balance at this high level of maths is at least 5 to 1. At Trinity College Cambridge I think it is about that, though I haven’t checked recently.
    I am waiting for Jonathan Wai to reply to me, and I will get him to look at your comments particularly.
    best
    James

    • Replies: @res
  22. @Michael Watts

    Thanks. I am beginning to feel we need to demand a recount. Will see what Jonathan Wai says.

  23. @res

    I had discussed it before, but not in detail about provenance.

    • Replies: @res
  24. @dearieme

    Gerd Gigerenzers books are not only more rewarding than Kahnemann and Twerp’s, but are also more rewarding than Nassim Taleb’s, methinks.

    His findings are stunning often times to say the least. Such as the fact, that doctors usually can’t properly interpret statitic findings of medical tests = m o s t of the doctors are statistically illiterate – and many of them make more mistakes than bright kids make.

    (cf. “iatrogenesis” (and Victor Frankl)).

    @ Dearieme – that’s the second – if not the third time – we sing the Gigerenzer-praise together on unz.com, isn’t it?

    • Replies: @dearieme
  25. DFH says:
    @Pericles

    They don’t have the single word paper anymore, unfortunately

  26. @dearieme

    A point well raised. What is wrong with “effect” as a noun, or “affect” when misused as a verb.
    It follows that “negative impact” is less than meaningless.
    For those of you who find this difficult to follow, this is what used to be known as “grammar”.
    I’m showing my age, I know.

  27. dearieme says:
    @res

    “I expect you will be scandalized by …”: why so?

    • Replies: @res
  28. dearieme says:
    @Bardon Kaldian

    To my taste the best novelist in English at present is Hillary Mantel. She’s a her.

  29. dearieme says:
    @Dieter Kief

    Repetitive, moi? Guilty as charged. If you’ve read his more recent books would you like to recommend one particularly?

    • Replies: @Dieter Kief
  30. res says:
    @Yan Shen

    Re: “balanced”, I assume everyone here realizes “more balanced” is meaningful only in the context of a reference group. One can say two groups have different balances, but saying one is more balanced than the other depends on the reference. The precise numerical scores are an artifact of the scoring chosen by the test designers and if I recall correctly that actually changed a bit (~50 points) at the high end with the 1995 SAT recentering.

    Your male/female and European/East Asian M/V analogy is interesting. If you write that up it would be interesting to talk about spatial ability and/or Jewish tilts as well.

    Looking forward to your article. And the comments ; )

  31. res says:
    @James Thompson

    Not sure if you are referring to the case-control GWAS or the preprint. I remember talking about the former here not that long ago, but if you meant the latter I missed it somehow. Regarding the Duke TIP in general, I know you have mentioned it in many posts. I debated adding some links to those but did not see one which was a good overview of the study relevant to Factorize’s context.

  32. res says:
    @dearieme

    I assumed you would not like that the purer definition was considered less common (only known at maximum fluency). On rethinking this conversation, that was pretty much your original point, but I thought making it more concrete (specific proficiency levels) would have an additional impact force. (in all seriousness, I find using impact very natural in that sentence. How would you write it?)

    • Replies: @dearieme
  33. dearieme says:
    @res

    I thought making it more concrete would would make it bigly pizzazful.

    • LOL: res
    • Replies: @res
  34. res says:
    @James Thompson

    Thanks! I think you chose correctly. One of the best things about your blog IMHO is your ability to identify and summarize the important points of a paper at a consistently good level of complexity. In contrast, I tend to overcomplicate the things I find interesting and underemphasize the others.

    I downloaded and started reading the paper just after your link and was very happy when I looked at Figure 1, said “that’s interesting!”, scrolled down in your post and saw you had included it with some discussion.

    I was wondering if a violin plot showing populations rather than density (i.e. normalized by sample sizes) would be better. What do you think?

    Thinking about that 0.01% figure 1 SAT plot some more, one thing that surprises me is that the high M male/female tilts are very similar (maybe a 10-20 point greater mean tilt for men). I think this is contrary to the assertion (which I happen to believe) that an important part of the male/female difference in STEM representation is caused by high M women typically having a higher V giving them more options. What do you think?

    Is it possible that observation is caused by the SAT math ceiling after all? I seem to recall a discussion a while ago (here or in iSteve?) where I estimated the post-1995 combined SAT ceiling for 13 year olds (+4 SD?) based on the ceiling for 17 year olds (~+3 SD?) and developmental curves (~1 SD?). IIRC the pre-1995 ceiling was an SD higher. But the 1995 change was more in V than M so not sure.

    More on that observation, arguably the 0.01% SAT V tilt side shows women having larger average tilts there. And the ACT version of the plot clearly shows the effect I expected on both the M and V sides (and at all ability levels!). I have been focusing on the SAT because the sample sizes are larger and I am more familiar with the SAT and its history. BTW, comparing the SAT and ACT 0.01% Figure 1 plots it looks like the M/V balance is different (i.e. the ACT V threshold is relatively easier to meet). Also, the intersection of the two 0.01% SAT sub-distributions is around a 0 tilt while for the ACT it is at about -5.

    Looking forward to hearing more from Jonathan Wai! I wonder if he could create versions of Tables 1 and 2 and Figures 1 and 2 with the high M and high V groups separated (there would be overlap in those subsets though). It would be interesting to compare those subsets with a combined group using single combined SAT/ACT thresholds (not sure what appropriate 13 year old combined thresholds would be though). I think the single combined threshold would be better for comparing tilts by sex and the extreme M and V groups good for seeing if those differ from the combined results.

    Perhaps rather than a pure combined SAT threshold (e.g. 1100) do something like either M or V > 700 OR M + V > 1100 so we are looking at true subsets?

    The reason I want to separate the groups is that I think the different male/female numerical balance in the high M/high V groups is what causes the almost 100 point sex coefficient in Table 1 for 0.01%. The Figure 1 SAT 0.01% plot shows more like a 10-20 point shift (for high M) as I observed above.

    • Replies: @James Thompson
    , @res
    , @EH
  35. @dearieme

    To my taste the best novelist in English at present is Hillary Mantel. She’s a her.

    Donna Tartt (The Goldfinch) is great, as are quite a few of the short stories of Annie Proulx (only ten or so of T. C. Boyle might be close or here and there even better as her best).

    But – Franzen and Wolfe aren’t topped by nobody (adrenovartis: This is Bavarian logic/grammar = quite meaningful in Bavaria – (when there’s a clear day, I can even see Bavaria from the town at the lake, I happen to live in, if that counts as an excuse).

  36. res says:
    @dearieme

    I used the wrong HTML tag (s instead of strike) and forgot to check how it rendered. It was supposed to be:
    I thought making it more concrete would have an additional impact force.

    • Replies: @dearieme
  37. dearieme says:
    @res

    I doff my cap: “additional force” seems ideal to me.

  38. Factorize says:

    res, what is your guess on what the plot would look like for high end East Asian populations?

    Smart fraction theory suggested roughly a 5 IQ math tilt for their entire population or one third SD or 30 SAT points. Would this linear translation of the East Asian tilt remain constant as higher ability were considered or might it be more of a magnifying Effect as occurs with extreme outliers when the Bell Curve is shifted even by only a few points?

    • Replies: @res
  39. res says:
    @Factorize

    This might be a better question for Yan Shen. I don’t have a good enough sense of the relative magnitudes of the East Asian/European IQ difference and East Asian and European male/female M/V tilts. This is made much more complex by the presence of different subpopulations (e.g. nationalities) as well.

    The linearity of the transform and its effects is a good question which needs data to answer. The big effect I see is the changing mean M/V scores producing tail changes in number who pass a given threshold. I don’t think this would affect the true tilt much (that is a big assumption though), but it would change the Table 1 and 2 regression analyses because of the issue I mentioned above with the differing balance of two separate subpopulations (high M and high V) dominating the actual tilts.

    • Replies: @Yan Shen
  40. Why is it that I used to read that at the eight-hundred level there were more boys that girls on the SAT?

  41. Factorize says:

    res, did you see it!
    Party Time!

    https://www.biorxiv.org/content/early/2018/03/02/274654

    They are talking 2 Meg on the way for height! Yeah!
    The current article is only a tiny 700K!

    We all should know that things could soon get mighty mighty interesting.
    2 Meg might even be more than is needed when using the right approach.
    (Should be able to clearly push through the phase transition at 2M).
    Perhaps they could do a redo of the compressed sensing; this time looking for
    interactions!

    Great great new!

    • Replies: @res
  42. @res

    Too late at night for me to comment. Tiny additional datum: Trinity College Cambridge maths undergraduates 3 to 1 men.

    • Replies: @res
  43. res says:
    @James Thompson

    Thanks for the additional datum. My guess would be that population is around 0.1% rather than 0.01% (in terms of IQ that would be about a full SD) which I think would make a difference. Does that seem reasonable?

    As a possible comparison, using Emil’s calculator http://emilkirkegaard.dk/understanding_statistics/?app=tail_effects
    with means of 105 and 100 we see the following ratios:
    2SD – 2.1
    3SD – 2.8
    4SD – 3.9
    Those align fairly well with our data. Does a third of an SD difference in math means with the same SD seem in the ballpark of real world male/female differences? BTW, also changing the lower mean SD from 15 to 14 changes the ratio to 13.5 at 4SD!

    • Replies: @James Thompson
  44. res says:
    @Factorize

    I am a bit confused because that paper does not seem to acknowledge the height compressed sensing work. Is that because the CS paper is still a pre-print? Or was the current paper submitted before the CS pre-print was available?

    Here is the relevant quote from the paper about the larger upcoming study:

    The present study is part of a larger effort led by the GIANT consortium, expected in the near future to yield one the largest GWAS ever conducted (N between 1.5 and 2 million).

    It looks like BMI and height are the primary variables in the GIANT data: https://portals.broadinstitute.org/collaboration/giant/index.php/GIANT_consortium_data_files
    It seems like a huge lost opportunity to have that many genomes and not gather more variables.

    Thanks for the link!

    • Replies: @Factorize
  45. Yan Shen says:
    @res

    Well not to get people’s hopes up too high, but as a relatively non-technical layperson, my upcoming article is much more along the lines of a popular blog post that one might expect from say Steve Sailer or Chanda Chisala rather than something produced by the likes of say James Thompson or Steve Hsu. :)

    Hopefully it’ll still kick off some interesting discussions though!

  46. @res

    The same informant giving the 3 to 1 figure (25% of the undergraduates were women) said that it was fewer in the following years. I am looking to get proper data on this in a few month’s time.

    As to how we actually calculate the figures for sex differences, it really does depend on

    A) whether you assume a 2 or 4 point male advantage in general intelligence
    B) whether you place that difference as, say 100 versus 104, or more sensibly 102 versus 98
    C) whether you put women’s sd at 14, which seems to make sense, but not in Bulgaria.

    Assuming 4 point difference, women at sd 14, then the observed preponderance of men is entirely as expected: no glass ceiling. All this yet to be proved, confirmed and accepted.

    • Replies: @res
  47. @dearieme

    It’s not just one language & not just personal taste.

    In past half a century there were/are Marguerite Yourcenar, J.C. Oates, ..But, on the other hand- Thomas Bernhard, Cormac McCarthy, Don DeLillo, John Updike, Czeslaw Milosz, G. Garcia Marquez, David Grossman, Salman Rushdie, Mario Vargas Llosa, ..

    This is not even comparable.

    • Replies: @dearieme
  48. dearieme says:
    @Bardon Kaldian

    De gustibus non disputandum est.

  49. res says:
    @James Thompson

    Thanks for the elaboration. One clarification. I was focusing on the math side only which was why I estimated 5 points.

    A good reference for this conversation is https://www.ets.org/Media/Research/pdf/RR-02-04-Dorans.pdf
    I think this is the definitive reference for the 1995 SAT recentering and it has distribution histograms by race and sex. The “Gender Comparisons” section begins on page 14. There we see numbers indicating a difference of about 4.3/11 = 0.39 SD on male/female SAT math scores in 1990. So I think my 1/3 SD working estimate for math ability difference is reasonable.

    Recentering brings male and female averages closer together, numerically, on SAT M, a mean difference of 3.8 (38) instead of 43, while leaving them virtually unchanged on SAT V, mean differences of 1 (10) and 10, as expected given the nature of the recentering conversions described earlier. Because the standard deviation of the 1990 Reference Group on SAT M is 11 (110) on the recentered scale subgroup, differences are numerically smaller there than they were on the original scale on which the 1990 Reference Group had a standard deviation of 123. SAT V subgroup differences are invariant to the scale shift because the standard deviation is essentially the same.

    Table 1 on page 15 gives both original and recentered summary statistics for the subtests by sex. There we see that the male and female SDs are also different, with the math difference being about 3-4x the size of the verbal difference. On the recentered scale, after multiplying their numbers by 10 to account for the funky 920-980 scale rather than 200-800, we see:
    Math SD 104/113 female/male (scaled this is 13.8 for females if males are 15)
    Verbal SD 109/112

    This excerpt from page 18 explains the 920-980 funkiness:

    (The 920-to-980 scale is used in this paper to facilitate comparisons that would have been difficult to make had both sets of SAT scores been described side-by-side on the same 200-to-800 scale.)

    Relevant to discussion of SAT math being easier than verbal:

    On the original 200-to-800 scale (top portion of each figure), the highly off-centered nature of the SAT V distribution is quite evident. For female students, the 50th percentile (or median) for SAT V is 410, while for male students, it is 420. On the recentered 920-to-980 scale, the median score for both gender groups is 950, the midpoint of the scale.
    On SAT M, the median for female students on the original 200-to-800 scale is 450, 40 points higher than the SAT V median of 410. On the recentered 920-to-980 scale, the median SAT M score for female students is 948, which is 2 (20) points lower than the SAT V median of 950.
    On SAT M, the median for male students on the original 200-to-800 scale is 500, 80 points higher than the SAT V median of 420. On the recentered 920-to-980 scale, the median SAT M score for male students is 952, which is only 2 (20) points higher than the SAT V median of 950.

    It is worth emphasizing that the Wai paper uses post-1995 SAT scores (earlier results are converted).

    I highly recommend the Dorans paper. The sex and race histograms (on both the original and recentered scales) are very informative. I just wish they had given the analogs of Table 1 for the racial differences. But that was probably a bridge too far with respect to crimethink. The medians can be extracted from the text, but unfortunately I see no way to get the SDs except by eyeball estimate.

    Also keep in mind the viewpoint behind the recentering and paper. From the conclusion:

    The college graduates in the year 2000 were the first complete cohort to have scores on the recentered SAT I scales. These students went through their last years in high school and their years in college with a markedly different perspective of their own abilities, particularly their verbal skills, than earlier generations had of their abilities. For decades, scores on the old SAT scales told us that both males and females were better in mathematical reasoning than in verbal reasoning. The new scales tell us that females are better verbally than they are in mathematical reasoning. The scale shift altered the way generations of students view themselves.

    Fascinating.

    P.S. The Dorans PDF is secured so text copy does not work. See https://www.wikihow.com/Unlock-a-Secure-PDF-File if you want to copy text like I did above. The Ghostscript/GSview approach worked very nicely for me.

  50. @dearieme

    Of course. But I doubt a well-read & informed female reader would put any 20th C female writer in the same category with Joseph Conrad, D.H. Lawrence, Marcel Proust, Franz Kafka, William Faulkner, Thomas Mann, Hermann Broch, Robert Musil, James Joyce, R.M. Rilke, …

    It goes beyond personal preferences.

  51. @dearieme

    Well, Charles Murray would say that if you allow 50 years to pass then the best writers will probably still be mentioned, the others not.

    • Replies: @Pericles
  52. writers says:

    Agatha Christie, Enid Blyton, Dorothy Sayers, Dianna Wynn Jones say you are very wrong. Flatulent frauds like Lawrence and Faulkner will long be forgotten while people are reading the sweet, reasonable and well written books of these ladies.

    • Replies: @dearieme
  53. dearieme says:
    @writers

    Lawrence I find laughably bad. As bad as Enid Blyton? Worse, because pompous. Conrad I’ll grant you. Faulkner: dunno, I’ve never felt moved to look.

    Hemingway is another I dismiss: as inauthentic as they come. Fitzgerald: dull, dull, dull.

    Women can even do broad humour. The Richmal Crompton “William” books are wonderful stuff.

    Naturally all sincere capitalists will accept J. K. Rowling as the best writer since … I dunno, Martin Luther?

    • Replies: @YetAnotherAnon
  54. Factorize says:
    @res

    Yes, you are completely correct about the analysis silo mentality that appears to happen with research. The signal does not seem to have reached them that such a massive sample might be beyond what would be required using the compressed sensing approach.

    I would think that at some point taxpayers might not be overly amused with paying for these huge samples when all the information possible is not being extracted.

    This is the year for the mega-Height GWAS! Having one highly polygenic trait under our belt will alert everyone that we have clearly entered into the polygenic age and have a robust method that can be used with confidence for more politically sensitive traits. I am waiting anxiously to see how accurately they will be able to predict height when all the research from genetics is put into the mix.
    Yet, we might need to wait for the 500K UKBB exome scans to be completed for the next big boost in predictive power.

    res, I was looking at past blog posts from about 8 months ago and I noticed that there was something of a heated discussion about using 9 SNPs to make predictions about group IQ clustering. Off hand, from my current understanding of the idea of polygenic adaptation, a subset of SNPs for, say IQ, should give a signal for the overall trait. There was a great deal of discussion about this at the time and the question seemed to have went unresolved. Do you have any update on this? Have even larger sets of IQ SNPs been analyzed? (The research discussed included 9; there are currently approximately 200).

  55. res says:

    res, I was looking at past blog posts from about 8 months ago and I noticed that there was something of a heated discussion about using 9 SNPs to make predictions about group IQ clustering. Off hand, from my current understanding of the idea of polygenic adaptation, a subset of SNPs for, say IQ, should give a signal for the overall trait. There was a great deal of discussion about this at the time and the question seemed to have went unresolved. Do you have any update on this? Have even larger sets of IQ SNPs been analyzed? (The research discussed included 9; there are currently approximately 200).

    I assume you are referring to the series of posts about Davide Piffer’s work in June? Like this one: http://www.unz.com/jthompson/piffer-replies-to-prof-posthuma/

    I am not sure what has happened since then. The last I knew Davide was incorporating an analysis of Linkage Disequilibrium (LD) into his work. I don’t know if he has included the new SNPs in his analysis. He seemed busy then.

    I share your curiosity.

    • Replies: @James Thompson
  56. Factorize says:

    res, this seemed like a fairly surprising and important discovery that he made. What is your latest take on it? Was he correct? Polygenic adaptation does at the very least at a basic intuitive level make a great deal of sense: the entire genome becomes activated to select IQ SNPs and so they at some level should act jointly.

    Did you see the Math/Verbal split article. Wow, very well written; almost disputes its own thesis. I think that’s why he threw in a few “bs”es. He covered the landscape fairly well.

    Would have liked to have seen a reference to psychohistory and the Foundation series. Perhaps what we really need is for the humanities to fully transition to a more mathematical basis. Instead of East Asians trying to play a weak hand by going verbal, they cold make a conscious effort to play to their strengths. If they could apply the same level of quantitative skill to economics, history, other social sciences etc., then they could make a substantial contribution to even these mostly verbal fields. Developing psychohistory to the point of actual calculations would be even more impressive than more obviously quantitative achievements. Actually making accurate predictions about our essentially irrational and sometimes frankly insane species would seem to be close to magic.

    Of course, the current big quantitative effort in GWAS is a major effort to make medicine less art and more science. Will be interested to see what it is the Chinese now actually select for. If Chinese verbal IQ actually is around 97, then selecting embryos that specifically increased this aspect of g could be the most strategically high return choice. If this were done than the entire nature of Chinese society could change.

    Lord Kelvin’s quote also came to mind: “I often say that when you can measure what you are speaking about, and express it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meagre and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely, in your thoughts, advanced to the stage of science, whatever the matter may be”

    Somewhat surprised that no mention was made of the idea that GWAS could be done that specifically searched out for the verbal and quant IQ SNPs. Given the shift to quant at higher IQ
    levels in East Asians, Whites and Indians (present topic of this blog) and presumably Africans,
    one would be lead to assume that the reason for this universal bias is simply that more such SNPs exist. Having better visuospatial skills has through evolutionary time been of higher importance than verbal skills.

    As the article notes it seems surprising that the Terman study did not actually consider beyond merely the question of g. One would have thought the question of g for what would have been thought relevant.

    I think it is also important to note that there have been so many critical junctures in human history in which NO ONE seemed able to articulate what was happening, and what should be done to respond. Verbal intelligence can be derided when we see people who use language to create their own system of a non-rational philosophy. Yet, many times we greatly need a leader who can actually
    words well. Fortunately for England and the entire world, there was someone on hand when needed in the 1930s, namely Churchill. It is not entirely obvious how history would have worked out without him.

  57. Pericles says:
    @James Thompson

    I wonder what authors from 50 years ago would still be read if it wasn’t for the thumbs on the scales by Skool and U. Off hand, of the Brits it’s Tolkien.

  58. @res

    Piffer, as far as I know, is not working on this at the moment. However, I will see if there have been any further developments.

  59. @dearieme

    Enid Blyton’s not a great writer, but she’s a great children’s fantasy writer. Not much in the way of style, but her imagination is wonderful. The Faraway Tree books are classics, if you can find the original unbowdlerised versions.

    (Crompton is great but for older kids, Blyton is great for ages 4-7).

    • Replies: @dearieme
  60. res says:

    res, this seemed like a fairly surprising and important discovery that he made. What is your latest take on it? Was he correct? Polygenic adaptation does at the very least at a basic intuitive level make a great deal of sense: the entire genome becomes activated to select IQ SNPs and so they at some level should act jointly.

    I agree with your assessment. There are some issues–like LD and possibly different causal variants in different populations (IIRC a good example of this is trying to use European height polygenic scores on African populations)–and the results seemed a bit too good to be true to me, but I think Piffer is onto something.

    Piffer looked at height in https://f1000research.com/articles/4-15/v3
    which was mentioned in http://www.unz.com/jthompson/piffer-replies-to-prof-posthuma/
    He is quite good about about including data so you might want to take a look. I wonder if there is any chance of the height work getting any traction since it is less controversial. It would be interesting to take another look at that paper in light of the Hsu Compressed Sensing height paper.

    Did you see the Math/Verbal split article. Wow, very well written; almost disputes its own thesis. I think that’s why he threw in a few “bs”es. He covered the landscape fairly well.

    I just read it after your prompt. He did cover a lot of ground. That was the first time I heard the Feynman/Unz anecdote. I like that he referenced the article so thoroughly.

    Also some good comments. Including from a brand new commenter “Though Criminal.”

    • Replies: @Factorize
  61. dearieme says:
    @YetAnotherAnon

    My test was entirely personal. I know I read heaps of Blyton’s books – I can remember not a thing about them except expression such as Famous Five and Secret Seven. With Crompton I can remember laughing my head off, repeatedly. Mind you, your point about ages may explain a fair bit of that.

  62. Factorize says:
    @res

    Thank you for the link to the height url!

    Yes, height would be a better phenotype to demonstrate new analytic techniques instead of starting out with IQ. The height research looks solid. I am not sure why there was such a blog brawl about this 8 months ago as the height paper had been published a year before. Shouldn’t that paper have been enough to counter any rebuttals? As you noted, the dog hasn’t barked for the last 8 months so I think we’re in the clear.

    So much exciting science to be done; not enough boat rowers. It is certainly surprising that this technique has not gained more traction. Probably like much of psychometrics and especially group differences it is more that the answers are not politically acceptable.

    I am also interested to see how the next act with Compressed Sensing will unfold. What will happen when we move from 500K to 2M? GWAS research has so much momentum and everyone wants to jump on board. 2018 should be a great year of genetic discovery!

    I found it exciting that the f1000 platform appears to have the data all there. Having the data, the software, the writeup all in one place opens the potential for a floodgate of citizen science opportunities. They have found before that when science is opened to the people solutions start to emerge that the experts had not even considered.

    Running the latest mega-GWAS in height through software linked to the article would be fun! Researchers would tap into crowd energy to amplify their research impact. Why hasn’t there been an open science movement about this issue? Don’t be a spectator in life be a participant! Using the whole bench could be a game changer.

  63. res says:
    @res

    The reason I want to separate the groups is that I think the different male/female numerical balance in the high M/high V groups is what causes the almost 100 point sex coefficient in Table 1 for 0.01%. The Figure 1 SAT 0.01% plot shows more like a 10-20 point shift (for high M) as I observed above.

    Can anyone here offer a second opinion on this observation? The RQ1, RQ2 and RQ3 discussions on page 3-4/78-79 are highly dependent on interpretation of the Table 1-4 coefficients:

    4.2. RQ1: Are there sex differences in ability tilt in the right tail of cognitive abilities?
    Results from the regressions for the different tests provide evidence that tilt is associated with sex. All beta coefficients for sex were statistically significant in all models (see Tables 1 through 4). To interpret these coefficients, a reader should be reminded that the beta coefficient for sex must be considered with the coefficient for intercept in mind. A positive beta coefficient is associated with greater tilt for males toward math. Conversely a positive intercept coefficient is associated with greater tilt for females toward math whereas a negative coefficient provides evidence for a greater tilt toward verbal. For example, as shown in Table 1, in the SAT (top 0.01%), the beta coefficient for tilt is 97.57 (SE = 4.52) and the intercept coefficient was 28.82 (SE = 3.79). This suggests that males and females in the top 0.01% are more tilted toward mathematics than verbal but that males are, on average, 97.57 points more tilted than females.
    4.3. RQ2: Do sex differences increase as ability tilt increases?
    The violin plots in Figs. 1 through 4 show that sex differences increase as ability tilt increases. Further, statistical evidence is provided through an examination of the regression coefficients. The regression coefficients for the top 0.01% for each testing measure are outside of the confidence intervals of the top 5% and top 1%. This provides strong evidence that the regression coefficients are statistically different and greater. For example, the confidence interval for the beta coefficient for the math tilt for sex in the regression for the top 0.01% of SAT takers is 88.23 to 106.11. Conversely, the confidence interval for the top 1% is 29.36 to 30.06. Clearly these confidence intervals do not overlap.
    4.4. RQ3: Do sex differences in ability tilt increase as ability selectivity increases?
    As shown in Tables 1 through 4, in all cases (save one), the beta coefficient for sex increased as selectivity increased. This suggests that, as ability selectivity increases (top 5%, top 1%, top 0.01%), males were increasingly tilted toward mathematics than females. This provides evidence that the presence of ability tilt in a population is associated with ability selectivity. This is illustrated in the split violin plots in Figs. 1 through 4; the shapes of the violin plots change as ability selectivity increases. In particular, the distribution of tilt shifts more extreme. These plots provide a visual
    confirmation that ability tilt increases as selectivity increases. The only case where the coefficient for sex did not increase as ability selectivity increased was for the ACT from the top 5% to top 1%. An examination of the violin plot does suggest a difference in the distribution of tilt. A further examination of coefficients suggests that the sex differences are explained through the interaction of sex and year given the relatively larger coefficient of 0.08.

    I don’t think these coefficients correctly represent the relative male and female ability tilts because the study population is not a homogeneous group. It is a combination of a high M and high V group. These subgroups have different tilt characteristics, and more importantly, different male/female population balances. This effect is most important for the top 0.01% group because it has the most extreme tilts AND the largest difference in male/female balance in the subgroups. If this effect exists it clearly matters for the sex coefficients. The effect on the other coefficients is unclear to me–which is why I think a high-V and high-M subpopulation reanalysis should be done. A look at the 1% and All panels of Figure 1 show a similar magnitude of tilt as my 25 point (or less) estimate below so I also have some concern about the strength of the relationship between tilt and selectivity.

    To be clear, my sense is that correcting for this issue would not change the abstract conclusions, but it might change the magnitude of the effects claimed. In particular, I think it is hard to reconcile a 97.57 tilt sex coefficient for the SAT top 0.01% in Table 1 with the violin plot for that group in Figure 1. The right side of that plot looks more like a 25 point (or less) sex difference in tilt based on the offset modes.

    Does anyone see problems with my assessment?

    P.S. A hypothetical example to make this more concrete below.

    [MORE]

    Assume two male and female subgroups with equal tilts but different populations:
    high-V: tilt for both sexes is -1 and populations are male 100, female 200
    high-M: tilt for both sexes is +1 and populations are male 200 and female 100
    I believe the apparent sex difference in tilt seen by a combination of these populations would be
    (based on the male average tilt of +1/3 and the female average tilt of -1/3 in the combined group)
    2/3 which is clearly spurious (and also happens to be of quite large magnitude!).

    This compares to the real data, where a male/female difference in tilts clearly exists, but I believe the size of that effect is less than claimed due to the effect elucidated above.

    Criticism welcomed.

    • Replies: @res
    , @res
  64. res says:
    @res

    Some additional thoughts which occurred to me while writing that comment.

    First, to help put the tilt magnitudes being discussed in perspective, looking at Table 1 of the Dorans paper linked above we see post-recentering average tilts of:
    male: 520 (M) – 505 (V) = 15
    female: 482 – 495 = -13
    and pre-recentering:
    male: 497 – 427 = 70
    female: 454 – 417 = 37
    So the tilt difference by sex shrank slightly with recentering from 33 to 28. Contrast that 28 with my 25 estimate above and the 97 Table 1 0.01% coefficient.

    The math and verbal SDs are around 110 so the tilt differences are on the order of 1/4 SD of the individual scores.

    Here is a look at SAT math scores over the last 40 years: http://www.aei.org/publication/2015-sat-test-results-confirm-pattern-thats-persisted-for-40-years-high-school-boys-are-better-at-math-than-girls/

    Second, I think it is important to consider that the TIP SATs were administered in 7th grade (around age 13) rather than the usual 11th grade (around age 17). Those are important developmental years and I think it is questionable to extrapolate sex differences observed at age 13 to either age 17 or adulthood. Does TIP have high ceiling tests for their sample administered later in life? It might be instructive to compare those results with the earlier tests (for tilt and more).

    Here are some papers on sex differences in children, but I am sure there are people here better acquainted with that literature than I am.

    https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2867482/

    https://www.sciencedirect.com/science/article/pii/S0160289616303154

  65. I had the pleasure of access to an unexpurgated versions n of Grimms’ Fairy Tales. Definitely a bays book. Enid Blyton had no appeal after that. Crompton was brilliant though. William Ballantyne (Coral Island) is a once best selling author who has disappeared from sight. His books described fisticuffs between boys and young men as exciting and desirable.

    • Replies: @dearieme
  66. @dearieme

    If you’ve read his more recent books would you like to recommend one particularly?

    Risk Savy – How to make Good Decisions, Penguin, New York 2013 is a valid entrance ticket into Gigerenzer’s arguments and style.

    Gigerenzer reminds me of Semmelweis because what he finds is often strinkingly simple, and very effctive.

    Unfortunately, the world still hasn’t reallyacknowledged the blessings of Laozi, Occam and Wittgenstein (and Brecht, even: The simple things, which are so hard to achieve…).

    • Replies: @dearieme
  67. dearieme says:
    @Dieter Kief

    Thank you. That should nicely complement my wife’s book: Risk Gravy – How to make Good Pies.

    • Replies: @Peripatetic commenter
  68. res says:
    @res

    I think the preprint Appendix C table in conjunction with the Figure 1 violin plots have enough information to evaluate my idea further with real numbers,but it would be helpful to hear any critiques of my analysis before I do this.

    One thing I found interesting is that in the preprint we see the topline conclusion: “As ability tilt increased, so did sex differences in ability tilt.” Despite that, the Table 5 RQ3 significance tests do not show this. Any thoughts on that oddity?

    Another thing I find odd is that to my eye the Figure 1 violin plots are consistent with a male/female difference in tilt of about 25 points across the ability ranges (close to the average tilt difference from Doran’s). However, the Figure 2 ACT violin plots do appear to show varying tilt differences with ability: From ~2 to ~5 to ~2-5 as ability increases. I am not sure what would cause the 5 differences (e.g. 1% modes and 0.01% left modes), but 2 is close to the average sex difference in ACT (M – AVG(V,E)) tilt indicated by this page: 1.1 – ((-0.8 + -0.4)/2) = 1.7

    http://www.aei.org/publication/gender-differences-on-the-act-test-boys-score-higher-on-math-and-science-girls-score-higher-on-english-and-reading/

  69. res says:

    Looking into this further, the most relevant TIP papers I see are:

    Wai, J., Putallaz, M., & Makel, M. C. (2012). Studying intellectual outliers: Are there sex
    differences, and are the smart getting smarter? Current Directions in Psychological
    Science, 21, 382–390. http://dx.doi.org/10.1177/0963721412455052

    Wai, J., & Putallaz, M. (2011). The Flynn effect puzzle: A 30-year
    examination from the right tail of the ability distribution provides
    some missing pieces. Intelligence, 39, 443–455. https://doi.org/10.1016/j.intell.2011.07.006

    Wai, J., Cacchio, M., Putallaz, M., & Makel, M. C. (2010). Sex differences in the right tail
    of cognitive abilities: A 30-year examination. Intelligence, 38, 412–423. http://dx.doi.org/10.1016/j.intell.2010.04.006

    I find it unclear if the 1995 SAT recentering was consistently accounted for across the entire series of papers. It is mentioned in the current and 2011 papers, but I did not see any mention in the 2010 and 2012 papers. The effects of the recentering can be seen quite clearly in the Appendix B table covering SAT verbal scores in the 2010 paper. The verbal section changed much more in difficulty and the number of people meeting the 700 verbal threshold in each 5 year bucket (for each sex) changed from 100 post-1995. Even if the recentering is accounted for it still biases the sex balance of the earlier samples (fewer females) because the verbal section was so much harder. I think that change in sex balance will have an effect on the time coefficients in the current analysis.

    It is worth mentioning that the >=700 SAT V group is almost equally balanced by sex. IMHO this is obscured by using density for the violin plots (density is better in some ways though).

    If I interpret the 2010 Appendix A and B tables correctly (the full sample >= 200 is intended to be top 5% ability, i.e. 1 in 20) then the overall selectivity thresholds were roughly:
    pre-1995 verbal – 1 in 200,000 (!!!)
    pre-1995 math – highly variable (did the test decrease in difficulty from 1981-1995?) males from 1 in 15,000 to 1 in 7,000, females from 1 in 200,000 to 1 in 20,000
    post-1995 verbal – 1 in 20,000
    post-1995 math – again variable, most recently males 1 in 3,000, females 1 in 12,000

    Is there an update of the 2010 Appendix A and B tables covering through the 2015 end date of the current paper?

    To emphasize how much sex balances of the high-V and high-M group might skew results over time lets take a look at the number meeting the thresholds by sex in two of the 5 year buckets. The later 5% samples were about 10% larger overall.

    bucket_____V 700 males__V 700 females__M 700 males__M 700 females
    1991-1995_____7___________9__________271_________70
    2006-2010____117_________134_________628_________164

    So we see both a dramatic drop in selectivity from 1 in 525 of the top 5% to 1 in 200 of the top 5% and an even more dramatic change in high-V vs. high-M balance with a resultant change in the sex balance. Even in the current more balanced groups we have 3x the number of high-M vs. high-V.

    • Replies: @Factorize
  70. dearieme says:
    @Philip Owen

    I remember “Coral Island”. And “Swiss Family Robinson” and “Children of the New Forest” and “Emil and the Detectives”. Or, more accurately I remember the titles, and that I enjoyed them and re-read them. I also remember The Water Babies, an absolute stinker.

    But we are digressing rather a long way from “tilt”. Happily Doc T is an accommodating host.

    • Replies: @YetAnotherAnon
  71. Factorize says:
    @res

    res, I have not went in depth into the methodology of the recent paper, though I wanted to make comments that you might find useful. I snipped the SAT plots into paint and used the pixel measure feature to have an accurate measure of the distribution peaks for males. I found 30.1, 52.7 and 169.4 male tilts for All, 1% and 0.01%. Due to the non-symmetric nature of the distributions for 1% and 0.01% the actual average math tilts would be less.

    I wonder why they did not include the distribution of the difference in the panels. When you casually look at the top panel plot for the SAT, it might not be immediately obvious that while the peak of the positive math tilt for men and women occurs near the same tilt level, though much of the actual distribution for women is pushed to a verbal tilt.

    • Replies: @Sparkon
  72. EH says:
    @res

    I think the figures are combining two separate tests, V and M, and giving a rarity cutoff using the combined score of the two despite them having very different ceilings. It used to be around a standard deviation higher ceiling for V, though it lost a lot of top when they dropped analogies. The top 1 in 10,000 that tilt toward verbal on the top chart would then be actually more like top in in 100,000 in observed rarity of verbal performance but perhaps only 1 in 100 in math.

    • Replies: @res
  73. Has everyone seen this?

    https://www.spring.org.uk/2013/12/connectivity-the-difference-between-mens-and-womens-brains.php

    A fascinating new study on the brains of 949 young people finds striking gender differences in the brain’s connectivity between males and females (Ingalhalikar et al., 2013).

    These may help explain some of the classic psychological differences between men and women.

    To me it is significant that there is something physically measurable.

    The differences in connectivity begin to show up around puberty.

    • Replies: @res
    , @another fred
  74. res says:
    @EH

    Some good observations, but I think rather than combining the tests they are using the same threshold (700 on the post-1995 SAT scoring scale, with the earlier scores translated to the later scale) for each individual test both before and after recentering. I focus on the top 0.01% because the extreme average tilts of each subgroup there means it most effectively demonstrates the population size effect I am proposing.

    It is worth emphasizing that post-recentering the SAT subtest scores are much closer to equivalent difficulty than they were before 1995. The point of the recentering was to set the mean performance for both tests at the center of the scoring range-500. However, the means have moved a bit since then: https://blog.prepscholar.com/average-sat-scores-over-time

    For the top 0.01% the current paper cites Lubinski et al. (2001)
    Lubinski, D., Webb, R. M., Morelock, M. J., & Benbow, C. P. (2001). Top 1 in 10,000: A 10-year follow-up of the profoundly gifted. Journal of Applied Psychology, 86, 718–729. http://dx.doi.org/10.1037/0021-9010.86.4.718
    and then says

    Initial score benchmarks were drawn for the top 0.01% from Lubinski et al. (2001), and translated into current cut scores for the SAT (SAT-M 700+ or SAT-V 700+; female = 1472, male = 3451).

    Here is what Lubinski et al. (2001) says about the original thresholds. I include a long quote because I think it is informative. Does anyone know if that 1998 J.C. Stanley personal communication has ever been published?

    All 320 participants (78% Caucasian, 20% Asian, 2% other) in this 10-year follow-up study secured scores that were either >=700 in the math portion of the SAT (SAT-M) or >=630 in the verbal portion of the SAT (SAT-V) before age 13 (1980-1983). For members of this age group, these cutting scores constitute a selection intensity of about 1 in 10,000 in mathematical and verbal reasoning ability, respectively. The IQs of the participants were estimated from sample statistics that were collected on hundreds of thousands of talent search participants compiled over the past two decades (raw data, J. C. Stanley, personal communication, June 1998). Talent search participants who took the SAT consisted of a sample of approximately the top 3% in general ability for ages below 13 years. Means (and standard deviations) for these adolescents’ SAT-M and SAT-V scores
    were approximately 430 (SD = 85) and 370 (SD = 75), respectively. We assumed that adding these two mean values (430 + 370 = 800) approximated the cutting score for the top 1% (z-score = 2.32) on the general factor. Given that the correlation between SAT-M and SAT-V for talent search participants is around r = .55, we estimated their standard deviation on general intelligence on the basis of their SAT-M + SAT-V composite to be as follows: [(85)2 + (75)2 + 2(.55)(85)(75)]“2 = 140.93. At this point, each student’s general ability level was estimated by subtracting 800 from their SAT composite, and dividing this difference by the standard deviation (140.93) to reflect the number of standard deviation (z-score) units that needed to be added to 2.32 to estimate their normative standing on general intelligence in z-score units. Finally, this value was multiplied by a conventional IQ standard deviation (viz., 16) and added to 100 to estimate IQ on the familiar metric. Once we performed these computations on our participants’ scores, we found the mean and standard deviation to be 186 and 11, respectively (with 99% of these estimates >160).

    Based on the official equivalence tables I think the new thresholds should have been 690/690: https://research.collegeboard.org/programs/sat/data/equivalence/sat-individual
    But I don’t know if the TIP group has better data or what.

    • Replies: @EH
  75. res says:
    @another fred

    Interesting. Thanks.

    As you observed, one point from that article which is relevant to my comment 65:

    They found that there were few differences between males and females before the age of 13, but that the different patterns of connectivity kicked in at puberty.

    • Replies: @dearieme
  76. dearieme says:
    @res

    Hormones are a social construct.
    (A friend once heard this said at a Social Sciences seminar.)

  77. I used to go to math competitions when I was a kid. And the winners were always boys. Once in a while a girl might get to the top three – but that was rare, and got rarer as we matured.

    What was particularly striking about this inescapable reality was that the boys and the girls took EXACTLY the same classes and were treated EXACTLY the same way. And competitions are anti-discriminatory by definition – mathematics can not be sexist or racist – you either solve the problems or you don’t.

    The inescapable – and obvious to anyone who cares to look – conclusion is, as Thompson says, that men are overwhelmingly superior to women at the top level of mathematics. And that conclusion carries strong implications – it explains why the tech industry is so male-focused. Women, of course, have their own areas of superiority. (Never hear anyone complain that most primary school teachers are women.)

    It is disastrous for modern society that it denies these obvious differences between men and women. You can’t change or deny human nature. It will always prevail.

    • Replies: @Bardon Kaldian
  78. utu says:

    When the Level of Ability (how is it defined?) goes to smaller fractions, say top 0.1%, top 0.01%,… the range of tilt=SAT_math-SAT_verbal should get narrower until we get Dirac delta at tilt=0 (which has zero support) where all geniuses are who maxed out on both SAT_math and SAT_verbal. I do not understand why the range of tilt seems to get wider here when you go from top 5% to top 1% to top 0.1%. What we see here might be possible if the Level of Ability is not measured by SAT=SAT_math+SAT_verbal but by some other measure that is skewed towards SAT_math. But then it would mean that in the top 1% verbal ability might be actually higher than in the top 0.1%

    Q: Isn’t this violin plot just a histogram or some pdf that smoothes the histogram?

    • Replies: @res
    , @res
  79. I think the vast majority of people are not intelligent enough to a) get into life situations where the female / male IQ differences matter b) actually observe that females might be on average (in this very small sub groups) less intelligent. So for the majority of people the right rule of thumb is that women and men are equally intelligent.

  80. @jimbojones

    It is disastrous for modern society that it denies these obvious differences between men and women. You can’t change or deny human nature. It will always prevail.

    It is not “modern society”, just its Teutonic-Celtic branch (Anglo-Saxony, the Netherlands, Germany, Scandinavia…).

    On one hand you got this:

    and, on the other- this:

    • Replies: @another fred
  81. Well,

    the numbers are partly going to skewed by population size.

    The number of x skilled mathematicians from which to sample are going to outsize women and while I am sure there’s a control for that, I don’t see it addressed in the study.

    Given both the socialization of men, in this case western men verses the socialization of women, whereby men are pressed to solve problems of a technical nature and historically women have had not had that expectation —I am not sure there is much to debate about.

    is a plumber smarter than a math instructor? That all depends on the depth of training, and experience they have at solving issues in their field.

    The volume of men concentrated in the field of math exceeds the number of women. It would not be a genius jump for a one to note that in the process of ironing sharpening iron more men would develop exceptional skills in greater numbers than women acquiring exceptional skills.

    I do all of the technology around here. And its really irritating at times to be yanked away from something I am doing to fix a computer glitch or reboot a system — install a program, use my housemates email to confirm some technical order, fix this or that only to be told I am not listening to why I was called in the first place.

    I have been around long enough to know that more essential problems are solved by improving communication skills, i.e. listening than math skills. As numerically, communication occurs far more than the application of mathematics. I am willing to grant that women are better at interpersonal kills than most men. And far fewer issues result from lack of mathematics than lack of the vast array of communication skills required to navigate life.

    It doesn’t really matter that the likelihood of being shot by another person of less than one percent, apparently, women still want to talk about the risks of gun ownership.

  82. JosephB says:

    Just a quick note that the comic is from xkcd, and the copyright (https://xkcd.com/license.html) requires a note of where you got it from.

    • Replies: @res
    , @James Thompson
  83. @Bardon Kaldian

    It is not “modern society”, just its Teutonic-Celtic branch (Anglo-Saxony, the Netherlands, Germany, Scandinavia…).

    2,000 years ago one of the Roman historians commented that one of the distinguishing (and odd) aspects of the Gauls and Germans was that they allowed women to speak in council.

    I’ve worked in 21 countries in Latin America, Europe (including FSU), the Middle and Far East and the North European cultures are by far the least “male dominated” (Chinese second).

    • Replies: @Bardon Kaldian
  84. @another fred

    For those who are not familiar, connectivity is being found to be strongly associated with IQ. This science is still in its infancy so how strongly and exactly how this “novel technique” is “associated” is “developing news”.

    “We saw a clear link between the ‘hubbiness’ of higher-order brain regions and an individual’s IQ,” Seidlitz said.

    “This makes sense if you think of the hubs as enabling the flow of information around the brain — the stronger the connections, the better the brain is at processing information.”

    http://www.sci-news.com/othersciences/neuroscience/iq-brain-connectivity-05607.html

    Also:

    http://bigthink.com/philip-perry/yale-neuroscientists-can-now-determine-human-intelligence-through-brain-scans

    • Replies: @res
  85. Sparkon says:
    @Factorize

    I have not went in depth into the methodology…

    go went gone

    I have not gone

    It’s a common mistake, but one that grates on grammarians.

  86. @another fred

    (Chinese second)

    Chinese? Aren’t they androcracy? No woman close to power, as far as I can see….

    • Replies: @another fred
  87. @Bardon Kaldian

    No woman close to power, as far as I can see….

    True, as far as power structure is concerned, but their voices are strong in the home and small businesses.

    • Replies: @Bardon Kaldian
  88. res says:
    @utu

    Q: Isn’t this violin plot just a histogram or some pdf that smoothes the histogram?

    Kind of. A “violin plot” is essentially a smoothed density histogram. Or looked at another way, a more sophisticated box plot (see variants at first link below which also include box plots). I like the variant they use with different samples back to back. Though some just prefer overlapping density plots. Does anyone know of a good way to simultaneously convey both density and absolute numbers in the comparison? That is the one shortcoming I see with the paper presentation. Perhaps just adding sample size numbers and a few grid lines on the plot?

    There is much more about violin plots on the net. Some sample links:

    http://www.sthda.com/english/wiki/ggplot2-violin-plot-quick-start-guide-r-software-and-data-visualization

    https://stats.stackexchange.com/questions/28431/what-are-good-data-visualization-techniques-to-compare-distributions

    • Replies: @utu
  89. @another fred

    Well, OK, but Chinese culture is historically typical patriarchy. It is not stupid machist, shallow swaggering macho rule, but still-men rule. Not as men, but as persons.

    By the way, what’s wrong with real females in power? Mrs. Thatcher was the best post-WW2 British prime minister. I say- all the power to real women & off with crazy feminists, faggots (not homos) & other deviants who are so determined by their sexuality & genitalia they don’t see there is a real world out there.

  90. @res

    Interesting paper. Thanks! I think the SAT tilt for the top 0.01% plot in Figure 1 above merits some additional discussion.

    You can say that again!

    I just glanced at the top violin plot and said “no way”. There’s a range restriction problem with the SAT. Smart–0.1% people–will by simple necessity of being the 0.1% be over 700 in both tests capping the skew at 100. (Actually i’m pretty sure they’d need to be over or near 750–my son was right on the cusp of this with a 740 verbal keeping his scores probably more at the 1/500 or so ballpark, rather than 1/1000.) So there is no way you can have anything like those “tilts”.

    Even if 0.1% was “they scored in *either* the math or verbal at 0.1% level” … you’d have a few math 800, not that good verbally nerds, but you still wouldn’t see such a high tilt.

    Then i read your comment below–13 year olds. LOL. That’s a completely parentally selected–”oh my kid’s a little genius”–population. And then with that, they took *either* at 700V or 700M as evidence of 0.1%. That either will inherently will tend to produce a bi-modal distribution. And, of course, the SAT math test doesn’t require super-duper math. (I would probably have gotten 700 or near it at 13.) And “my kid is really good at math” is the sort of marker for these “my kid should take the SAT now” “prodigies”. So it’s utterly unsuprising to see both a math tilt and to see it even for girls.

    ~~

    I do believe that in general, relative to the average, smart people are indeed smarter in math. I.e. it takes a little brain power to think mathematically, but once you get going in smarts you cover “more ground” so to speak mathematically than verbally as you add smarts. But we definitely have a lot of “verbalists” out there in our SAT selected elite. From the output of mainstream elite journalists and politicians it’s clear these are not people prone to thinking mathematically. (In fact, reading a lot of their output, you’d think they are screened for aversion to mathematical thinking.)

    However, because the math/verbal tests are separate, this “smart people more mathematical” distinction is actually somewhat arbitrary. (Relative to what?) And in terms of the SAT, it’s obvious that the math test is “easier” for a smart person. (More high scores.) To really be able to announce that smart people tilt more toward math ability, you’d want some test that is not range restricted where the mean and standard deviation were both the same, but then saw increasing skew toward math as you went into the right tail. But it’s still a bit hard to cleanly be able to determine that it’s the humans that are skewing and not just the difficulty of the “tougher” questions. The problem remains that the math and verbal tests are indeed separate.

    What you can tell from the SAT and lots of other tests is that relative to each other:
    – there are progressively more men as you move toward the right tail
    – on average men are quite a bit smarter than women in math at the right tail (i.e. several times more men with the highest scores or for say the 99%tile for each sex, the men are significantly better in math than the women)
    – men and women in the right tail are close to parity on verbal (i.e.if you take the 99%tile in verbal there’s a pretty equal distribution of the sexes, which is not true at lower levels)

    Boiled down:
    – men have a higher variance than women; more smart and dumb men than women
    – relative to each other men are better at math, women at verbal

    In other words, we know what we’ve known for a very long time.

    • Replies: @res
  91. res says:
    @utu

    When the Level of Ability (how is it defined?) goes to smaller fractions, say top 0.1%, top 0.01%,… the range of tilt=SAT_math-SAT_verbal should get narrower until we get Dirac delta at tilt=0 (which has zero support) where all geniuses are who maxed out on both SAT_math and SAT_verbal. I do not understand why the range of tilt seems to get wider here when you go from top 5% to top 1% to top 0.1%.

    I get your point about the limit, but practically speaking I don’t think that is true based on both the data and my intuition. M and V don’t correlate perfectly. 0.01% is about 4SD and those who are +4SD on both variables are even rarer than that (has TIP ever published how many of their subjects met both thresholds?). The data clearly show increasing tilt with single ability thresholds in the violin plots. I suspect any limit effect observed would just be an artifact of the test ceilings.

    One interesting question (RQ3 in the paper) is “Do sex differences in ability tilt increase as ability selectivity increases?” The paper concludes yes, but I think my comments above raise some question with that (particularly a visual inspection of the tilt seen in the 0.01% SAT-M portion of Figure 1). I think it would be appropriate to reanalyze the data separating the high-M and high-V subgroups to double check the answer to RQ3 and more accurately estimate the magnitude of that effect (and some of the effects in the other RQs).

    P.S. As mentioned multiple times above and described in the paper (see section 3.3 and associated references), “Level of Ability” is defined by frequency based score thresholds for the post-1995 recentering SAT subtests and similar thresholds for other tests.

    • Replies: @utu
  92. utu says:
    @res

    Any thought about my observation on the range of the tilt? How is it possible that in top 0.1% group math and verbal SATs can differ by that much?

  93. res says:
    @JosephB

    Good point. I had trouble telling if it was sufficient to embed their image URL:

    Or if the license implies more is needed. The comic page itself is at https://xkcd.com/385/

  94. res says:
    @another fred

    Thanks for the interesting links. But aren’t MRI studies like this considered somewhat unreliable at this point due to the small sample sizes?

    From your first link:

    In the study, the researchers compared the brains of 296 typically-developing adolescent volunteers.

    Their results were then validated in a cohort of a further 124 volunteers.

    They showed that if two regions have similar profiles, then they are described as having ‘morphometric similarity’ and it can be assumed that they are a connected network.

    They verified this assumption using MRI data on a cohort of 31 juvenile rhesus macaque monkeys to compare to ‘gold-standard’ connectivity estimates in that species.

    From your second link:

    126 participants, all a part of the Human Connectome Project, were recruited.

    Here is a 2010 mention of reproducibility problems with MRI studies: https://www.nature.com/news/2010/100317/full/news.2010.129.html

    Does anyone here have a good sense of the current state of the art in MRI and fMRI brain studies linking to other variables?

    • Replies: @James Thompson
  95. res says:
    @AnotherDad

    I just glanced at the top violin plot and said “no way”. There’s a range restriction problem with the SAT. Smart–0.1% people–will by simple necessity of being the 0.1% be over 700 in both tests capping the skew at 100. (Actually i’m pretty sure they’d need to be over or near 750–my son was right on the cusp of this with a 740 verbal keeping his scores probably more at the 1/500 or so ballpark, rather than 1/1000.) So there is no way you can have anything like those “tilts”.

    It’s 13 year olds. The SAT has more ceiling there. And the pre-1995 verbal SAT had a great deal of ceiling for that group. How old was your son when he got those scores? Before or after 1995?

    The frequency tables in the appendices of
    Wai, J., Cacchio, M., Putallaz, M., & Makel, M. C. (2010). Sex differences in the right tail of cognitive abilities: A 30-year examination. Intelligence, 38, 412–423. http://dx.doi.org/10.1016/j.intell.2010.04.006
    make clear how uncommon the over 700 scores are. An excerpt from my comment 70 above:

    bucket_____V 700 males__V 700 females__M 700 males__M 700 females
    1991-1995_____7___________9__________271_________70
    2006-2010____117_________134_________628_________164

    The sample sizes (and keep in mind these numbers were the preselected top 5%!) were:
    years_______males__females
    1991-1995____93662___93787
    2006-2010___103868__103559

    In no 5 year sex group did more than 100 score 800 on either subtest. In 1981-1985 there were exactly 0 800s. In 1981-1995 there were exactly 0 verbal 800s, 0 female math 800s, and 8 male math 800s.

    That’s a completely parentally selected–”oh my kid’s a little genius”–population.

    I know someone who was screened for the study (i.e. took the SAT at 13). The initial selection (top 3 or 5%) was teacher and test score (relatively lower ceiling in-school tests) based. The final ability levels were determined by the SAT tests. A 5% initial threshold means that anyone close to 0.01% should pass the initial selection.

    And then with that, they took *either* at 700V or 700M as evidence of 0.1%. That either will inherently will tend to produce a bi-modal distribution.

    I strongly agree with this (except it is 0.01%). It is the fundamental point underlying most of my comments above.

    And, of course, the SAT math test doesn’t require super-duper math. (I would probably have gotten 700 or near it at 13.)

    I did well on my SAT, but I think my lack of geometry and higher level algebra would have hurt me at 13. I am much less confident about my ability to score 700 then. The accelerated schooling aspect is one area where I think your parental comments are on target (another area is around the 5% initial threshold). I think the SMPY/TIP selection methodology has some undesirable dependence on early schooling. But that is a small issue with what is the best study I know of high intellectual ability people. I might seem overly critical in this thread, but that is because I think this data is the best chance we have of answering some of these research questions and I want it to be analyzed as correctly as possible.

    Boiled down:
    – men have a higher variance than women; more smart and dumb men than women
    – relative to each other men are better at math, women at verbal

    In other words, we know what we’ve known for a very long time.

    IMHO the present paper asks questions which go beyond that. I think the answers they gave are basically on target but I have concerns about the effect size accuracy. Reproducing their original research questions from section 2.:

    RQ1 : Are there sex differences in ability tilt in the right tail of cognitive abilities?
    RQ2 : Do sex differences increase as ability tilt increases (distance between math and verbal scores increases)?
    RQ3 : Do sex differences in ability tilt increase as ability selectivity increases (top 5%, top 1%, top 0.01% of academic ability)?
    RQ4 : Have sex differences in ability tilt changed over time?
    RQ5 : Do sex differences in ability tilt vary as a function of measure and cultural context?

  96. Factorize says:

    res, this is a great figure from an infoproc post on April 19,2016 showing how top level M/V pushes people in certain college and career tracks.

    • Replies: @res
  97. AndrewR says:
    @Anon

    How did all those genius men come to find themselves utterly henpecked by Strong Independent Wymyn Who Don’t Need No Men?

  98. this is super rare. the comments section of this article is as good as the article it self.

    thanks you all.

    • Replies: @James Thompson
  99. res says:
    @Factorize

    It is a great figure. I assume you noticed that the arrows represent spatial ability?

    Those plots along with additional discussion appear in this 2009 paper by Wai, Lubinski, and Benbow: https://my.vanderbilt.edu/smpy/files/2013/02/Wai2009SpatialAbility.pdf

    One quote from that paper which intrigues me:

    Finally, sex differences in relative levels of interests are important to take into consideration. Although the covariance structure of specific abilities and interests is comparable for males and for females, the sexes display mean differences in a number of interests; for instance, spatially talented females tend to be more interested in artistic pursuits than are spatially talented males and the inverse is true for engineering and mechanical activities (Lubinski & Benbow, 2006; D. B. Schmidt et al., 1998). These mean sex differences in interests correspond to findings, shown in
    Figure 8, that spatially talented females were more likely than similarly talented males to pursue artistic domains. These proclivities can and do change over time, but relative levels of interests (and competing interests) are always important to take into account (Geary, 1998, 2005; Gottfredson, 2003, 2005)

    I wonder if anyone has looked at math/spatial tilts across different groups. I wonder if that might be part of what is driving their observation above (e.g. both math and spatial needed for engineering) or if that is all caused by non ability related interests.

    P.S. Another paper which discusses sex differences, tilt, spatial ability, and those plots: https://my.vanderbilt.edu/smpy/files/2013/02/ScienceSexDifferences.pdf
    Much interesting material in that paper. This caught my eye (page 26):

    It should be noted that despite the significant sex difference in spatial performance, most women in R.C. Gur et al.’s (1999) study performed comparably to the men on the spatial tests. As suggested in several sections of this monograph, it is possible that (some) females may achieve high levels of spatial performance using different strategies than males and possibly by using different regions of the brain. Haier et al. (2005) also found that males and females may solve some complex problems, such as items on IQ tests, differently, with females showing a greater use of language-related brain regions and males showing greater use of spatial-related brain regions.

  100. For any mathematical issue, my wife refers to me (but I did run a derivatives desk). She, however , speaks three times as many words a day as I do.

  101. utu says:

    Hello, Anybody here could explain the range of tilt for top 1% . According to

    https://blog.prepscholar.com/sat-percentiles-and-score-rankings

    99% score covers SAT composite score range is 1550-1600. This means that the range of tilt must be within -50 to +50. One can’t have tilt=100 if one is in top 1%. Now look at Fig. 1 from Wai et al. and you can see that his “violin” plot for top 1% has values ranging form -200 to +300. And the range for top 0.1% is even wider from -300 to +400. How is this possible? How one can be in top 0.1% and be 300 points more stupid in math than verbal or vice versa?

    The range of tilt should be getting narrower as Level of Ability increases (x in top x% is getting smaller). So, what is wrong with Wai? Or what am I missing?

    • Replies: @res
  102. utu says:
    @res

    No, the paper is screwed up. You can’t be in to 1% and have difference larger then say 50 between maths and verbal.

    The data clearly show increasing tilt with single ability thresholds in the violin plots.

    The data shows that results are impossible. Wai screwed up! He fell in love with violin plots and forgot what he was actually doing. He even gives citations to violin plots as if it was important. This guy may not know what his priorities should be.

    • Replies: @res
  103. utu says:

    Wai’s Trick Explained

    I have already explained why bimodal distributions in Fig. 1 are impossible when one defines the Level of Ability on the basis of

    Level of Ability=SAT composite score= SAT_math+SAT_verbal

    But one will get a bimodal distribution if Level of Ability is defined as an alternative: either verbal or math like for instance this

    Level of Ability=max(SAT_math, SAT_verbal)

    This guarantees the distribution to be strongly bimodal. So in Fig. 1 top 1% are all those who are top 1% in math or all those who are to 1% in verbal. This means that there is a lot of “morons” in this group who are very bad at math or very bad in verbal test. This explains wide range of tilt in Fig 1.

    I did search of Wai paper for the Level of Ability to see if he defines it. I could not find it. In Section 3.3 he states that top 1% were SAT-M 430+ or SAT-V 450+ which suggests that indeed he used a logical sum of scores not an arithmetic sum of scores.

    Why Wai is so shy about explaining what he is doing when his striking result (bimodal shapes) hinges on the definition of the Level of Ability? I am not saying that his work is incorrect but clearly he is sneaky and tries to hide the fact that his result is forced by the definition. The result is just a manifestation of the definition. It could not be otherwise. But apparently nobody pays attention to it because everybody is concentrating on the claim that girls are different than boys. But what is really the conclusion of top 0.1% graph? That in this group there are many girls who are awful at math and even more boys who are awful in verbal test. But can one generalize this when thinking about smart boys and smart girls? No, because we do not consider somebody smart who is hopeless in math or in verbal tests. So I think Wai guy is a manipulative cheat. Another sad case for the so called science.

    This seems to be another paper with dubious ethical standards (Davide Piffer’s was previous one) that Mr.Thompson decided to popularize.

  104. res says:
    @utu

    Or what am I missing?

    The SMPY/TIP uses SATs administered at 13 years old rather than 17. The top 1% cut scores for this paper are SAT-M 430 and SAT-V 450. So it is possible to have a +-300 point tilt and still be in the top 1% on the worse performing test.

    For a concrete example of an extremely large tilt consider Terence Tao. At 8 years 10 months he took the SAT and scored 760 math and 290 verbal for a tilt of 470!

    http://www.davidsongifted.org/Search-Database/entry/A10116

    One thing is that he took the test in 1983 so his converted scores

    https://research.collegeboard.org/programs/sat/data/equivalence/sat-individual

    would be 770/370 for a tilt of 400 as computed by this paper.

    Two additional comments.

    1. The SAT portion of the study has an effective sample size of about 25 million people (1.3e6 * 20x initial selection factor). With that many people you can see some extreme cases.

    2. As you observed earlier, the violin plots are smoothed histograms so it is unclear what the exact behavior in the tails really is. The preprint version of the paper (linked in an earlier comment) has more numbers than the final version. I was unable to find the max/min tilts observed, but it does have counts for the number over 300 and under -300 for each of the groups (see Appendices A-C).

    • Replies: @utu
  105. I think many of are considering stats from the days when calculated the numbers via a calculator. Given the degree that computers distribute the numbers it is unlikely said differences can be explained without creating another model and even then —

    These are precise computer calculations, with variances so precise and minuscule, trying to decipher or explain the differences — require another set of equations of multiple variables considered. Intelligence, and critical thinking is just not static enough to make certain conclusions as narrow as some of you desire.

    But then the model must be accurate. Maybe I am missing the confidence testing data.

  106. res says:
    @utu

    No, the paper is screwed up. You can’t be in to 1% and have difference larger then say 50 between maths and verbal.

    See my comment just above.

    Can you please chill on the arrogance?! How many times does someone have to explain to you why you are wrong about something before you figure out that maybe you should moderate your language about other people when you don’t fully understand something?

  107. bjondo says:
    @Habakkuk Mucklewrath

    Larry Skunk should be fired, removed from America.

  108. utu says:
    @res

    It is not “SAT-M 430 and SAT-V 450″ but it is “SAT-M 430 or SAT-V 450.” This trick gives him strongly bimodal distributions and very wide tilt range. See my previous comment.

    • Replies: @res
  109. res says:
    @utu

    Fair enough on the “or” correction ( I was referring to the cut scores, not how they were used, but more clarity is good).

    I agree with you about the bimodal distribution (but not about it being a “trick”). I discussed that a long time ago in comment 12. And Factorize mentioned it even earlier in comment 8.

    The wide tilt range is a feature of the data. Using separate thresholds emphasizes that by removing the group of people with similar overall ability by combined cut score who should have smaller tilts.

    I am most concerned with the effect of the separate thresholds and lumping together the disparate high-M and high-V groups on the regression analysis results. I really wish someone in this thread would engage with that point.

    • Replies: @utu
  110. utu says:
    @res

    I agree with you about the bimodal distribution (but not about it being a “trick”)

    The wide tilt range is a feature of the data.

    This (bimodality and wide range) is not a feature of the data but of any data that is processed the Wai’s way by mixing two partially disjoint populations: top x% SAT_m and top x% SAT_v. Each of this populations will have its own histogram centered on opposite sides of tilt=0 on tilt axis. When you combine them you get bimodal pdf and the wide range. And the range will get wider as x in “top x%” gets smaller.

    Imagine 2-D Gaussian pdf(SAT_m, SAT_v) which means very bland data with no features, with nothing special about it. Two independent variables w/o functional interdependence. This Gaussian pdf subjected to Wai’s way of data torture will produce bimodal distribution. Are we suppose to exclaim WOW or Eureka? Did we discover anything? No. We might be surprised but only because of our ignorance because we haven’t thought of it.

    Before I thought it through I was wondering that it was interesting that bimodality shows there is a dip at tilt=0 which implies that people who are equally talented verbally and in math are less numerous than people who are talented either verbally or in math. But this is a false insight that has nothing to do with math or verbal talents or the nature of human intelligence. This is the result of Wai’s way data torture. He misleads and makes it appear way much more interesting than it really is. So I insist that this is really a trick and I do not think that Wai was not aware of the triviality his presentation that he tricked himself as well. No, he is way too shy about defining the Level of Ability as if it was one ability. In his case the Level of Ability is rather the level of one of two abilities and the populations of two different abilities are mixed. He crossed the line to the ethically challenged territory.

    • Replies: @utu
    , @res
  111. utu says:
    @utu

    But this is a false insight that has nothing to do with math or verbal talents or the nature of human intelligence. This is the result of Wai’s way data torture.

    Clarification: What I meant here is not that is is not a fact that “people who are equally talented verbally and in math are less numerous than people who are talented either verbally or in math” but that this does not deserve to be called an insight and that this fact does not tell us anything about the nature of intelligence. Because this fact is a mathematical necessity like the fact that a matrix has less diagonal elements than elements that are above or below the diagonal.

    • Replies: @res
  112. @JosephB

    I should have made it explicit, but assume that my readers know about xkcd, and also know of my high opinion of xkcd.
    Nonetheless, thanks for pointing this out.

  113. @res

    Human connectome is now at 1200+ subjects. Try Dimitri van der Linden in search bar.

    • Replies: @res
  114. @Astuteobservor II

    Yep. Should hoover it up into a new, joint, post. Or just keep reading and enjoying it.

    • Replies: @Astuteobservor II
  115. res says:
    @James Thompson

    Thanks! Searching for Dimitri van der Linden gave four hits (most recent first):

    http://www.unz.com/jthompson/womens-brains/?highlight=%22Dimitri+van+der+Linden%22

    (above is the most relevant, from last year, has an extended discussion in the comments)

    http://www.unz.com/jthompson/have-you-jewish-personality/?highlight=%22Dimitri+van+der+Linden%22

    http://www.unz.com/jthompson/adopt-child-but-discard-illusion/?highlight=%22Dimitri+van+der+Linden%22

    http://www.unz.com/jthompson/lci14-conference-proceedings-dimitri/?highlight=%22Dimitri+van+der+Linden%22

    Searching for Human Connectome adds two more hits:

    http://www.unz.com/jthompson/iq-brain-map/?highlight=%22human+connectome%22

    (expressing a desire to reproduce other results with the Human Connectome data)

    http://www.unz.com/jthompson/connectivity-matrix-predicts-fluid/?highlight=%22human+connectome%22

    (random connection, this blog post uses fingerprint matching as an analogy, some discussion of computer fingerprint matching by Jeffrey Ullman)

    More about the HCP from a general web search:

    http://www.humanconnectomeproject.org/

    https://www.humanconnectome.org/ (different from above, long list of subsidiary projects)

    https://en.wikipedia.org/wiki/Human_Connectome_Project

    Any thoughts on the current state of MRI and fMRI study reproducibility in light of the HCP?

    • Replies: @James Thompson
  116. res says:
    @utu

    This (bimodality and wide range) is not a feature of the data but of any data that is processed the Wai’s way by mixing two partially disjoint populations: top x% SAT_m and top x% SAT_v.

    Sigh. The wide tilt range IS a feature of the data (as I asserted). Those people exist irrespective of the dual threshold technique. Presumably they are represented in the All plot, but the numbers are so small as to be invisible.

    The bimodality and general overemphasis of the extremes (mostly in the 0.01% plot) is a characteristic of the thresholding as you assert. Presumably a population of the combined single threshold 0.01% would give roughly normal curves (male and female curves offset by average tilt difference at that level) with smaller far tails because the people in the dual threshold far tails might not meet the single threshold (perhaps this is the point you are making?). I would very much like to see that plot. I think it would give a better sense of the typical 0.01% tilts and how males and females compare. I suspect it would look a lot like the All plot, but it would be interesting to see the differences (e.g. do the tails appear wider at that ability level? Is the average tilt offset different?).

    I believe some of this could be seen if the All and 0.01% data in Figure 1 were plotted as log population histograms instead of density. The 250x difference in sample sizes obscures the far tail behavior of the All plot.

    My impression is that the All group (the initial 5% screen) was more of a single threshold screen (the difference does not really matter as much at that level in any case). Does anyone know if this is so? I don’t recall having seen that detail of the initial screen in the voluminous SMPY/TIP literature, but I imagine it is there somewhere.

    • Replies: @utu
  117. res says:
    @utu

    I’m not sure I am parsing your multiple negatives correctly, but I think you were right the first time about this: It is likely there are more equally talented people than people with extreme tilts–even at the 0.01% level. Your Gaussian argument is on point, but the two variables are correlated which makes a difference. I find visualizing that multivariate distribution with different thresholds applied very helpful.

    Of course, that depends on how top 0.01% is defined. I just think it is hard to argue that someone who scores 690/690 is less talented than someone who scores 700/200. In reality the boundary is fuzzy of course (for a variety of reasons, noise, developmental differences, etc.).

    I wonder if the dual thresholds were chosen specifically to look at the extremes given that intensive followup of the smaller sample is expensive per capita.

    Where I think you go far wrong is with the intemperance of your language and conclusions. We might disagree with the best way to interpret this data, but the paper is clear in its procedures. And is accompanied by a vast earlier literature with many more details available. Calling this research ethically challenged is a gross misstatement IMHO.

    • Replies: @utu
  118. utu says:
    @res

    the combined single threshold …. dual threshold

    Let’s define your terminology:

    Combined single threshold is the section of 2D pdf(X,Y) defined by conjunction (X>x0) AND (Y>y0)

    pdf(Z=Y-X) will have narrow range that will get only narrower as the threshold is getting tighter. In the limit the pdf is Dirac delta with zero range. pdf is not bimodal!

    Dual threshold is the section 2D pdf(X,Y) defined by alternative (X>x0) OR (Y>y0)

    pdf(Z=Y-X) will have wide range. And pdf is bimodal.

    the difference does not really matter as much at that level in any case – correct

    • Replies: @res
  119. utu says:
    @res

    It is likely there are more equally talented people than people with extreme tilts–even at the 0.01% level.

    You can see it is not true. But what I was saying that this is almost always not true when the 2D pdf is more or less normal and not deranged like Dirac delta: when pdf(Y,X)=d(Y-Z) (d-suppose to be delta)

    I think you go far wrong is with the intemperance of your language and conclusions.

    I am concur on the intemperance but as far as conclusions or should I say imputations I think I am correct. And if Mr. Wai encountered more intemperance while still being a puppy on seminars of conferences I think he would not obfuscate the definition of the Level of Ability which as we know now is of utmost importance and warrants some discussion. He would not BS us with references to why violin plots are superior instead.

    • Replies: @res
  120. res says:
    @utu

    Let’s define your terminology:

    Good call.

    Combined single threshold is the section of 2D pdf(X,Y) defined by conjunction (X>x0) AND (Y>y0)

    That’s not what I meant. I was thinking of X + Y > threshold.

    Dual threshold is the section 2D pdf(X,Y) defined by alternative (X>x0) OR (Y>y0)

    Yes.

  121. @res

    At 1200 subjects this is as good as an IQ standardisation sample, though not a large birth sample. However, I think it has power, as a very large MRI sample, so should be taken seriously.

    • Replies: @res
  122. res says:
    @utu

    You can see it is not true.

    I am lost in a sea of references (e.g. “it”) and negatives.

    According to Appendix A4 in this paper: https://www.ets.org/Media/Research/pdf/RR-99-02-Dorans.pdf
    the correlations for the SAT subtests are as follows. This is for the normal test taking group and I am not sure how comparable these numbers are for 13 year olds.
    _________SAT-V__SAT-M___Combined
    SAT-V______1_____0.71______0.92
    SAT-M_____0.71_____1______ 0.93
    Combined___0.92___0.93_______1
    If the correlations for the current paper are anything like this then I think it is safe to say relatively equal scores are most common overall. This may not be true for the 0.01% group in the paper though for the reasons you mention. I do think relatively equal scores would be most common for a 0.01% group defined by combined threshold. This could be easily checked by anyone with access to the data.

    What I don’t understand is why you insist on imputing negative motives to people and calling them names (e.g. puppy here). I really don’t see you having justification for doing that. And doing so makes you look like a non-serious and ill tempered troll IMO.

    As far as obfuscation, do you honestly think the paper failed to be clear on its methodology?

    FWIW I think the split violin plot was an excellent way to present the data (compare to the tables in the preprint!). Without the Figure 1 plots I would not have realized how much the separate thresholds mattered for the regression analysis.

    • Replies: @utu
  123. res says:
    @James Thompson

    Good comparisons. Another way of looking at this is whether the MRI data is as reliable (e.g. test-retest) as IQ tests. I don’t have a good feel for the answer to that. I suspect the MRI data is not as reliable, but have no quantitative sense of the difference.

    • Replies: @James Thompson
  124. utu says:
    @res

    This may not be true for the 0.01% group in the paper though for the reasons you mention.

    There is no maybe or perhaps. By writing “maybe not true” implies you are still not getting it. Once the dual threshold ( (X>x0) OR (Y>y0)) gets high enough equal score cases will be less numerous than unequal scores cases. There is no other possibility.

    I do think relatively equal scores would be most common for a 0.01% group defined by combined threshold.

    Obviously. The pdf(X-Y) will be a concave (single peak) function with support that gets narrower with increasing threshold approaching Dirac delta.

    As far as obfuscation, do you honestly think the paper failed to be clear on its methodology?

    Absolutely. (1) lack of definition of Level of Ability, (2) not being clear enough that the dual threshold is used , (3) not pointing out that bimodality is a direct consequence of the dual threshold method and not a special feature of data, (4) not pointing to the fact that high Level of Ability in his approach means that for high threshold you are ending up with many one sided “geniuses” but also with one sided “morons.”

    I would prefer histograms instead of smooth pdf’s that he generated . It is possible that for 0.01% where samples are small (<1,500) the histogram looks lousy. BTW, why he did not show 0.1%? How are these smooth pdf's defined? To what extent they are synthetic? The ruggedness of histogram is hidden.

    For some reason he does not calculate threshold from his sample but gets them forma some other publication. There should be a table of top x% with SAT- M and SAT-V values for males and females including sample sizes. We could is also how cuf off point differs between male sand females.

    Why he even brings up violin plots? From what I gathered violin plots are an improvement on box plots to show approximation of pdf which usually are plotted vertically around data points. While here he is really plotting pdf. Violin plot is a box blot where instead of box there is pdf. If there was no such thing as violin plot would he insist on plotting a box plot? He never heard about pdf or histogram?

    Why even mention violin plot in this case? Because he used software with that name to get pdf? It is possible that the software produces very lousy approximation of pdf that suppose to be congruent with raw data histogram than more dedicated software that generates pdfs.

    A decent reviewer should make him drop violin plots and refer to plots as pdf or histograms and explain how the pdf was constructed from histogram or raw data. That he put on one graph two pdfs of which one is up side down is nice but it has nothing to do with the violin plot. He could have plot two pdfs in different colors with out up side downing one.

    I do not know if this paper was even necessary. Probably it is all well know within the field. But if he decided to write it, the paper could have been much shorter and packed with much more information if he did not go through the ritual of bowing to all possible reviewers and signaling that he is cool while doing some obfuscating.

  125. @res

    Rich Haier and Rex Jung answered that question when I put it to them at an ISIR conference some years ago. They have test-retest data, though I do not have the figures myself. I think that the better labs, like theirs, will have these measures, and are reasonably happy with them.

    • Replies: @res
  126. EH says:
    @res

    Thanks for that. The selection procedure has much the same problem with cutoff levels that using a combined score would have. It unnecessarily throws away the information from the different and not particularly highly correlated tests (.55, but even less correlated at the high end) to produce a single number that doesn’t take account of the still pretty significant differences in difficulty at the top between the V and M tests. Basing the whole thing on the guess that 800 V+M was a top 1% score for 13 year-olds is particularly shaky assumption, and the resulting deviation/rarity IQ estimate of 180 average is not believable.

    There is no way to equate old verbal to new verbal scores at the top end because the harder types of questions such as analogies have been replaced with questions that few people get “right” only because they are somewhat matters of opinion: “which answer is closest to the author’s intent in this passage” sort of questions. Different scores above ~700 are not distinguishing levels of ability but are merely luck now. I don’t much credit that College Board table equating old and new scores at the high end as it lists scores which do not occur – e.g. on an otherwise perfect V or M, skipping one question loses 20 points, one question incorrect loses 40 points. Missing one question on late-’80s SAT-V scored 760, which was about a 1 in 2000 score overall, I believe 1600 V+M is not much rarer now.)

    • Replies: @res
  127. res says:
    @James Thompson

    Good to know. Perhaps we have turned the corner on MRI study reliability? Similar to the way GWAS became so much more reliable after the researchers got better at multiple hypothesis testing.

  128. res says:
    @EH

    Thanks for your reply!

    not particularly highly correlated tests (.55, but even less correlated at the high end)

    Agreed about the high end. The sample giving 0.55 correlation in the top 3% was only the first few years of the SMPY as well. It would be interesting to see an analysis of the correlations in the current full TIP sample–including sex differences and by ability group. Has anything like that been published? Would probably want to try different definitions of the ability groups due to the dual threshold issue discussed above.

    Basing the whole thing on the guess that 800 V+M was a top 1% score for 13 year-olds is particularly shaky assumption

    At this point they should have enough data to be the best source for validating that number. My look at the relative prevalence of the different ability groups above makes me think the estimates are fuzzy.

    the resulting deviation/rarity IQ estimate of 180 average is not believable.

    I strongly agree. And it was 186!!! The original quote: “we found the mean and standard deviation to be 186 and 11, respectively (with 99% of these estimates >160).”

    I find that incredibly problematic. Given the tail characteristics of the normal curve a hard threshold in the far tail (say at IQ 160) should have a much lower average. A quick look in R gives:

    > IQs IQs_over_160 = 160]
    > length(IQs_over_160)
    [1] 311
    > summary(IQs_over_160)
    Min. 1st Qu. Median Mean 3rd Qu. Max.
    160.0 161.0 162.2 163.2 164.7 175.7

    Notice that the mean is only 3.2 points above the threshold! And the MAX was under 176! Could they have been using ratio IQs?

    Also, 4SD is about 1 in 30,000 so 160 deviation IQ is above their targeted threshold of 0.01%, yet 99% of their subjects measure above that?!

    There is no way to equate old verbal to new verbal scores at the top end because the harder types of questions such as analogies have been replaced with questions that few people get “right” only because they are somewhat matters of opinion: “which answer is closest to the author’s intent in this passage” sort of questions. Different scores above ~700 are not distinguishing levels of ability but are merely luck now.

    Has anybody looked at the data? It seems that effect should show up in test-retest reliability of high end scores.

    I don’t much credit that College Board table equating old and new scores at the high end…

    I wonder about it as well. But it is the best we have. Another thing which bothers me is that the Math table simultaneously shows the post-1995 SAT 20 points easier at old 770-780 while being 10 points harder at 660-710. That harder result seems particularly difficult to believe.

    • Replies: @EH
  129. Paw says:
    @Anon

    Tests , IQ , do not show all capabilities , art of thinking , creating and acting.
    Men had to hunt and acquired/in process/ so many abilities , learn to solve problems and take care of everything../Family, tribes, food, protection etc./
    Then our ancestors saw , they had to protect women /sometimes against their will/.
    Arranged marriages had so many advantages and we can see so many of it now.

  130. @dearieme

    Water Babies is a strange, Freudian book. But then Kingsley was a pretty strange guy.

    Not sure “Coral Island” would be in many school libraries these days, with its descriptions of cannibalism and sacrifice, and its assumption that Victorian Christianity is the highest form of civilisation to which the world must be led.

  131. @dearieme

    The great regret of living in the US is that you simply cannot get good meat pies or sausage rolls!

  132. atom says: • Website

    Fertility clinics looks to the both males and females for diagnosis of fertility problems. Diagnosis has shown that fertility problems both of human affected and solve diagnosis solutions.

  133. hyperbola says:
    @Habakkuk Mucklewrath

    Larry Summers was fired from Harvard for his participation in corruption by a small, racist, foreign sect that abuses Americans. It is the same corruption that landed us with a Kagan NeoCon fraud on our Supreme Court. It is rampant at Harvard.

    Why Larry Summers lost the presidency of Harvard

    https://mathbabe.org/2012/03/11/why-larry-summers-lost-the-presidency-of-harvard/

    Some people still think Larry Summers got fired from being the president of Harvard because of the ridiculous comments he made about women in math (see my post about this here) or because of the comments he made about Cornel West. Actually, the truth is something worse, and for which he should actually be in jail. It’s also something that makes Harvard look bad, so maybe that’s why it’s less known….

    Let’s set the record straight: Summers was directly involved with defrauding the U.S. Government (see below) and Russia. He admitted to not understand conflict of interest issues (see below). It is particularly appalling, …….

  134. hyperbola says:

    This is more wanking of the usual deceptive nature. With a sample size of 2 million and a cutoff of o.01% (1/10,000), the number of individuals in the 0.01% group is 200 – split between males and females. There is NO useful (social or other) conclusion based on such numbers. Above all, there is no justification whatseoever for treating the other 1,999,800 members of the sample differently based on their sex.

    • Agree: utu
  135. Dr. Thompson, I am curious about your thoughts of the meta-studies from Rosalind Franklin University supposedly disproving any meaningful differences between male and female brains. The first thing that occurred to me is that they did not control for ethnicity at all.

  136. Have you a link, please.

    • Replies: @Daniel Chieh
  137. @James Thompson

    Yes, of course.

    No difference in hippocampus:

    https://www.sciencedirect.com/science/article/pii/S1053811915007697

    No difference for amygdala:

    https://www.sciencedirect.com/science/article/pii/S1053811916307431

    Suspicious(to me) that it keeps coming from one university and from the same people, but very wildly popularized and I’ve seen no significant debunking/pushback. Of course, they could be experts in this specific field which there are few rivals, admittedly. Maybe.

    Official announcement is that these do not merely dismiss false beliefs in these brain structures, but indicate there are minimal gender differences in the brain:

    https://www.rosalindfranklin.edu/news/male–female-brain-differences-big-data-says-not-so-much/

    Also, while on this:

    Sex differences in the human corpus callosum: myth or reality?

    https://www.ncbi.nlm.nih.gov/pubmed/9353793

    Not Rose Franklin university, but finds opposite result that men have larger corpus callosum. Concludes that this means that there are no differences.

    The widespread belief that women have a larger splenium than men and consequently think differently is untenable. Causes of and means to avoid such a false impression in future research are discussed.

    • Replies: @James Thompson
  138. @Daniel Chieh

    thanks. some sophistry here.

    No difference in hippocampus:

    https://www.sciencedirect.com/science/article/pii/S1053811915007697

    Er, no. They say there is no “disproportionate” difference, once you “correct” for the smaller size of the female brain, but happily admit that men have a bigger hippocampus.

    Same argument for the amygdala.

    Third paper is better, and also shows that male brain is better, and of course male corpus callosum is bigger.

    So, yep, sex differences, but not always proportion differences once you allow for brain size.

    • Replies: @Daniel Chieh
  139. @James Thompson

    Thanks. IIRC the argument was that absolute difference means little because the increased neurons are just to accommodate/manage the increased body mass of males.

    The papers themselves are paywalled to me – did they make an effort to control for ethnicity?

    Still, throws some water on extreme male brain theory for autism?

    • Replies: @James Thompson
  140. @Daniel Chieh

    Sorry, haven’t read them them, but usually such measures are restricted to European populations, so as not to invite comparisons.

    • Replies: @Daniel Chieh
  141. @James Thompson

    Much appreciated for your time and work in general, Dr. Thompson. I’ve been following you since before you came to Unz, incidentally, and am quite happy that you’ve found a home here.

  142. Thanks for your kind words.

  143. EH says:
    @res

    The 186 IQ could be a plausible measure of relative ability on an equal-interval scale (a point in the 180s representing the same increment of ability that a point in the 100s represents) assuming a log-normal distribution – 186 on a 15-point s.d. log-normal distribution has the same rarity as a 163.5 normal-distribution IQ (1 in 87,460), within 0.3 points of the 163.2 predicted average. A 1 in 10,000 score would be 172 LN-IQ or 155.7 regular IQ. The 175.7 max predicted IQ would be the same rarity as a 210 LN-IQ. (A 210 IQ is about the same 6 s.d. above an average 120 IQ college professor as the college professor is above an average chimpanzee or 3.75 year-old child). Still, I think the numbers don’t come out quite right – the average is nearly 10x rarer than the cutoff 1 in 10,000 rarity.

  144. Factorize says:

    Anyone consider how genetic engineering will differentially affect IQ differences between men and women? If the X Chromosome has IQ enhancing alleles, as it likely does, then women will receive a double dose of such additive variants. Men will receive a single dose. The Y Chromosome probably has few variants of interest for IQ and the autosomes are shared equally between men and women. What does this suggest? It suggests to me that we are moving to the age of women power! It is not easy to see why this would not occur.

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