Friday 29 July 2016

Detterman’s 50 years of seeking satisfaction.

 

 

 

 

When I talk of people I admire, some readers assume that they are dead. Doug Detterman is alive and well, but I am going to say nice things about him anyway. Don’t talk ill of the dead, and don’t postpone speaking well of the living.

Doug is a quiet guy, who has entirely ignored the American habit of self-promotion, but has gently put modern intelligence research on the map. Almost unseen, in 1977 he founded and edited one publication Intelligence from precarious obscurity, to fragile partial visibility, to its present position as the leading journal on intelligence research.  He only got his freedom from the editorial coal mine last year. He also founded the International Society for Intelligence Research (ISIR) in 2000.

Now he looks back at 50 years of intelligence research, and avers that it is much more important than curing cancer, controlling global warming or ending poverty. He also regards teachers and schools as over-rated, since they only account for 10% of pupil achievement. Five decades dedicated to finding a satisfactory answer to a simple question: why are some people smarter than others?

His answer: a traffic jam. All the modules of the brain have to go through a central hub, and the poorer the connection the lower the intelligence.

His lecture is a treasure house: wise, instructive, and great fun to read.

https://drive.google.com/file/d/0B3c4TxciNeJZcHZnbmFHb1hVNHc/view?usp=sharing

21 comments:

  1. I presume it was an error to say that drop-out rates correlate positively with IQ?

    Pity about the antipodean slides at the end.

    Interesting stuff, though I'm not sure why it makes sense to say that g is not a thing and then conjecture a thing that he hopes will explain g.

    All in all, high marks for Mr Detterman.


    A question for you, doc. What is the inwardness of "Most environmental effects are not shared by children growing up in the same family"?

    And another: what results are now so well established that there is little sense in flogging away at the same topic?

    And a third: so what should people work on who are not biochemists, geneticists, and the like?

    ReplyDelete
    Replies
    1. On g not being a thing, and then explaining how it comes about, I agree with you. Family environmental influences turn out to be small and transitory as far as intellect is concerned. A surprise.
      Even well-established findings can do with enquiry and extension.
      What should non-biochemists and non-geneticists work on? Artificial intelligence, data analysis and theoretical modelling.

      Delete
    2. elijahlarmstrong30 July 2016 at 22:16

      There is a difference between g being an explanandum and g being a unitary construct. For instance, SES variables like income and education show a positive manifold, and it's legitimate to explore why this is. But this positive manifold is probably the result of many different causal factors, such as inherited wealth, cognitive and personality traits, access (by virtue of credentials) to good jobs, pure luck, etc.

      Delete
    3. "Pity about the antipodean slides at the end."

      :)

      Delete
  2. Noteworthy how a man like this feels the need to introduce his work with:

    "This is my personal view.
    Others might see things
    differently. Take what you
    like and leave what you
    don’t."

    while I am sure Jay Gould, Lewonting, "Montagu", and the rest of the cultural gang bang never did.

    Very tough to be a free one, in a social species environment, like academia, where groups survive and gangs dominate.

    ReplyDelete
    Replies
    1. Yes. One person with a good argument and solid reproducible observations should be able to up-end a cartfull of charlatans. Often doesn't happen though.

      Delete
  3. "wise, instructive, and great fun to read"

    Even when it lists "World conservatism" in a list of "negative things" that "have been found to correlate" with intelligence?

    "Criminality, accidents, world conservatism, death [sic], ..., religiosity [obviously meaning any cult save for rationalism/scientism/technologism, marketism, you bet)"
    There is also "ilegitimacy[sic]"

    Then we have "falsification (lie score)": are we sure?
    Deception is an evolutionary tool, and it's easily arguable (and has resulted from psychometric testing) that the more the intelligence the better the deception.
    The "lie score" doesn't measure quality but quantity, and, very conveniently, "lies" instead of deception.
    Indeed he can be right: less intelligent people need more lies to achieve the same degree of deception (if they can ever).

    The ludicrous list goes on, mixing science and literature perceived as science.
    "Authoritarianism, dogmatism."
    The fellow(s), clearly, take(s) only the dogmatism he doesn't adhere to as "dogmatism."
    Dogmatism comes from intellectual narcissism, and is fostered by intelligence.
    Talk of today's forbidden truths with an average group of teens at the park and 10 Harvard professors, and see where dogmatism is more prevalent (A funny instance: http://www.unz.com/sfrancis/sex-summers-and-the-return-of-human-nature/).

    And then, icing on the cake, we have "racial prejudice."
    A finding that agrees with our daily experiences — we know that it's the less intelligent ethnic/racial groups to be more racist, with the blacks topping the ladder.
    Still, one could wonder why measurements were made for "racial prejudice" and not a set of prejudices... it was probably due to some prejudice.


    ReplyDelete
    Replies
    1. The slide summarises many detailed studies, and the descriptors are too brief to give the explanatory detail, but merely shows the range of correlated variables. The intention is to show the nature of the nexus of behaviours and attitudes related to mental ability.

      Delete
  4. "Modify schools
    Improve teachers
    These efforts have had little success
    Why?
    Because schools and teachers have only small
    effects on academic achievement!"
    -------------
    "22,923 took SAT and 4-5 years later GRE

    Selected 7,954 students at 292 colleges or universities
    with > 10 students

    Used SAT-Math + major + gender to predict GRE-Math

    R2 = 0.93


    No more than 7% of variance could be due to teachers

    Doesn’t matter where you go to college.

    What matters is student SAT – Quality of college =
    Mean SAT"
    -----------------------------

    Schooling methods and teachers' level of education and intelligence (I am speaking of the average teacher in pre-university school) have dropped constantly over the last 70 years.
    Their quality changes little for very high-intelligence people (who learn their own) and low-intelligence people (who can't learn, anyhow).
    But schooling quality, both in terms of methods and teacher intelligence changes a lot in the educational life of the mid-high intelligence (IQ = 110-125).
    He won't become a researcher or scholar (if that's what's meant by "academic achievement), but he'll be a far more knowledgeable person.

    As a person in that range of intelligence (110-125) I can claim that swapping a mediocre teacher following new-style methods with an intelligent and motivated old-style one never failed to affect my understanding of the subject deeply. It also affects motivation a lot, especially in a youth.

    Schools have been harmed and ruined for 70 years (effect of democracy, or it wouldn't have occured quite uniformly in the entire West), and this gentleman speaks of the ineffectiveness of "teacher improvement" and "teaching methods modifications."

    Teachers have modest wages, and only who thinks they can't do any other better paid job want to become teachers.
    Where does the gentle man see the improvement over the last decades?

    What a puzzle.

    I hope I didn't misunderstand his point, still.
    There's no doubt that the genetic factor is incomparably more decisive than teacher quality and even teaching method; like there's no doubt teacher quality and teaching methods have dropped and been disfigured by progressively... progressive societies.

    In Europe, you heard real questions, asking not only study but intelligence, asked in university exams until the late 90s.
    Then was a huge campaign to increment graduate rates, and have as many graduates as possible.
    What did they change? They started asking questions that were no actual questions, but memory/conscentious study tests.
    "Parrots" and mindless swots rejoiced: you can often hear them recite the relevant page(s) of the relevant books by heart at such speed that listening is nigh impossible, and be awarded, after the show, their much-lusted full marks (to which, naturally, people of their kind attach utmost importance).

    It was simple. Remove intelligence from the requirements to pass exams, graduate, and even graduate cum laude.
    This is what was done around 15 years ago.

    I really can't but keep wondering what "improvements" this gentleman spoke about.

    ReplyDelete
    Replies
    1. I think that Doug's point would be that, given access to reasonably good education (as provided by countries with above $16,000 incomes) the variance is accounted for by intelligence. Poor countries would have more school based variance, because of lack of schools, resources and teachers. Remember, analysis of variance depends on samples and time frames tested.

      Delete
  5. elijahlarmstrong30 July 2016 at 22:13

    It's a good lecture. I agree with Detterman that g is probably not a natural kind. Few thinkers analyze it as such (Mike Anderson and John Carroll being some exceptions); even Jensen explicitly argued that g would be explained by many cognitive and neural kinds. But what does it mean, exactly, for g to be nonunitary? A cognitive process can be nonunitary at one level and unitary at others; working memory, for example, is generally treated as a natural kind by psychologists, but it is decomposable into other processes. Indeed, every cognitive process except primitive Boolean operations, or whatever the lowest level of cognition is, is nonunitary in some sense. The philosophy of psychology needs a convincing account of how processes are individuated.

    All this being said, I would give fairly little weight to SLODR, unlike Detterman. It is of modest effect size, and it might be explained by social differentiation (high-IQ people have stronger intellectual interests and more leisure in which to develop specific skills). It's also worth noting that you can find a (weaker) positive manifold amongst ECTs –– and, I would guess, neural parameters –– which suggests that the sampling of these parameters can't wholly explain g. Shared genes à la Britt Anderson (including assortative mating) might do some work. Likewise the environment: Some environmental factors, like formal education, seem to jointly improve a variety of cognitive abilities. How much of g do they account for?

    Some frontiers that Detterman does not touch on: (1) The cognitive psychology of reasoning processes is still fairly primitive; for instance, we don't know why context-embedded reasoning is so much easier than decontextualized reasoning. (Is it because the context "activates" the same process, or because there are different processes entirely?) We know a lot about working memory, which instantiates reasoning, but not so much about reasoning itself. Sternberg's componential analysis was an attempt to analyze reasoning processes, but it seems to have been abandoned. (2) There is a lot we do not know about the evolutionary psychology of cognitive abilities. For example, what is the phylogeny of g? It is probably present in most mammals. Is it present in reptiles? In insects? In prokaryotes? Why, given strong selection pressure for cognitive abilities, do they show significant additive heritability? Is it because of GxE interactions, or because they are mutational targets, or because of balancing selection (for instance, the adaptive value of a social hierarchy), or because of deficiencies in our methods? (3) What is the nature of the effects of cognitive stimulation, like formal education and Head Start, on cognitive abilities? They seem to be rather dissimilar to the sum of the causes of test covariance, i.e., to g. How much do they affect processing speed? Working memory? Basic neural parameters like myelination and cortical size?

    ReplyDelete
  6. Intelligence is more important than curing cancer, controlling global warming or ending poverty because only high intelligence can solve these problems and all the other hard problems. Average intelligence is only important to the extent that it determines how many are in the highest reaches of the right tail; what matters for solving hard problems is how many people you have over IQ 140, 160 or even higher for the hardest problems. Only in this high range is intelligence more important than curing cancer or ending poverty, yet it is one of the least-studied areas of the field.

    The normal distribution does not fit the frequency of scores in the high tail, the really important part, yet seldom do you see anything else used in research papers, even ones on high intelligence. It has long been known that the log-normal distribution fits the high tail much better than the normal distribution - log-normal predicts more high raw scores than normal.

    Why isn't the log-normal distribution used more often? In the form that it's implemented in spreadsheets and other widely-used software, it doesn't work at all like the normal distribution - you can't just plug in 100 for the mean and 15 for the standard deviation and get any kind of sensible numbers out. To convert the mean and s.d. you have to use the formulae here.

    Log-normal mean for 15 point standard deviation, 100 avg.
    =LN(100)-0.5*LN((15/100)^2+1)
    = 4.59404488


    Log-normal standard deviation for 15 point s.d., 100 avg.
    =(LN((15/100)^2+1))^0.5
    = 0.14916638

    so that for instance one could calculate the true rarity of a given high IQ score (top 1 in n) in a spreadsheet using:
    =1/(1-(LOGNORMDIST((IQ_score);LN(100)-0.5*LN((15/100)^2+1);(LN((15/100)^2+1))^0.5;1)))
    (replace 100 and 15 in the equation for other means and std. devs.)

    or the percentile for that score using:
    = 100 * LOGNORMDIST( IQ_score ; 4.59404488; 0.14916638 ; 1)
    (as with NORMDIST, the final "1" specifies the cumulative distribution)

    I suspect using the wrong distribution is the source of a good deal of the problems with high-range testing. To make more predictive intelligence measures in the high range it will be essential to use a curve that matches the true distribution reasonably well. More exotic curves have been proposed, e.g. Burt fit a Pearson type IV, but the log-normal works quite well. See Bob Seitz's article for more on using the log-normal distribution for IQ. (Bob Seitz is a past officer of the 1 in 1,000,000 IQ Mega Society. Here are his other articles on intelligence. The ones on the Terman study, Leta Hollingworth's Children above 180 IQ and Willam Sidis I found quite interesting.)

    ReplyDelete
    Replies
    1. elijahlarmstrong31 July 2016 at 21:15

      IQ is forced to fit a normal distribution these days. Ratio IQ is log-normal, but ratio IQ is a rather poor concept for a variety of reasons.

      Delete
    2. ''Intelligence''

      When all people talk about '' be 'smart' '' on social networks

      '' Not to be racist ''

      '' Do not be evil ''

      '' The 'intelligence' can solve all these problems [that were caused by the same intelligence it, lol] ''


      ... I think they mean '' wisdom ''

      so why not be less vague *

      Delete
    3. Elijah, ratios of mental to chronological age aren't useful, but measures on a "ratio scale" (equal-interval measures with a true zero) are, being the aim of Rasch measures and Item Response Theory (IRT), and I'd be surprised if these aren't log-normally distributed. Actually these methods involve ratios of logarithms - see my post here and the pdf Measurement Essentials textbook at rasch.org.

      Delete
    4. elijahlarmstrong3 August 2016 at 02:25

      IRT is not my field of expertise (to say the least!), but thanks for the refs.

      Delete
  7. You used to be able to buy log-normal graph paper.

    ReplyDelete
    Replies
    1. Really? Neat. Just a few days ago I was searching for normal distribution graph paper which led me to Prof. Michael Friendly's Gallery of Data Visualization / The Best and Worst of Statistical Graphics which is marvelous.

      Delete
    2. Somewhere in London or Paris there will be an elderly widower whose stationery shop still stocks all those ancient specialised graph papers. All you have to do is find him before he goes to join his maker and the children throw out his stock.

      I used to particularly enjoy equilateral triangular graph paper.

      Delete
  8. Well-Explain Well-Thinking and brilliant twist in message on how to think about free credit report topic! Awesome post, Leslie!..>

    ReplyDelete
  9. The information which you have provided is really nice and awesome. I want to share something more about this as well.

    With Regards, Depression Treatment in Sydney

    ReplyDelete