Science as racism?

[andyandy]

A non-IQ test on a small sample size. You think that counts?

Show me where they adminster an actual IQ test to children who could not speak English.

But no, toss a few tests of any kind at a small sample, and just extrapolate the results. "We know what they would test as because we know the test is right. And how do we know the test is right? Because it says what they would test as!"

1000 is the magic number for sample sizes based on a normal distribution.....where you can be 95% confident that the results will be +/- 3 %
hence the reason it's used for polling purposes, and for a lot of surveys....

it's not sufficient to dismiss it as a "small sample" - unless perhaps you don't believe reaction times are normally distributed.....

 

[Steven Howard]

it's not sufficient to dismiss it as a "small sample" - unless perhaps you don't believe reaction times are normally distributed.....

Can we assume a priori that they are?

 

[Dave1001]

does it conclude that both H sap sapiens (100,000-c.30,000 years ago) and H neanderthals (c. cultural explosion) had a higher IQ than modern day H sap sapiens?

Isn't that an ad absurdum argumentative question, kind of like if bpesta asked you if you conclude that the difference between abstract reasoning in bonobos and humans is cultural?

 

[bpesta22]

I agree, just the mention that brain size correlates with IQ sounds stupid and junk scientish.

Then, when one reads the actual peer-reviewed article wherein the findings are discussed, the rationale provided (why and what it means) it no longer seems like junk science (In fact, I have a pub with the guy who did the meta-analysis I was referencing--though ours is not on IQ).

Indeed, that's my main motivation for posting so much and so fiercely as the defender of IQ tests.

The contrast between what the science reveals (i.e., peer reviewed journal articles), and what most intelligent educated people /skeptics know and think about the science, is vast.

This annoys me, as merely reading the actual science-- something anyone can do-- paints a vastly different picture.

I dunno, maybe I underestimate how difficult it is for non-psychologists to understand this stuff.

A sample size of 1000 might be needed for a pollster worried about how 300 million people in america might vote. That N size is vast overkill in a scientific experiment where one can introduce control; repeated measures variables, multiple measures, many trials, etc.

I bet the inspection time differences would emerge as significant with just 30 per group.

At any rate, sample size issues are NOT the same for people doing polling versus people trying to demonstrate an effect in a controlled experiment.

I can forgive Yahzi for not knowing this. I can't forgive him for thinking it's a valid criticism (using only an N of 1000!) for the research I cited. Nor can I forgive him for complaining about sample size being too small when one gets significant results (sample size issues relate to statistical power and type ii errors, which are irrelevant when one finds an effect-- obviously, the sample size was big enough, given significant results).

I've got 10 years of teaching experience. I've answered tons of questions from students. I can tell based on the question whether the student is really grasping the whole lecture or missing the point entirely. IMO-- perhaps I am wrong-- Yahzi just don't understand the science (independent of whether it's right or wrong).

 

[Jeff Corey]

But these are correlational studies. By definition, you can't "introduce control" in a study with no true independent variable.
As you know, in a real experiment, the scientist manipulates an independent variable to see if it changes the dependent variable. This other crap is mere correlational masturbation with the third variable problem as the two ton gorilla lurking in the corner, with Hammy and a few other prehominids.

 

[Yahzi]

I agree, just the mention that brain size correlates with IQ sounds stupid and junk scientish.

Because it is.

The contrast between what the science reveals (i.e., peer reviewed journal articles), and what most intelligent educated people /skeptics know and think about the science, is vast.

And yet you are absolutely cluless about the effects of social cues in primates.

This annoys me, as merely reading the actual science-- something anyone can do-- paints a vastly different picture.

And it offends me to see people present black/white IQ scores from the Army in 1917 with a straight face. Do you honestly think there was no bias in the measurement or collection of that data? Just how naive can you pretend to be?

I dunno, maybe I underestimate how difficult it is for non-psychologists to understand this stuff.

The only thing hard to understand is how you can pretend you're doing science when you ignore every fact I bring up.

A sample size of 1000 might be needed for a pollster worried about how 300 million people in america might vote. That N size is vast overkill in a scientific experiment where one can introduce control; repeated measures variables, multiple measures, many trials, etc.

1. 1,000 is the absolute smallest sample size you have presented yet. Even your 1917 data is 23,000. When I asked you for an example of Asian testing on actual Asians, this was the best you could do.

2. You still have not demonstrated any IQ tests applied to non-English speaking people. Even while you freely assert what those tests will show.

I can forgive Yahzi for not knowing this.

You don't need to. I know it.

I can't forgive him for thinking it's a valid criticism (using only an N of 1000!) for the research I cited.

See above, for why your small sample size matters. That you could only mention one study, of such a small size, while you natter on about hundreds of studies dating back to 1917 with tens of thousands of subjects... do you begin to see the point here?

No, of course not.

But please, answer this question: what is the sample size of non-English speaking Asians who have taken IQ tests? (Not reaction tests, but actual IQ tests - you know, the test that measures the number you have assigned to the Asian population).

Why... it's zero.

Is that small enough for you?

I can tell based on the question whether the student is really grasping the whole lecture or missing the point entirely.

Then why do you avoid so many of my questions?

IMO-- perhaps I am wrong-- Yahzi just don't understand the science (independent of whether it's right or wrong).

This from the guy who suggested that if the courts allow it, it must be true.

This from the guy who thinks it's plausible to measure psychometric data to 3 digits of precision over 20 years. No, that data wasn't faked; Old Burt just got it right the first time!

This from the guy who dismissed the gorilla example of social cues affecting physical development because he found it "inane."

You're just a beacon of science, you are.

You do realize, don't you, that you have never actually responded to the gorilla example? That you have dismissed it every time I have brought it up, but have never actually said why it cannot matter. You have expressed your personal incredulity that nuerological development can be culturally influenced; and when faced with observations in other primates that show it to be true, have simply dismissed the data.

Why can't you explain what startling logic allows you to observe massive biological effects from culture in primates, but dismiss it out of hand in humans?

Why can't you explain how psychometric testing is as accurate 100 years ago as it today, even while the definition of what was being measured has completely changed? (Remember how you started off spouting on about "G" instead of IQ?) Are you honestly asserting that IQ tests are so easy to create that you can make one even when you don't know what it is you are measuring? (In which case... where are all those Oriental IQ tests?)

Why can't you see that your IQ field is riddled with bad assumptions, racist motivations, spurious data, lousy methodology, and inadequate definitions?

Well, the obvious answer is: because it tells you what want to hear.

Theology.

 

[Yahzi]

it's not sufficient to dismiss it as a "small sample"

It's only small in the context of all the other studies he mentions.

My point being that the people who do IQ studies have no problem assigning an IQ score to half the planet because they tested 1000 kids with a reaction test that they think they can correlate to an IQ test.

Is there a scientific term for "hearsay?"

 

[andyandy]

Isn't that an ad absurdum argumentative question, kind of like if bpesta asked you if you conclude that the difference between abstract reasoning in bonobos and humans is cultural?

it was a slightly facetious post hence the smilie.....

nevertheless if one is going to make a case for a correlation between IQ and cranial size (and implied brain size) it would indeed be relevant to examine anthroplogical evidence

there have been two significant spurts of brain enlargement -

1) the first 2.0-1.5 million years ago, around the crossover of Australopithecus, homo habilus and homo erectus.....

2) the second about 500,000-200,000 years ago with homo erectus and neanderthal....

the first spurt (1) is tentatively attributed to toolmaking - but archaeologists can find no major change in the archaeological record correlating with the second period of brain expansion (2)....as the same hunter-gatherer lifestyle with limited tools was maintained....

so there is no strong relationship between brain size expansion and the known changes in homo behavior - and as such there's no simple relationship between brain size and "intelligence" (if that is what people who measure IQ wish to demonstrate)

 

[andyandy]

Can we assume a priori that they are?

no....

having thought about it (and looked up a couple of sites ), they will be skewed to the right - however they still appear to give a rough normal distribution approximation....(especially if one looks to eliminate outliers)

http://ecl.ucsd.edu/Ratcliff_1993.pdf (graph p3)

 

[andyandy]

At any rate, sample size issues are NOT the same for people doing polling versus people trying to demonstrate an effect in a controlled experiment.

sure - the underlying point is that it's significance rather than n that one should look at.....

 

[Dancing David]

it's not sufficient to dismiss it as a "small sample" - unless perhaps you don't believe reaction times are normally distributed.....

That is a question of how the sample was gathered more than anything. Was it randomly generated from as large a pool as possible, or was it chosen for an ability to be sampled more easily. It is then still a strech, but a better one to say that a trait is randomly distributed. First the sample must be randomly distributed.

 

[Dancing David]

But these are correlational studies. By definition, you can't "introduce control" in a study with no true independent variable.
As you know, in a real experiment, the scientist manipulates an independent variable to see if it changes the dependent variable. This other crap is mere correlational masturbation with the third variable problem as the two ton gorilla lurking in the corner, with Hammy and a few other prehominids.

Interesting, I think the 'control' in this case would be the sample distribution.

In the case at hand the independant variable is allegedly the racial origins of the humans, and the dependant variable is the IQ.

Correlational masturbation, a great term, does it become correlational intercourse when the correlation is above 68%?

 

[bpesta22]

Interesting, I think the 'control' in this case would be the sample distribution.

In the case at hand the independant variable is allegedly the racial origins of the humans, and the dependant variable is the IQ.

Correlational masturbation, a great term, does it become correlational intercourse when the correlation is above 68%?

But in an inspection time task, the correlational stuff is the subject's demographic characteristics (race, age, gender, etc).

Obviously, one can't randomly assign people into the black or white groups, so indeed causal inferences are iffy.

This applies to a solid majority of experiments within social science. Things like gender, age, political party, education level, etc. are worthy of study even if we can't use them as randomly assigned IV's.

For example, any study showing that teachers treat girls who raise their hands differently from boys who raise their hand would suffer the same problem (one couldn't claim that student gender caused the difference in teacher behavior, as the kids weren't randomly assigned to be boys or girls). Should we not study this important area because any significant results would be correlational masturbation?

What about the correlation between obesity and heart disease? Is that masturbation? I can't imagine a study where people are randomly assigned to be obese or normal-weight, and then followed around to see who gets heart disease.

So, if you define good science as only those experiments which use random assignment, thus allowing causal inferences, you'd be throwing out many babies from the correlational bathwater.

***

My point was that n = 30 per group would likely give enough statistical power to show a difference on inspection time. Certainly, n =1000 is overkill (if you're just interested in showing the difference, given diminishing marginal returns that n size gives to stat power). If you'd like, I can do the calculations that plug in n size and effect size to determine how many subjects one needs to reject the null (across whatever power levels you think are appropriate).

30 per group is good enough because of the controls and standardization of the inspection time task (e.g., not one trial but many; using a staircase method to hone in on the shortest exposure time that still results in accuarate performance, etc.).

 

[bpesta22]

It's only small in the context of all the other studies he mentions.

My point being that the people who do IQ studies have no problem assigning an IQ score to half the planet because they tested 1000 kids with a reaction test that they think they can correlate to an IQ test.

Is there a scientific term for "hearsay?"

Yahzi if you'd actually read even the small section I quoted (versus the whole article) you'd see that the reaction time studies included the raven's matrices-- an untimed, widely used IQ test, and perhaps the single best measure of g in existence.

Another example of a smug but invalid criticism on your part, resulting simply from not reading what's before you. Whatever "truth" is about IQ, it's clear that the study I referenced-- and you dismissed-- indeed used a real IQ test (together with the RT tasks).

 

[Dave1001]

How could one study white vs. black IQ difference and not look at biracial people who are white+black in their analysis? Has this been done?

 

[Dancing David]

But in an inspection time task, the correlational stuff is the subject's demographic characteristics (race, age, gender, etc).

Obviously, one can't randomly assign people into the black or white groups, so indeed causal inferences are iffy.

I think we are in agreement in all this.

This applies to a solid majority of experiments within social science. Things like gender, age, political party, education level, etc. are worthy of study even if we can't use them as randomly assigned IV's.

I agree, but I should have said 'something like the independant variable'.

For example, any study showing that teachers treat girls who raise their hands differently from boys who raise their hand would suffer the same problem (one couldn't claim that student gender caused the difference in teacher behavior, as the kids weren't randomly assigned to be boys or girls). Should we not study this important area because any significant results would be correlational masturbation?

My point was that the level of coorelation is imporatant (IE coorelational intercousre), some coorelations are more significant than others.

What about the correlation between obesity and heart disease? Is that masturbation? I can't imagine a study where people are randomly assigned to be obese or normal-weight, and then followed around to see who gets heart disease.

I was not the one making the random assignment claim, Tthat is why i said that matching demographics is about the closest you can come to control

So, if you define good science as only those experiments which use random assignment, thus allowing causal inferences, you'd be throwing out many babies from the correlational bathwater.

T'weren't me that said it. i am saying that there is a level where coorelation (68% or thereabouts) means more.

***

My point was that n = 30 per group would likely give enough statistical power to show a difference on inspection time. Certainly, n =1000 is overkill (if you're just interested in showing the difference, given diminishing marginal returns that n size gives to stat power). If you'd like, I can do the calculations that plug in n size and effect size to determine how many subjects one needs to reject the null (across whatever power levels you think are appropriate).

Assuming that it is a representative sample and that it is a large enough portion of the larger group.

30 per group is good enough because of the controls and standardization of the inspection time task (e.g., not one trial but many; using a staircase method to hone in on the shortest exposure time that still results in accuarate performance, etc.).

 

[Dancing David]

How could one study white vs. black IQ difference and not look at biracial people who are white+black in their analysis? Has this been done?

Good point and how do you figure the 'high yellow' or pale skinned people, especialy since the indigenous people to southern africa are 'caucasian' in morphology but not in descent.

Where does asia begin, Turkey?

 

[Dave1001]

Good point and how do you figure the 'high yellow' or pale skinned people, especialy since the indigenous people to southern africa are 'caucasian' in morphology but not in descent.

Where does asia begin, Turkey?

The study of genetic markers for subpopulation and admixture has been going on long enough that I think IQ researchers must have been able to start using these instead of racial self-identification or morphology. Anyone have links to any studies that do this?

 

[hammegk]

But these are correlational studies. By definition, you can't "introduce control" in a study with no true independent variable.
As you know, in a real experiment, the scientist manipulates an independent variable to see if it changes the dependent variable. This other crap is mere correlational masturbation with the third variable problem as the two ton gorilla lurking in the corner, with Hammy and a few other prehominids.

Darn behavorists. The strangest crap comes spewing out at random.

Stick it in one ear & pull it out the other; clean out those cobwebs in the empty space.

 

[Jekyll]

How could one study white vs. black IQ difference and not look at biracial people who are white+black in their analysis? Has this been done?

Yes, one of the studies found a 9 point drop from mixed race(black/white) couples with a white mother compared to those mixed race couples with a black mother.

The study of genetic markers for subpopulation and admixture has been going on long enough that I think IQ researchers must have been able to start using these instead of racial self-identification or morphology. Anyone have links to any studies that do this?

Yup, they find IQ correlates much more strongly to skin colour than genetics.
Nisbett goes over both of these points in his rebuttal of the paper Bpesta is defending. You can track down citations for all of the papers from there.