"N = 23; Full Scale IQ"
Enough for me to pretty much stop reading right there. Isn't 23 far too small? If there is signal there, you probably need more samples to draw it out.
Enough for me to pretty much stop reading right there. Isn't 23 far too small? If there is signal there, you probably need more samples to draw it out.
For example, N = 1 is enough if the rat can solve crossword puzzles. Effect size is so large that probability that you could guess which rat is in the control group and which is not is close to 1.0.
In this case, such large effects can be rejected pretty soundly; with the sample sizes given, 23 vs 1058, you have 80% power to detect `pwr.t2n.test(n1=1058, n2=23, power=0.80, alternative="greater")` ~> d=0.52 (which is, for example, about half the size of the black-white IQ gap, and considered a 'medium' effect size).
In addition, the longitudinal cohort studies of the gifted like Terman, Hunter, TIP, or SMPY, do not find any signs of severe emotional/behavioral/social difficulties, the genetic correlations (https://en.wikipedia.org/wiki/Genetic_correlation#Intelligen...) are all in the wrong direction, adults are more able than kids to self-select into stimulating environments (possibly why heritability of intelligence increases with age), and so on. A new review paper which is relevant: "From Terman to Today: A Century of Findings on Intellectual Precocity" http://emilkirkegaard.dk/en/wp-content/uploads/From-Terman-t... , Lubinski 2016.
From a statistical perspective, this result doesn't tell us much about gifted adults not because the sample size is far too small, but because IQ scores at age 5-6 inherently don't predict adult IQ sufficiently well. This subsample of 23 kids with mean IQ of 134.6, selected because they are past the 130 threshold, are going to regress to the mean as they grow up. With the usual r=0.5 of 5yo/adult scores, their adult mean IQ will be more like 117 - which is above average but not 'gifted' by anyone's criterion. To try to study gifted adults by starting with them as kids, you would need to start another standard deviation up, around 150 IQ, which would be extremely difficult. (The correlation rapidly increases up to middle school, which is why TIP/SMPY don't suffer from this issue as much: much higher threshold and also much more reliable tests.)
> "With the usual r=0.5 of 5yo/adult scores, their adult mean IQ will be more like 117"
As a total layperson, this is surprising! I would have thought that smart kids would have tended to stay smart, rather than have whatever effect be subject to a regression to the mean. Is this relationship well-established?
What do you think of that argument, and do you know of any relevant research on that aspect of the question? It anecdotally resonates with my experience, and seems to be a prominent subsection of the responses in most threads of this type, but I would like to know what formal attempts have been made to measure it, if any exist.
Also, again IIRC, I thought Towers reassessed one of Terman's longitudinal studies and found that there was a correlation between the highest IQ individuals and social/emotional mal-adaption. Do you disagree with Towers' assessment?