Quote:
Originally Posted by Arouet
I read the article. What I meant was, is 6 points close enough that it might be in the margin of error to make the two numbers statistically the same? I'm not great with stats, but PZ seems to make the same point.
PZ's concerns:
Quote:
Show me the error bars on those measurements.
I think these comments have more to do with the sampling of a single individual, not the collective sampling of multiple individuals. In particular, one's score on an IQ test varies with all sorts of variables (fatigue, time of day, time since the last meal, etc) which are very likely not controlled in the sampling. This means that you can take the test twice and get two different scores. I couldn't find any information about the level of variation specifically on IQ tests based on these factors, but I do know that there have been a multitude of studies on the performance of children on tests that shows a significant shift in their performance based on the conditions.
Given Dr. Kanazawa's apparent inability to properly analyze data, it's not at all clear whether his claim of statistical significance is mathematically valid, let alone methodologically sound.
Quote:
Show me the reliability of IQ as a measure of actual, you know, intelligence.
This connects to the question of methodologically sound, and the commentary I linked to on the other post. It seems that the assumptions he makes regarding the nature of intelligence are not consistent with the field (in particular, not empirically supported), so that his conclusions regarding the connection between IQ and intelligence may be completely meaningless.
Quote:
Show me that a 6 point IQ difference matters at all in your interactions with other people, even if it were real.
I think this is the statement that links to your original comment:
Quote:
What I mean is is it statistically significant within the confines of a study? PZ raises the same point. Should we really consider these results to mean anything?
It's not so much about statistical significance, but rather it's whether the statistically significant data means anything relative to the claims being made. Here's an example:
In town A, a sample is chosen and the average household in the sample makes $50,000. In town B, a similar study is conducted, and the average household makes $100,000. This study turns out to be statistically significant in the sense that the samples were large enough it's unlikely that the error will make a significant change in the computed averages. One might try to claim that this shows that the people in town B are wealthier.
But the reality is that the cost of living in town B is several times larger than the cost of living in town A, so that those living in town A actually have more spending power (ie, wealthier) than those living in town B.
This type of distinction is important when you're considering sociological data (psychometrics, in this case) because the thing you're measuring may turn out to be different from the thing that you're really trying to address.