[parse_tree]

The only wierd thing to me is that the strength of the relationship appears to gradually weaken as income gets higher (which would make sense), but then sort of jumps up at $200,000. That's surprising, I would have thought the highest would have been around the upper middle class - comfortable enough to be able to dedicate oneself to school, but poor enough to still be hungry.

I'd like to see the relationship between extremely high SAT scores (e.g., 99th percentile) and family income. I suspect middle to upper-middle class students would proportionately outnumber the rich.

[te_platt]

Maybe. My guess is there is a correlation between {income and emphasis on education} and {emphasis on education and SAT scores}. It looks like you are saying at some score the SAT becomes something of an IQ test and IQs are more evenly distributed. I'm not sure that is the case.

[parse_tree]

No. I think SAT score would be correlated with IQ, but not too strongly on its own (many not particularly brilliant people get high scores by preparing excessively). I think measured IQ and family income would, together, be quite good predictors of SAT performance. But what leads me to believe that also leads me to believe that predictive strength would tail off as family income (or IQ) got extremely high. E.g., for very smart people, the SAT is probably not a challenge, so their score wouldn't benefit much from 10 more IQ points. Similarly, very rich people can't use their money to do much for their kids SAT score than those making $150k / year can.

[Retric]

I think people that make a lot of money tell their children to focus on the important aspects of education. And prepping for the SAT has a much better payoff than most extracurricular activities.

[stavrianos]

I think the jump is because of the cut-off. If the X were allowed to continue smoothly past 200k, the scores might have followed the same curve all the way up.

[parse_tree]

But presumably average score for each group is used. I wouldn't think the difference in average score between the 200k group, and the 220k+ group would be great enough to cause that (and I don't think there'd be too many data points the farther you went out to have much influence on the average).

Unless they were using some type of weighted average for each interval and then skewed the $200k group by adding its weighted average in with the intervals beyond it, then averaging that. Which wouldn't surprise me given the appearance of the chart.