The above graphic, from an old Wall Street Journal article, is just one of many similar graphs that get floated around on social media when people rail against the SAT. “Look!” they shout. “It’s just an income test! The SATs just tell you who the rich kids are.” Well, elite colleges are very invested in harvesting rich kids, yes, but the SAT is not the problem. This particular graphic is specifically misleading in the scale of its Y-axis - these are 800-point tests, and yet the total range of the Y-axis is 180 points, making the effect seem misleadingly large. But it’s also just weird to try and represent a simple correlation with a graph in this context. We have specific statistics that reflect what these graphs try to depict, the amount of variation in one number (in this case SAT scores) that is predictable from another number (family income). In particular, the coefficient of determination or r² is designed for exactly this purpose. What does r² tell us in this instance? Well, Sackett et al (2012) is the largest and most representative dataset of this type of which I’m aware. And across 150,000 students, they observe an r² of .0625 when regressing SAT scores and family income. That is to say, if we imagine a big pool of 100% of the variance in SAT scores, we can only predict 6.25% of that variance based on family income. And yet this condition is constantly used to insist that the SAT is “just a wealth test”!
The above graph is no doubt pulled from a different dataset, and you do see slightly higher r² figures in some research. But all of the data of which I’m aware is more like Sackett than not - consistently, familial income just isn’t a very strong predictor of SAT scores, no matter what Twitter tells you. And, given that SAT scores are not easily modified with tutoring, it’s truly strange that the very-real phenomenon of rich students having a leg up in college admissions is blamed on the test. If you can get a “college counselor” (read: mercenary who you pay to get your kid into college) to talk to you frankly, they’ll tell you that the SAT is the hardest element for rich families to game. But because our discussions on race and class privilege are always tissue-thin, and because so many people harbor personal resentments for the SAT because of the stress of taking it themselves, it’s powerfully difficult to get the establishment media to reflect on any of this.
Of course, there’s an added layer: the “antiracist” worldview holds that racial and class inequality amount to powerful obstacles for Black and poor students, which certainly is true. But if the SAT reflects race and class effects - much smaller than critics think, but real enough - isn’t that exactly what left-leaning people should expect? If you think our society is full of racial and socioeconomic privilege, wouldn’t it be bizarre to expect our educational tests to fail to reflect that reality? And wouldn’t getting rid of such tests amount to voluntarily abandoning a powerful tool for demonstrating the continuing salience of such inequality?
Correction: Got my numbers jumbled. Commenter Sean points to
In the 2006 national population of test takers, the correlation between SES and composite SAT score was .46. Therefore, 21.2% of variance in SAT scores is shared with SES, as measured here as a composite of mother’s education, father’s education, and parental income. Thus, SAT scores are by no means isomorphic with SES, although the source of the SES-SAT relationship is likely due to some combination of educational opportunity, school quality, peer effects, and other social factors.
Broader point remains - no reason to use graphs when stats are more accurate, relationship is much weaker than people think and certainly not consistent with “just an income test,” and some income effect is exactly what left leaning people should expect. Also there's the question of where the causation goes the other way and rich people are smarter, which again fits with a progressive economic perspective.
The information captured by r^2 is completely hidden in this chart; we don't even get error bars. It's compatible with nearly any strength of association between the variables. A scatterplot would have been more informative on variance but probably concealed the association from visual inspection.
Variance in often omitted (I doubt it's intentional hiding) from popular reporting, which is fine for most readers because few have learned about it to begin with. Nearly everyone can get this gist of a simple chart or graph like the one above. Never mind that variance is one of the most important concepts you'll need to understand and evaluate anything beyond summary statistics.
Even more broadly than your second point - our society, broadly speaking, tends to reward smart people who go to good colleges with higher-paying jobs - not all the time, not consistently, but as a broad trend. If there is even a tiny bit of genetic OR home environmental cause for those traits we should expect people with high incomes to have smarter and/or better-at-studying kids - not because the income itself magically causes it, but because, well, it would be weird if two mathematicians had no higher chance of having a kid who was good at math than two assembly line workers. Again, broadly speaking, general trends, stereotyping, etc, etc., but that's why it's a 6% r^2 like you said and not 100. And again, it's not necessarily even genetic, it's also just two college-educated professionals being able to help a student plan for and prioritize academics and having houses full of books and so on compared to two parents juggling fast-food jobs. It would be surprising if that didn't have any effect.