Wapo claims that “the least-gifted children of high-income parents graduate from college at higher rates than the most-gifted children of low-income parents,” basing “giftedness” on polygenic scores for educational attainment. The author is either intentionally being deceptive or hasn’t thought through regression toward the mean. Polygenic scores are currently very imperfect, meaning that the true genetic potential of someone who scores high or low will on average be closer to (regress toward) the mean. If there are two groups with two different means, then they will regress toward different means. Thus, a high-scoring individual in a group with higher genetic potential should be expected to be, on average, genetically smarter than an individual with the same score but whose group has lower genetic potential.
To give a concrete example, suppose you take a group of people who are, on average, 5 feet tall and measure their height, but add some random “error” onto the measurements of each person. Then you take the “tallest” people in the group using your measure, who are between 5’10” and 6″2. On average, these individuals will have gotten “lucky” in both actual height and the “error” which was added to their measured height, so their real heights will be lower than the measurement though still higher than the average for their group. But suppose you do this for a group which averages 6 feet tall, from which you select the group of people measured to be between 5’10” and 6″2. Someone in this group will be just as likely to be a tall person who got inches knocked off as short person who got inches added on, so his true height will be about 6 feet.