The only thing that the [standardized] tests are sensitive to is SES and racial/ethnic characteristic. If those two variables were partialled out statistically, the results would show that schools are pretty feckless instructionally.
That's not "news." It's been around since the Coleman Report of the 1960's. But the popular conclusion is that we have to "change society." The "obvious conclusion" has been overlooked--change instruction. When you do, you find that the correlation with accomplishments and SES is near-zero. That's empirical reality, not a statistical manipulation.
At least at the elementary level and if we don't include comprehension with uncontrolled vocabulary, and if we teach the higher-SES kids like we do currently, but his larger point is valid. SES matters quite a bit as long as instruction is rotten.