ELL is a bucket that has people with learning disabilities mixed with immigrants and children of immigrants. The fact that a 14-year-old Korean kid who just moved to America can’t remember the English word for “photosynthesis” is not a reflection of poor skills in science but poor skills in English. I argue that the intent of a science test is to test knowledge of science, and that giving immigrants poor marks for subjects they are good at is nothing more than a failure of testing. This failure should be known and understood when analyzing test results—no test is perfect, but we should be able to account for known sources of error. Simpson’s paradox is the classic example of what wrong conclusions you can draw when you fail to account for variables which are known to affect the result.
I agree with you as to the Korean immigrant. But ELLs as a class are not "behind native English speakers only when you measure them in English". They are native English speakers themselves, and they are behind the non-ELL class because they are selected for poor performance.
This does not reflect the reality of US practice, in which students may be labeled "ELL" even if they prefer to speak English at home and/or with their friends. You graduate from the ELL designation by passing a standardized test, so it's more of a designation for "who's stupid" than for "who has trouble with English". It's not really surprising that a group defined by their inability to score well on standardized tests turns in poor scores on other standardized tests.