Victor Lavy, Olmo Silva, Felix Weinhardt, The Good, the Bad and the Average: Evidence on the Scale and Nature of Ability Peer Effects in Schools (abstract):
We study the scale and nature of ability peer effects in secondary schools in England. In order to shed light on the nature of these effects, we investigate which segments of the peer ability distribution drive the impact of peer quality on students’ achievements... We find significant and sizeable negative peer effects arising from students at the very bottom of the ability distribution, but little evidence that the average peer quality and the very top peers significantly affect pupils’ academic achievements... We further provide evidence that the effect of the very best peers substantially varies by the ability of other pupils. On the other hand, the effect of the very worst peers is similarly negative and significant for boys and girls of all abilities.
I don't have access to the full paper, but intuitively, this sounds about right, and provides a hook for discussing peer effects in charter schools. Gifted and talented programs, some magnet schools, and other exclusive schools are designed to "cream" the best students. Some charters also "cream," explicitly or through more subtle means (depending on local rules), but my perception is that the bulk of lottery-based charters, intentionally or not, have a process better described as decanting:
Decantation is a process for the separation of mixtures. This is achieved by carefully pouring a solution from a container in order to leave the precipitate (sediments) in the bottom of the original container. Usually a small amount of solution must be left in the container, and care must be taken to prevent a small amount of precipitate from flowing with the solution out of the container.
That is, it is not about pulling the best off the top, but about leaving the most extremely difficult and needy students behind. This does not need to be as intentional or active as the term "decantation" would suggest though. Say you're selling your food stamps to buy drugs or considering the possibility of fondling your step-daughter: you're not likely to sign the kids up for a charter school that will lead to lots of calls and perhaps unexpected house-visits from nosy white 20-somethings. I'm not saying that's a big percentage of parents in high poverty neighborhoods, but issues like these play a big role in the lives of the very highest need kids, which, as this research would seem to verify, play an outsized negative role in the life and success of a school.
And, I guess one underlying issue I have with the whole NCLB accountability regime, which I don't focus on that much, is simply that it strongly discourages creating a school to specifically addressing the very lowest achievers. You're just not going to get the test scores you need; you can't pull up the lowest acheivers far enough, and at that extremity (e.g., 16 year olds with "third grade" reading skills) I'd imagine that growth models will be especially funky. And having those kids in the building will make the rest of your scores lower than an equivalent school without them.