Is There Any Cream to Skim? Sorting, Within-School Heterogeneity, and the Scope For Cream-Skimming

Critics of school choice argue that cream-skimming will worsen outcomes for those left behind in public schools, a dynamic that relies on a substantial degree of withinschool heterogeneity. Since “high quality” families may have already sorted themselves, or may represent a small fraction of the total, this paper will examine whether existing within-school heterogeneity leaves any scope for cream-skimming to operate. The first empirical section shows that the assumptions made by simulation studies over-estimate within-school heterogeneity by at least 20% to 40%, thus inflating the cream-skimming effect. The second empirical section asks, “given the current level of within-school heterogeneity, how strong would peer effects have to be to significantly worsen outcomes for those left behind?”. In order for cream skimming to lower math test scores by a decile, the peer effect would have to be larger than the effect of converting both parents from college graduates to high-school dropouts. In order for cream skimming to substantially worsen dropout rates or college attendance rates, the peer effect would have to be two to three times larger than the strongest estimated predictor of these outcomes. The required peer effects would be smaller, but still unreasonably large, if family types started from a uniform distribution. These results indicate that current levels of within-school heterogeneity are so low that peer effects would have to be unrealistically strong to give cream skimming any bite. Acknowledgements: My dissertation committee Jim Hines (Chair), John Bound, and Dan Silverman has provided valuable advice and feedback, as have Julie Cullen, Charlie Brown, and Lucas Davis. Fellow graduate students Mike Lovenheim and Jon Lanning have also provided valuable feedback and advice. All errors are of course my own.