A Statistical Search for Unusually Effective Schools.

A statistical analysis of data on Michigan, New York City, New York State, and Project Talent schools found evidence of schools that consistently produce outstanding students even after allowance is made for the different initial endowments of their students and for chance variation. Methodologically, like many previous studies, this report uses regression analysis of achievement data, but focuses on statistical outliers rather than central tendencies. Three tools of analysis were used to examine the residuals: (1) Histograms of residuals, showing no immediate evidence of extreme overachievers. (2) Comparisons, over different grades and years, of the number of schools that consistently overachieved with the number expected assuming all residual variation was random. Evidence of unusually effective schools was found. (3) Comparisons of background characteristics of the hypothesized overachieving schools with those of the average school. Outstanding Michigan schools tended to have smaller class sizes, more teachers earning over $11,000, and more teachers with greater than five years' experience. 90 pp. Ref.