Technical Brief
Predicting Individual School Performance

Predicting Individual School Performance
Regression equations produced by the models were used to create expected achievement scores (dependent variables) on the basis of selected known values of the school context (independent variables). These expected scores provide us with guidelines about what we can statistically expect for individual school performance in the 98-99 school year state assessments. Note that we cannot reliably predict the 1999-2000 state assessment performance from the 98-99 data. The techniques are applied here only in a retrospective analysis not for the purpose of predicting future performance. Standard (random) errors were computed in addition to the specific point estimates for the expected scores. The point estimates combined with the errors generated a band within which we would expect individual school performance to lie. The band reflects a confidence level of 95%, i.e., there is only a 5% or less chance that a school's performance would sit outside the band due solely to chance. Because of employing the 95% confidence intervals, schools with both a preponderance of the five variables we considered in our two-block model and those who have small numbers for those variables within their entire student body have much larger bands. These schools at the two extremes are prone to more errors (recall the outlier discussion above), because they are outliers in the total population of RI schools. In other words, they are least like the other schools in the entire RI sample. Due to the fact that for all tests this year we used two years worth of testing data, we were able to considerably reduce the amount of error inherent in the model. The result is that performance bands this year are smaller in size than those associated with last year's report.

Back to top || Return to the Information Works Home Page