Take the DDK2011 dataset. Create a variable testscore which is totalscore standardized to have mean zero and variance one. The variable tracking is a dummy indicating that the students were tracked (separated by initial test score). The varible percentile is the student’s percentile in the initial distribution. For the following regressions cluster by school.

(a) Estimate a linear regression of testscore on tracking, percentile, and percentile^2. Interpret the results.

(b) Suppose the scores were censored from below. Create a variable ctest which is testscore censored at 0. Estimate a linear regression of ctest on tracking, percentile, and percentile^2. How would you interpret these results if you were unaware that the dependent variable was censored?

(c) Suppose you try and fix the problem by omitting the censored observations. Estimate the regression on the subsample of observations for which ctest is positive.

(d) Interpret and explain the differences between your results in (a), (b), and (c).