Interval and Band Estimation for Curves with Jumps.

Jump points in curves arise when the conditions under which data are generated change suddenly, for example because of an unplanned change in a treatment. If the size of the jump in the treatment effect exceeds the range of the error distribution then the design interval within which the jump point lies may be determind concisely. In other cases, however, estimators of the jump point are subject to sampling error. This paper suggests bootstrap methods for quantifying that error, and for constructing a confidence interval for the jump point and a confidence band for the curve.