A u-i approach to retrospective testing for shifting parameters in a linear model

A union-intersection-based procedure for detecting parameter shifts in a linear model is studied. The advantage of this procedure over those in the literature is that it allows unlimited (possibly data-directed) investigation of the nature of any detected shift via testing of sub-hypotheses with a controlled error rate. Weak convergence theory is used to obtain suitable critical values and to study local power. Monte Carlo methods are employed to study fmed-alternative power.