A new statistical test for trends: establishing the properties of a test for repeated binomial observations on a set of items
D.C. Howard · The Aphasiology Archive (University of Pittsburgh) · 2007
Statistical testing of trends during therapy have been limited because such tests are limited when there is serial dependency in the data. This paper describes a trend test for repeated measurements on a set of items where responses to individual items are correct or incorrect - the typical form of data in therapy studies. Simulations show hat this test produces valis results even with high item serial dependency, can be used to compare performance across sets of items, and can produce a measure of the number of items acquired per therapy session that is suitable for meta-analysis. The simulations suggest that multiple probing of performance during therapy is unnecesary and often unproductive. Power is much improved by larger numbers of items in treatment sets.