NON‐PARAMETRIC TREND TESTS FOR LEARNING DATA

Abstract

Many current stochastic models of learning imply stationarity of portions of individual response sequences. Non‐parametric tests are presented for detecting overall group trend and homogeneity of individual trends in a sample of protocols. The protocols may differ in length and consist of any responses that take ordered values, such as errors and successes, ratings, or response latencies. The power of the test for overall group trend is discussed in relation to both linear and logistic alternatives.