ENBIS: European Network for Business and Industrial Statistics

Run orders in factorial designs to protect against a possible time trend

22 September 2009, 15:05 – 15:25

Abstract

Submitted by

Adrian Wilk

Authors

J. Kunert, A. Wilk

Affiliation

Technische Universität Dortmund

Abstract

The results of an experimental design realised in a time sequence depend on the run order in the presence of a time trend. For example, imagine the situation that a monotone time trend exists and that from a factor with two levels the low level runs are all executed before the first high level run is performed. In doing so the factor would be liable to be declared to have a significant influence on the response variable by a statistical test, although the factor is not active in reality. This is due to the fact that the effect estimator is highly correlated with the time trend. Hence the task is to get a grip on this difficulty by a suitable run order.

Different strategies exist to compensate a possible time trend. We compare three strategies with respect to their power in the presence of a time trend. The power itself is quantified by the probability to identify truly active effects as active effects. A further criterion describes the keeping of the level of significance. We use simulation to quantify the influence of time trend on the various designs and observe good performance of randomised orders. In our simulation studies we confine ourselves to analysing factorial designs of length n = 8, n = 16 and n = 32.

View paper

Return to programme