Keywords

autocorrelation, control chart, empirical evaluation, power analysis

 

Authors

  1. Borckardt, Jeffrey J. PhD
  2. Nash, Michael R. PhD
  3. Hardesty, Sue MD
  4. Herbert, Joan MS
  5. Cooney, Harriet MSN
  6. Pelic, Chris MD

Abstract

Recently in this journal, F. Alemi (Qual Manage Health Care. 2004;13(4):216-221) proposed the use of Tukey's Control Chart because of the unique benefits associated with nonparametric inferential statistics including robust performance with (1) low N sizes, (2) nonnormal data, and (3) the presence of outliers. However, an assumption that applies to virtually all inferential statistical procedures (parametric and nonparametric) is that the observations in question are independent from each other. Unfortunately, there is good reason to suspect violation of this assumption when evaluating quality processes over time (as is often the case in health care settings). In this article, the power (ability to detect real changes in a data process) and type I error (probability of false positives) performance of Tukey's Control Chart technique is empirically evaluated. When observations are not independent, Tukey's Control Chart technique demonstrates unacceptable type I error performance. Additionally, regardless of whether observations are independent, the technique demonstrates low power to detect real effects unless the effects are quite large.