A parametric control chart adjustment for handling serial dependence in health care quality data

Numerous tools are available to health care quality managers geared toward helping them make data-based inferences about quality processes. Recently, in this journal, Tukey's control chart technique was promoted as a good option for handling short streams of time series data when the assumption of data normality cannot be confirmed. Although this technique does not appear to perform well with serially dependent (or autocorrelated) data, an autocorrelation-corrected version of the technique is now available. However, when managers wish to capitalize on the superior power of parametric control charts (ie, when data sets are large and normality can be confirmed), there are currently few options available in the way of statistical process control techniques that appropriately handle autocorrelated data. In this article, the authors report the empirical false-positive rates and power performance of the mean-sigma (X-S) control chart technique under various levels of autocorrelation. Results indicate that this popular technique offers poor false-positive control with autocorrelated data. Next, the authors describe a method for autocorrelation correction and finally compare the autocorrelation-corrected X-S chart with the original X-S technique. The autocorrelation-corrected X-S chart demonstrates better type I error control with similar power to the original chart and may offer quality managers an important tool for appropriately handling autocorrelated quality data.