Shewhart control charts are based on a fundamental assumption that quality characteristic variables are independent and normally distributed when a process is in a state of statistical control, or in control. In practice, however, the distribution of the variables being measured may be strongly skewed. The bootstrap method is a powerful computer-based method that can provide control charts without the usual normal-theory assumptions. This paper proposes techniques for constructing control limits of X^^--s^2 charts from well-known bootstrap confidence intervals, such as the BC_a (bias-corrected and accelerated), ABC (approximate bootstrap confidence intervals) and Bootstrap-t method. These bootstrap methods have not been used in published papers for bootstrap control charts. We provide simulation studies and evaluate the performance of two conventional control charts (the standard and simple percentile bootstrap control charts) and these bootstrap control charts based on the difference between the estimated control limits and the true control limits. The results show that the percentile and BC_a methods for X^^--s^2 charts perform comparatively better than the standard method when the process distribution is skewed. The control limits estimates for the ABC and Bootstrap-t methods are less accurate than those for the other methods that were examined. In particular, the Bootstrap-t method estimates for the s^2 charts is found to be unreliable
View full abstract