Evaluating Estimation Methods of Missing Data on a Multivariate Process Capability Index

Quality of products has been one of the most important issues for manufacturers in the recent decades. One of the challenging issues is evaluating capability of the process using process capability indices. On the other hand, usually the missing data is available in many manufacturing industries. So far, the performance of estimation methods of missing data on process capability indices has not been evaluated. Hence, we analyze the performance of a process capability index when we deal with the missing data. For this purpose, we consider a multivariate process capability index and evaluate four methods including Mean Substitution, EM algorithm, Regression Imputation and Stochastic Regression Imputation to estimate missing data. In the analysis, factors including percent of missing data (k), sample size (m), correlation coefficients (rho) and the estimation methods of missing data are investigated. We evaluate the main and interaction effects of the factors on response variable which is defined as difference between the estimated index and the computed index with full data by using General Linear Model in ANOVA table. The results of this research show that the Stochastic Regression Imputation has the best performance among the estimation methods and the percent of missing data (k) has the highest effect on response variable. Also, we conclude that the sample size has the lowest effect on response variable among the mentioned factors.