The joint estimation (JE) outlier detection method of Chen and Liu [6][7] is an effective on-line process control method for time series of length x_n >= 47 [15][16]. Because of the need for process control methods for time series of less than 50 observations we investigate the effectiveness of JE for time series of short length. We show that JE has a higher frequency of detecting outliers when they occur with a lower false alarm rate than EWMA. We also show that as the length of the series is extended beyond the period where the outlier occurs. JE can often correctly identify the type of the outlier.
展开▼