The Descriptive Statistics Report provides a summary of the descriptive statistics calculated for the cluster features. This report is used to evaluate the effects of trimming 2.5% (or the specified value) off the right-hand tails of the distributions on the sample averages and standard deviations. A sample report is shown in Figure 11-25. In Section 11.4.3 we discussed trimming a small percentage of the right-hand tail from the distribution to minimize the estimate of the standard deviation that was made for the population. For example, if the average value for CPU time in the full sample is 16.6 with a standard deviation of 175.1 seconds, then the coefficient of variation (that is, the standard deviation divided by the mean) is 10.55. As a general rule, a high coefficient of variation (that is, greater than 2.5) indicates that the outliers are still present in the trimmed distribution. If, for example, 2.5% of the observations are excluded from the calculation of the sample statistics, the average and the standard deviation would be 5.8 and 11.1 seconds, respectively. These values result in a coefficient of variation of 1.9. In this batch job class study, the statistical behavior of the observations would be significantly improved by excluding 2.5% of the observations. If the coefficient of variation for one of the features exceeds the recommended range, you should decrease the percent of the distribution to be included in the trimmed statistics by adjusting Observation trimming (percent) on the Statistical Analysis Parameters screen shown in Figure 11-22.
DESCRIPTIVE STATISTICS NUMBER OF OBSERVATIONS: 2000 ----------SAMPLE---------- ------TRIMMED 97.50%------ FEATURE MIN MAX AVERAGE STD DEV CV AVERAGE STD DEV CV -------- --------- --------- --------- --------- ---- --------- --------- ---- JOBMXNTA 0 6 0.2135 0.54687 2.6 0.154951 0.391913 2.5 JOBTCBTM 0.01 7115.71 16.6509 175.139 10.5 5.80658 11.0867 1.9 JOBNLR 0 334275 2327.42 12277.9 5.3 1019.77 2339.31 2.3
Figure 11-25. Descriptive Statistics Report
|
Copyright © 2014 CA.
All rights reserved.
|
|