[理学]ExperimentalDesignchapter2-2012.ppt

Introduction In this chapter we consider experiments with one single factor. Two models, ANOVA and Regression, are introduced. The least squares and maximum likelihood estimators are given. For the data of Example 2.1, k=5,m=10, SSE SSLof Now we have two unbiased estimators of s 2: sE 2 and sLof2. If these two estimators have a significant difference, it indicates the lack of fit. The test for model adequacy via the lack of fit approach is given by: Model should be improved Regression model is ok Model is ok, not perfect Choose other models * See p.14 in the Chinese book. If Flof is rejected If the F-value given by the lack-of-fit shows the significant differences, then we can conclude that the regression model doesn’t fit the data set, and the F-value for the regression is not good enough. Therefore, we should look for another model to fit this data set. Table 2.6: 2.5 Multiple comparisons A set of means: 9.8, 15.4, 17.6, 21.6, 10.8 We want to test: or In general: A linear combination of treatment totals such that with restriction such linear combinations are called Contrasts. A. The Least Significant Difference (LSD) Method. that is called the least significant difference, where MSE is the estimation of variance of random error. By using the LSD, we can see the above table results are mostly significant at all levels except between (15.4 17.4) and (9.8 10.8). Thus can reject the null hypothesis. If We conclude levels i and j have a significant difference. ** ** * * * ** ** * B. Bonferroni Test The LSD test is valid for testing one pair of treatments. Suppose that m such tests are performed. It is easy to show that, under H0, the probability of declaring at least one pair of treatments significantly different (which is called the experimentwise error rate) exceeds αfor m1. For larger m, the experimentwise error rate is higher. Bonferroni method for testing , if In the case of the one-w