This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Stat 201A HW5 Solution 1. (a) The correlation between A and DF is 0.5. Main effect A is partially aliased with 16 2factor interactions (2fi’s) which form 4 groups: BC = DG = EF = HK, BF = CE = DK = GH, BG = CD = EK = F H, BK = CH = DF = EG The correlation between A and the group of BG is . 5 while the correlation between A and other groups is 0 . 5. A is not aliased with other 3 groups: BD = CG = EH = F K, BE = CF = DH = GK, BH = CK = DE = F G (b) The main effects analysis suggests D and C are important and followed by F, E, H . But the residuals plot reveals the model is not adequate. Since A and DF is not fully aliased, it is possible to include them in the same model. In the analysis, we can include all 9 main effects plus 6 2fi’s. We can include one 2fi from the 3 groups that are not aliased with A , say BD, BE, BH . For the 4 groups that are partially aliased with A , we can include at most 3 2fi’s from the 4 groups. Of course, we don’t know which to include and have to do variable selections here. There are many procedures. I apply the Dantzig selector to a model with 9 main effects and 7 2fi’s ( BC, BD, BE, BF , BG, BH, BK ). By inspecting the profile plot, I conclude C, D, E, F, H, BG = CD, BH = DE, BK = DF are important (using δ = 20). As we are interested in testing whether A is negligible, I include A in the model, the summary is Estimate Std. Error t value Pr(>t)Estimate Std....
View
Full Document
 Spring '10
 wu
 Correlation, Degrees Of Freedom, Normal Distribution, Trigraph, Errors and residuals in statistics, Residual standard error, freedom Multiple Rsquared

Click to edit the document details