Page 221 - Statistics II for Dummies
P. 221

Chapter 11: Finding Your Way through Two-Way ANOVA     205


                                temperatures 1, 2, and 3, respectively. Each boxplot shows the results of the
                                whiteness levels for the five shirts washed under that combination of detergent
                                and water temp.

                                Looking at these plots you can see that each detergent reacts differently with
                                different water temperatures. For example, Brand A does best in cold water
                                (water temp level 1) and worst in warm water (water temp level 2), while
                                Brand C is just the opposite, having the highest scores in warm water and the
                                lowest in cold water. Each detergent does best/worst under a different com-
                                bination of water temperatures. You can really see why the interaction term
                                in this model is significant!

                                Now which combination of detergent and water temperature does the best?
                                If you look at the plots, Brand B in cold water looks really good, and so does
                                Brand C in warm water, closely followed perhaps by Brand D in hot water.
                                This is where Tukey’s multiple comparisons come in.
                                Running multiple comparisons on all 12 combinations of detergent and water
                                temperature, you confirm that the three top combinations identified are all
                                significantly higher than all the others (because their sample means were
                                higher and their differences from all the other means had p-values less than
                                0.05). But the top three can’t be distinguished from each other (because the
                                p-values for the differences between them all exceed 0.05). Tukey also tells
                                you that the three worst combinations are Brand A in warm water, Brand B
                                in hot water, and Brand D in cold water. And they’re all at the bottom of the
                                barrel together (their means are significantly lower than all the rest but can’t
                                be distinguished from each other). So no single combination can claim all the
                                bragging rights or shoulder all the blame.

                                You can imagine the many other comparisons that you could make from here
                                to put the other combinations in some sort of order, but I think the best and
                                worst are the most interesting for this case. It’s like the fashion police com-
                                menting on what the stars wear on awards night. (Whatever they do wear,
                                let’s hope their statistician told them which brand of detergent to use and
                                what water temperature to wash it in!)


























          17_466469-ch11.indd   205                                                                   7/24/09   9:44:19 AM
   216   217   218   219   220   221   222   223   224   225   226