Page 207 - Six Sigma Advanced Tools for Black Belts and Master Black Belts
P. 207

OTE/SPH
 OTE/SPH
          August 31, 2006
 JWBK119-12
                          Introduction to the Analysis of Categorical Data
        192              2:58  Char Count= 0
        Table 12.14 Comparison of possible logistic regression models with only main effects.
                                                                       95% CI of
                                                                       odds ratio
                                               SE          p-  Odds
        Predictors                  Coefficient Coeff.  Z  value ratio Lower Upper
        Constant                    −19.761   6.515 −3.03 0.002

                   Dummy variable
                     combinations
        Spine
        condition  c 21     c 22

        2          1         0      −0.962    1.397 −0.69 0.491  0.38  0.02  5.91
        3          0         1      −2.097    1.055 −1.99 0.047  0.12  0.02  0.97
        Width                        0.837    0.256  3.27 0.001  2.31  1.40  3.82





        variable is added. Here, a model with spine and carapace width is adopted. The logis-
        tic regression table for the refined logistic regression model with spine and carapace
        width is shown in Table 12.14.
          From Table 12.14, it can be observed from the p-value for the coefficient estimates
        that the dummy variable representing spine condition 2 (p-value 0.491) appears to
        be unnecessary. Spine conditions 1 and 2 can potentially be grouped together. It is
        expected that this will not affect the predictive power of the simplified model. Hence,
        a simplified model is fitted with only one dummy variable to differentiate crabs with
        spine condition 3 from the others. The parameter MLEs together with their corre-
        sponding p-value and odds ratio are shown in Table 12.15. From equation (12.16), this
        model gives a likelihood ratio statistic of 21.964 with 2 degrees of freedom. Compared
        with the earlier model with an additional dummy variable distinguishing spine con-
        dition 2, the difference in the likelihood ratio statistic, G D , is 0.479. The p-value based
                              2
        on an approximate null χ distribution with 1 degree of freedom, is 0.5 implying that
        this simplified model is no different from the earlier model with 2 dummy variables
        differentiating spine conditions 1, 2 and 3.




        Table 12.15 Comparison of possible logistic regression models with only main effects.
                                                                     95% CI of
                                                                     odds ratio
                                      SE             p-    Odds
        Predictors       Coefficient  Coeff.  Z     value  ratio  Lower   Upper
        Constant          −20.871    6.339  −3.29   0.001
        Spine Condition 3  −1.732    0.857  −2.02   0.043  0.18    0.03    0.95
        Width                0.866   0.251    3.44  0.001  2.38    1.45    3.90
   202   203   204   205   206   207   208   209   210   211   212