Page 53 - Contributed Paper Session (CPS) - Volume 4
P. 53

CPS2119 Noor Azlin M. S. et al.
              TABLE 3.  Goodness of fit of the ordinal regression model with different link function using
                                          Pearson chi-square
                    Ordinal Model
                   with Selected Link   Pearson chi-         df          p-value
                                          square
                       Functions
                  Logit                  4357.87           2693           0.000
                  Probit                 7810.00           2693           0.000
                  Clog-log               4083.16           2693           0.000
                  Nlog-log              269025.68          2693           0.000
                         TABLE 4. Pseudo R-square Statistics with different link function
                      Ordinal Model with                Pseudo R-square
                    Selected Link Functions      Cox and Snell       Nagelkerke
                  Logit                              0.444             0.573
                  Probit                             0.433             0.559
                  Clog-log                           0.431             0.556
                  Nlog-log                           0.400             0.517
                The  result  of  goodness-of-fit  demonstrated  that  all  models  with  four
            different link functions are not well fitted (p< 0.000). Literature has mentioned
            that chi-square is likely to be highly significant with a large sample size, which
            in this study should be achievable as 2,808 samples has been used (Agresti,
            2007; Lall, 2004).  Moreover, chi-square is sensitive to empty cells especially
            when dealing with large number of categorical predictors. Therefore, other
            methods of assessing the goodness of fit, such as measures of association, like
                          2
                                                            2
            the pseudo R , are instead used. The results of R  value as shown in Table 4
            indicated that the ordinal model with logit link function has the highest value
                2
            of R  which is 0.573.
              TABLE 5.  Parameter estimates of determinants of level of family well-being using ordinal regression
                                      model with different link functions
            Variables               Ordinal Regression Model with Selected Link Functions
                               Logit          Probit          Clog-log        Nlog-log
                          Estimate   SE   Estimate   SE   Estimate   SE    Estimate   SE
            Strata
               Rural (Ref)   -       -       -       -       -        -      -       -
               Urban        0.032   0.155   0.009   0.062   0.092   0.085   -0.038   0.054
            Ethnic
               Indian (Ref)   -      -       -       -       -        -      -       -
               Chinese      0.296   0.189   0.162   0.95    0.198   0.136   0.039   0.091
               Bumiputra   0.364*    0175   0.199   0.103   0.283*   0.127   0.050   0.084
            Education Level
               Primary (Ref)   -     -       -       -       -        -      -       -
               Tertiary     0.256   0.186   0.118   0.083   0.227   0.135   0.0130   0.087
               Secondary    0.213   0.354   0.142   1.00    0.204   0.115   0.116   0.071
               No Formal
               Education    0.275   0.248   0.172   0.136   0.185   0.188   0.339   0.117
            Type of Family
               Nuclear (Ref)   -     -       -       -       -        -      -       -
               Extended     0.216   0.254   0.096   0.083   0.074   0114    0.064   0.070
               Blended      0.074   0.274   0.006   0.150   0.022   0.190   -0.105   0.137
               Single-Parent   0.061   0.558   0.067   0.305   -0.058   0.354   0.111   0.293

                                                                42 | I S I   W S C   2 0 1 9
   48   49   50   51   52   53   54   55   56   57   58