In Example 7.3, we assessed the uniform association between the diagnosis of probands and informants. Perform the following alternative approaches for assessing uniform association and compare the results with that of Example 7.3

(a) Fit a saturated log-linear model and test whether the association is uniform.

(b) Treating one of row and column variables as the response and the other as the predictor, apply an appropriate generalized (or adjacent) logistic model to assess the uniform association between the two variables.

Transcribed Image Text:

Example 7.3 Both the row and column variables in Table 2.9 are ordinal, and thus we may use model (7.14) to assess uniform association of proband vs informant "diagnoses". If we apply goodness-of-fit tests to the log-linear model for uniform association, the Pearson chi-square statistic is 1.18, and the deviance statistic is 1.21, both with three degrees of freedom. The corresponding p-values are 0.7587 and 0.7512, respectively. Both tests indicate that there is no sufficient ground to reject the null hypothesis that the association between the diagnoses of probands and informants is uniform. Further, A is significantly positive, indicating the two diagnoses are positively associated. If the row variable is not ordinal, we may still be able to assess the row-effect association. If model (7.15) is applied to Table 2.9, the Pearson chi-square statistic is 0.5487, and the deviance statistic is 0.5446, both with two degrees of freedom. The corresponding p-values are 0.760 and 0.762, respectively. Both tests support the row-effect association.