In a previous study, the performance of goodness-of-fit test based on clustering partitioning strategy for multinomial logistic regression has been investigated by using different type of clustering techniques. It is known that the clustering technique involves distance calculation to identify which cluster a data belongs to. There are several distance matrices available, but the most popular one is the Euclidean distance. Thus the previous study only considered the Euclidean distance in the process of clustering the data. In this study, the effect of different distance matrix has been investigated. Four different distance matrices, which are Maximum, Manhattan, Minkowski and Canberra were used to investigate the effect on the performance of the test. The results were then compared with the results from previous studies, which used Euclidean distance. The results show that Canberra distance produced different mean value and rejection rate while the other distance measurements produced the same mean value and rejection rate as Euclidean distance.
