Abstract: | The problem of categorial data analysis in survey sampling arises because of non‐independence of sample elements of the sample obtained through imposed sampling design. In this article the performance of modified χ2 statistic for testing independence of attributes have been evaluated for small sample sizes with the help of log‐linear models with respect to its achieved level of significance for fixed nominal level at 5%, through simulation technique. It hase been observed that the perfiormance of these test statistics depends on average and coefficient of variation of eigen values of design effect matrix. The first order corrected statistic is able to capture the effect of sampling design to a great extent but the performance of second order corrected statistic is much better. Further, these modified χ2 test statistics were applied to a real survey data and their performance were evaluated with respect to their achievied level of significance. |