Abstract:
One of the test of hypothesis which is popular to the Sociologist is the test of independent between two variables. This method can be applied by recording the frequencies in the two way frequency distribution or which is called the contingency table. The statistical value for the test will be the total of the squared difference between the frequency from the test and the frequency of the assumption devided by the frequency of the assumption. The statistical value for this test will approximately distribute as chi-square distribution under the above assumptions. The statistical value will be concluded the test of hypothesis whether the result will be accept or reject the independent hypothesis of the two variables. This can be done only by comparing with the critical value which is determined only by level of significance and the degrees of freedom. The solution from this method is still debatable whether it is appropriate or not. It is doubted that the statistical value which is the result can explain linear relationship between two continuous variables, a case of the objectors. The size of samples, tables and the classification of data whether or not will effect the suitabilits of the test. These issue can be regcognized by more statistical calculation so that it can be reliable. For example, the calculation of contingency coefficient which bring sample size to the consideration expressed by c = √w / w+n and the Cramers V which bring both sample sizes and size of tables to the consideration expressed by คือ v² = w / n . min (r-l, c-l) where w is the statistical value for the test n is sample sizes r is member of data classify by row c is member of data classify by column For answering the problem, we determine the study by simulation method which will generate random numbers which are bivariate normal and bivariate multinomial. The sample sizes have been set up for studying are 20, 30, 40, 50, 75 and 100 and sizes of table level are set up at 2x2 2x3 2x4 2x5 3x3 3x4 3x5 4x4 4x5 and 5x5. In case of bivariate normal, we shall determine only absolute value correlation coefficient from 0.00 to 0.98 by classify level 0.00 to 0.40 with difference 0.02, 0.40 to 0.60 with difference 0.01 and 0.60 to 0.98 with difference 0.02. In case of the bivariate normal distribution the influence to the value of Chi-square of the sample sizes, sizes of table and classification data has been studied. In the aspect of testing hypothesis, it is found that sample sizes and sizes of table make simulation significance level higher than theoretical significance level and simulation critical value differs from each other sample sizes at the same significance level and the same degrees of freedom. In the case of identifying the relation between correlation coefficient and the value of Chi-square, it is found that, when sample size and size of table increase, the expected value of correlation coefficient will diminish. Different grouping in some cases of study will make the means and variances of Chi-square values differ in each grouping. In case of the independent multinomial distribution, it is found that all sample size and all size of table, the means of Chi-square values are consistent with of theoretical value and the variances tend to be less than the theoretical values. As when the sample size and size of table increases, the values of simulation significance level are likely to be less than theoretical significance levels. But, if both variables are correlated, it is found that the mean and variance of Chi-square value tend to increase. When sample size increases, the simulation significance levels appear to be higher than theoretical significance levels in each size. Only in the case of bivariate normal, Chi-square value calculated from models and table mentioned above may be used to indicate the confidence interval and the expected value of the correlation coefficient. Moreover, if the size of sample varie, estimation of the expected value of correlation coefficient may also be found from the simple linear regression equation which has the constant value (a) and regression coefficient (b) provided at the specified values of Chi-square and size of table.