Application of Pearson’s X^2 Test of Independence with Small Expected Cell Frequencies. Abacus

Abstract

Pearson’s X^2 is re-examined within the context of small expected cell frequency (e_ij< 5), for the Pearson’s X^2 statistic to satisfy the asymptotic approximation to Chi-square distribution. This paper proposes scalar multiplier theta > 0, such that theta(e_ij') ≥ 5, where e_ij' is the smallest expected cell count in the contingency table under consideration. The product of the sample size ‘n’ and ‘theta ’ results in each cell count becoming (theta)n_ij, which does not cause any change in the cell probabilities. Thus the assumption of independence is thereby satisfied. This approach guarantees the safe application of Pearson’s X^2 for test of independence under small expected cell counts with the degrees of freedom also multiplied by theta.