Squared differences between observed and expected frequencies are used in chi-square tests because variances always use squared quantities. differences between observed and expected frequencies sum to zero. the squaring operation weights each cell appropriately. nominal data categories require it