PUB TYPE Reports Evaluative (142)

When the same parameters are estimated by data from several independent samples, it may happen that, for any pair of samples, even though the test for parameter discrepancy is statistically significant, the two individual confidence intervals overlap. To overcome this potential contradiction, a new type of one-sample confidence intervals is developed. Their evaluation will lead to the same statistical decisions reached by the two-sample test for parameter discrepancy. Moreover, the simultaneous decisions on parameter estimation, statistical inference, and directional prediction can be made with specified confidence coefficients and error rates by simply comparing a pair of comparable confidence intervals. In contrast with conventional confidence intervals, the comparable new confidence intervals have narrower widths, disjoint or overlap depending on whether the parameter discrepancy is statistically significant or not. The proposed procedure can be applied to both simple and multiple a-priori comparisons o