An Introduction to Power Analysis, N = 1 1. H: 100 H1: > 100 N = 1, = 15, Normal Distribution For = .10, Zcritical = 1.28, Xcritical = 100 + 1.28(15) = 119.2, therefore: we reject H if X 119.2. Our chances of rejecting the H are 10% or less if H is true: 10% if = 100, less than 10% if < 100. That is, with our -criterion set at .10, we have no more than a 10% chance of making a Type I error when H is true. What if H is false? What are our chances of making a Type II error? Assume the Truth is that = 110. We shall fail to reject the false H when X < 119.2. P(X < 119.2 | = 110, = 15) = P(Z < (119.2 - 110) / 15) = P(Z < .61) = approx .73 or 73% = . Our chances of correctly rejecting this false H, POWER, = P(Z .61) = 27%. That is, were the H wrong by 10 points, 2/3 , we would have only a 27% chance of rejecting H. ----------------------------------------------------------------------------------------------------------------------2. Raise to .20: Zcritical = .84 Xcritical = 100 + .84(15) = 112.6 Power = P(Z (112.6 - 110) / 15) = P(Z .17) = 43% Increasing Raises Power, but at the expense of making Type I errors more likely. ----------------------------------------------------------------------------------------------------------------------3. Increase The Difference Between H And The Truth: = 120 P(Z (112.6 - 120) / 15) = P(Z -.49) ==> approximately 69% Power Big Differences (Between H & Truth) Are Easier To Find Than Small Differences Copyright 2000, Karl L. Wuensch - All rights reserved. power1.docx 2 4. Lower by using more homogeneous subjects and by holding constant various extraneous variables: = 5 Xcritical = 100 + .84(5) = 104.2 P(Z (104.2 - 120) / 5) = P(Z -3.16) = 99.9% = Power Lowering Increases Power ----------------------------------------------------------------------------------------------------------------------5. Directional vs. Nondirectional Hypotheses A. Nondirectional H: = 100 H1: 100 N=1 = .10 = 15 Zcritical = 1.645, that is, reject H if |Z| 1.645 Xcritical 100 + 1.645(15) 124.68 100 - 1.645(15) 75. 32 Assume that the true = 110 P(X 124.68 OR X 75.32) = POWER P(X 124.68) = P(Z (124.68 - 110) / 15) = P(Z .98) = .1635 P(X 75.32) = P(Z (75.32 -110) / 15) = P(Z -2.31) = -----Power = 17% .0104 ----------------------------------------------------------------------------------------------------------------------B. Directional, correct prediction by H1 See Example 1 on the first page, Power = 27% 3 C. Directional, incorrect prediction by H1 H: 100 H1: < 100 Zcrit = -1.28 Xcrit = 100 - 1.28(15) = 80.8 Power* = P(X 80.8 | =110) = P(Z (80.8 - 110) / 15) = P(Z -1.95) = 3% Thus, if you can correctly predict the direction of the difference between the truth and the H, directional hypotheses have a higher probability of rejecting the H than do nondirectional hypotheses. If you can't, nondirectional hypotheses are more likely to result in rejection of the H . *One could argue that the probability here represented as “Power” is not Power at all, since the H ( 100) is in fact true ( = 110). Power is the probability of rejecting H given that H is false, not true. Copyright 2000, Karl L. Wuensch - All rights reserved.