# Calculating power for a logistic regression analysis

I'm wanting to calculate the level of power achieved in a logistic regression analysis in G*Power using alpha of .05 and two-tailed test. My research question was looking at whether scores on an IAT (IV1 - normal distribution) could predict insufficient or sufficient levels of exercise engagement (binary DV), over and above 2 explicit measures of motivation (IV2; IV3 - negative skews) and whether this relationship was moderated by another variable (Mod - normal distribution; including the interaction term in the analysis). Total number of covariates in the analysis was 6 (including the interaction term).

With a sample size of N = 99, I found the model was a significant fit 𝜒²(6, N = 99) = 23.85, p < .001, correctly classifying 82.8% of cases. But only IV2 was a significant predictor (OR = 2.55, 95% CI 1.07-6.07). How can I determine and report whether or not I had sufficient power in this test?

• If you have a significant result, you had enough power. If you don't, you didn't. – Jeremy Miles Dec 12 '16 at 3:28
• Okay - I ran an a priori power analysis to see whether I had enough participants and it appears that I did: testing for a logistic regression analysis predicting physical activity engagement with an odds ratio (OR) of 2.0 (Salmon, et al., 2003) at 80% power and a two-tailed significance level of .05, a sample of 92 was needed. What if my main predictor was a non-significant predictor, but had an odds ratio of (OR = 11.82, 95% CI .01-25887.58), that seems very large? – user140660 Dec 13 '16 at 23:23
• That looks like you have a separation problem - some cells are too small. (Also, don't use the classification table, it's misleading.) – Jeremy Miles Dec 14 '16 at 1:23