tag:blogger.com,1999:blog-6894866515532737257.post4119414902572947109..comments2024-09-18T04:06:40.413-07:00Comments on Probably Overthinking It: Internet use and religion, part fiveAllen Downeyhttp://www.blogger.com/profile/01633071333405221858noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-6894866515532737257.post-62775660818006783322015-12-09T05:50:58.299-08:002015-12-09T05:50:58.299-08:00Yes, that would probably be a good idea. I think ...Yes, that would probably be a good idea. I think I am getting the same effect by perturbing one variable at a time and standardizing the effect size. But it might have been simpler to transform all the variables at the beginning.Allen Downeyhttps://www.blogger.com/profile/01633071333405221858noreply@blogger.comtag:blogger.com,1999:blog-6894866515532737257.post-49149893649517815272015-12-09T04:23:45.276-08:002015-12-09T04:23:45.276-08:00Very nice analysis indeed Allen! I would only sugg...Very nice analysis indeed Allen! I would only suggest you to scale and normalise all predictors and the two outcomes you're studying. In particular I would use a technique I've learned from Gelman & Hill (2006) of subtracting the mean from each observation and then dividing by two times the standard deviation, so that a 1-unit change in the rescaled predictor corresponds to a change from 1 standard deviation below the mean, to 1 standard deviation above. This is to maintain coherence when considering binary input variables.Anonymousnoreply@blogger.com