tag:blogger.com,1999:blog-8223454891323668732.post4083648686048045994..comments2024-03-19T22:57:32.924+11:00Comments on Eco-Stats Research Blog: Eco-Stats Lab, March 2014 - Measurement Error modeling using SIMEXDavid Wartonhttp://www.blogger.com/profile/05231788956034178551noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-8223454891323668732.post-12357396567151950362014-03-28T14:36:50.325+11:002014-03-28T14:36:50.325+11:00par(mfrow=c(1,2))
plot(Y~w,log="y")
abli...par(mfrow=c(1,2))<br />plot(Y~w,log="y")<br />abline(coef(mod_naive),col="purple")<br /><br />plot(Y~x,log="y")<br />abline(coef(mod_simex),col="burlywood")<br /><br />The second plot should have a steeper line on it, that one has the SIMEX fitDavid Wartonhttps://www.blogger.com/profile/05231788956034178551noreply@blogger.comtag:blogger.com,1999:blog-8223454891323668732.post-38960216135473980382014-03-28T14:30:09.334+11:002014-03-28T14:30:09.334+11:00To see the made-up data for example 2,
par(mfrow=c...To see the made-up data for example 2,<br />par(mfrow=c(1,2))<br />plot(Y~w)<br />plot(Y~x)<br /><br />The relationship between Y and x is stronger (because it has been measured without error) and when we fit a measuremnt error model we are trying to recover this relationship between y and x even though we have only observed x with error (AKA w)David Wartonhttps://www.blogger.com/profile/05231788956034178551noreply@blogger.comtag:blogger.com,1999:blog-8223454891323668732.post-78954290827202219902014-03-28T14:27:25.792+11:002014-03-28T14:27:25.792+11:00You would probably have quite different measuremen...You would probably have quite different measurement errors for height and girth - girth you can get by wrapping tape around the trunk, height is a bot if guesswork. Let's say the sd of measurement error for girth is 1.5 inches, and for height it is 4 metres.<br />mod_simex4 = simex(mod_naive, SIMEXvariable = c("Girth","Height"), measurement.error = cbind(1.5,4), B=200)<br />summary(mod_simex4)David Wartonhttps://www.blogger.com/profile/05231788956034178551noreply@blogger.com