Timeline for Back-transformation and interpretation of $\log(X+1)$ estimates in multiple linear regression

8 events

when toggle format	what		by	license	comment
Aug 8, 2015 at 20:16	answer	added	Paul		timeline score: 3
Jul 8, 2015 at 9:07	answer	added	landroni		timeline score: 5
Nov 21, 2011 at 14:52	comment	added	Charlie		@whuber, True, if you undo the transformation and try to interpret that result, it may not have a normal distribution; I forgot about that part of the question. But I'd recommend bootstrapping that distribution, rather than forcing yourself to find a model that has errors that are normally distributed.
Nov 21, 2011 at 13:58	history	edited	whuber♦	CC BY-SA 3.0	added 18 characters in body
Nov 21, 2011 at 13:57	comment	added	whuber♦		Actually, @Charlie, you do need normality of some statistics if you want the t-tests and confidence intervals to be correct. The log transformation is strong enough to raise doubts unless the dataset is quite large: the approximate normality of an estimated coefficient of a log response will imply the non -normality of the coefficients for the response itself and vice versa.
Nov 21, 2011 at 11:29	history	edited	user88	CC BY-SA 3.0	edited title
Nov 21, 2011 at 7:31	comment	added	Charlie		The coefficients would be easier to interpret if you used the natural log. Also, you should worry about the linearity assumption seeming to make sense. You don't need normality of anything (variables or error terms) so long as you have enough data---the central limit theorem typically comes to your rescue.
Nov 21, 2011 at 6:44	history	asked	Glenn	CC BY-SA 3.0