Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: R: How to "reverse" log transformed result


From   Austin Nichols <austinnichols@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: R: How to "reverse" log transformed result
Date   Wed, 28 Sep 2011 08:23:29 -0400

Carlo Lazzaro <carlo.lazzaro@tin.it>:
The formula you give relies on very strong assumptions about the error
terms, while the -glm- approach does not.
This is mentioned elliptically on page 8 ("pernicious retransformation
problem") of
http://stata.com/meeting/boston10/boston10_nichols.pdf
(references given at the end), which is the presentation from which
the blog post
http://blog.stata.com/2011/08/22/use-poisson-rather-than-regress-tell-a-friend/
(which Nick pointed to) was spawned.

For those who insist on ignoring good advice, and pursuing the
retransformation, see
findit levpredict
and
findit predlog
for two options.

However, I doubt any of these approaches is a actually a good idea, if
the original poster has data where some people have not yet been
granted a pension (censored durations), in which case a hazard model
is clearly indicated (-help st- in Stata, or -findit hshaz-).

It's not clear to me what the value of the multilevel piece is, so I
ignore the word, but perhaps the poster can illuminate that value.

On Wed, Sep 28, 2011 at 8:04 AM, Carlo Lazzaro <carlo.lazzaro@tin.it> wrote:
>
> Dear Morten,
> I do share the previous comments in that without knowing what you typed is
> difficult to advise.
> However, for what it worths, back transforming from a log transformation,
> the mean on the original scale can be obtained by exp(lm+lv/2), where lm and
> lv are the mean and the variance on the log scale, respectively.
>
> See as a useful reference: Briggs, A. and Nixon, R. and Dixon, S. and
> Thompson, S. (2005)Parametric modelling of cost data: some simulation
> evidence. Health Economics 14(4):pp. 421-428.
> http://eprints.gla.ac.uk/4151/.
>
> Kindest Regards,
> Carlo
>
>
> -----Messaggio originale-----
> Da: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] Per conto di Morten Støver
> Inviato: mercoledì 28 settembre 2011 9.41
> A: statalist@hsphsun2.harvard.edu
> Oggetto: st: How to "reverse" log transformated result
>
> I'm doing an multilevel linear regression analysis where I try to
> investigate  if there are variation in the lenght of the rehabilitation
> process before people are being granted a disability pension. I measure
> this in days, and since my data are very skewed, I've done a log
> transformation. Now I wonder how I can transform the results back to the
> original scale of measurement. As an example, this are the results for
> the different types of diagnosis.
> "Other" diagnosis:    (ref)
> Mental disorders:    0.1993938
> Musulosceletal:        0.0840664
>
> If I now try to transform the data back using di exp(.1993938) I get the
> result 1.2206626.
> If I try to analyse the data without log transforming them, I find that
> the mental disorders group have 166 days (95% CI: 75.5-265.6) longer
> rehabilitation time before being granted a disability pension than the
> "other" diagnosis group.
> I guess that the di exp is not the right way to transform the results
> back, but I don't know any other way to do it.
> I'm using Stata 11.
> Thank you for your help

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index