Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: boostrapping from a log regression
From 
 
Maarten buis <[email protected]> 
To 
 
[email protected] 
Subject 
 
Re: st: boostrapping from a log regression 
Date 
 
Thu, 2 Sep 2010 11:09:24 +0000 (GMT) 
--- On Thu, 2/9/10, [email protected] wrote:
> Its the actual Beta Coefficient im having difficulty
> trying to obtain in natural units not the predicted
> values of the dependent variables (y).
That is exactly the same issue, so you should use 
-glm-
 
> I had understood from a colleague that it is possible
> to obtain this by bootstrap. 
There are a variety of ways to "fix" the problems you
get when you log transform your dependent variable, but
why try to fix something when you can prevent it?
> I have actually tried a number of models including GLM
> models, but the AIC on the log transformed model is so
> much lower that i`d prefer to use it if it is at all
> possible.
That difference is artificial, and just due to the 
difference in scaling of your dependent variable (which
happens to be the very problem you try to solve...).  
*IC measures can help in comparing some non-nested models,
but that does not mean you can use them to compare all
models with one another. 
> Failing that, does anyone know if i would need to perform
> any similar transofrmations on beta coeficients from a GLM
> -Gamma- link(identity)-, or -Gamma- -link(log).
You should not use the -link(identity)- option here, than
you just get a variation on linear regression (key difference
is that the residual variance changes as the mean changes).
The key option for your application is the -link(log)- option.
Consider the example below:
*--------- begin example ----------
sysuse auto, clear
sum mpg, meanonly
gen c_mpg = mpg - r(mean)
gen byte baseline = 1
glm price c_mpg foreign baseline, ///
    link(log) family(gamma)       ///
	eform nocons
*---------- end example -----------
(For more on examples I sent to the Statalist see: 
http://www.maartenbuis.nl/example_faq )
The coefficient of baseline is the constant, i.e.
the price (in dollars) when both foreign and c_mpg
are zero. c_mpg is 0 when mpg is average, and foreign
has the value 0 when the car is domestic. So a domestic
car with average mpg costs about $5529,-. This price
changes by a factor 1.30 (i.e. 30%) when the car is
foreign, and changes by a factor 0.96 (i.e. -4%) for
every extra mile per gallon.
Hope this helps,
Maarten
--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany
http://www.maartenbuis.nl
--------------------------
      
*
*   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/