Re: st: transformation of variables

[Nick Cox <njcoxstata@gmail.com>]

Even with several predictors you can set up a slide show. Some technique:

sysuse auto
gen gpm = 1 / mpg

regress gpm headroom trunk weight length turn displacement gear foreign
foreach v in headroom trunk weight length turn displacement gear foreign {
        favplot `v'
        more
}

Here -favplot- from -favplots- (SSC) is a tweak to make results more
readable. You may naturally use -avplot- in official Stata .

Nick


On Sat, Jun 4, 2011 at 8:43 AM, Nick Cox <njcoxstata@gmail.com> wrote:
> In what sense is your model too big?
>
> -avplots- can help (also -favplots- from SSC) with up to say 20
> predictors, but with say 200 the graphs are not likely to be readable.
> But -ladder- is at best partly relevant: the transform that most
> nearly symmetrises a marginal distribution is not necessarily one that
> is appropriate for transforming a predictor.
>
> From another point of view, if your model is too large to tinker with,
> it is too large!
>
> Nick
>
> On Sat, Jun 4, 2011 at 8:26 AM,  <cohaag@staff.mail.uni-mannheim.de> wrote
>
>> I am running a regression model and the Rampsey Reset Test tells me that my
>> model is not correctly specified.
>> I am guessing that there might be a non-linear relation between my dependent
>> variable and some of my independant variables. Since my model is too big I
>> cannot use graphs to find the "right" relation or transformation for my
>> variables.
>> Is there any way that I can use the ladder-command to get a suggestion from
>> stata how to transform my variables in order to meet the linearity
>> assumption?
>>
>> Or do I just have to try all the possible transformations and re-run the
>> regression and the Reset-Test?
>

*
*   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/