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 on April 23, and its replacement, statalist.org is already up and running.


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

Re: st: How to find the best transformation for each variable in 120 periods


From   Nick Cox <njcoxstata@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: How to find the best transformation for each variable in 120 periods
Date   Fri, 22 Feb 2013 17:55:03 +0000

Closer to home in a couple of senses are cube roots:

http://www.stata-journal.com/article.html?article=st0223

Odd integer roots are equally well defined for arguments of any sign,
although you have to go in Stata (e.g.)

sign(x) * (abs(x))^(1/3)

I've been working with glacier advance and retreat rates. Occasionally
glaciers advance a lot or retreat a lot but most values are much
closer to zero, so cube roots are a non-standard but useful way to
plot them. Haven't tried regressions, as that is not the nature of the
problem.

Nick

On Fri, Feb 22, 2013 at 5:28 PM, JVerkuilen (Gmail)
<jvverkuilen@gmail.com> wrote:
> On Fri, Feb 22, 2013 at 10:40 AM, Xixi Lin <winnielxx@gmail.com> wrote:
>> Hi JVerkuilen,
>>
>> You are right that the key assumption is Gaussian errors, but my data
>> does not have Gaussian errors, and I wanna use t statistics, so I
>> wanna to fix the non-normal residuals by transforming the variables (I
>> don't know what else to do to fix the Gaussian errors). Is there any
>> better way to deal with it? Or Is there any papers that I can read
>> about it?
>
> Well as I said, the Atkinson book is excellent. What's the nature of
> the non-normality? Outliers? Are they symmetric? You can't really
> decide this without thinking of what the likely problems are.
>
> The inverse hyperbolic sine is an under-used transformation for
> long-tailed data with zero or negative values. It was originally part
> of the Johnson system of distributions, which are based on
> transformations of the Gaussian. It behaves like the log for large
> magnitudes and like the square root for smaller ones.
>
> Burbidge, John B., Lonnie Magee and A. Leslie Robb. 1988 "Alternative
> Transformations to Handle Extreme Values of the Dependent Variable."
> Journal of the American Statistical Association, vol. 83, 123-127.
>
> http://worthwhile.typepad.com/worthwhile_canadian_initi/2011/07/a-rant-on-inverse-hyperbolic-sine-transformations.html
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


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