[R] transformation/link glm in R
Robert A LaBudde
ral at lcfltd.com
Tue Aug 18 17:51:26 CEST 2009
1. It sounds as if you say "left skewed" when you mean "right skewed"
and v.v. Right skewed distributions have a long tail to the right.
2. Your 3rd variable may be censored, as you indicate it is bounded
at 8 hr and that is also the mode. If so, this will complicate your analysis.
3. Your variables would at first sight appear to be continuous, yet
you are modeling them as counts. Why? Did you discretize the times?
4. I suggest you collect the standard transforms related to the
different distributions, and do some EDA where you plot the
distributions of the transformed data. How much does unimodal
symmetry improve? This might point you towards the right family.
Using the Box-Cox method to find a power transform may be useful in
down-selecting.
5. You might consider the gamma distribution with a 1/x link for the
first two varibles.
6. A beta distribution could probably model all 3 of your variables.
You have given insufficient information regarding your experiment and
its response variables to allow accurate advice to be supplied. Where
do the "bounds" come from? Are they arbitrary experimental cut-offs,
causing censoring? If so, would you be better to use survival
analysis (e.g., coxph() instead of glm)? Etc.
At 06:52 AM 8/18/2009, Mcdonald, Grant wrote:
>Dear sir,
>
>I have 3 different time response variables that are in the form
>seconds. All three response variables are not normally
>distributed. They are in the form of mate latency, the first two
>responses are bounded to 30mins and thwe third is bounded to 8
>hours. Frequency plots of raw data show that the first two are
>heavily skewed to the left and the third (bounded at 8hrs) is
>heavily skwewed to the right with most data points being 8 hours
>long. I am unsure of using the appropriate transformations in R and
>have only found appropriate trandformations for count data
>(glm(time~x*x1*x2, family=poisson) and proportion data
>(glm(time~x*x1*x2, family=binomial). Would it be possible to point
>me in the right direction of the appropriate glm family to use for
>such data or should I use some transformation seperately and use
>anova in R of which i am more confident of the code
>
>
>Sorry if this is an innapropriate question but I would
>greatly appreciate advise,
>
>G. Colin
>
>______________________________________________
>R-help at r-project.org mailing list
>https://stat.ethz.ch/mailman/listinfo/r-help
>PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
>and provide commented, minimal, self-contained, reproducible code.
================================================================
Robert A. LaBudde, PhD, PAS, Dpl. ACAFS e-mail: ral at lcfltd.com
Least Cost Formulations, Ltd. URL: http://lcfltd.com/
824 Timberlake Drive Tel: 757-467-0954
Virginia Beach, VA 23464-3239 Fax: 757-467-2947
"Vere scire est per causas scire"
More information about the R-help
mailing list