Bookmark and Share

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: Mixed effects model for asymmetric data


From   Richard Goldstein <[email protected]>
To   [email protected]
Subject   Re: st: Mixed effects model for asymmetric data
Date   Wed, 26 Sep 2012 14:46:20 -0400

Hi,

re: use of poisson, I suggest you look at the following blog:

http://blog.stata.com/2011/08/22/use-poisson-rather-than-regress-tell-a-friend/

Rich

On 9/26/12 2:15 PM, Ana Beatriz FS wrote:
> Thanks,  JVVerkuilen,
> 
> Unfortunately my variable is not a count one, I work with levels of hormones.
> 
> Following your suggestion, I assessed the quality of the model by the
> residuals and it's really really bad.
> 
> With respect to the transformations, they produce quite different
> distributions. I haven't find one that would fit all my points in
> time, even if not perfectly. I do think I have a problem here!
> 
> Best regards,
> 
> Ana Beatriz
> 
> 
> 2012/9/26 JVerkuilen (Gmail) <[email protected]>:
>> On Wed, Sep 26, 2012 at 1:13 PM, Ana Beatriz FS
>> <[email protected]> wrote:
>>> Dear Statalisters,
>>>
>>> I was performing multilevel mixed-effects linear regression but I
>>> realized my data is not normally distributed. Is there an equivalent
>>> model for asymmetric data in stata?
>>>
>>> I've tried to transform my data but I could not accomplish because my
>>> data require different transformations in each point of follow-up. My
>>> sample is composed by pregnant women in the three trimesters of
>>> pregnancy, but values of my dependent variable in first trimester
>>> require log transformation for normalization, in the second, sqrt
>>> transformation and so on.
>>
>> What are the measures? For instance, if they are counts, you might be
>> better off with -xtmepoisson-, which assumes that you have multilevel
>> Poisson distributed data. If the transformations are all fairly close,
>> you can probably get away with choosing one, so if the right answer is
>> sqrt and you log instead it won't be *that* far off. Also keep in mind
>> that the marginal distribution before conditioning on regressors can
>> be rather far from normal, so it's not really clear you need to do any
>> transformation. What do the residuals look like?
>>
>>
>> --
>> JVVerkuilen, PhD
>> [email protected]
*
*   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 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index