Statalist


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

Re: st: Overfitting?


From   TA Stat <tastat@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Overfitting?
Date   Sun, 15 Feb 2009 14:10:43 +0700

It seems to me that you don't have two covariates in the model.  It is
more like 3 dummy variables (year 1, 2 and 3).  So you have only one
variable with 4 categories.  Is it right? Each group should be
compared with the referenced group only.

The flip (or not flip) of the results is the way you have to explain
your results with some theories or results from previous studies.
Just another classic example of flipped result, the association
between alcohol consumption and lung cancer without controlling for
smoking.  Without controlling for smoking status, alcohol consumption
is positively associated with lung cancer.  After controlling for
smoking status, alcohol consumption is not associated with lung
cancer.

Cheers

On Sun, Feb 15, 2009 at 4:52 AM, Kieran McCaul
<kamccaul@meddent.uwa.edu.au> wrote:
> It's difficult to interpret this without seeing the univariate and
> multivariate results for both variables.
>
> ______________________________________________
> Kieran McCaul MPH PhD
> WA Centre for Health & Ageing (M573)
> University of Western Australia
> Level 6, Ainslie House
> 48 Murray St
> Perth 6000
> Phone: (08) 9224-2140
> Fax: (08) 9224 8009
> email: kamccaul@meddent.uwa.edu.au
> http://myprofile.cos.com/mccaul
> http://www.researcherid.com/rid/B-8751-2008
> ______________________________________________
> The fact that no one understands you doesn't make you an artist.
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Shell makka
> Sent: Saturday, 14 February 2009 6:54 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: Overfitting?
>
> Just for making it more clear in the cox model, its hazards ratio in
> univariate was like:
> categories      HR      CI                      P       whole p
>  1           0.67 (0.56 - 0.82) <0.001
>  2          0 .51 (0.4 - 0.67) <0.001
>  3          0.36 (0.22 - 0.6)  <0.001  <0.001
>
>
> and in the multivariate model is like:
>
>    1        0.78 (0.63 -0.97)  0.03
>    2        0.82 (0.56 - 1.21) 0.32
>    3       0.94 (0.46 - 1.92)  0.88    0.13
>
> So in the multivariate model the hazard is growing as year is
> increasing (year categories:1, 2, 3)
>
> Many thanks,
> [Redacted]
>
>
>
>
> On Sat, Feb 14, 2009 at 8:26 PM, Shell makka <shell.makka@gmail.com>
> wrote:
>> Thanks Maarten for your response, I will only use one of these
>> covariates in my final model.
>>
>>
>>
>>
>>
>> On Sat, Feb 14, 2009 at 7:59 PM, Maarten buis
> <maartenbuis@yahoo.co.uk> wrote:
>>> --- On Sat, 14/2/09, Shell makka wrote:
>>>> Would you please let me know that when effect of one
>>>> covariate flips in the multivariate model, is it only
>>>> because of overfitting or it can have other reasons.
>>>> This variable was significant in the univariate
>>>> analysis but when I used it in multivariate model which
>>>> contained another covariate that these too could be
>>>> also colinear (one is calander year and the other is
>>>> year of starting the study) it became insignificant and
>>>> its effect reversed.
>>>
>>> I would not call this overfitting, it just means that most
>>> of the effect of calender year occurs through year of
>>> starting the study. However, you do need a good theoretical
>>> reason for including both variables though. It is possible
>>> that such a good theoretical argument exist, but it is
>>> definitely not automatic.
>>>
>>> -- Maarten
>>>
>>> -----------------------------------------
>>> Maarten L. Buis
>>> Department of Social Research Methodology
>>> Vrije Universiteit Amsterdam
>>> Boelelaan 1081
>>> 1081 HV Amsterdam
>>> The Netherlands
>>>
>>> visiting address:
>>> Buitenveldertselaan 3 (Metropolitan), room N515
>>>
>>> +31 20 5986715
>>>
>>> http://home.fsw.vu.nl/m.buis/
>>> -----------------------------------------
>>>
>>>
>>>
>>>
>>> *
>>> *   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/
>>>
>>
> *
> *   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/
>
>
>
> *
> *   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/
>
*
*   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 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index