Re: st: Why do the stcox CI differ when using margins?

[Maarten Buis <maartenlbuis@gmail.com>]

The difference is discussed here:
<http://www.stata.com/support/faqs/statistics/delta-rule/>. So the
argument underlying the CIs from -margins- and -marginsplot- is
internally consistent, and applicable to a very large number of
situations. The CIs returned by -stcox- are a bit better, because
those are not so generic and can thus use more specific information
from the model. If your dataset is large and your parameters not too
extreme than both should be fine.

-- Maarten

On Mon, Feb 25, 2013 at 3:06 PM, Radoslaw Panczak <r.panczak@gmail.com> wrote:
> Nick,
>
> Thank you for prompt feedback.
>
> The reason of attempting margins with stcox came after I stumbled upon
> the mentioning of such possibility on Stata's website (>> Home >>
> Products >> Stata 12 >> What’s new in survival data).
>
> Website briefly mentioned: "... Graphs of margins, marginal effects,
> contrasts, and pairwise comparisons. Margins and effects can be
> obtained from linear or nonlinear (for example, probability)
> responses. New command marginsplot is available after stcox, stcrreg,
> and streg. ..."
>
> I was interested in using marginsplot to graph HRs and was puzzled by
> symmetrical CIs.
> Am I correct in assuming that there is no direct way to call
> marginsplot after stcox then?
>
> Best,
> Radek
>
>
> On 25 February 2013 14:16, Nick Cox <njcoxstata@gmail.com> wrote:
>> This is not my field, but my impression is that you are expecting
>> -margins- to give a sensible answer to what is here a dubious
>> question.
>>
>> -margins- is doing what you asked for, which implies _symmetric_
>> confidence intervals for the hazard ratios following a delta-method
>> calculation. Manifestly, that is not what -stcox- does. But -margins-
>> starts with the -stcox- estimates as basis and does what it can to
>> summarize the evidence in its own terms.
>>
>> In general, different criteria for confidence intervals will lead to
>> different results, especially with any inbuilt asymmetry or
>> nonlinearity or small sample sizes.
>>
>> It's the sheerest of flukes that none of the intervals displayed in
>> your example go negative. Increase the confidence level, and that will
>> happen.
>>
>> Nick
>
> *
> *   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/



-- 
---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------

*
*   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/