Re: st: Bonferroni corrected CIs (multiple comparisons issue?)

[Nicole Boyle <nicboyle@gmail.com>]

So I tried -pwcompare- for Bonferroni corrected CIs (using Stata 12)...

Do you know how to set the reference group in -pwcompare-? Right now,
it's testing all combinations, and I'm afraid that this is going to
result in a Bonferroni OVERcorrection due to the inflated DF.

Take a 4 level variable "var", coded 0 (reference), 1, 2, and 3.
I'd like to only adjust for the following three tests (aka: 3 DF):
1 vs 0
2 vs 0
3 vs 0

However, the following output suggests that pwcompare is adjusting for 6 DF:

.                 stcox i.var
.                 pwcompare i.var, mcompare(bonferroni) effects

---------------------------------------------------------------------------
Pairwise comparisons of marginal linear predictions

Margins      : asbalanced
-------------------------------------
             |    Number of
             |  Comparisons
-------------------------------------
    var     |            6
-------------------------------------
-----------------------------------------------------------------
                |   Results...etc....
----------------+-----------------------------------------------
    var        |
     1 vs 0  |   ...etc.....
     2 vs 0  |   ...etc.....
     3 vs 0  |   ...etc.....
     2 vs 1  |   ...etc.....
     3 vs 1  |   ...etc.....
     3 vs 2  |   ...etc.....
-------------------------------------------------------------------


Any suggestions on how to set a reference group?

Thanks,
Nicole

On Sat, Sep 7, 2013 at 3:35 PM, Nicole Boyle <nicboyle@gmail.com> wrote:
> Fantastic, thanks Nick!
>
> Nicole
>
> On Sat, Sep 7, 2013 at 3:23 PM, Nikolaos Pandis <npandis@yahoo.com> wrote:
>> Nicole try
>>
>>
>> 'pwcompare cat_var, mcompare(bonferroni) effects'
>>
>> after fitting stcox
>>
>> check "pwcompare" command as other options are avilable
>>
>>
>> Nick
>>
>>
>>
>> ________________________________
>> From: Nicole Boyle <nicboyle@gmail.com>
>> To: statalist@hsphsun2.harvard.edu
>> Sent: Sunday, September 8, 2013 1:07 AM
>> Subject: st: Bonferroni corrected CIs (multiple comparisons issue?)
>>
>>
>> Is anyone aware of a method of obtaining Bonferroni corrected
>> confidence intervals from a Cox regression?
>>
>> I am aware of the post-estimation command for Bonferroni corrected
>> p-values of, say, an exposure variable ("var") with three categories
>> (0.var, 1.var, 2.var), and therefore two tests against the reference
>> group (0.var)...
>>             stcox i.var
>>             test 1.var 2.var, mtest(bonferroni)
>> ...but since I feel that confidence intervals are more informative
>> than p-values, I'd like to explore the CI option.
>>
>> To clarify the intent behind this, I've been requested to obtain
>> Bonferroni p-values in order to help avoid the possibility of multiple
>> comparisons bias. If anyone has any recommendations for presenting
>> multiple comparisons results, I'd greatly appreciate your thoughts.
>>
>> For the record, of the four variables in my model that might warrant
>> Bonferroni corrections, they have (respectively) four, three, two, and
>> two tests against their reference groups.
>>
>> Thanks for your consideration,
>> Nicole
>> *
>> *   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/
>>
>> *
>> *   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/
*
*   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/