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]

st: Logit with FW vs Logit PA


From   Simone Peart Boyce <[email protected]>
To   Stata <[email protected]>
Subject   st: Logit with FW vs Logit PA
Date   Wed, 31 Jul 2013 06:08:40 -0700 (PDT)

Hi,
 
I have an unbalanced panel data set of 256 observations comprised of medical practices.  I am trying to calculate adjusted and unadjusted odds ratios of the probability of a practice achieving special recognition contingent on the number of educational sessions attended. My data are arranged such that once a practice achieves recognition, it is no longer in the sample.
 
Practice   n_mnths  Recognition educ_session
1               1            0                     0
1                2           0                      1
1               3            0                      1
1               4            1                      2
2               1            0                      1
...
2              10            1                     3
....
 
I dichotomized educ_session into educ_session<=1 or educ_session>1 (deduc).
 
I have tried different methods of calculating the unadjusted OR, with differing outcomes.
 
For the first, I collapsed the dataset and then calculated the OR.  My code is below:
 
collapse (count) n_mnths (max) recognition deduc, by(OrganizationName)
cc recog deduc [fw=n_mnths]
 
I get the following result
 
 
                                                
                                                         Proportion
                 |   Exposed   Unexposed  |      Total     Exposed
-----------------+------------------------+------------------------
           Cases |        82          80  |        162       0.5062
        Controls |        40          54  |         94       0.4255
-----------------+------------------------+------------------------
           Total |       122         134  |        256       0.4766
                 |                        |
                 |      Point estimate    |    [95% Conf. Interval]
                 |------------------------+------------------------
      Odds ratio |          1.38375       |    .8039282    2.387452 (exact)
 Attr. frac. ex. |         .2773261       |   -.2438922    .5811434 (exact)
 Attr. frac. pop |         .1403749       |
                 +-------------------------------------------------
                                  1-sided Fisher's exact P = 0.1323
                                  2-sided Fisher's exact P = 0.2433
 
 
 
When I do something similar but maintaining the longitudinal structure, I get
 
xtlogit dpcmh dLLS if ProjectName =="RPC", pa or
 
GEE population-averaged model                   Number of obs      =       256
Group variable:                   practice      Number of groups   =        24
Link:                                logit      Obs per group: min =         1
Family:                           binomial                     avg =      10.7
Correlation:                  exchangeable                     max =        32
                                                Wald chi2(1)       =      5.46
Scale parameter:                         1      Prob > chi2        =    0.0195
------------------------------------------------------------------------------
       dpcmh | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        dLLS |   3.253334   1.642671     2.34   0.019     1.209318     8.75219
       _cons |   .0631542    .019269    -9.05   0.000      .034729    .1148453
------------------------------------------------------------------------------
 
Why are these results different?
 
Thanks in advance,
Simone

*
*   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index