# RE: st: RE: Stata's logistic vs. SAS CATMOD WLS model.

 From VISINTAINER PAUL <[email protected]> To "'[email protected]'" <[email protected]> Subject RE: st: RE: Stata's logistic vs. SAS CATMOD WLS model. Date Thu, 23 Oct 2003 15:42:16 -0400

```You can interpret it by looking at your outcome over time, within each level

+----------------------------+
| time   intern   it       y |
|----------------------------|
1. |    0        0    0    -.08 |
2. |    1        0    0       0 |
3. |    0        1    0   -.447 |
4. |    1        1    1    .708 |
+----------------------------+

So, in the placebo group, Y changes from -.08 to 0, over time.  In the
intervention group, Y changes from -.447 to .708, over time.  In other
words, the placebo group has a near 0 slope over time, while the
intervention has a significantly positive slope over time.  That's what the
significant interaction term is telling you.

Paul

P.S.  If you plot these, remember that these are linear logits, if you
exponentiate them, values are no longer linear.

-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Ricardo Ovaldia
Sent: Thursday, October 23, 2003 12:10 PM
To: [email protected]
Subject: Re: st: RE: Stata's logistic vs. SAS CATMOD WLS model.

. logit meter time intervention it, cluster(id)

Iteration 0:   log pseudo-likelihood = -277.17887
Iteration 1:   log pseudo-likelihood = -268.85134
Iteration 2:   log pseudo-likelihood = -268.84209
Iteration 3:   log pseudo-likelihood = -268.84209

Logit estimates
Number of obs   =        400
Wald
chi2(3)    =      23.65
Prob
> chi2     =     0.0000
Log pseudo-likelihood = -268.84209
Pseudo R2       =     0.0301

(standard errors
----------------------------------------------------------------------------
--
|               Robust
meter |      Coef.   Std. Err.      z    P>|z|
[95% Conf. Interval]
-------------+--------------------------------------------------------------
--
time |   .0800427   .1964444     0.41   0.684
-.3049812    .4650666
intervention |  -.3672695   .2872475    -1.28   0.201
-.9302643    .1957252
it |   1.075455   .3101891     3.47   0.001
.4674952    1.683414
_cons |  -.0800427   .2006625    -0.40   0.690
-.4733339    .3132485
----------------------------------------------------------------------------
--

Which is not exactly what SAS produces, but like SAS,
it gives a significant interaction and a
non-significant intervention effect. How do I
interpret the interaction in this context?

Thank you again,
Ricardo.

--- VISINTAINER PAUL <[email protected]>
wrote:
> I haven't' tried this, but I think it will work.
>
> Set up your data as:
>
> Meter usage:  0 - no, 1 - yes
> Time:	0 pre, 1 is post
> Intervention: 0 - no; 1 yes
>
>      	   meter      time        intervn   id
>   1.         0          1          1 	1
>   2.         0          0          0	1
>   3.         0          1          0	2
>   4.         1          0          0	2
>
> 		 . . . etc.
>
> Then, use either xtlogit or logit with cluster(id).
> You can generate an
> interaction term between intervention and time.
> Something like:
>
> .gen it = intern*time
> .logit meter time intervention it, cluster(id)
>
>
>
> Paul Visintainer
>
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On
> Behalf Of Ricardo Ovaldia
> Sent: Thursday, October 23, 2003 9:29 AM
> To: [email protected]
> Subject: st: Stata's logistic vs. SAS CATMOD WLS
> model.
>
> Dear all,
>
> Last week I posted a question and did not received
> any
> replies. I would I appreciate any comments regarding
> the logistic model that I used. Is there a better
> way
> to do this in Stata. I rather not have to use SAS.
> Thank you in advance. Ricardo
>
> In an intervention study geared to teach diabetics
> about glucose monitoring, 100 patients were
> randomized
> to receive a standard educational method, and 100
> patients to receive a new method. One of the
> outcomes
> of interest is whether or not the patient could use
> the glucose-meter correctly or not, as determined by
> comparing their reported values with those obtained
> by
> a trained laboratory tech.
>
> Each patient was tested twice; before the
> intervention
> and two weeks after the intervention. Here is some
> of
> the data excluding covariates.
>
> . cl
>
>      interve~n     before      after
>   1.         0          1          1
>   2.         0          0          0
>   3.         0          1          0
>   4.         1          0          0
>   5.         1          1          1
>   6.         1          0          1
>
>
> I analyzed this data using -logistic- by including
> -before- as a RHS variable:
>
>   . logistic after before intervention
>
> Logistic regression
> Number of obs   =        200
>                                                   LR
> chi2(2)      =      46.68
>
> Prob
> > chi2     =     0.0000
> Log likelihood = -112.38704
> Pseudo R2       =     0.1720
>
>
----------------------------------------------------------------------------
> --
>        after | Odds Ratio   Std. Err.      z
> P>|z|
>    [95% Conf. Interval]
>
-------------+--------------------------------------------------------------
> --
>       before |   8.061982    2.90929     5.78
> 0.000
>    3.974411    16.35351
> intervention |   2.971752   1.009034     3.21
> 0.001
>    1.527546    5.781374
>
----------------------------------------------------------------------------
> --
>
>
> which indicates to me that the new method is
> superior
> to the standard method. When I presented the results
> one of the researchers suggested I use SAS's CATMOD
> Weighted Least Squares procedure to analyze these
> data. Following an example in the SAS manual I
> obtained:
>
>              Analysis of Weighted Least Squares
> Estimates
>
>                                            Standard
>
>     Chi-
> Effect             Parameter    Estimate      Error
>
>   Square    Pr > ChiSq
>
> Intercept               1         0.5100     0.0293
>
>   302.44        <.0001
> intervention            2        -0.0200     0.0293
>
>     0.47        0.4952
> time                    3        -0.0750     0.0184
>
>    16.63        <.0001
> intervention*time       4         0.0650     0.0184
>
>    12.49        0.0004
>
> Now, the time-by-intervention is significant but not
> the intervention term. Not being a SAS user, or
> familiar with CATMOD, I am not sure whether or not
> these results contradict my prior analysis. Is there
> any way to do what SAS is doing using STATA? Any
> help
> would be greatly appreciated. Here is the SAS code I
> used:
>
> proc catmod order=data;
> response marginals;
> model before*after=intervention| _response_;
> repeated time;
>
> Thank you,
> Ricardo.
>
>
> __________________________________
> Do you Yahoo!?
> The New Yahoo! Shopping - with improved product
> search
> http://shopping.yahoo.com
> *
> *   For searches and help try:
> *
> http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
> *
> *   For searches and help try:
> *
> http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

=====
Ricardo Ovaldia, MS
Statistician
Oklahoma City, OK

__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```