[Rocio Aguilar]

Clyde, please could you tell your opinion about the next code?

//Full interacted model taking (a030_4==3 as reference level)
. xtlogit a006 (c.a016 c.a020_nac i.a018 i.a019 i.a023 i.a021_imput c.a041_l2_imput c.a002_new  c.a024_l2 i.anyo)##ib3.a030_4, nolog i(ident)

// Coeff. of a016 for a030_4==3
. lincom a016
 ( 1)  [a006]a016 = 0

------------------------------------------------------------------------------
        a006 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         (1) |    .238141   .0756697     3.15   0.002     .0898312    .3864509
------------------------------------------------------------------------------

// Coeff. of a016 for a030_4==4
. lincom a016 + 4.a030_4#a016

 ( 1)  [a006]a016 + [a006]4.a030_4#c.a016 = 0

------------------------------------------------------------------------------
        a006 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         (1) |    .353059   .1461166     2.42   0.016     .0666757    .6394423
------------------------------------------------------------------------------
// Comparison between a016 for a030_4==3 and a016 for a030_4==4
. lincom 4.a030_4#a016

 ( 1)  [a006]4.a030_4#c.a016 = 0

------------------------------------------------------------------------------
        a006 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         (1) |    .114918   .1639872     0.70   0.483     -.206491     .436327
------------------------------------------------------------------------------

// The same results (larger code)
. lincom 3.a030_4#a016 + 4.a030_4#a016

 ( 1)  [a006]3b.a030_4#co.a016 + [a006]4.a030_4#c.a016 = 0

------------------------------------------------------------------------------
        a006 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         (1) |    .114918   .1639872     0.70   0.483     -.206491     .436327
------------------------------------------------------------------------------

:
Thanks in advance,
Best
Rocio

[Clyde Schechter]

Looks correct. I don't quite understand the last one. Since 3 is the reference level for a030_4, 3b.a030_4#coa.016 is just zero, so it serves no purpose. And as you note, it gives the same result as the preceding -lincom 4.a030_4#a016-.

[Rocio Aguilar]

Thanks Clyde,

 . lincom 4.a030_4#a016

is more "elegant"... Thanks for your help! Really, your advice saved our paper. Best, Rocio

[Meng Yu]

I used this model with "bs4rw, mse double rw(WT3R*) verbose:" command to apply bootstrap weights. It worked. However, the "lincom" command following it did not work. Can anyone help with this problem? Thank you.
I am pasting some of my results for this Participation variable. These are odds ratios. Do I have to use coefficients for using lincom command?
Before using lincom command, can I conclude something from this result? For instance, can I say "Compared with nonparticipants, males who participate everyday/2 or 3 days have lower odds of having mental health problems (dependent variable)."?

not participate # male	1.00
	(0.00)
not participate # female	1.00
	(0.00)
every day/2 or 3 days # male	0.63*
	(0.14)
every day/2 or 3 days # female	0.82
	(0.18)
once a week,2/3 times a month # male	1.00
	(0.10)
once a week,2/3 times a month # female	1.30**
	(0.12)

[Clyde Schechter]

Please read the Forum FAQ for excellent advice about how to post questions in ways that maximize your chances of getting a helpful and timely response. In particular:

1. This question is unrelated to the original topic of this thread; the only commonality is that the -lincom- command is mentioned in each. But in your question it is tangential. Please start a new thread for this question so that people who come to the Forum and search existing posts for answers to their questions a) will find your question under an appropriate title of its own, and b) will not waste their time reading this question if they were looking for something about comparing coefficients between different xtlogit models.

2. Stata code and output should be displayed between code delimiters.

3. The output you show is not even direct Stata output. It is output that has been laundered through -outreg- or -estout- or one of those pretty-print programs. But it is not the actual Stata response to your regression command. By providing incomplete information you make it difficult or impossible for people to answer your question. Show the complete and exact regression command(s) you used and show the complete and exact output that Stata gave you for it.

4. Never say that something "didn't work." There are many, many ways in which something might not work, and we have no way to know which happened to you. So when something "doesn't work," show the exact and complete command and then show the exact and complete output you got. Then, unless it's obvious, explain how what you got differs from what you wanted.

If you follow the advice in the FAQ, it will be easier for people to understand what your problem is and it becomes possible for them to help you. As written, your questions are unanswerable because you do not provide adequate information.