Looking at the ado code for -binreg-, it seems that -binreg- is a
wrapper for -glm-, and that the call to -glm- does not preserve the "by
varlist:" that you specify.
This should be a quick thing for SstataCorp to fix. In the mean time,
you can just call -glm- directly, which does properly support "by
varlist:". The (undocumented?) "cmd" option of -binreg- seems to tell
you the necessary -glm- syntax.
Nick Winter
> -----Original Message-----
> From: Lee Sieswerda [mailto:[email protected]]
> Sent: Wednesday, December 04, 2002 3:31 PM
> To: Statalist (E-mail)
> Subject: st: -by- with -binreg-
>
>
> Dear Statalisters:
> For some reason I cannot get -by- to work with -binreg-. I
> can get around
> the problem using -if-, but according to the manual I should
> be able to use
> -by-. Here is my log:
>
> . table pre_post sex, c(mean q7cat)
>
> ------------------------------
> | sex
> pre_post | F M
> ----------+-------------------
> pre | .3238095 .4953271
> post | .6 .6296296
> ------------------------------
>
> * Both sexes together
> . binreg q7cat pre_post, rd cluster(id)
>
> Iteration 1 : deviance = 570.9529
> Iteration 2 : deviance = 570.9529
>
> Residual df = 423 No. of obs =
> 425
> Pearson X2 = 425 Deviance =
> 570.9529
> Dispersion = 1.004728 Dispersion =
> 1.349771
>
> Bernoulli distribution, identity link
> Risk difference coefficients
> (standard errors adjusted for
> clustering on
> id)
> --------------------------------------------------------------
> --------------
> --
> | Semi-Robust
> q7cat | Coef. Std. Err. z P>|z| [95% Conf.
> Interval]
> -------------+------------------------------------------------
> --------------
> --
> pre_post | .2046461 .0391639 5.23 0.000 .1278862
> .281406
> _cons | .2057312 .0651479 3.16 0.002 .0780438
> .3334187
> --------------------------------------------------------------
> --------------
> --
>
> * Try to get separate estimates by sex
>
> . bysort sex: binreg q7cat pre_post, rd cluster(id)
>
> ______________________________________________________________
> ______________
> ___
> -> sex = F
>
> Iteration 1 : deviance = 570.9529
> Iteration 2 : deviance = 570.9529
>
> Residual df = 423 No. of obs =
> 425
> Pearson X2 = 425 Deviance =
> 570.9529
> Dispersion = 1.004728 Dispersion =
> 1.349771
>
> Bernoulli distribution, identity link
> Risk difference coefficients
> (standard errors adjusted for
> clustering on
> id)
> --------------------------------------------------------------
> --------------
> --
> | Semi-Robust
> q7cat | Coef. Std. Err. z P>|z| [95% Conf.
> Interval]
> -------------+------------------------------------------------
> --------------
> --
> pre_post | .2046461 .0391639 5.23 0.000 .1278862
> .281406
> _cons | .2057312 .0651479 3.16 0.002 .0780438
> .3334187
> --------------------------------------------------------------
> --------------
> --
>
> ______________________________________________________________
> ______________
> ___
> -> sex = M
>
> Iteration 1 : deviance = 570.9529
> Iteration 2 : deviance = 570.9529
>
> Residual df = 423 No. of obs =
> 425
> Pearson X2 = 425 Deviance =
> 570.9529
> Dispersion = 1.004728 Dispersion =
> 1.349771
>
> Bernoulli distribution, identity link
> Risk difference coefficients
> (standard errors adjusted for
> clustering on
> id)
> --------------------------------------------------------------
> --------------
> --
> | Semi-Robust
> q7cat | Coef. Std. Err. z P>|z| [95% Conf.
> Interval]
> -------------+------------------------------------------------
> --------------
> --
> pre_post | .2046461 .0391639 5.23 0.000 .1278862
> .281406
> _cons | .2057312 .0651479 3.16 0.002 .0780438
> .3334187
> --------------------------------------------------------------
> --------------
> --
>
> * Since that gave the same results as if I hadn't used
> * -by- at all, try -if- instead
>
> . binreg q7cat pre_post if sex=="M", rd cluster(id)
>
> Iteration 1 : deviance = 290.7011
> Iteration 2 : deviance = 290.7011
>
> Residual df = 213 No. of obs =
> 215
> Pearson X2 = 215 Deviance =
> 290.7011
> Dispersion = 1.00939 Dispersion =
> 1.364794
>
> Bernoulli distribution, identity link
> Risk difference coefficients
> (standard errors adjusted for
> clustering on
> id)
> --------------------------------------------------------------
> --------------
> --
> | Semi-Robust
> q7cat | Coef. Std. Err. z P>|z| [95% Conf.
> Interval]
> -------------+------------------------------------------------
> --------------
> --
> pre_post | .1343025 .0536422 2.50 0.012 .0291657
> .2394394
> _cons | .3610246 .0910578 3.96 0.000 .1825546
> .5394945
> --------------------------------------------------------------
> --------------
> --
>
> . binreg q7cat pre_post if sex=="F", rd cluster(id)
>
> Iteration 1 : deviance = 273.5711
> Iteration 2 : deviance = 273.5711
>
> Residual df = 208 No. of obs =
> 210
> Pearson X2 = 210 Deviance =
> 273.5711
> Dispersion = 1.009615 Dispersion =
> 1.315246
>
> Bernoulli distribution, identity link
> Risk difference coefficients
> (standard errors adjusted for
> clustering on
> id)
> --------------------------------------------------------------
> --------------
> --
> | Semi-Robust
> q7cat | Coef. Std. Err. z P>|z| [95% Conf.
> Interval]
> -------------+------------------------------------------------
> --------------
> --
> pre_post | .2761905 .0566061 4.88 0.000 .1652445
> .3871365
> _cons | .047619 .0911676 0.52 0.601 -.1310661
> .2263042
> --------------------------------------------------------------
> --------------
> --
>
> * Suppose I use -regress- instead of -binreg-. Then can I use -by-?
> . by sex: regress q7cat pre_post, cluster(id)
>
> ______________________________________________________________
> ______________
> ___
> -> sex = F
>
> Regression with robust standard errors Number of obs =
> 210
> F( 1, 104) =
> 23.69
> Prob > F =
> 0.0000
> R-squared =
> 0.0767
> Number of clusters (id) = 105 Root MSE =
> .48134
>
> --------------------------------------------------------------
> --------------
> --
> | Robust
> q7cat | Coef. Std. Err. t P>|t| [95% Conf.
> Interval]
> -------------+------------------------------------------------
> --------------
> --
> pre_post | .2761905 .0567421 4.87 0.000 .1636688
> .3887121
> _cons | .047619 .0913865 0.52 0.603 -.1336038
> .2288419
> --------------------------------------------------------------
> --------------
> --
>
> ______________________________________________________________
> ______________
> ___
> -> sex = M
>
> Regression with robust standard errors Number of obs =
> 215
> F( 1, 108) =
> 6.24
> Prob > F =
> 0.0140
> R-squared =
> 0.0183
> Number of clusters (id) = 109 Root MSE =
> .49378
>
> --------------------------------------------------------------
> --------------
> --
> | Robust
> q7cat | Coef. Std. Err. t P>|t| [95% Conf.
> Interval]
> -------------+------------------------------------------------
> --------------
> --
> pre_post | .1343025 .053768 2.50 0.014 .027725
> .2408801
> _cons | .3610246 .0912713 3.96 0.000 .1801091
> .5419401
> --------------------------------------------------------------
> --------------
> --
>
> So in this case -regress- works with -by-, but -binreg- does
> not. I also
> have sex as a numeric variable, but that doesn't alleviate
> the non-working
> -by- (NPI). Any ideas?
>
> Regards,
> Lee
>
> Lee Sieswerda, Epidemiologist
> Thunder Bay District Health Unit
> 999 Balmoral Street
> Thunder Bay, Ontario
> Canada P7B 6E7
> Tel: +1 (807) 625-5957
> Fax: +1 (807) 623-2369
> [email protected]
> www.tbdhu.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/
