[Bruce Weaver]

Clyde, the example in your #8 post just helped me solve a problem I had not been able to crack. But how did you figure out where to place the binomial(denominator) option? I.e., how did you know to do this...

melogit numerator independent_variables || id:, binomial(denominator)

...rather than this?

melogit numerator independent_variables, binomial(denominator) || id:

I have just searched the -melogit- documentation for binomial( again, and I still find nothing about where to place it, and no examples. So, how did you know where to put it?

Thanks,
Bruce

[Clyde Schechter]

Well, I've been using this approach for a long time (it can really speed up execution by orders of magnitude), and I'm not sure how I originally found out about it--it might have been from reading somebody else's post here. But the syntax for the -binomial()- option in -melogit- is in the help file:

Syntax

melogit depvar fe_equation [|| re_equation] [|| re_equation ...] [, options]

where the syntax of fe_equation is

[indepvars] [if] [in] [weight] [, fe_options]

and the syntax of re_equation is one of the following:

for random coefficients and intercepts

levelvar: [varlist] [, re_options]

for random effects among the values of a factor variable in a crossed-effects model

levelvar: R.varname

levelvar is a variable identifying the group structure for the random effects at that level or is _all representing one group comprising all observations.

fe_options Description
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
Model
noconstant suppress constant term from the fixed-effects equation
offset(varname) include varname in model with coefficient constrained to 1
asis retain perfect predictor variables
---------------------------------------------------------------------------------------------------------------------------------------------------------------------

re_options Description
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
Model
covariance(vartype) variance-covariance structure of the random effects
noconstant suppress constant term from the random-effects equation
fweight(varname) frequency weights at higher levels
iweight(varname) importance weights at higher levels
pweight(varname) sampling weights at higher levels
---------------------------------------------------------------------------------------------------------------------------------------------------------------------

options Description
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
Model
binomial(varname|#) set binomial trials if data are in binomial form
...

Note, in particular, the distinction between options and fe_options (and re_options).

[Bruce Weaver]

Thanks Clyde. I see it now.

[Bruce Weaver]

One more thing. AFAICT, -logit- has no -binomial()- option. So if one wished to estimate a model with no random effects, presumably the code would be:

melogit numerator independent_variables, binomial(denominator)

[Clyde Schechter]

Yes.

Or there is another way. -logit-, unlike -melogit- allows -fweight-s. So given a data set where the DV is a fraction and the total denominator is known, you could do something like this:

gen long obs_no = _n
expand 2
by obs_no, sort: gen new_DV = _n-1
by obs_no: gen new_denominator = round(total_denominator* cond(new_DV, fractional_DV, 1-fractional_DV), 1)
logit new_DV indep_variables [fweight = new_denominator]

This approach will, except perhaps for tiny discrepancies in distant decimal places, produce the same results as the -melogit- approach. It is, evidently, more cumbersome than -melogit-, and there is also the risk that the rounding to an integer in the creation of new_denominator can go wrong in a few cases. So I don't really recommend this approach. I raise it only a a fallback position in the event that -melogit- fails to converge. The -logit- estimator is more robust in that regard.