Re: st: gradient in -ml- producing type mismatch?

[Misha Spisok <misha.spisok@gmail.com>]

Thank you, Professor Buis.  Now I have more questions--hopefully a bit
better than the previous one.

Summary of questions:
1. What does `lnf' do when it precedes -mlvecsum-?
2. What else can go in that position?
3. In the syntax of -mlvecsum-, what does `scalarname_lnf' do?
4. In general, what, precisely, does -mlvecsum- implicitly do? (my
suspicion is that it multiplies the row vector, `x', of "independent
variables by `exp')
5. How can I build up my gradient function?

Believe it or not, I had read the [R] manual on this, as well as a few
presentations by Prof. Baum.  One source of confusion (among others,
to be sure) is that I read somewhere (else?) that -mlvecsum-
implicitly multiplies the row vector `x' (of the "independent"
variables) by what follows the equals sign; I thought this might have
been contingent on the `lnf' that seems to precede every instance of
-mlvecsum- that I can find.  So a first question is, what does `lnf'
do when it precedes -mlvecsum-?

Re-reading the manual section after your comment helps (I think),
clarify some things, but I still have questions.

Ultimately, I want a gradient that looks like this (please pardon the
misuse and abuse of syntax):

$ML_y1*`x' - ($ML_y1/`sumvij')*sum(by group)`x'*exp(`theta')

where `x' is a row vector, which seems to be handled by -mlvecsum-.
The [R] manual states that the syntax for -mlvecsum- is

mlvecsum scalarname_lnf rowvecname = exp [if] [, eq(#)]

What, precisely, does "scalarname_lnf" mean?  I've only seen `lnf' in
this position.

"rowvecname" seems straightforward; it's the name of the row vector to
be generated.  "exp" is a little less clear.  My understanding is that
whatever is in "exp" is (1) assumed to be a scalar and (2) multiplies
the row vector of independent variables (`x') by that scalar.  The
syntax in the manual isn't clear to me.

It seems that the first term, $ML_y1*`x', could be written as,

mlvecsum ? `partone' = $ML_y1.

I put "?" because I've only seen `lnf' follow -mlvecsum-, but I don't
want the row vector multiplied by `lnf'.

The second part is trickier (for me).  Misusing and abusing syntax,
I'd like to do something like,

by group: sum(mlvecsum ($ML_y1/`sumvij') `parttwo'  = exp(`theta')))

Or, thinking about another way to misuse and abuse syntax,

mlvecsum ? `xexb' = exp(`theta')
by group: egen double `sumxexb' = sum(`xexb')
mlvecsum ? `parttwo' = ($ML_y1/`sumvij')*`sumxexb' /* but I don't want
it multiplied by `x'... */

matrix `g' = `partone' - `parttwo'

I started to think that it's impossible to calculate my gradient due
to a violatation of the necessary functional form, but I don't think
so because the log-likelihood function is, in fact, the sum of each
observation's log-likelihood function.  I imagine (hope?) that a
little insight and experience might confirm or deny this and, further
(if possible), suggest some approaches to solving my problem.  I found
some lecture notes from MIT that mentioned this very model as a
"difficult" exercise, which suggests that it's "doable."

Below is an (admittedly ugly, notationally cumbersome, and somewhat
tedious) illustration of what I'm trying to do (focusing on
`parttwo').  (This may explain why I'm trying to work with -mata- in
my other post, as I can do this in Matlab, but would like to implement
it in Stata to take advantage of some of Stata's features; plus, in
spite of my frustration at this point, it's kinda fun.)

Thanks again.

Misha

Suppose this is my data:
Data	group	choice	y	x1	x2	x3
	1	1	y_{1,1}	x1_{1,1}	x2_{1,1}	x3_{1,1}
	1	2	y_{1,2}	x1_{1,2}	x2_{1,2}	x3_{1,2}
	1	3	y_{1,3}	x1_{1,3}	x2_{1,3}	x3_{1,3}
	1	4	y_{1,4}	x1_{1,4}	x2_{1,4}	x3_{1,4}
	1	5	y_{1,5}	x1_{1,5}	x2_{1,5}	x3_{1,5}
	1	6	y_{1,6}	x1_{1,6}	x2_{1,6}	x3_{1,6}
	1	7	y_{1,7}	x1_{1,7}	x2_{1,7}	x3_{1,7}
	1	8	y_{1,8}	x1_{1,8}	x2_{1,8}	x3_{1,8}
	1	9	y_{1,9}	x1_{1,9}	x2_{1,9}	x3_{1,9}
	1	10	y_{1,10}	x1_{1,10}	x2_{1,10}	x3_{1,10}
	2	1	y_{2,1}	x1_{2,1}	x2_{2,1}	x3_{2,1}
	2	2	y_{2,2}	x1_{2,2}	x2_{2,2}	x3_{2,2}
	2	3	y_{2,3}	x1_{2,3}	x2_{2,3}	x3_{2,3}
	2	4	y_{2,4}	x1_{2,4}	x2_{2,4}	x3_{2,4}
	2	5	y_{2,5}	x1_{2,5}	x2_{2,5}	x3_{2,5}
	2	6	y_{2,6}	x1_{2,6}	x2_{2,6}	x3_{2,6}
	2	7	y_{2,7}	x1_{2,7}	x2_{2,7}	x3_{2,7}
	2	8	y_{2,8}	x1_{2,8}	x2_{2,8}	x3_{2,8}
	2	9	y_{2,9}	x1_{2,9}	x2_{2,9}	x3_{2,9}
	2	10	y_{2,10}	x1_{2,10}	x2_{2,10}	x3_{2,10}

If I understand correctly, by defintion, `theta' is as follows:
group	choice	theta
1	1	x1_{1,1}b1 + x2_{1,1}b2 + x3_{1,1}b3
1	2	x1_{1,2}b1 + x2_{1,2}b2 + x3_{1,2}b3
1	3	x1_{1,3}b1 + x2_{1,3}b2 + x3_{1,3}b3
1	4	x1_{1,4}b1 + x2_{1,4}b2 + x3_{1,4}b3
1	5	x1_{1,5}b1 + x2_{1,5}b2 + x3_{1,5}b3
1	6	x1_{1,6}b1 + x2_{1,6}b2 + x3_{1,6}b3
1	7	x1_{1,7}b1 + x2_{1,7}b2 + x3_{1,7}b3
1	8	x1_{1,8}b1 + x2_{1,8}b2 + x3_{1,8}b3
1	9	x1_{1,9}b1 + x2_{1,9}b2 + x3_{1,9}b3
1	10	x1_{1,10}b1 + x2_{1,10}b2 + x3_{1,10}b3
2	1	x1_{2,1}b1 + x2_{2,1}b2 + x3_{2,1}b3
2	2	x1_{2,2}b1 + x2_{2,2}b2 + x3_{2,2}b3
2	3	x1_{2,3}b1 + x2_{2,3}b2 + x3_{2,3}b3
2	4	x1_{2,4}b1 + x2_{2,4}b2 + x3_{2,4}b3
2	5	x1_{2,5}b1 + x2_{2,5}b2 + x3_{2,5}b3
2	6	x1_{2,6}b1 + x2_{2,6}b2 + x3_{2,6}b3
2	7	x1_{2,7}b1 + x2_{2,7}b2 + x3_{2,7}b3
2	8	x1_{2,8}b1 + x2_{2,8}b2 + x3_{2,8}b3
2	9	x1_{2,9}b1 + x2_{2,9}b2 + x3_{2,9}b3
2	10	x1_{2,10}b1 + x2_{2,10}b2 + x3_{2,10}b3

`sumvij' is
group	choice	sumvij: note that sumvij_{i,j} = sumvij_{i,k} for all j and k
1	1	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
1	2	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
1	3	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
1	4	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
1	5	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
1	6	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
1	7	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
1	8	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
1	9	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
1	10	exp(theta_{1,1}) + exp(theta_{1,2}) + … + exp(theta_{1,10})
2	1	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})
2	2	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})
2	3	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})
2	4	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})
2	5	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})
2	6	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})
2	7	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})
2	8	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})
2	9	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})
2	10	exp(theta_{2,1}) + exp(theta_{2,2}) + … + exp(theta_{2,10})

As an intermediary step, I want this row vector, which I'll call
`sumxexb' = (`sumxexb1', `sumxexb2', `sumxexb3'), noting that
sumxexb_{i,j} = sumxexb_{i,k} for all j and k.
group	choice	corresponding to x1: call it `sumxexb1'; note that
sumxexb1_{i,j} = sumxexb1_{i,k} for all j and k	corresponding to x2:
call it `sumxexb2'; note that sumxexb2_{i,j} = sumxexb2_{i,k} for all
j and k	corresponding to x3: call it `sumxexb3'; note that
sumxexb3_{i,j} = sumxexb3_{i,k} for all j and k
1	1	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
1	2	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
1	3	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
1	4	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
1	5	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
1	6	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
1	7	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
1	8	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
1	9	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
1	10	x1_{1,1}*exp(theta_{1,1}) + x1_{1,2}*exp(theta_{1,2}) + … +
x1_{1,10}*exp(theta_{1,10})	x2_{1,1}*exp(theta_{1,1}) +
x2_{1,2}*exp(theta_{1,2}) + … +
x2_{1,10}*exp(theta_{1,10})	x3_{1,1}*exp(theta_{1,1}) +
x3_{1,2}*exp(theta_{1,2}) + … + x3_{1,10}*exp(theta_{1,10})
2	1	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})
2	2	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})
2	3	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})
2	4	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})
2	5	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})
2	6	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})
2	7	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})
2	8	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})
2	9	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})
2	10	x1_{2,1}*exp(theta_{2,1}) + x1_{2,2}*exp(theta_{2,2}) + … +
x1_{2,10}*exp(theta_{2,10})	x2_{2,1}*exp(theta_{2,1}) +
x2_{2,2}*exp(theta_{2,2}) + … +
x2_{2,10}*exp(theta_{2,10})	x3_{2,1}*exp(theta_{2,1}) +
x3_{2,2}*exp(theta_{2,2}) + … + x3_{2,10}*exp(theta_{2,10})

The final step for the second part is

group	choice	
1	1	(y_{1,1}/sumvij_{1,1})*sumxexb_{1,1}
1	2	(y_{1,2}/sumvij_{1,2})*sumxexb_{1,2}
1	3	(y_{1,3}/sumvij_{1,3})*sumxexb_{1,3}
1	4	(y_{1,4}/sumvij_{1,4})*sumxexb_{1,4}
1	5	(y_{1,5}/sumvij_{1,5})*sumxexb_{1,5}
1	6	(y_{1,6}/sumvij_{1,6})*sumxexb_{1,6}
1	7	(y_{1,7}/sumvij_{1,7})*sumxexb_{1,7}
1	8	(y_{1,8}/sumvij_{1,8})*sumxexb_{1,8}
1	9	(y_{1,9}/sumvij_{1,9})*sumxexb_{1,9}
1	10	(y_{1,10}/sumvij_{1,10})*sumxexb_{1,10}
2	1	(y_{2,1}/sumvij_{2,1})*sumxexb_{2,1}
2	2	(y_{2,2}/sumvij_{2,2})*sumxexb_{2,2}
2	3	(y_{2,3}/sumvij_{2,3})*sumxexb_{2,3}
2	4	(y_{2,4}/sumvij_{2,4})*sumxexb_{2,4}
2	5	(y_{2,5}/sumvij_{2,5})*sumxexb_{2,5}
2	6	(y_{2,6}/sumvij_{2,6})*sumxexb_{2,6}
2	7	(y_{2,7}/sumvij_{2,7})*sumxexb_{2,7}
2	8	(y_{2,8}/sumvij_{2,8})*sumxexb_{2,8}
2	9	(y_{2,9}/sumvij_{2,9})*sumxexb_{2,9}
2	10	(y_{2,10}/sumvij_{2,10})*sumxexb_{2,10}

Or, less succinctly,
1	1	(y_{1,1}/sumvij_{1,1})*sumxexb1_{1,1}	(y_{1,1}/sumvij_{1,1})*sumxexb2_{1,1}	(y_{1,1}/sumvij_{1,1})*sumxexb3_{1,1}
1	2	(y_{1,2}/sumvij_{1,2})*sumxexb1_{1,2}	(y_{1,2}/sumvij_{1,2})*sumxexb2_{1,2}	(y_{1,2}/sumvij_{1,2})*sumxexb3_{1,2}
1	3	(y_{1,3}/sumvij_{1,3})*sumxexb1_{1,3}	(y_{1,3}/sumvij_{1,3})*sumxexb2_{1,3}	(y_{1,3}/sumvij_{1,3})*sumxexb3_{1,3}
1	4	(y_{1,4}/sumvij_{1,4})*sumxexb1_{1,4}	(y_{1,4}/sumvij_{1,4})*sumxexb2_{1,4}	(y_{1,4}/sumvij_{1,4})*sumxexb3_{1,4}
1	5	(y_{1,5}/sumvij_{1,5})*sumxexb1_{1,5}	(y_{1,5}/sumvij_{1,5})*sumxexb2_{1,5}	(y_{1,5}/sumvij_{1,5})*sumxexb3_{1,5}
1	6	(y_{1,6}/sumvij_{1,6})*sumxexb1_{1,6}	(y_{1,6}/sumvij_{1,6})*sumxexb2_{1,6}	(y_{1,6}/sumvij_{1,6})*sumxexb3_{1,6}
1	7	(y_{1,7}/sumvij_{1,7})*sumxexb1_{1,7}	(y_{1,7}/sumvij_{1,7})*sumxexb2_{1,7}	(y_{1,7}/sumvij_{1,7})*sumxexb3_{1,7}
1	8	(y_{1,8}/sumvij_{1,8})*sumxexb1_{1,8}	(y_{1,8}/sumvij_{1,8})*sumxexb2_{1,8}	(y_{1,8}/sumvij_{1,8})*sumxexb3_{1,8}
1	9	(y_{1,9}/sumvij_{1,9})*sumxexb1_{1,9}	(y_{1,9}/sumvij_{1,9})*sumxexb2_{1,9}	(y_{1,9}/sumvij_{1,9})*sumxexb3_{1,9}
1	10	(y_{1,10}/sumvij_{1,10})*sumxexb1_{1,10}	(y_{1,10}/sumvij_{1,10})*sumxexb2_{1,10}	(y_{1,10}/sumvij_{1,10})*sumxexb3_{1,10}
2	1	(y_{2,1}/sumvij_{2,1})*sumxexb1_{2,1}	(y_{2,1}/sumvij_{2,1})*sumxexb2_{2,1}	(y_{2,1}/sumvij_{2,1})*sumxexb3_{2,1}
2	2	(y_{2,2}/sumvij_{2,2})*sumxexb1_{2,2}	(y_{2,2}/sumvij_{2,2})*sumxexb2_{2,2}	(y_{2,2}/sumvij_{2,2})*sumxexb3_{2,2}
2	3	(y_{2,3}/sumvij_{2,3})*sumxexb1_{2,3}	(y_{2,3}/sumvij_{2,3})*sumxexb2_{2,3}	(y_{2,3}/sumvij_{2,3})*sumxexb3_{2,3}
2	4	(y_{2,4}/sumvij_{2,4})*sumxexb1_{2,4}	(y_{2,4}/sumvij_{2,4})*sumxexb2_{2,4}	(y_{2,4}/sumvij_{2,4})*sumxexb3_{2,4}
2	5	(y_{2,5}/sumvij_{2,5})*sumxexb1_{2,5}	(y_{2,5}/sumvij_{2,5})*sumxexb2_{2,5}	(y_{2,5}/sumvij_{2,5})*sumxexb3_{2,5}
2	6	(y_{2,6}/sumvij_{2,6})*sumxexb1_{2,6}	(y_{2,6}/sumvij_{2,6})*sumxexb2_{2,6}	(y_{2,6}/sumvij_{2,6})*sumxexb3_{2,6}
2	7	(y_{2,7}/sumvij_{2,7})*sumxexb1_{2,7}	(y_{2,7}/sumvij_{2,7})*sumxexb2_{2,7}	(y_{2,7}/sumvij_{2,7})*sumxexb3_{2,7}
2	8	(y_{2,8}/sumvij_{2,8})*sumxexb1_{2,8}	(y_{2,8}/sumvij_{2,8})*sumxexb2_{2,8}	(y_{2,8}/sumvij_{2,8})*sumxexb3_{2,8}
2	9	(y_{2,9}/sumvij_{2,9})*sumxexb1_{2,9}	(y_{2,9}/sumvij_{2,9})*sumxexb2_{2,9}	(y_{2,9}/sumvij_{2,9})*sumxexb3_{2,9}
2	10	(y_{2,10}/sumvij_{2,10})*sumxexb1_{2,10}	(y_{2,10}/sumvij_{2,10})*sumxexb2_{2,10}	(y_{2,10}/sumvij_{2,10})*sumxexb3_{2,10}


On Sat, May 8, 2010 at 5:01 AM, Maarten buis <maartenbuis@yahoo.co.uk> wrote:
> --- On Sat, 8/5/10, Misha Spisok wrote:
>> I'm getting the following error (-ml- program and example
>> that reproduces error is included below
> <snip>
>>     gen double `obsg' = `partone' - $ML_y1*`myfrac'
>>     matrix `g' = `obsg'
>
> `obsg' is a variable and `g' is a matrix, and the two
> don't go together.
>
> For a solution, see -help mlvecsum-.
>
> Hope this helps,
> Maarten
>
> --------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
>
> http://www.maartenbuis.nl
> --------------------------
>
>
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/