I cannot find anything in the manual. I have a problem where the respondent allocates 100 points into K outcomes to show relative preferences. The dependent variable is a vector of K entries. Each entry is a count that sums to a constant (which does not need to be the same from row to row).
Basically, the log-likelihood for each row is:
ll = y * log(p)+y * log(p)+…+y[K] * log(p[K]);
Can someone provide a complete example STAN code? I believe the “multinomial” distribution is the way to go? Thanks in advance!
y ~ multinomial§
Does it matter how y is declared, as a matrix[N,K] y or a row_vector[K] y[N]?
How does it ensure that p represents Pr(y=1|) for all the allocation to the first outcome?
And the encoding for
p is built in and defined in the manual.
For anyone else who thought the N in
was the sum of y1, y2, …, yk from the multinomial distribution in the manual, it isn’t. This
example helped me and might help others.
Maybe it was a mistake, but when I coded that the first time, I realized the sum of
y wasn’t a necessary input for the distribution as it’s implicit in the output variable.
Maybe it wasn’t. I just didn’t get it!
I know, that made it worse!
I’m just glad I figured it out myself!
can you share the example ?