Multinomial Logit: probability of choice of a soccer action

Modeling
, ,
14

Right, that’s what I thought when writing the Python code above.

The difference between categorical and multinomial is that

model {
    // three events
    1 ~ categorical_logit(beta);
    1 ~ categorical_logit(beta);
    2 ~ categorical_logit(beta);
}

is the same as

model {
    // two 1s, one 2, no 3s
    {2,1,0} ~ multinomial_logit(beta);
}

The predictor beta depends only on columns 2-6. Structure your dataset so that each row corresponds to a unique beta and instead of event_id you have three columns that count the number of shots, passes, and dribblins for that beta. The number of rows to loop over in this format is at most 515\times12\times7\times2\times3=259560 which is about half as many as in the per-event format.

2 Likes