LL(start) seems to ignore the weights
Posted: 19 Aug 2022, 12:18
Hello,
I am estimating an MNL model with weights and was quite confused to observe that LL(0) and LL(start) were different, although all the initial values of my parameters were 0. I first thought that I had not properly understood the definition of LL(0) but after spending a bit of time in the source code, I believe that the reason is that LL(start) is calculated by ignoring the weights, while LL(0) accounts for them. I am not very familiar with the discrete choice models but I guess this is not desired. If it is desired, then a hint in the manual would be helpful for people like me...
That's all - thank you very much for your response and for this great package!
Raphaël
I am estimating an MNL model with weights and was quite confused to observe that LL(0) and LL(start) were different, although all the initial values of my parameters were 0. I first thought that I had not properly understood the definition of LL(0) but after spending a bit of time in the source code, I believe that the reason is that LL(start) is calculated by ignoring the weights, while LL(0) accounts for them. I am not very familiar with the discrete choice models but I guess this is not desired. If it is desired, then a hint in the manual would be helpful for people like me...
That's all - thank you very much for your response and for this great package!
Raphaël