1
$\begingroup$

The initial function of log-likelihood of multinomial probit model with $J$ alternatives:

$ ln \ell=\sum_{i=1}^N\sum_{j=1}^{J-1} y_{ij} \cdot ln \Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})+ ({n_i-\sum_{j=1}^{J-1}y_{ij}}) \cdot ln[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]$

Then, the first derivative of

$\dfrac{\partial}{\partial\beta_{kj}}\sum_{i=1}^N \left\{y_{ij} \cdot ln \Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})+ y_{iJ} \cdot ln\left[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})\right]\right\}$

would be

$\sum_{i=1}^N \left\{y_{ij} \dfrac{x_{ik}\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})}{\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})}- y_{iJ} \dfrac{x_{ik}\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})}{1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})}\right\}$

Can somene help me with the second derivative? I am aware that for each $\beta_{kj}$ I need to differentiate wrt every other $\beta_{kj}$.

This is my attempt of doing second derivative of the probit log-likelihood function which should have the following form:

$\dfrac{\partial^2}{\partial\beta_{kj}\beta_{k'j'}}=\sum_{i=1}^N\left\{y_ij\dfrac{\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})\cdot\varphi'(\sum_{k=1}^Kx_{ik}\beta_{kj})-\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})\cdot\Phi'(\sum_{k=1}^Kx_{ik}\beta_{kj})}{\Phi^2(\sum_{k=1}^Kx_{ik}\beta_{kj})}+y_{iJ}\dfrac{[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]\cdot\varphi'(\sum_{k=1}^Kx_{ik}\beta_{kj})-\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})\cdot[1-\sum_{j=1}^{J-1}\Phi'(\sum_{k=1}^Kx_{ik}\beta_{kj})]'}{[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]^2} \right\}$

Hence, the derivatives of the numerator and denominator will depend whether $j'=j$ (Case 1) or $j'\ne j$ (Case 2).

Case 1 $j'=j$

$\varphi'(\sum_{k=1}^Kx_{ik}\beta_{kj})=-x_{ik'}\cdot\sum_{k=1}^Kx_{ik}\beta_{kj}\cdot\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})$

$\Phi'(\sum_{k=1}^Kx_{ik}\beta_{kj})=x_{ik'}\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})$

$[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]'=-x_{ik'}\cdot\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})$

2) case $j'\ne j$

$\varphi'(\sum_{k=1}^Kx_{ik}\beta_{kj})=0$

$\Phi'(\sum_{k=1}^Kx_{ik}\beta_{kj})=0$

$[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]'=-x_{ik'}\cdot\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj'})$

By substituting these terms in the main equation for the case (1) yields

$=-\sum_{i=1}^N\left\{y_{ij}\dfrac{(\sum_{k=1}^Kx_{ik}\beta_{jk})\cdot\phi(\sum_{k=1}^Kx_{ik}\beta_{jk})}{\Phi(\sum_{k=1}^Kx_{ik}\beta_{jk})}+y_{ij}\dfrac{\phi^2(\sum_{k=1}^Kx_{ik}\beta_{jk})}{\Phi^2(\sum_{k=1}^Kx_{ik}\beta_{jk})}+y_{iJ}\dfrac{(\sum_{k=1}^Kx_{ik}\beta_{jk})\cdot\phi(\sum_{k=1}^Kx_{ik}\beta_{jk})}{[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]}+y_{iJ}\dfrac{\phi^2(\sum_{k=1}^Kx_{ik}\beta_{jk})}{[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]^2}\right\}\cdot x_{ik}x_{ik'}$

And by solving for case (2), first fraction is equal to $0$

$\sum_{i=1}^N\left\{y_{ij}\dfrac{\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})\cdot0-\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})\cdot0}{\Phi^2(\sum_{k=1}^Kx_{ik}\beta_{kj})}+y_{iJ}\dfrac{[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]\cdot0-\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})\cdot(-x_{ik'})\cdot\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj'})}{[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]^2} \right\}$

so the result is

$=-\sum_{i=1}^N\left\{y_{iJ}\dfrac{\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj})\cdot x_{ik'}\cdot\varphi(\sum_{k=1}^Kx_{ik}\beta_{kj'})}{[1-\sum_{j=1}^{J-1}\Phi(\sum_{k=1}^Kx_{ik}\beta_{kj})]^2} \right\}$

Is this correct?

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.