I have a linear model $Y \sim N(X\beta,\sigma^2*I)$

The mean is parameterized into the form


Where $\gamma=B\beta$ and $Z=XB^{-1}$

I have to show that the maximum likelihood estimator of $\gamma$ is given by:


I have tried to calculate the mle by differentiating the log likelihood function of the normal distribution in relation to gamma, I set the equation equal to zero and then solve for gamma. I then substitute the expressions for Z and $\gamma$ into the equation, but from this point I am stuck.

Does someone know if this is the right way to solve this problem, or do I just have to check if the estimator is biased in some way?


I guess this is self study so I will try to give some hints rather than a full answer.

Under the reparametrization you have the following normal model

$$ Y = Z\gamma + \sigma^2I_n \sim N(Z\gamma, \sigma^2I_n) $$

You probably know by now that the ML estimate $\hat{\gamma}$ is equal to the OLS estimate $(Z^TZ)^{-1}Z^TY$

Now substitue back $Z = XB^{-1}$ and try to simplify this expression.

  • $\begingroup$ Thanks alot for your answer! This is where I previously went before going to a hault. I end up with this expression: (XB)^T.Y=γ(XB^-1)^T.XB^-1 . (Hope you can read this). But from this point I become very unsure, because my matrix algebra is a bit rusty. I assume I somehow how to end up with the linear model that you wrote. But I'm not quite sure how to reach that point? $\endgroup$ – Christoffer GT Jan 29 '18 at 20:18
  • $\begingroup$ Don't multiply by $Z^TZ$ - keep in mind that $(AB)^{-1} = B^{-1}A^{-1}$ and try to get rid of some inverses since $AA^{-1} = I$ of course $\endgroup$ – Łukasz Grad Jan 30 '18 at 8:31
  • $\begingroup$ Thanks again. I think I end up with something usefull now, but I'm still unsure wheter X is invertible or how commutative my matrices are?: ((XB^-1)^T.XB^-1)^-1.(XB^-1)^T.Y=γ => ((X^-1B)^T.X^-1B).(XB^-1)^T.Y => I.B^T.B(XB^-1)^T.Y => IXB.Y=γ => IY=XB^-1γ => Y=XB^-1γ => Y=Zγ . $\endgroup$ – Christoffer GT Jan 30 '18 at 11:16
  • $\begingroup$ I think I got it now!! ((XB^-1)^T.XB^-1)^-1.(XB^-1)^T.Y=γ => ((X^-1B)^T.X^-1B).(XB^-1)^T.Y => X^-1B.Y=γ => Y=XB^-1γ which is equal to: Y=Zγ => Y=XB^-1Bβ => Y=X β, Sorry for the mess, but does it look right? $\endgroup$ – Christoffer GT Jan 30 '18 at 13:09

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.