Let $X_1,\ldots, X_n$ be i.i.d. $N(µ, C)$ random $p$-vectors. Derive the Likelihood Ratio Test for $H_0: C = σ^2(1 − ρ)I_p + ρ1_p1_p^T$, where $1_p = \begin{pmatrix} 1 \\ 1 \\ \vdots \\ 1 \end{pmatrix}_{p\times1}$,
against the alternative hypothesis $H_1$ that the covariance matrix $C$ is not of this form. Here $\mu$ is unknown and $\sigma$, $\rho$ are free parameter.

I have tried to calculate the LRT directly: Let $C_0 = σ^2(1 − ρ)I_p + ρ1_p1_p^T$, then the LRT has the form $$ \frac{sup_{H_0}L(C,\mu)}{supL(C,\mu)} = \frac{L(C_0,\hat{\mu}_{ML})}{L(\hat{C}_{ML},\hat{\mu}_{ML})} = |\frac{1}{n}A|^{n/2} |C_0|^{-n/2} exp^{-\frac{1}{2} (trace(C_0A^{-1}) -p^2)} $$ since $\hat{C}_{ML} = \frac{1}{n} \sum_{i=1}^{n} (X_i - \hat{X})(X_i - \hat{X})^T =\frac{1}{n} A$. Here i don't know how to continue the computations

I also wonder, if I know a LRT for spherical variance ($H_0: C= \lambda I_p$, for some $\lambda \in R$), can I use it somehow?


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.