2
$\begingroup$

If x = (x_1,x_2,...,x_n) is a vector whose components have a distribution that is a finite mixture of multivariate normals, is the expected value of x_1 still a linear function of the other components, as is the case when the components have a multivariate normal distribution?

$\endgroup$
1
$\begingroup$

If$$(x_1,\ldots,x_p) \sim \sum_{i=1}^k \pi_i {\cal N}_p(\mu_i,\Sigma_i)$$then$$f(x_1,\ldots,x_p)\propto\sum_{i=1}^k \pi_i \exp\{-(x-\mu_i)^\text{T}\Sigma_i^{-1}(x-\mu_i)/2\}\big/|\Sigma_i|^{1/2}$$and $$f(x_1|x_2,\ldots,x_p)\propto \sum_{i=1}^k \pi_i \varphi(x_{-1};\mu_i,\Sigma_i)\exp\{-(x_1-\mathbb{E}[X_1|x_{-1},\mu_i,\Sigma_i])^2/2\sigma_{i1}^2\}\big/|\Sigma_i|^{1/2}$$ meaning that the posterior expectation of $X_1$ given $X_{-1}$ is a non linear combination of the componentwise conditional expectations.

$\endgroup$

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.