1
$\begingroup$

Given three ordered-categorical variables: $u_1, u_2, u_3$ with $K$ categories, I'm trying to derive their expected variance-covariance matrix using their marginal probabilities, thresholds, and polychoric correlation matrix.

I can derive their variances using the marginal probabilities:

Where:

$\bar{u_1} = p_1*1 + \cdots + p_k*K$

$V(u_1) = p_1*(1-\bar{u_1}) + \cdots + p_k*(K-\bar{u_1})$

But I'm stuck on how to derive $Cov(u_1, u_2)$. I've been told I'll need the joint probabilities of $u_1$ and $u_2$, but I'm not sure how to get those from the information that I have:

  • Marginal probabilities
  • Thresholds
  • Polychoric correlation matrix

Any help greatly appreciated. Thanks!

$\endgroup$

1 Answer 1

1
$\begingroup$

Asked for help too soon sorry, I found the solution in Olsson (1979, p.447):

The joint probability $p_{ij}$ can be derived via:

$p_{ij} = \Phi_2(a_i,b_j) - \Phi_2(a_{i-1},b_j) - \Phi_2(a_i,b_{j-1}) + \Phi_2(a_{i-1},b_{j-1})$

Where $a_i,b_j$ are the thresholds and $\Phi_2$ is the bivariate normal PDF with (polychoric) correlation $p$

Olsson, U. Maximum likelihood estimation of the polychoric correlation coefficient. Psychometrika 44, 443–460 (1979). https://doi.org/10.1007/BF02296207

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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