I am attempting to conduct a robust correlation estimation by adapting This Post to 4 outcome variables (my real data have 9 outcomes). I’m a little stuck on how to write the likelihood to estimate the additional scales and degrees of freedom parameters. Some fake data: sigma = c(10, 15, 10, 20) …

Does multi_student_t allow a vector for the argument nu?

It looks like mu can be a vector , nu is a real, and cov (sigma) is a matrix

Just checked and no, nu can’t be a vector. When the doc says real it means a single real, not a vector.

Thank you! Setting nu to a single real fixed the issue. Would it make any sense to put a linear formula on nu to to estimate a real for each outcome and then loop over those estimates?

[image] JLC: to estimate a real for each outcome and then loop over those estimates? That’s a really neat idea I’ve never seen/thought of before. Give it a shot with some fake data and check how it performs.

That looks much more reasonable than the ugly hack I was thinking of. I thought maybe I could just tuck a line into transformed parameters (and I hope I don’t need anything Jacobian) or alternately put something into the model block.

Oh! When you were talking a loop I thought you were thinking of looping over the N rows with a nu per row thereby making nu a kind of outlier-ness quantity to label each row

I’m mostly interested in the rho estimates and having those be robust to possible outliers. Looping over rows makes more sense! I was thinking about it incorrectly. I’m using this application to try to wrap my head around correlation estimates in raw Stan and to see if they align with the estima…

If dealing with likert/ordinal data, be sure to check out this and this .

Estimating Robust Correlations for Multiple (>2) Outcomes

Modeling

mike-lawrence June 2, 2021, 3:30pm 7

Note also that there’s description here how to hack a different-nu-per-dimension. I wish it was a built in thing though.

Topic		Replies	Views
Is there a more efficient way to code this robust correlation approach Modeling rstan , techniques , specification	8	956	October 31, 2021
Cholesky Decomposition with Ordinal Outcome Modeling rstan , techniques , specification	9	694	May 27, 2021
Estimating correlations between fitted parameters Modeling cmdstan , specification	20	1186	June 17, 2024
Multiple linear regression with t student Modeling	5	646	September 22, 2018
Remove/estimate correlation of parameters (or get to know if correlation is real) Modeling	34	3374	April 16, 2020

Estimating Robust Correlations for Multiple (>2) Outcomes

Related topics