Hello all, I am not an expert in statistics but I have used some models in STAN before. Currenlty, I am trying to run a multivariate outcome logistic regression model (multivariate outcome logit model). The below link contains information on how I can perfom this for probit model and linear regres…

[image] tomas.rossetti: You mean a logit regression with categorical outcomes? I mean the multivariate probit model. The documentation description is quite detailed, but for the purpose of getting useful responses here it may help to summarized the extensive description into the core of what …

Thanks for your response. No, I do not mean categorical outcome. Consider the below system of models: y1 ~ x1 + x2 + … y2 ~ x1 + x2 + … y3 ~ x1 + x2 + … y4 ~ x1 + x2 + … Where all y are binary and correlated. You can think of yi as person’s response to a number of life satisfaction questions. …

The problem with coding this model with logits and not probits is that Stan doesn’t have a multivariate logistic distribution to replace the multivariate normal. I think it should be possible to take a multivariate normal, rescale it so that its margins are logistic rather than normal and use that …

@caesoma the model in the doc is equivalent to a multivariate probit because the inverse probit is the standard normal cdf. So to sample y from a bernoulli distribuiton with probability parameter equal to the inverse probit of x is the same thing as sampling from a standard normal distribution and …

Thanks, @jsocolar . Would you know if there is any package that does this task in R without requring me to write the code of the model myself? (not necessarily in the Bayesian form)

@tomas.rossetti , can you provide any guidance for how one might estimate such a model with these packages? Since blavaan can use Stan on the backend, it seems likely that if it can estimate this model we could generate the Stan code. @MaryBo I don’t know of any particular packages for this, but t…

I just remembered that blavaan only implements probits for ordinal/binary outcomes, but lavaan does have the logit link. I guess it’s a starting point. If I understand the model correctly, x1 + x2 + ... is the structural equation of latent variable that represents life satisfaction, according to wh…

Cool! This is a neat and potentially relevant approach. FWIW, I’m pretty sure that it is not the same model presented in the users guide. The SEM approach looks like it’s inducing correlation in the y by modifying the probabilities underlying each y in the same direction when sampling the latent …

[image] Multivariate outcome logit Modeling I believe you could use blavaan or lavaan to estimate this model. Thanks, @tomas.rossetti . I know that SEM models can be used to estimate a system of models (similar to my example). But, I would likt to use my model for pred…

Multivariate outcome logit

Modeling

tomas.rossetti January 25, 2023, 12:58pm 9

I believe you could use blavaan or lavaan to estimate this model.

Topic		Replies	Views
Learning how to write multivariate outcomes models Modeling	2	1637	February 6, 2018
Help calculating log-likelihood with binomial_logit_glm to use with loo Modeling fitting-issues , specification , loo	22	1489	August 3, 2022
Multiple outcomes hurdle model - exists? Modeling	11	1703	March 13, 2019
Multivariate logistic model and Cholesky correlation matrix Modeling rstan , techniques	3	295	December 3, 2024
Efficient computation of multi-logit Modeling	4	177	December 16, 2024

Multivariate outcome logit

Related topics