I am trying to write winbugs codes for a latent variable w.
My outcome y follows Bernoulli distribution and I want to use the latent one (w) in my analysis.
y=1 if w>0 and y=0 if w=
Seems like you can use the step(e) function in your program. Initial values do not matter in most cases.
okay. Thanks Andrey Davydenko . How will be the structure of the data for Z (latent variable)?
####Data
z?
Do I need data for y (binary variable)?
Hi Thierno Barry ! Actually, if you want to implement Bayesian logistic regression, you do not need that normally distributed variable and the step function.
I strongly recommend you to see Page 19 from this tutorial: http://www.medicine.mcgill.ca/epidemiology/Joseph/courses/EPIB-668/bayesreg.pdf
Good luck with WinBUGS modelling, I believe it's a great tool. But you can use R instead, there will be more people who may help as R has wider community. HTH!
Okay my Dear Andrey Davydenko , but I would just to use latent variables in my analysis.
Previously, I used binary variables (y1 and y2). Now, I need the latent variables Z for my analysis.
Just, how to transform?
If you want to use a normally distributed variable (w), it seems you then cannot use the Bernulli-modelled variable at the same time. You need to use either Bernulli distribution or normal distribution. Otherwise the model cannot be correctly specified.
But what is the purpose of introducing this normally distributed variable?
- Badges
- Science topic
- Similar topics
- Statistics
- Survey