Hello, I’m playing around with shifted log normal distributions (for reaction time data), trying to get my head around the parameter estimates and also how to do contrasts with them using the emmeans package (which from brms v2.13 and later, has its own emmeans support according to the documentatio…

Thanks for the context. I think @rvlenth wrote a lot of what I wanted to say, while I was preparing my answer - I basically describe the same stuff, but in a bit different language (and I didn’t want to delete my draft, so I am posting it as well :-). I fear you are misunderstanding multiple things…

Hi, some of the response depends on the exact formula (and potentially other code) you used to fit your model, could you share those, as well as the complete model output? [image] nahorp: Is the above understanding of sigma_Intercept as exp(5.21-0.97) correct? Not really. First, focusing to…

Hi @martinmodrak , thank you for your response. could you share those, as well as the complete model output? Surely, these are as below, [Screen Shot 2020-08-27 at 4.09.39 pm] What are your thoughts now based on the above? Is sigma_Intercept interpreted as as exp(5.21-0.97) or is there anot…

Well, I have to say I am confused by these questions. I think the parameters are being misinterpreted. As I read the output, the three parts of the model predict as follows: mu = 5.21 + 0.41*age_group1 + 0.01*bias_type1 – 0.22*trial_type1 log(sigma) = -0.97 + 0.15*age_group1 log(ndt) = 5.05 + 0.03…

BTW, I’m now not sure if the answer I am quoted in in Q.3 is correct. It depends on how mu and ndt are to be interpreted. It may not be at all appropriate to force a log link on the mu model.

Thanks for your response @rvlenth . With regards to the first question, I understand your response and the use of type = "response" in the emmeans call but exp(-0.97) is 0.379. If this is on the response scale (which is milliseconds), 0.379 milliseconds for the first age group (age group is a cate…

Well, these are things I can’t say much about., but I am quite sure I interpreted the model coefficients correctly. What I don’t know, and what you need to find out, is what is meant by the parameters mu, sigma, and ndt. I know that, often, a lognormal distribution is parameterized by \mu and \sigma,…

Thank you so much @rvlenth and @martinmodrak for pointing me in the right direction! All of it makes so much more sense now. I did a bit more reading and now know how the parameters from the model relate to the mean and the variance of the distribution ( this website was also great in that regard). …

Contrasts for distributional parameters in a shifted log normal distrbution

Interfaces brms

Topic		Replies	Views
Understanding shifted lognormal parameters and priors brms interpret-results	8	6857	December 1, 2020
Shifted lognormal stan conversion and mean function brms	4	610	December 5, 2020
Modelling sigma Modeling	13	1645	December 1, 2020
Lognormal brms model: looking correlations between its re-parametrized estimates (median and SD)? Modeling	2	718	December 7, 2020
Shifted log normal distribution with varying shifts by participant brms	12	3528	November 23, 2020

Contrasts for distributional parameters in a shifted log normal distrbution

Related topics