0

I have a longitudinal data, with different follow-up number for individuals. I have considered measurements for each individual as a curve which I already smoothed them, then calculated the area under curve and divided by the interval length to find the average of my outcome which is functional ability over a year. My aim is to see if the area is different among individuals with and without disease (my exposure). I do not want to remove the zeros. I tried the model below, but it is not approprite.

f = lmer(Area ~ statusDisease+age+gender+education+(1 | sibling id), data = dat, REML = FALSE)

Then I am advised to do Tobit model? but is it appropriate in my case? and how to death with the dependency? like the random factor in lmer(). Which model in appropriate in this case? Any help is appreciated.

user358238
  • 129
  • Please explain what you are actually trying to model. Understanding the variables is important in determining the modelling approach. – mkt Jul 30 '22 at 14:49
  • @mkt I already read the link, but did not get anything out of it for my case, please see the update, where I explain more my aim in the model – user358238 Jul 30 '22 at 14:59
  • I'm afraid this does not clarify the situation sufficiently, nor is it clear why the answers at the linked thread do not address your case. – mkt Jul 30 '22 at 15:01
  • Can you please explain why the proposed duplicate is not one? Based on your description, it seems like it answers your question. – Stephan Kolassa Jul 30 '22 at 15:02
  • 1
    Have you tried the approaches in the top voted answer to the duplicate? If they did not work, can you explain where the problem is? If we can clearly differentiate the two questions, we will happily reopen is. – Stephan Kolassa Jul 30 '22 at 15:06
  • @ Stephan Kolassa it is because some of the areas (outcome) are zero. And I do not want to remove the individuals with zero area. – user358238 Jul 30 '22 at 15:06
  • 1
    The top voted answer to the duplicate does not propose removing zeros, which I agree would be a bad idea. Have you tried zero-inflated or hurdle models? – Stephan Kolassa Jul 30 '22 at 15:08
  • @ Stephan Kolassa one of the suggestion is :glmmTMB will also do 'zero-inflated'/hurdle models for Beta or Gamma responses), but could you please clarify kindly the details? for example how should I model? should I choose, family = beta? is that sufficient? – user358238 Jul 30 '22 at 15:12
  • @ Stephan Kolassa Could you please kindly clarify these models? "Have you tried zero-inflated or hurdle models?"which package can I use? and which parameter should be set in order to get a correct model in my case? zero-inflated can work for NOT count outcome? thanks for your guidance – user358238 Jul 30 '22 at 15:26
  • @ Stephan Kolassa could you please answer the question with the update? – user358238 Aug 05 '22 at 17:30

0 Answers0