Multilevel ordinal logistic model using longitudinal data

[haspanos]

Hi all,

My data set consists of individual responses, nested in regions, for 5 different years. Since my dependent variable has a scale from 0 to 10 and because I would like to capture both individual and regional effects I understand that I have to run a multilevel ordinal logistic regression.

However, I cannot find if there is any command in Stata that I can use in this case. I know that -xtologit- command is used for random-effects ordered logistic models and -meologit- command runs a multilevel mixed-effects ordered logistic regression but none of them is appropriate for my case.

I am aware of this book [Multilevel and Longitudinal Modeling Using Stata, Third Edition] but since I cannot buy it immediately, I would appreciate if you have any additional information.

Thank you in advance.

[Richard Williams]

If the categories are more or less evenly spaced, treating an 11 category ordinal variable as continuous may not be too bad.

[haspanos]

Hi Richard,

Thank you for your comment. I see your point. Actually the categories are totally evenly spaced.

[Clyde Schechter]

I know that -xtologit- command is used for random-effects ordered logistic models and -meologit- command runs a multilevel mixed-effects ordered logistic regression but none of them is appropriate for my case.

I can see that -xtologit- is not appropriate because you have three levels, and the -xt- commands only support two levels. But why isn't -meologit- suitable for your case?

That said, Richard's suggestion of treating the 0-10 scale as if it were an ordinary continuous variable sounds like good advice to me given that you say the categories are evenly spaced.

[haspanos]

Hi Clyde,

Thank you for your response and your comments. I thought that -meologit- was not suitable for my case because I could not specify any time variable. But if I understood correctly, this is not necessary since time could be considered as one level. So level-1 is the individuals level-2 is the regions and level-3 is the different years?

Since you also agree with Richard, if I treat my dependent variable as continuous then I have to use -xtmixed- command right? In this case, I am wondering if the fact that my individual data are collected from a social survey which means that I do not observe the same individuals over time would be problematic. Do you may know if I should create pseudo-panels as I read somewhere? Thank you in advance.

[PRIVARD BRIN]

Hi all,

I am looking for a way to do multilevel ordered modeling for longitudinal.

My data set consists of individual responses, nested in 20 group, for 4 different sector.

I have 3-level for ordinal weighted logit models and longitudinal data (11 times). Indeed, my dependent variable has a score between 0; 0,25; 0,50; 0,75; 1 and because I would like to capture both individual and group effects in theirs longitudinals decisions.

I understand that I have to run a multilevel ordinal logistic regression. I found the commands ologit, meologit, gllamm following the examples of Sophia Rabe-Hesketh and Anders Skrondal (multilevel and long ordinal responses). I use stata 12 and the last two commands don't seem to work.

My question is:

- what command or procedure would be appropriate for fixed or random effects model estimation for this version of stata -stata 12.
- How to consider the longitudinal character of my dependent variable (which is of hierarchical character).
- At what point do you integrate the specific variables at level 1, 2 and 3

Thanks for your responses.