Multilevel regression where level 2 error term follows an AR 1 process

[Solange Sardan]

Hello everyone!
Currently i'm working on a multilevel model that follows the framework from DiPetre and Grusky, where they assume that level 2 corresponds to time points and level 1 to individuals. Now i'm struggling with the assumption that time dependence is modelated as an autorregresive structure between random effects, and this is why i'm asking for your help. Is there an specific command to to this? or should i treat the errors term separately?
Thank you a lot!

[Clyde Schechter]

The -mixed- command has a -residuals()- option, which you can specify as ar# (replace # by an appropriate number). See -help mixed- and the corresponding manual section for more details.

[Solange Sardan]

I read that one, but i thought it could be used only for the lowest level of the model, may i use it for the highest level too?
And thank you for the response, i really appreciate it.

[Clyde Schechter]

I thought it was the lowest level of the model that you wanted autoregressive error structure for. (The use of terms level 1 and level 2 is not helpful, because different people use them to refer to different, indeed, opposite, things.) The -residuals()- option only is used for the bottom level. There is a somewhat analogous -covariance()- option that applies to higher level random effects, but it does not encompass an autoregressive model. I'm not sure Stata can accommodate what you want. I don't know of a way to do it, but perhaps somebody else does.

Your chances of getting a better response will improve if you provide a complete reference for DiPetre and Grusky so that somebody who doesn't recognize it instantly and is motivated to look it up can actually find it. This is a multi-disciplinary forum, and references that may have risen to folklore status in one discipline may be completely unheard of in others.

[Solange Sardan]

Thank you very much Clyde! you're right, i'll add more information about the paper.
The paper is called "The Multilevel Analysis of Trends with Repeated Cross-Sectional Data" by Thomas A. DiPrete and David B. Grusky (1990)

[Joseph Coveney]

Wouldn't a complete reference look more like the following?

DiPrete, Thomas A., Grusky, David B. 1990. The multilevel analysis of trends with repeated cross-sectional data. Sociological Methodology. 20, 337–68.

[Solange Sardan]

Yeah, that's the one.

[Bingqing GUO]

I thought it was the lowest level of the model that you wanted autoregressive error structure for. (The use of terms level 1 and level 2 is not helpful, because different people use them to refer to different, indeed, opposite, things.) The -residuals()- option only is used for the bottom level. There is a somewhat analogous -covariance()- option that applies to higher level random effects, but it does not encompass an autoregressive model. I'm not sure Stata can accommodate what you want. I don't know of a way to do it, but perhaps somebody else does.

Your chances of getting a better response will improve if you provide a complete reference for DiPetre and Grusky so that somebody who doesn't recognize it instantly and is motivated to look it up can actually find it. This is a multi-disciplinary forum, and references that may have risen to folklore status in one discipline may be completely unheard of in others.

Thank you so much Clyde! Your solution is very helpful.