estimation of spatial multilevel model

[gtrocky1982]

Dear All,
I just joined this active forum. Thank you Charlton and George for all their help in this process. I’m a Ph.D. student and writing my dissertation at the moment. One of my essays aims to assess the factors influencing child malnutrition in Nepal. I have data on the Z-scores measured for total kids of 7500 (surveyed in 2006 and 2011 but not the panel). More than one kids were sampled from a household and these households are located in various districts of the country. I plan to model it as a three-level multilevel model: 1) Kids, 2) household, 3) District. At the district level, I’d like to take account of spatial error correlation across the neighboring districts. My dependent variable (Z-scores) is a continuous variable and not the discrete outcomes. Can you suggest me how to estimate such models? Thank you.
Sincerely,
Ganesh

[Paings1987]

In my opinion, here it is, as far as reasonably described

Code: Select all

http://www.bristol.ac.uk/media-library/sites/cmm/migrated/documents/multilevel-spatial-models.pdf
http://www.envplan.com/abstract.cgi?id=a45398

I hope that I helped a little bit though.

[gtrocky1982]

Thanks for the papers. Hope I'll be able to estimate the model.
Ganesh

[GeorgeLeckie]

Dear Ganesh,

Thank you for your email

Suggest, you first fit a standard three-level random-intercept model ignoring district-level spatial dependency, then extend the model to include either a multiple membership or CAR random effects structure to allow for district-level spatial dependency. See the following...

Chapter 17 in our MCMC manual
http://www.bristol.ac.uk/cmm/media/migr ... -print.pdf

Best wishes

George