Hi:
I have a query I'm hoping someone can advise me on. I've looked at journal papers, textbooks, other forums, and have tried people at my own institution (those who are familiar with area don't have time to help!), and while I can see what I want to do can be done, I don't have any clear advice on how to proceed...
Here's a broad outline of my problem:
I'm fitting a multilevel model using an historical data set. I then want to use this model to make future predictions of the outcome variable. As well as predicting the outcome for clusters included when fitting model, I want to predict the outcome for new clusters (i.e. clusters not included when fitting the model).
To do this I first need to predict the values of the random effects for each cluster. In Stata, this is simple enough for clusters included when fitting the model (i.e. Stata can give me random effects as point estimates and errors). But, I don't know how to do it for new clusters. I see this is possible from an explanation given by Andrew Gelman ('Data Analysis Using Regression and Multilevel/Hierarchical Models') but he uses R and WinBUGs: at the moment I don't have time to learn how to use these so need to stick with Stata. A key paper using Stata is by Skrondal and Rabe-Hesketh ('Prediction in multilevel generalized linear models') but they don't cover how make make predictions of random effects for new clusters (at least as far as I understand the paper).
In sum, my question is:
After fitting (estimating) a multilevel model in Stata, how (e.g. commands; particular restructuring of the data) do I predict random effects for new clusters?
(Note that, the reason I have new clusters is that I have historical data for the outcome variable for a sub-set of the clusters, but projection data for the predictors for all clusters).
Any help would be greatly appreciated.
Thanks
S
I have a query I'm hoping someone can advise me on. I've looked at journal papers, textbooks, other forums, and have tried people at my own institution (those who are familiar with area don't have time to help!), and while I can see what I want to do can be done, I don't have any clear advice on how to proceed...
Here's a broad outline of my problem:
I'm fitting a multilevel model using an historical data set. I then want to use this model to make future predictions of the outcome variable. As well as predicting the outcome for clusters included when fitting model, I want to predict the outcome for new clusters (i.e. clusters not included when fitting the model).
To do this I first need to predict the values of the random effects for each cluster. In Stata, this is simple enough for clusters included when fitting the model (i.e. Stata can give me random effects as point estimates and errors). But, I don't know how to do it for new clusters. I see this is possible from an explanation given by Andrew Gelman ('Data Analysis Using Regression and Multilevel/Hierarchical Models') but he uses R and WinBUGs: at the moment I don't have time to learn how to use these so need to stick with Stata. A key paper using Stata is by Skrondal and Rabe-Hesketh ('Prediction in multilevel generalized linear models') but they don't cover how make make predictions of random effects for new clusters (at least as far as I understand the paper).
In sum, my question is:
After fitting (estimating) a multilevel model in Stata, how (e.g. commands; particular restructuring of the data) do I predict random effects for new clusters?
(Note that, the reason I have new clusters is that I have historical data for the outcome variable for a sub-set of the clusters, but projection data for the predictors for all clusters).
Any help would be greatly appreciated.
Thanks
S
