Multiple membership model

[Frankfeng]

Hi Sir,

I'm looking at the spatial-template models. In my dataset, years nested within region. I also try to looking at the multiple membership effect. After I built up the model. The IGLS runs ok. However, when I change to MCMC and run the model, it says: the 2th MM unit for obs 1 is present but has zero weight.

For each weight, what I calulate is the current neighbourhood divided by the total number of the neighbourhoods.

Could you let me know which part I can fix the model? And I also attached the print screen. I find the equation in Mlwin 2.27 is a bit different from the MCMC manual. Is this also the matter?

Thank you!
Frank

[ChrisCharlton]

When setting up a multiple membership model in MLwiN it requires all of the ID columns to be consecutive, i.e. if the first ID column for the classification was in c10 then the others would be in c11,c12, etc. This is also true for the weight columns. You also need to check that for each non-zero ID the corresponding weight is non-zero and for each ID set to zero (i.e. empty) the weight needs to be set to zero. I'm not sure why some of the symbols of your equation aren't displaying correctly, although it may be an issue with the fonts on your machine.

[Frankfeng]

When setting up a multiple membership model in MLwiN it requires all of the ID columns to be consecutive, i.e. if the first ID column for the classification was in c10 then the others would be in c11,c12, etc. This is also true for the weight columns. You also need to check that for each non-zero ID the corresponding weight is non-zero and for each ID set to zero (i.e. empty) the weight needs to be set to zero. I'm not sure why some of the symbols of your equation aren't displaying correctly, although it may be an issue with the fonts on your machine.

Hi Chris,
Thank you very much for the information!

Yes, the weight columns are consecutive. And the weight had been transfered from "missing value" to zero if the place dosen't have the neighbourhood (of course, each region has at least one neightbourhood).

I also tested whether the Year has problems (it is from 1995 to 2010), I change them from 1to 16. But the warning still show up.

I wonder how I can trace the " the 2th MM unit for obs 1 is present but has zero weight."?

Thank you!
Frank

[ChrisCharlton]

The code that is generating this error is simply looping over each of your multiple membership ID columns and checking that the corresponding weight column is as expected. In this case it is saying that the first row of your second multiple membership ID column has a non-zero value however the first row of your second weight column has a zero value. If this isn't the case then I would need to see the data to be able to determine why this is being triggered.