www.cmm.bristol.ac.uk/forum

Hi everyone

I'm relatively new to MLwiN, but hopefully this isn't a silly problem. I've asked a few colleagues, but haven't been able to get to the solution yet.

I have a dataset with 2.7million cases. They are counts of deaths aggregated by age group, year of death, and area of death.
I have four levels:
Country (1,2) (l)
Area (1, ... 25,253) (k)
Year (1990 - 2010) (j)
ID (sequential ID number for each count of deaths for 15 age groups; 1, ... 2.7million) (i)

I've set up a Poisson model, with an area population size offset.
I've included all of the relevant fixed terms (14 variables + constant + offset) into the model, however, the area level measurements have all four subscripts, rather than just k and l as expected. This has happened to me before, and I needed to sort the data to fix it. However, when I try sort the data in MLwiN (using the four variables as sort keys, and applying the sort to all variables in the dataset), the program goes to "Not Responding", and then eventually crashes.


Further information:
- The Hierarchy Viewer is blank (completely empty, no variables at all) if that could offer any explanation? (see https://www.dropbox.com/s/255xabupktgu6 ... r.JPG?dl=0)
- The area variable was originally alpha-numeric (eg '00AAFA' or 'G12 8RZ'). I have converted this so that each area code is now a unique number (ranging 1-25,253). We had hoped that this was the problem, but it hasn't helped.
- I'm running MLwiN 2.34

Any ideas and suggestions would be greatly appreciated!
Thanks,
Megan

When you sort the data through the graphical interface it firstly makes a copy of the data that will be overwritten so that it can offer the possibility of undoing it afterwards. As you have a large data file it may be that there isn't space to do this. If you use the command line version of the SORT command (see Help->Contents and Index) this does not happen and you may therefore be more successful. If this doesn't work either then I can attempt to trace the problem if you were able to provide some data that exhibited the same problems in a predictable manner.

Normally when the hierarchy view doesn't display anything it means that the model is not completely set up (the equation window would probably also say 0 out of 0 cases in use). If that isn't the case then I would need to see the worksheet in question before providing an explanation.