Bootstrap, big samples and small sizes.

[Silke Burestam]

Two questions:

We have a big sample and need to calculate models for the whole sample and for a lot of subgroups of different sizes. It´s a kind of satisfaction index with three questions measuring the satisfaction and a lot of questions measuring different aspects of satisfaction like “satisfaction with the service”.

For the confidence intervals the only way is the bootstrap procedure. But to do this 1000 times for each model and then manually sort the data to get the quantiles is obviously not an easy stuff. Is there a smarter way?

Second question is when we have samplesizes , that means subgroups to small for a own model. Think of it as a city and a model for the city. Then in the city there are areas, like streets, to small to make a model of their own. We have heard of a way not to make models but to use the coefficients from the “citymodel” to estimate the outcome for the “streetmodell”. What we want is to estimate the satisfactionindex for the street using the subsample for this street but to use weights / coefficients from the citymodell. Have you done this and can you describe how to do this?

Thanks a lot!

[Diogenes]

Hi,
it looks multilevel problem
see viewtopic.php?t=482&highlight=multilevel
I am not able to help you at this moment.
Best regards
Bido