Can this be solved using a binary logistic multi-level model?

by Anony Mouse   Last Updated August 14, 2018 20:19 PM

Hello everybody!

Is it possible to solve the following task by using a binary logistic multi level regression? If not, how can you solve it?

The concept as a diagram:

enter image description here

  1. I have the location of individual stores and data on locational factors based on the stores exact location.
  2. I know to which retail chain each individual store belongs but have only little data for the retail chains.
  3. I have different retail centers that contain the individual stores and I have data for locational factors based on the retail centers.

So, by just looking at that, it looks like I have 3 levels, BUT there are problems:

  • The dependent variable is on the level of the retail chain: insolvent? (yes or no). I want to predict if a retail chain is more likely to get insolvent, based on locational factors of the stores and based on locational factors of the retail centers, in which these stores lie in.

  • I think the levels aren't strictly hierarchical: Each individual store is only inside one retail center, BUT a retail chain contains multiple stores. Because of that, a retail chain is present in multiple retail centers.

We had the idea of having 3 levels in a multi level model, but I think one can't say retail chains are nested in retail centers, because they are, through their stores, present in multiple retail centers. The only thing on this train of thought I could come up with is having the retail chain as level 3, then have the retail centers at level 2 and the stores at level 1. And I just managed to get some first results that way through the SPSS generalized linear mixed models function, but I am very much unsure, if this makes sense / is allowed.

At first, I just wanted to make this a two level analysis, where I would just say that my dependent variable is on the level of the store: does this store belong to an insolvent retail chain? (yes or no) and then have independent variables (the locational factors) on the level of the store (level 1) and on the level of the retail center (level 2) in which each store lies. I would then look at the difference in aggregated predicted stores between the group of insolvent retail chains and not insolvent retail chains to conclude something out of that.

A smart person now told me, that the two level version would be improper if done like this, because being insolvent is not really a variable that is on the level of the store, but really on the level of the retail chain.

Is there any way to solve this?

Thanks a lot if you read all or part of it! Any ideas would be much appreciated.




Related Questions


How do you solve your nested data analysis?

Updated August 07, 2017 02:19 AM