Endogenous problem in multilevel analysis

[shkingdom]

Dear all,

I have to conduct a two-level model (level 1=respondents and level 2=cities). The variable of my interest is a level 2 variable.
My concern is that the variable of my interest is likely to be endogenous (i.e., it is likely to be associated with unobserved level 1 and 2 factors). Accordingly, I want to apply an instrumental variable approach.
I've read several readings but I was not able to find any information regarding the level 2 endogenous variable. For example, many readings deal with the endogenous problem of level 1 variables.
Additionally, the fixed effects model is not suitable as it cannot estimate level 2 variables.
It would be very much appreciated if someone tells me how to handle this issue in MLwiN.

Thanks in advance,
Wander

[joneskel]

The are a lot of things going on in your question

This paper discusses one type of endogeneity and what to do about it

Explaining Fixed Effects: Random Effects modelling of Time-Series Cross-Sectional and Panel Data
Andrew Bell, Kelvyn Jones

available from

https://www.researchgate.net/publicatio ... Panel_Data

This next paper (which is not in finished form!) discusses modelling and estimating level 2 variables

Do multilevel models ever give different results?.
Kelvyn Jones, Caroline Wright, Andrew Bell
in preparation 01/2013;

available from https://www.researchgate.net/publicatio ... ev=prf_pub

Here is the abstract

It is sometimes said that multilevel models make no substantive difference to interpretation and represent something of a fuss over nothing. This paper demonstrates that this is not always the case with a simple example. It is shown that differences between a single ­level and multilevel model will depend on the nature of the between and within group relations. We demonstrate the equivalence of the within­ between model and the Mundlak model of contextual effects and argue that such models should be routinely used as the standard multilevel model conflates two potentially different relations in one term. We also consider the multilevel model as a measurement model and show how the multilevel property of precision weighted estimates can usefully be exploited as a simple way of dealing with measurement error in the calculation of group means. This results in less biased estimation of the contextual and between relations in a model of cross­level interactions. We provide an Appendix that gives a step ­by­ step demonstration of how these models can readily be specified, estimated, and displayed in the MLwiN software.

and here finally is the data on which the modelling is done in the second paper

https://www.researchgate.net/publicatio ... ev=prf_pub

[francescoaiello]

Hi. I wonder to know if the Mundlak (1978) procedure to control for level 2 endogeneity has been extended to a 3-level multilevel model in a longitudinal setting. The case is as follows: time is the first level of the hierarchy, firms are at the secodn level and regions are the third level.