Can I use CRE in cross-sectional data?

[Hope Muronga]

Hi Statalists,

I am working on household survey data, cross sectional. The survey is repeated on different households each year. My question is whether it is correct to use Correlated Random Effects on my data.

I am trying to look at the impact of social interaction on the experience of crime. My dependent variable is a binary variable (victim of crime or not). My data includes different types of crimes, Household Characteristics and Community Interaction factors.

Thank you!

[Clyde Schechter]

If you have cross-sectional data, where different households appear in each wave of the survey, I don't see what use correlated random effects would be. CRE is one approach to separately identifying within- and between-panel effects in longitudinal data. But your data won't have any such effects? Unless I am missing something like you have multiple respondents within households, or some other nesting structure in the data that you haven't mentioned, there isn't anything to apply CRE to.

[Hope Muronga]

Thank you very much for your response Clyde Schechter

I am fairly new to Stata and the following is how my data looks like, I also have data on Household characteristics and community interaction factors. I have read on logit and clogit but my data is on Household characterisics is not across crimes. I have had trouble knowing how to implement it. Your suggestions will be greatly appreciated.

Household	Crimes	Victim of Crime	Crimes most likely to occur	Actual crime level in the area
1	Murder	0	0	1365
1	Homerobbery	0	0	1543
1	Burglary	1	0	17654
2	Murder	0	0	35
2	Homerobbery	0	0	15
2	Burglary	0	0	1876
3	Murder	0	0	2357
3	Homerobbery	0	1	123
3	Burglary	1	1	78

[Clyde Schechter]

I think you are not at the stage of writing code to estimate logistic models yet. You need a clearer picture of what your question(s) is(are) and what variables you think bear on them. For example, you state that victim of crime or not is your dependent variable. But you actually have three outcomes for that for each household, one each for murder, home robbery, and burglary. For all I know you have other crimes there as well. Do you want to do separate models for each crime? Do you want to instead synthesize a variable for "victim of any of these crimes" and analyze that? I can only imagine what your social interaction variables are, but are all of them to be used, or, if you are modeling each crime separately, are some types of social interaction more salient and others perhaps irrelevant? If you are going to study each crime separately, do you also plan to compare or contrast things across crime categories?

I raise these questions because you cannot begin to write code until these are settled issues. Moreover, the particular data organization that will be most suitable will also depend on the answers to these. As I have no expertise in either sociology or criminology, I cannot advise you on any of this. If you are not comfortable with your subject-matter expertise to make these decisions, then you should consult a colleague in one or both of those disciplines. When you have a concrete set of research questions spelled out, then you can start to worry about how to code the analysis in Stata.

[Hope Muronga]

Thank you very much. I really do appreciate your time!