Interpreting multilevel output

[Valentin Stoian]

Hello.
I am working on a project aiming to study whether the perception of corruption at individual level, in different sectors (police, health, education) is influenced more by individual-level variables such as whether the individual paid a bribe in any of these sectors (a binomial variable) or region-level level variables (crime/capita, number of beds). I am using a combination of the Quality of Government Institute Dataset EQI dataset http://qog.pol.gu.se/data/datadownloads/qog-eqi-data and a selection of their regional-level variables http://qog.pol.gu.se/data/datadownlo...euregionaldata. For now, I am trying to predict the perception of corruption in police at the individual-level, as a function of whether the person paid a bribe to police or not (level 1) and of the crime rate( at NUTS2 regional level). I entered the following model (by point-and-click):


. mixed corruption_prevalent_police crime_per_capita, || nuts_c: R.crime_per_capita, covariance(identity)
and got this output.

I have read articles http://journals.sagepub.com/doi/abs/...nloadContainer where the second-level variable is presented as having a p-value and becoming significant in some models. Could you please advise me on how to read this output? No advanced math please, I am a Stata newbie, but have some experience with SPSS.
Valentin

[Damian Tojeiro]

Hi Valentin, it seems that your mixed model has sense instead of a simple OLS as you can see at the bottom of the table. I have read that you should allow the covariance being unstructure in order to have a correlation between your random intercept and random slope unless you have specific reason for not doing it. In order to check if your dependent is different with respect to the regional dimmension you can use the intraclass correlation that will give you what I understood you want, that is, the proportion that your level-1 and level-2 variables are explaining (estat icc is the command).
Hope this help. For sure, people from this forum know more than me on this and they will tell you more.
Best.

[Marcos Almeida]

Welcome to the Stata Forum/ Statalist,

Please take some time to read the FAQ. There you will find advice on how to share command, output and data so as to entice a truly helpful reply.

That said, mixed models, you know, tend to be more complex than multiple regression. Hence, it is quite a jump and core knowledge should be grasped before delving with ‘crossed’ slopes.

Perhaps you should start , say, with the null model, then go on with the additional covariates. Using the R. notation must come with a solid argument as well.

Last but not least, you may find interesting information in the Stata Manual as well as tips on the interpretation.

Hopefully that helps.