Levels Problem in Regression

[Moeen Butt]

I hope you are doing well.

I am facing a problem with levels in model specification. I have an independent variable at the restaurant level and an outcome variable at the health district level. There are many restaurants at the health district level. Is there any statistical approach/regression model that corrects for such levels issue? I would appreciate your help.

Moeen

[Rich Goldstein]

are restaurants nested within health district levels (i.e., does each restaurant appear in one and only one health district)?

[Carlo Lazzaro]

Moeen:
we need to know a bit more about your dataset.
I'm not clear if your data can be analysed with OLS invoking a clustered standard errors at health district level or if you can take the whole stuff a notch up with -mixed- (if you have a time variable).

[Moeen Butt]

are restaurants nested within health district levels (i.e., does each restaurant appear in one and only one health district)?

Yes, restaurants are nested within health districts.

[Moeen Butt]

Moeen:
we need to know a bit more about your dataset.
I'm not clear if your data can be analyzed with OLS invoking a clustered standard errors at health district level or if you can take the whole stuff a notch up with -mixed- (if you have a time variable).

My dataset comprises restaurants level inspections. The restaurants are located in health districts. There are total 20,000 restaurants located in 26 health districts. My independent variable is restaurant inspections (i.e., number of times a restaurant is inspected per year). My outcome variable is Foodborne illnesses, which is at the health district level. Therefore there is a level issue between my independent and outcome variables.

[Carlo Lazzaro]

Moeen:
do your data refer to one wave of data (that is, the 20K restaurants are inspected only once during a given year; cross-sectional data) or to different ones (that is the same restaurants are inspected in different occasions, equally spaced, more or less, across time (panel data)?

[Moeen Butt]

Hi Carlo, I have four years of data (panel data). Restaurants can be inspected multiple times per year. Some restaurants are inspected once per year, whereas some are inspected three or more times per year.

[Carlo Lazzaro]

Moeen:
assuming that your regressand is categorical (Foodborne illnesses yes/no), you may want to consider -xtlogit- with (conditional) fixed effect.
As Stata is likely to complain about -repeated time values- you can -xtset- your data with -panelid- only (restaurant, I assume), being aware that this fix comes at the cost of making time-series operators (such as lags and leads) unavailable.