Sounds like you require a many-to-one merge.

You should get your information on each individual and their county and/or time observed. Then you can do a .

Just be forewarned you will need to have variation on county OR time in your observations in order to glean any insights from these variables.

Dear Jeffrey,

Thank you very much for your response. I have individual data for 25 countries. Shall I center to mean for every variables in the model? For example I have age as a variable, then shall I center it to mean? If you could add more on this point, it would help me greatly.

Regards

Prafulla

The path to follow is merge.

If you have opened your low-level database, you have to use merge m:1 varlist using high-level database

For estimation you have several options and it critically depends on the type of data you have. If you have panel data, you can use fixed-effects + clustered standard errors at the highest level (conservative approach) or more complex multilevel models (require more assumptions, see, e.g., https://www.stata.com/bookstore/multilevel-longitudinal-modeling-stata/). If you have cross-sectional data, you will have to choose between fixed-effects and clustering (with less ability to pick up the effects at certain levels) and other multilevel models where you explicitly model the effects at different levels... In Economics, for instance, the practise is to use the conservative strategies with fixed effects and clustering.

You do not need to recenter the variables, as far as I know.

Thank you so much for your kind suggestion. I will look into the things you suggested.

Prafulla