Uncertainty about the -mixed- method

[Joakim Dahl-Nielsen]

Hi,

I am new to Stata and hope you can help. I am currently writing my thesis on how country level data affects firm level outcomes over time. I argue that the time level is nested in the firm level which is nested in the country level. I have chosen the -mixed- command for this. The time period is 2006-2017 (with variable name "t"), I have 5816 unique firms (with variable name f_id) and 33 unique countries (with variable name c_id). It is clear to me that the data should be in the long format, but I am in doubt as to whether the data structure I am using is correct, and I hope you could help with this. I apologize if the format is incorrect.

Should the data be structured as:
a) I only have 1 set of country observations. That is, I only have one observation for country 1 at 2006, one observation for country 2 at 2006 etc.

c_id f_id t
1 1 2006
1 1 2007
1 1 2008
2 2 2006
2 2 2007
2 2 2008


or

b) I repeat the country observation for every firm observation. That is, I have repeat country observations for country 1 at 2006 next to every firm from country 1 at 2006.
c_id f_id t
1 1 2006
1 1 2007
1 1 2008
1 2 2006
1 2 2007
1 2 2008
... ... ...
2 10 2006
2 10 2007
2 10 2008
2 11 2006
2 11 2007
2 11 2008

Thank you in advance

[Clyde Schechter]

Structure b) is what you need.

[Joakim Dahl-Nielsen]

Thank you Clyde Schechter!

If I want to compute the model using -mixed, ml- such that time is level 1, firm observations are at level 2, and country observations are at level 3, should the syntax then be:

-mixed sps roa rd sales cfs tdta mvb fata lngdpcap trade inflation avg1inverted clusters regulationburdeninverted mktcapcred eduquality unions wageflexibilityinverted publictrust avg2inverted i.t || c_id: ||f_id:, ml variance

Where sps is the dependent firm-level variable, roa through fata are firm level controls, lngdpcap through avg2inverted are country-level variables and i.t is a time dummy?

[Clyde Schechter]

Note that since at least version 15, ml is the default estimation for -mixed-, so you don't have to specify that unless you are using an older version of Stata.
Your code looks basically right. I have a few little concerns. Other than t, or any of your variables discrete? If so, they, too should have an i. prefix? Also when I see avg1inverted and avg2inverted, I worry that the latter is the square of the former. Is it? If so, it should not be a separate variable, it should be c.avg1inverted#c.avg1inverted.