Can I test country level data and firm level data in one regression

[Martijn van Gijlswijk]

Hello everyone,

I'm trying to find the effects of culture on the existence of teamwork. My dataset consists of 29 countries and of around 30.000 companies. My dataset is cross-sectional.

Dependent variable
Team = a dummy variable that equals 1 when there is teamwork present in the company. (Firm level)

Independent variables
Individualism = A national cultural score which is defined on country-level (range 1-100).
Masculinity = A national cultural score which is defined on country-level (range 1-100).
Power = A national cultural score which is defined on country-level (range 1-100).
Female = The percentage of females in the company used as a proxy for altruism (Firm level)

Control variables
Size = This variable equals the amount of employees in the company (Firm level)
Sector = 0-5 depending on type of sector (dummy). (Firm level).

What is the best way to analyze this matter.? Using a regression on country level with the averages of the firm level variables comes at the cost of the amount of observation (29 instead of 30.000).
To be able to analyze this matter on firm level, I have to make the assumption that all cultural scores are exactly the same for each company in the same country, which feels like a strong assumption. However, I would have much more observations (30.000 instead of 29).

What is the best way to analyze this matter?

Thank you in advance and best regards,

[Carlo Lazzaro]

Martijn:
welcome to this forum.
I would go -logistic- or -logit- with standard errors clustered at -country- level.

[Martijn van Gijlswijk]

Thank you for your answer Carlo.
However, isn't clustering on country level problematic since i only have 29 clusters?
I understand from Nichols (2007) and Angrist and Pischke that i should pick a dimension with which i get at least 42 to 50 different clusters.

[Carlo Lazzaro]

Martijn:
I see your concern.
As fa as that topic is concerned, I would point you to the excellent reply #4 provided by Clyde Schechter at: https://www.statalist.org/forums/for...es-in-data-set.
With 29 clusters your're probably somewhere in between 15 and 50.
Another option may consider creating a -newid- that included firms and country:

egen newid=group(country firm)

[Karel Novak]

If I may kindly hijack this thread: After reading the original question, I have immediately thought of using multilevel model (multilevel mixed-effects logistic regression). Is it wrong idea?