regression interpretation

[Giorgio Carlini]

Hello, I've a panel model at country level (I used xtivreg) . The dependent variable is the number of patent per inhabitants and my regressors includes social capital index and other predictors. Social capital values are positive is some countries and negative in others. My estimated coefficient on social capital is positive, so I guess that on average higher social capital should translate into higher patent. Am I right?
Is this also true in those countries where social capital is negative? In other words, how to interpret the impact on patents?
Should I need to add use an interaction term between social capital and country code for this purpose?
Thanks

[Carlo Lazzaro]

Giorgio:
how could ever listers help you out if you do not:
1) expain why you switched to instrumental variable panel data regression;
2) share what you typed and what Stata gave you back (as per FAQ).
Thanks.

[Giorgio Carlini]

Hello Carlo,
I apologize for not being clear. I switched to IV as one of my dep var (migration flows) might be endogeneous. The time period is 1995-2012 and the unit of analysis is the 20 italian regions (cod_reg). So my code is:
xi: xtivreg2 patents (immigrants = IV_immigrants) social_capital firm_size R&D i.year, fe first robust
As I said, I'm interested here on the interpretation of the social capital coefficient, which is positive, as social capital values are positive in the North and negative in the South.
1- Does this mean that higher social capital will raise patents in both the North and the South, or as i guess only on average?
2- Or to see the different impact of social capital on patents between the northern and southern regions should I use an an interaction term such as c.social_capital##i.cod_reg?
Thanks

[George Taylor]

A positive coefficient on social capital implies that higher social capital is associated with more patents, though it is not immediately clear that this relationship is causal.

You should use an interaction term if you believe that there are characteristics of the northern countries that would make the effect of social capital different in these countries to the effect of social capital in the southern countries - c.social_capital##c.North, where North is a dummy variable.

Are you interested in the coefficient on immigrants? If not, you may not need to use IV...

[Giorgio Carlini]

Thanks for the reply.
yes I'm interested in the coefficient of immigrants but I want to investigate the impact of social capital too.
When I type
xi: xtivreg2 patents (immigrants = IV_immigrants) c.social_capital##c.North firm_size R&D i.year, fe first robust
I get the following message :
c.social_capital##c: operator invalid
what is the problem?

[Carlo Lazzaro]

Giorgio:
please double-check whether the community-contributed module -xtivreg2- supports -fvvarlist- notation.
If not, you should create the interaction by hand

gen interaction=social_capital*North

with your code changing as follows:

xtivreg2 patents (immigrants = IV_immigrants) social_capital North interaction firm_size R&D i.year, fe first robust

[Giorgio Carlini]

Hello Carlo,
I changed the code as you suggested, however in the output the var North is omitted. I guess because I'm using fe. Am I right?
Giorgio

[Carlo Lazzaro]

Giorgio:
if, as it would be reasonable to expect, -North- is a time-invariant predictor, the -fe- estimator wipes it out due to demeaning.

[Giorgio Carlini]

Yes unfortunately. Thus, how to find an answer to my initial question, i.e., asses the social capital impact on patents between north and south?

[Carlo Lazzaro]

Giorgio;
the -fe e estimator is not the right tool to anwer your question.