[Maria Kohnen]

Dear Carlo,
Thank you for your reply.
I did check what similar studies used for models. Unfortunately this is a field that has not been researched a lot.
In general:
does it makes sense to use that regression

 xtreg RDlog POST_FINE_DUMMY, fe vce (robust)

with the dummy codes as mentioned to check if there is a stat difference between the R&D spendings in the two periods?

[Carlo Lazzaro]

Maria:
it might be OK, but your regression is at risk of omitted variable bias if some time-variant predictor is omitted.
Check this issue with your supervisor.

[Maria Kohnen]

Hi Carlo, thanks for the reply.

Actually the regression is

 xtreg RDlog POST_FINE_DUMMY i.year, fe vce(robust)

so this is how i hope to cover the time variant aspects.
is that better?
i mean, does the pre-post dummy make sense?

also, I did not quite understand what you meant in your previous post, that these are essentially two very different regressions...

1) I just want to see if there is a sig difference between the two periods
2) i introduce the interactions level of fine and leniency:yes or no

I mean, with all your knowledge of statistics, is this a model that can work/is legitimate?

1)

 xtreg RDlog POST_FINE_DUMMY i.year, fe vce(robust)

2) introducing interactions for leniency yes/no and level of fine s/m/l

 xtreg RDlog POST_FINE_DUMMY##leniency_dummy POST_FINE_DUMMY##i.index_level_of_fine i.year, fe vce (robust)

is this valid?

[Carlo Lazzaro]

Maria:
the first model is simpler to get, whereas the second one is more comprehensive, but leaves room to rising difficulties in disseminating its results.
Which one is better is difficult to say (for me, at any rate). However, I would prefer the model that gives a truer and fairer view of the data generating process.

[Maria Kohnen]

Hi Carlo,

one more question about the xtoverid function.
do I include i.year if i test for it?

xi_xtreg RDlog Post_fine_dummy i.year, re vce (robust)

?
pr leave i.year out fr that test

[Carlo Lazzaro]

Maria:
yes, I would include it.

[Maria Kohnen]

Hi Carlo,
I hope I expressed my question the correct way.

Including the i.year was in regard to the

 xtoverid

funtion to choose between re and fe.

So, is it okay to include i.year like this:

xi: xtreg RDlog Post_fine_dummy i.year, re vce (robust)

then

 xtoverid

?
The test statistic is very lrge (around800...) and the sig 0.0000

[Maria Kohnen]

Dear Carlo,
i just checked autocorrelation for the first time with the

xtserial

command and it was pretty significant with a value around 10!
i only have dummies as predictors. What can Ben done?
i already introduced robust se

is vce (robust) and cluster id the same?
thank you

[Carlo Lazzaro]

Maria:
- if -xtoverid- outcome reaches statistical significance, you should go -fe-;
- under -xtreg-, both robustified and clustered standard errors do the same job, that is dealing with heteroskedasticity and/or autocorrelation.