Variable ceo_tenure is omitted because of collinearity in an xtreg

[Sanne Bergh]

Dear reader,

I am looking into the moderating effect of CEO tenure on the direct effect between several inter-organizational relationships and managerial attention (with panel data)

When preparing my final dataset, I am executing the following steps.

** To determine the final sample, a few steps have to be undertaken 

**Step 1: drop unncessary observations
drop if missing(firm_name) // drop unnecessary observations, 7,086 were deleted

**Step 2:  Identify time variable & unit identifyier
xtset id_firm_num year_num // When working with panel data, an xtset must be performend to identify a time variable and unit identifyier. 
save dataset_masterthesis_sannevandenbergh, replace 

**Step 3: run a full regression model 
ssc install outreg2

xtreg digital_transformation_keywords c.AC_keywords##c.ceo_tenure c.alliance_count##c.ceo_tenure employees_num firm_performance rnd_intensity organizational_slack ceo_total_compensation lengthtenk i.year_num i.industry_num  // number of observations = 17,940  


gen sample = e(sample)
keep if sample == 1 // create final sample // 7,629 observations were deleted 
drop sample 
save dataset_masterthesis_sannevandenbergh, replace // save final dataset for the data preperation file

However when performing the xtreg it says note: ceo_tenure omitted because of collinearity..
How can I resolve this problem?

Best,
Sanne

[Carlo Lazzaro]

Sanne:
collinearity (and related omission) is the result of your model specifcation.
Therefore, the only way to get rid of it is to change the set of your prdictors in the right-hand side of your regression equation.
What I see is that you have two interactions: are they informative?
Last but by no means least, please share what you typed (as you did) and what Stata gave you back (as you didn't) via CODE delilmiters. Thanks.

[Sanne Bergh]

Hi Carlo,

Thanks for your fast reply.
What do you mean with change the set of predictors in the right-hand side of the regression equation? and what are informative interactions?

[Carlo Lazzaro]

Sanne:
1) I meant plugging different predictors in the right-hand side of your regression equation. They may be, in part, identical to the present ones. You may aslo want to consider adding one predictor at a time and see when the perfect collinearity issue creeps up;
2) I meant: does the interacted terms reach statistical significance? I'm not a p-value fan, but if this is not the case, you can focus on a more parsimonious model.
3) the last point of my previous reply (neglected, so far ) still holds.