Dummy variables with missing values
Posted: Tue Mar 22, 2011 6:46 am
Hi,
In our panel data we have a dummy variable that has a lot of missing values (approximately 60% of the values missing, the other 40% have the value 0 or 1). It's no problem to import these data (we import them from excel) but once they are imported, EViews seems to indicate that these values are not-numeric (with an 'abc'-logo before the variable).
I tried to import these data from excel and indicating that they were numbers (not characters) but then all values are displayed as NA.
When trying to do a regression with this dummy it says 'insufficient observations'.
We have another dummy variable in the model which has no missing values and with no problems when doing a regression. Is our problem that our dummy has missing values and that it's not allowed? This dummy is a dummy for management capabilities and is the result of firms' sales/#employees being larger (value 1) or smaller (0) than the average. But, for a lot of firms there is no information on the number of employees so it's not possible to give a value.
If this is our problem, is there a way to circumvent it?
Thanks in advance,
Andreas Standaert
In our panel data we have a dummy variable that has a lot of missing values (approximately 60% of the values missing, the other 40% have the value 0 or 1). It's no problem to import these data (we import them from excel) but once they are imported, EViews seems to indicate that these values are not-numeric (with an 'abc'-logo before the variable).
I tried to import these data from excel and indicating that they were numbers (not characters) but then all values are displayed as NA.
When trying to do a regression with this dummy it says 'insufficient observations'.
We have another dummy variable in the model which has no missing values and with no problems when doing a regression. Is our problem that our dummy has missing values and that it's not allowed? This dummy is a dummy for management capabilities and is the result of firms' sales/#employees being larger (value 1) or smaller (0) than the average. But, for a lot of firms there is no information on the number of employees so it's not possible to give a value.
If this is our problem, is there a way to circumvent it?
Thanks in advance,
Andreas Standaert