Unbalanced panel data - missing values problem

[SGarbrecht]

Dear Sir or Madam,

I have facing difficulties with EViews when running a panel regression. I have unbalanced panel data (6 independent variables, 9 control variables). For some of my control variables, there are a few values missing. Moreover, one of my control variable only covers 4 out of my 7 sample years. Unfortunately, EViews excludes all observations for which a have missing value, resulting in a small sample in my regression. Do you know how I can get EViews to include these observations?

Thanks for your help,
S.

[startz]

What value would you like EViews to use for the variable that is missing?

[SGarbrecht]

I would like EViews to ignore the missing value, i.e. treat it as if it was not there.

[EViews Gareth]

That's what it is doing. If the value is missing, the observation is not included (I.e. It isn't there)

[SGarbrecht]

But why is the entire observation excluded if only one single value is missing over the entire sample period out of every independent variable I have?

[EViews Gareth]

Because that's the way the math works. You can't NA out a single value without NAing out the entire observation (row).

[startz]

What EViews does is called "listwise deletion." It's the standard approach to missing data.

Think of it this way. Suppose you are regressing Y on X and Z and you have 2 million observations...but the first half are missing the X observations and the second half are missing the Z observations. You don't have any observations with which to take account of the correlation between X and Z.

There are imperfect ways of adjusting for missing data. These include imputing missing values and using dummy variables to control for missings. None of these are perfect and none of these are built into EViews.

[EViews Gareth]

Although imputting the missing prior to estimation is built in.

[SGarbrecht]

Many thanks!