fixed effect regression with unbalanced panel

[Stefano Scibilia]

Dear Statalisters,

I would like to do a fixed effect regression using 28 European countries across a 1969-2016 timeframe. my problem is that I have a very unbalanced panel. My independent variable is an index which is computable only for the years shown below while my dependent variable would be available every year.

xtdescribe, pattern(50)

      id:  1, 2, ..., 28                                     n =         28
    year:  1967, 1969, ..., 2016                             T =         44
           Delta(year) = 1 unit
           Span(year)  = 50 periods
           (id*year uniquely identifies each observation)

Distribution of T_i:   min      5%     25%       50%       75%     95%     max
                         2       2       6         8         9      13      27

     Freq.  Percent    Cum. |  Pattern
 ---------------------------+----------------------------------------------------
        1      3.57    3.57 |  ...........................................1..1...
        1      3.57    7.14 |  ...........................................1..1..1
        1      3.57   10.71 |  .......................................1...1..1..1
        1      3.57   14.29 |  .....................................1..1..1......
        1      3.57   17.86 |  .................................1...1..1..1..1...
        1      3.57   21.43 |  .................................1...1..1..11.1111
        1      3.57   25.00 |  ..............................1.1....1..1..1.1....
        1      3.57   28.57 |  ............................1....1...1..1..1..1...
        1      3.57   32.14 |  ............................1.1...................
        1      3.57   35.71 |  .........................1...1.......1..1..1..1...
        1      3.57   39.29 |  .........................1...1.....1.1..1..1..1...
        1      3.57   42.86 |  ........................1..1....1.....1.1.1..1..1.
        1      3.57   46.43 |  ....................1......11.1..1...1..1..1..1...
        1      3.57   50.00 |  ....................1......111...1...1..1..1......
        1      3.57   53.57 |  ....................1....1..1....1...1..1..1..1...
        1      3.57   57.14 |  ....................1...1...1....1...1..1..1..1...
        1      3.57   60.71 |  ...................1.....1..1...1....1..1..1..1..1
        1      3.57   64.29 |  ...................11.1.1.1.1..1.1...1...1.1...1..
        1      3.57   67.86 |  ..................1.....1..1..1..1...1..1..1..1...
        1      3.57   71.43 |  ..................1..1...1..1.1..1................
        1      3.57   75.00 |  ................1...1..1..1.....1....1..1..1..1...
        1      3.57   78.57 |  ...............1.........1.......1.1.1..1..1..1...
        1      3.57   82.14 |  .............1....1....1....1....1...1..1..1..1..1
        1      3.57   85.71 |  ............1......1....1...1....1...1..1..1..1...
        1      3.57   89.29 |  ...........1.....1....1....1.....1....1....1......
        1      3.57   92.86 |  ......1....1..1.11..1.1.1..11..1.1111111111111111.
        1      3.57   96.43 |  ..1....1....1......1....1..11...1....1..1..1..1..1
        1      3.57  100.00 |  1.......1.....1.....1....1..1....1....1...........
 ---------------------------+----------------------------------------------------
       28    100.00         |  X.X...XXX..XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

Is it possible to run the fe regression? How should I deal with the missing data?
Thanks for the help!

[Ahmad Alsuliman]

if a variable is missing, stata will drop this observation.

it is up to you and your research assumptions as to how to deal with missing data. You may want to drop these observations if you have reasons to believe that the distribution is not to be affected by the dropped observations.

you can also:

1. replace the missing values with averages (but that is generally problematic and again you should have your reasoning ready!)

2. you can also impute the missing data (again imputation has assumptions)


I'm not sure I understand the structure of your data and the nature of the variables. but generally, you should able to run fixed effect even if you have missing data!!