Variable mean in panel data

[Beatrice Rotta]

Hello everyone!
I have a panel model - 14 individuals over a time period of 31 years.
I need to perform some simple descriptive statistics: in particular, I need to observe the mean for each variable, specific to each individual (e.g. mean of variable A for individual 1, 2, 3 and so on).

By typing the following command

sum if id == 1

I obtain the following output:

------------------------------------------------------------------------------------------------------------------------
Variable | Obs Mean Std. Dev. Min Max
-------------+---------------------------------------------------------
Tab | 0
id | 31 1 0 1 1
t | 31 16 9.092121 1 31
A | 31 205.4194 149.3472 2 432
B | 31 300.4194 99.29444 126 412
---------------+---------------------------------------------------------
C | 31 139 15.79029 114 165
D | 31 257.4839 18.91185 216 282
E | 31 150.7742 125.7905 7 389
F | 31 111.8387 78.85011 8 236
G | 31 180.4516 146.6683 3 403
---------------+---------------------------------------------------------
H | 31 37.80645 25.08442 4 73
I | 31 33.32258 22.42972 3 78
--------------------------------------------------------------------------------------------------------------------------

The same result is obtained also running the command

bysort id: egen mean_A = mean(A)

However, these results are not the actual mean.
I can determin it by simply observing values in browse table, where, for instance, the mean value of variable B for individual 1 is around 5000.

Is there a different way in order to get accurate results about variable means, specified for single individual?

Thank you in advance!

[Andrew Musau]

I can assure you that Stata's computation of the mean is correct. What are your rules? Do you want missing values to be treated as zero values? That is a restriction, but not necessarily a correct one. See the variable "wanted2" below.

webuse grunfeld, clear
keep company time invest
keep if time<5
keep if company<4
replace invest=. if time<3
bys company: egen wanted1= mean(invest)
bys company: egen wanted2= mean(cond(!missing(invest), invest, 0))

Res.:

. l, sepby(company)

     +---------------------------------------------+
     | company   invest   time   wanted1   wanted2 |
     |---------------------------------------------|
  1. |       1        .      1    334.15   167.075 |
  2. |       1        .      2    334.15   167.075 |
  3. |       1    410.6      3    334.15   167.075 |
  4. |       1    257.7      4    334.15   167.075 |
     |---------------------------------------------|
  5. |       2        .      1     366.1    183.05 |
  6. |       2        .      2     366.1    183.05 |
  7. |       2    469.9      3     366.1    183.05 |
  8. |       2    262.3      4     366.1    183.05 |
     |---------------------------------------------|
  9. |       3        .      1      60.9     30.45 |
 10. |       3        .      2      60.9     30.45 |
 11. |       3     77.2      3      60.9     30.45 |
 12. |       3     44.6      4      60.9     30.45 |
     +---------------------------------------------+

[Beatrice Rotta]

Thank you for your answer and for your clarification about missing values. I didn’t think about it.

My concern is due to the fact that, for example, my A variable is GDP, which is a very big number, with no missing values. However the final result of the mean is 205, as you can see.

I have difficulties in figuring out how to obtain a more representative result.

thank you again!

[Andrew Musau]

Have you browsed through the data? The mean for variable "B" and id=1 is indicated as 300.4194. If you believe this not to be correct, copy and paste the result of the following:

dataex id t B if id==1

[Nick Cox]

Very weird results sometimes arise because data that arrived as string variables were put through encode when destring should have been used.
.