Issue with xtsum output

[Ronald Biefinger]

Hello,

I tried to figure out a problem that came up when I used xtsum yesterday to inspect a new panel data set.

Within variation for var "c_id" should be "0" and it is. But for var "year" the between-variation was not "0" what it should be. There is no value like 2008.5 allowed - only full years. Also my dummy variable for the year 2004 does not look right...

Do you have an explanation for that? Also T-bar looks awkward. I analyse a 8 year period. I already visually inspected the data and checked for any year values that might not be integers, but maybe there is another way to check for those deviations?

Thank you!

Stata 11.2

sort c_id year
xtset c_id year
xtsum c_id year dummy04

Variable         |      Mean   Std. Dev.       Min        Max |    Observations
-----------------+--------------------------------------------+----------------
c_id     overall |  1319.154   1049.926          1       4611 |     N =   64340
         between |             1048.901          1       4611 |     n =    4611
         within  |                    0   1819.384   1819.384 | T-bar = 6.67492
                 |                                            |
year     overall |  2007.994   2.026894       2004       2011 |     N =   64340
         between |             .4408717     2006.5     2008.5 |     n =    4611
         within  |              1.97871   2004.137   2011.851 | T-bar = 6.67492
                 |                                            |
dummy04  overall |  .0438763    .204824          0          1 |     N =   64340
         between |             .0592206          0         .2 |     n =    4611
         within  |             .1954326  -.1561237   .9188763 | T-bar = 6.67492

[Ronald Biefinger]

Is there necessarily a problem with my dataset / year values or did I use the the xtsum command wrong?
Appreciate your help

[Nick Cox]

Clearly it's axiomatic that an identifier is constant within panels. You can expect year to show no variability between if your dataset is balanced, and occasionally otherwise, but variation in year between is diagnostic of an unbalanced panel. 2006.5 is the lowest mean year in your case: you should expect mean year to be something.5 if the number of years is even and all years are present, and sometimes otherwise. This can be seen from an example of a two-year panel for 2013 and 2014 in which case the mean year for any panel with both years present must be 2013.5.

Try experiment in a sandbox:

webuse grunfeld
xtsum
drop in 5/10
xtsum

After dropping a few observations balance is lost and the report for year changes.

[Ronald Biefinger]

Thank you a lot for your answer. Yes, it is an unbalanced panel, so your explanation makes perfect sense.

[Shahla Akram]

please anyone here tell me how i calculate within min and max value in panel data.