Stata says the panel is strongly balanced but there are many observations missing

[Jamie Holmes]

The xtdescribe command tells me the panel is strongly balanced, however, many countries in my sample have missing data for entire period for some variables. How can a panel be balanced if observations are missing?

[Joe Canner]

Here is the definition of panel balance from the xtset documentation:

The terms balanced and unbalanced are often used to describe whether a panel dataset is missing
some observations. If a dataset does not contain a time variable, then panels are considered
balanced if each panel contains the same number of observations; otherwise, the panels are unbalanced.

When the dataset contains a time variable, panels are said to be strongly balanced if each panel contains the
same time points, weakly balanced if each panel contains the same number of observations but not the same
time points, and unbalanced otherwise.

My reading of this is that you don't need to have complete data for every variable to have a strongly balanced data set. However, you should provide the complete results from xtdescribe so that we can comment more intelligently.

[Sebastian Kripfganz]

xtdescribe (and xtset) analyse the data structure based on the panel and time identifiers only. As long as you have one row in your data set for each panel-time identifier combination (without missing values in these identifier variables), than those commands will tell you set your data set is strongly balanced (irrespective of any missing values in your remaining variables).

[Jamie Holmes]

summarize

Variable | Obs Mean Std. Dev. Min Max
-------------+--------------------------------------------------------
country | 6256 68.5 39.2619 1 136
year | 6256 1985.5 13.27698 1963 2008
incomeineq~y | 3732 42.10842 7.101112 20.57831 59.95708
pppconvert~g | 5481 6469.255 9427.672 49.07564 121189.6
realintere~s | 2976 6.688341 25.7861 -97.81207 789.799
-------------+--------------------------------------------------------
inflationg~r | 4903 36.66549 347.9358 -30.18327 13611.63
externalde~i | 2303 63.14364 84.38922 .1 2687.7
grosscentr~p | 2303 53.81016 59.59731 2.3 1209.3
humancapit~o | 3877 61.73231 33.83302 .18163 162.3487
tradeopenn~s | 1440 7.549694 9.867625 .47 254.58
-------------+--------------------------------------------------------
v11 | 0

. xtset country year
panel variable: country (strongly balanced)
time variable: year, 1963 to 2008
delta: 1 unit

. xt describe
unrecognized command: xt
r(199);

. xtdescribe

country: 1, 2, ..., 136 n = 136
year: 1963, 1964, ..., 2008 T = 46
Delta(year) = 1 unit
Span(year) = 46 periods
(country*year uniquely identifies each observation)

Distribution of T_i: min 5% 25% 50% 75% 95% max
46 46 46 46 46 46 46

Freq. Percent Cum. | Pattern
---------------------------+------------------------------------------------
136 100.00 100.00 | 1111111111111111111111111111111111111111111111
---------------------------+------------------------------------------------
136 100.00 | XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

[Jamie Holmes]

OK so how do I analyse the balancedness/unbalancedness of the data or from my observations can I assume that it is unbalanced?, I want to know whether to use ANOVA/ML methods to estimate the variance components. Thanks...

[Roberto Ferrer]

I follow Joe Canner's citation with some examples:

* load data
webuse pig, clear
keep if inrange(id, 1, 3)
xtset id week

* strongly balanced
list, sepby(id)
xtset

* still strongly balanced
replace weight = . in 1
list, sepby(id)
xtset

* unbalanced
drop if missing(weight)
list, sepby(id)
xtset

* load data (again)
webuse pig, clear
keep if inrange(id, 1, 3)
xtset id week

* weekly balanced
replace week = week + 1 if id == 1
list, sepby(id)
xtset