Unbalanced panel data analysis

[Liyuwork Dana]

Hi everyone,

I am struggling how to deal with unbalanced panel data. I have an unbalanced panel dataset because some respondents dropped out from the survey (since time series, high probability of getting this issue). I red a book that notes unbalanced panel data has to be checked for whether it is completely randomly or not before we run any analysis.

So my question is that is there any means I can check whether my unbalanced panel dataset is completely at Random (MCAR) or not? Or is there any thing I need to tell STATA that my panel data is unbalanced so that not bias my analysis (whether I proceed random or fixed method or OSL)?

Thanks in advance!

[Carlo Lazzaro]

Liyuwork:
while is recommended to investigate the mechanism underlying the missingness of your data, it is also true that Stata can handle both unbalanced and balanced panel without any problem.
Hence, you can run your panel data regression on the unbalanced panel (base case analysis) and then consider investigating your the missing data mechanism(s) and deal with missing data accordingly (see -mi- entries in Stata .pdf manual and -search mcartest- for an useful user-written programme) (sensitivity analysis).
As an aside, please note that (pooled) OLS rarely outperforms -xt- panel data regression commands (I'm intentionally not specific on this point, as you do not state if your dependent variable is constinuous or else).

[Liyuwork Dana]

Thanks Carlo,

So is this mean we don't have to worry about this and just proceed considering the data is balanced panel data? In other cases there is a default tick box to consider our missed data when data run the analysis rather than trying to impute data, which sometimes might be necessary but can I analyses data with imputing the missed data? The other thing is how I investigate wether missing data is completely at Random (MCAR) or not? I know how to test and interpret p-vales checking MCAR in SPSS but not in STATA, can any one help me on this too, please?. I use STATA 15.1 SE.

Many thanks,
Liyu

[Carlo Lazzaro]

Lyuwork:
- Stata can perform panel data regression on an unbalanced panel, too. It does not mean that Stata converst the unbalanced panel into a balanced one, but that unbalancedness is not an issue that you should be worried about;
- Stata will automatically omits observations with missing values in any of the variables (listwise deletion). You can run the regression on the unbalanced panel and then (if feasible) -mi- the missing values (the pre-requirement of any approach aimed at dealing with missing values is investigating the reason of the missingness);
- type -search mcartest- and read https://www.stata-journal.com/articl...article=st0318.

[Liyuwork Dana]

Thanks Carlo

Can anyone tell me what is the difference between GLS random-effect and ML random effect estimators, and when we choose them? From my experience other than panel data analysis, ML (Maximum likelihood) is to better estimates our coefficients taking missing values into consideration. So, is ML option in the current case (panel data) to take dropout participants into consideration? If this is the case, is there any option we consider those of dropout (resulted unbalanced dataset) using fixed-effect estimator? The other question is that I have some independent variables not collected in all panel period, i.e. collected at time 1 then not collected at T2, and then after collected at T3 and T4. Is there any means I could include those variables in my panel data analysis? Is there any means to impute these kind of missed information?


Appreciate your support.

[Carlo Lazzaro]

Liuwork:
1) from an empirical point of view, -xtreg, mle- mirrors -mixed- results.
I fail to get your point about the preferability for -xtreg,mle- (vs what?) to deal with drop-outs. Be as it may, I would first investigate whether the dop-outs are informative or not. That said, there are estimators for weighting the observed data in order to take account of the drop-outs in panel data (Wooldridge, J. M. 2002. Inverse probability weighted M-estimators for sample selection, attrition, and stratification.Portuguese Economic Journal 1: 117–139); a worked-out example (Stata code) is reported in https://www.stata.com/bookstore/appl...lth-economics/ (pages 286-290).
1) you can include those independent variables if they contribute to give a fair and true view of the data generating process. If they are not available for all the data waves, it simply makes your panel unbalanced.

[joy yakubu]

Hi,
please I have an unbalanced panel data of 47 countries, when I run the xtreg command for the total number of groups in the regression drops to 44. how can i detect the countries that were dropped in the regression? I am using stata 15.1.

Thank you.

[Carlo Lazzaro]

Joy:
see -rowmiss- function available from -egen-.