Balanced vs. Unbalanced Panel Data for pooled time series cross section

[Miena Haider]

Dear Stata Community,

I need to decide whether I want to use an unbalanced or a balanced panel data set for my analysis.

I have a pooled data set.

Balancing the dataset (one observation wave per ID) gives me better estimates while the reverse holds for the unbalanced data set.

Can you guys suggest any literature that can help me with the decision?

Thank you,
Miena

[Carlo Lazzaro]

Miena:
I would go unbalanced (by the way, Stata can handle both balanced and unbalanced panel datasets with no problems).
Conversely, any approach aimed at discarding missing data and creating a balanced panle from an unbalanced one, ends up making-up your original sample and biasing your analysis (especially if the missingness is informative).