Panel data or Pooled Cross Section?

[Lia Rodriguez]

Hello. I am not sure if the data I am working on are pooled cross section or panel data.The database includes 60 national elections of 14 countries in which non-residents were allowed to vote. The elections were held among 1993 and 2016. Variables include aggregate data such as electoral turn out of citizens living abroad, for each election in each country. My ID variable is country and my time variable is year. In some countries emigrants could vote once during that period while in other countries they could vote more than ten times. For most of the countries my database includes every election held were emigrants could vote, while for others I have missing data. Understanding if I have a panel or pooled cross sections is important to see if I have to deal or not with serial correlation residuals. Can anyone help me?

[Carlo Lazzaro]

Lia:
I would consider your data as a panel dataset.

[Lia Rodriguez]

Thank you very much Carlo!

[Ana Duran]

I have a somewhat similar question. I am trying to study the effects of a policy on diet using a repeated survey (3 years) of students within schools and within cities. Cities are repeated across time.

Using the following strata specification:

svyset [pw=pw], strata (strata) psu(psu)

where strata = cities and psu = schools

I am wondering if a pooled OLS adjusted for year would be adequate to estimate what I need. Here is the stata command I am using:

svy: reg fruit_5 i.meal i.year sex i.age_3_cat i.cat_edu

I would very much appreciate your help.

Best regards,

Ana

[Carlo Lazzaro]

Ana:
my previous reply #2 still holds.
That said, another option might be -mixed- (due to nesting).
Eventually, OLS sounds like a poor approach in your case; anyway, you should consider -cluster()-ed standard errors.
For the future, please start a new thread. Thanks.

[Ana Duran]

Thank you for your reply. I started a new thread as recommended, thank you for your help.

However mixed or xt... may not be appropriate in this case because cases are not followed across time - these are repeated cross-sectional surveys... they are not the same students... What do you think?

And I do not think I need to use -cluster()-ed standard errors because using the survey mode, the standard errors are already clustered at the strata I informed. In fact, Stata will not allow -cluster()-ed standard errors when using svy.

[Carlo Lazzaro]

Ana:
thanks for clarifications.
Most about the best methodological approach depends on what is well rooted in your research field.