quasi-panel made out of cross-sectional data

[Kasia Kiki]

Hi all,
Does anybody know how to create a quasi-panel data out of the cross-sectional data. I have a few surveys for a few years, nevertheless the individuals are not the same people, therefore I can not just merge the data. I can create a quasi-panel data, aggregating data at the regional level. I would like to get some hint how to perform it ( my data is a few separate surveys from about 6 years with same variables but different individuals).
Thanks!,

[Jared Greathouse]

Hey Kasia Kiki . What do you mean by quasi-panel? i've never heard this term before. Also, can you show an example of your current data so I can have something to work with?

[Fei Wang]

Kasia, what you described is called "synthetic panel" in my mind. First, you may use -append- to combine different datasets (make sure the same variables have the same names across datasets.), then you may use -collapse- to generate the synthetic panel from individual data. Some example code as below.

* Combine datasets
    use data1, clear
    append using  data2 data3 data4...

* From individual pooled cross-sectional data to synthetic panel
    collapse (mean) varlist (rawsum) weight [pw=weight], by(regionid year)

In the collapse part, "varlist" means the list of variables that you'd like to aggregate at the region level, "weight" means the original individual sampling weights -- if individuals are equally weighted, then you may generate weight = 1 before collapsing. In the synthetic panel, a weight variable is necessary to reflect the total individual weights within a synthetic group.

[Kasia Kiki]

Thank you a lot! it worked, nevertheless, I do not know if the "mean" approach is correct, because my dependant variable is a categorical one. So I am not sure if it makes sense with that. I think the other approach is needed since while collapsing on a geohraphical level, dependant variable took values of 0.54, 0.45 etc, which doesn't make sense.

[Fei Wang]

Thank you a lot! it worked, nevertheless, I do not know if the "mean" approach is correct, because my dependant variable is a categorical one. So I am not sure if it makes sense with that.

Right, averaging categorical variables may not make sense (except for happiness indicators, etc.). Then I would suggest that, in the original data, you generate dummies for the categorical variable, like

tab var, by(var)

While collapsing, calculate the mean of each dummy, and you will get the fraction of individuals in a specific category within a region-year -- That's something meaningful for further analysis.