Missing handling for pooled cross-sectional data (?)

[Chiara De Siena]

Hi all,

I am using pooled cross-sectional data for a period of 14 years (it is not a panel!). I will perform multinomial and logistic regressions, but I do not know how to manage missing values with pooled cross sectional data .
I usually do a listwise deletion, but I think it does not fit for such dataset. Do you have any suggestions?

Thank you all in advance,
Have a nice day!

Chiara

[Carlo Lazzaro]

Chiara:
listwise deletion may work in a limited set of situations (see https://statisticalhorizons.com/list...n-its-not-evil).
That said, the usual first step to take when dealing with missing values is a diagnostic one: is their missingness mechanism ignorable or not?

[Chiara De Siena]

Honestly, I have no idea. I found that "This data set includes a high amount of missings which were imputed with model based imputation methods. Since a high amount of missings are not MCAR one has to think about which variables should be included for imputation", but I don't understand if this is true only for a country (- Austria, which I don't have in my dataset) or for all.

What can I do to know if I have MCAR, MAR, MNAR?

[Carlo Lazzaro]

Chiara:
thanks for clarifying.
If missing data were imputed, you should have now 10 or 20 complete datasets in addition to the original one.
As far as the differences among MCAR, MAR, MNAR are concerned, see -mi- entry in Stata .pdf manual and relalated references.
See also the community-contributed command -mcartest- and related paper.