Dear Statalister,
This if my first post so please forgive any etiquette mistakes...
I'm trying to figure what matching method to use for my study.
I'm using Stata/IC 14.0, and the user-written program psmatch2
I have a large dataset (68 000 treated, 1 500 000 untreated). For the characteristics I'm controlling for, there is not much bias (age, gender, geographical distribution, month...).
I would prefer matching without replacement (easier to explain the procedure and calculations to an average audience). When I do that, my untreated group has the exact same characteristics as my treated group, which is nice.
But I still feel it's a shame to loose the information from all these other untreated observations.
I understand psmatch2 cannot perform matching without replacement with multiples neighbors.
Would you rather oversample the untreated group with replacement, or perform a 1:1 matching without replacement ?
Thanks in advance,
Léon-Paul Schwab
This if my first post so please forgive any etiquette mistakes...
I'm trying to figure what matching method to use for my study.
I'm using Stata/IC 14.0, and the user-written program psmatch2
I have a large dataset (68 000 treated, 1 500 000 untreated). For the characteristics I'm controlling for, there is not much bias (age, gender, geographical distribution, month...).
I would prefer matching without replacement (easier to explain the procedure and calculations to an average audience). When I do that, my untreated group has the exact same characteristics as my treated group, which is nice.
But I still feel it's a shame to loose the information from all these other untreated observations.
I understand psmatch2 cannot perform matching without replacement with multiples neighbors.
Would you rather oversample the untreated group with replacement, or perform a 1:1 matching without replacement ?
Thanks in advance,
Léon-Paul Schwab
Comment