Dear all,
I am working on evaluating the effectiveness of two university outreach programmes. The programmes are not randomly assigned to students/ schools so the evaluation designs involve running a difference in differences model on a pre-matched sample. What I want to do is to identify appropriate/ comparable control groups of students/ schools and trim the sample so that I run the difference in differences estimation only on the matched sample. Specifically, I want:
(a) For evaluation 1, to create a control group of students by matching treated students with similar untreated students.
(b) For evaluation 2, to create a control group of schools by matching treated schools with similar untreated schools.
In my previous attempt I performed a nearest neighbour matching and adjusted the weights. The sample was reduced to what I understand is "the matched" sample. I used the code below:
I'm not sure if this is the correct approach to follow for what I aim to do. I am thinking that I shouldn't necessarily have to include the outcome variable (univ_apply) at the matching stage. Essentially, I want to perform the matching so that I can obtain balance in the observable characteristics of the students/ schools in the treated and control schools and not necessarily to identify the "nearest neighbour" of each individual student or each individual school.
Any advice on how I could perform this difference in differences estimation on a pre-matched sample in Stata would be greatly appreciated.
Many thanks for your help, it is greatly appreciated!
I am working on evaluating the effectiveness of two university outreach programmes. The programmes are not randomly assigned to students/ schools so the evaluation designs involve running a difference in differences model on a pre-matched sample. What I want to do is to identify appropriate/ comparable control groups of students/ schools and trim the sample so that I run the difference in differences estimation only on the matched sample. Specifically, I want:
(a) For evaluation 1, to create a control group of students by matching treated students with similar untreated students.
(b) For evaluation 2, to create a control group of schools by matching treated schools with similar untreated schools.
In my previous attempt I performed a nearest neighbour matching and adjusted the weights. The sample was reduced to what I understand is "the matched" sample. I used the code below:
Code:
qui psmatch2 treatment $student_characteristics, out(univ_apply) common pstest $student_characteristics, both reg univ_apply i.programme_on##i.treatment $student_characteristics i.year i.yearofstudy i.ssid [aweight=_weight], vce(cluster ssid)
Any advice on how I could perform this difference in differences estimation on a pre-matched sample in Stata would be greatly appreciated.
Many thanks for your help, it is greatly appreciated!
