New ID var based on combining two ID vars

[Ellen vdBerk]

Hi all,

I am working with panel data, students nested in classes. I want to cluster on class.
Since I am interesting in certain effects measured on the class level, I want to take into account the students who stayed together in the same class in both year 1 and 2.
There are two ID variables, classyear_1 and classyear_2. But the problem is that values in year 1 and year 2 have completely different ranges. See an example of how it looks below. As you can see there have been quite some students who switched class between year 1 and 2. That's okay, but I want to at least have the students that were in the same class for both years have the same values in a new variable.
I think I'll have to do something with foreach, and then get the mimumum/maximum for each classyear_1, but I don't exactly know what I should do.
I hope maybe someone could help me. Thanks in advance!

id	classyear_1	classyear_2
1	1	434
2	1	436
3	1	436
4	1	437
5	1	437
6	2	436
7	2	437
8	2	437
9	6	332
10	6	332
11	7	332
12	8	331
13	8	331
14	8	334

[Nick Cox]

egen group = group(classyear_1 classyear_2), label

puts people who stayed together in the same group.

See also https://journals.sagepub.com/doi/pdf...867X0800700407

[Ellen vdBerk]

Thank you for your quick response Nick.

This works and I can use the grouped variable as a cluster variable. However, for respondents who are only included in 1 year it creates a missing variable, logically.
So for those respondensts, I tried to extract information from either year 1 or two by doing the following:

egen class_full = group(classyear_1 classyear_2), label
replace class_full = classyear_1 if class_full==.

However, the class_full variable now makes up random class_full values for those respondents. So the data looks like this. For id 138 it works fine, but 148 is incorrect:

id	classyear_1	classyear_2	class_full
138	8	331	8 331
148	9	.	3 435

Is there any way to solve this?

[Nick Cox]

That's not to work well, because the point of group() is to create identifiers 1 up without regard to the values of the variables that are input.

The help for egen indicates a missing option which treats missing values directly. So, I would throw out that variable and try again with the missing option also specified.

But note that, other details aside, people missing on one class year will be lumped together with people also missing in the same class year, with naturally no guarantee that their unknown class years were really identical. Or that they were even in the school?