Social relations model or round robin data in Stata

[Chris Martin]

I'm going to have some round robin data from teams that I need to analyze. This means within each team, each member will rate every other member. Thus member A will rate B, C, and D. Member B will rate A, C and D, and so on. In psychology, these are usually called round-robin data and analyzed using the "social relations model" but there may be other terms used in econometrics and other fields. One thing that I'll need to do is convert the data from wide to long. And then I'll need to analyze it using the social relations model. I know there's a package in R called TripleR that does this, so I could use Rcall to run it in Stata. That may turn out to the most efficient way. However, I'm curious about whether there's a way to do this without calling R.

This is what round robin data look like in wide format when there's no missing data:

## Rater A B C D E
## A NA 3 1 0 5
## B 2 NA 5 4 1
## C 4 1 NA 6 4
## D 0 1 0 NA 4
## E 2 2 5 3 NA

'

I looked around and wasn't able to find a Stata package that specifically does this, which is why I'm posting here.

[Daniel Schaefer]

I don't know that there is a build-in command for this, but the social relations model is an SEM model, so one could implement the model using Stata's -sem- command.

Wow, the wide to long procedure is (as you say) equivalent to transforming your adjacency matrix to an edge list. Neat!

[Chris Martin]

That's a good point. I should also mention that I'm using Qualtrics to collect the data. If anyone has advice on naming variables in Qualtrics to make analysis easier in this situation, let me know.

[Mike Lacy]

I could imagine something like the Bradley-Terry model used to analyze the power of sports teams being relevant. That's designed for binary "ratings" (win/lose), but perhaps there is an extension to ordinal ratings. I'd be interested if you track something down. Search for "Bradley" at https://stats.idre.ucla.edu/stata/ex...matched-pairs/.