I wanted to share a paper/project I’ve been working on. The short version is that we’ve used AM Scaling to estimate voter perceptions of candidate ideology over a decade and a half. I put together a website that has information about the measure as I think it will be useful to a lot of people.
I also put together a discussion of how the model is fit in Stan (the technical details). I’ve learned a lot from reading this forum and I am hoping that this might help people like me in the future.
Thanks for sharing this!
Cool work!
A thought on the model: have you considered a hierarchical structure for the individual-level alpha and beta parameters? A partially-pooled model could help regularize the unpooled estimates, which can be noisy, to yield more robust measures of individual perception.
Jørgen Bølstad’s paper (Hierarchical Bayesian Aldrich–McKelvey Scaling | Political Analysis | Cambridge Core) on Hierarchical AM models tackles this directly, showing it improves the stability and accuracy of these individual estimates (his implementation is in the hbamr R package). Interestingly, while his current model is metric, he’s mentioned he’s developing ordinal versions, which aligns with your approach.
One final thought: Bølstad’s paper also emphasizes estimating latent respondent positions from self-placements. Applying this to your dynamic framework could be a powerful extension, allowing for analyses of how voter ideology distributions shift across demographics or geography over time.