I had some trouble finding the code[1] and the paper[2]. The Stan code itself is [3]
I think this is quite a strong model - if I was looking to work on it I'd like to see it incorporate some economic data (and probably previous election poll trends) as a way of attempting to improve the early-election-cycle forecasts.
The hardest part of an open source poll would be anonymization. Demographic information in some areas could be enough to come pretty close to uniquely identifying an individual.
The rest of it: Likely Voter Models, Demographic Weighting, etc would be pretty easy to have open source and reproducible once you've solved the issues.
Yes, although is reproducibility even possible in polls? People's opinions change on a whim. In fact people's opinions can change simply because they were asked a question so the answer might be different if asked the question again,
I guess you could conduct the same poll with multiple sets of people and see how the results compare, but in a way that's what poll aggregators do.
ptrs: Ben Adida worked on a cryptography-based open-audit voting system called HELIOS in 2008 -- could be used as part of anonymous yet auditable open-source polling / voting system:
I think this is quite a strong model - if I was looking to work on it I'd like to see it incorporate some economic data (and probably previous election poll trends) as a way of attempting to improve the early-election-cycle forecasts.
[1] https://github.com/pkremp/polls
[2] http://votamatic.org/wp-content/uploads/2013/07/Linzer-JASA1...
[3] https://github.com/pkremp/polls/blob/master/state%20and%20na...