-----

How do people rank their preferences? That is a well-studied question in behavioral economics, but interestingly there seems to be little work specifically focused on voting for political representation. When it comes to ranking behavior in an electoral context, the first task is descriptive: if we call the full set of rankings a "preference profile," we can start by coming up with ways of summarizing or characterizing profiles. The second task and the main focus of this talk is generative:  just as large language models respond to a prompt by generating language, we'll look at models that take some parameters and generate preference profiles. In this talk, Duchin will review existing models (from the computational social choice literature), introduce some new ones, and discuss efforts to validate against real-world voting. The endgame is democratic design: with descriptive and generative tools in hand, we can subject random preference profiles to various voting rules and see what comes out, giving us new tools for comparing systems of election.

-----