About the Conference

Conference on Ideal Point Models
May 1–2, 2015, at MIT


The Conference on Ideal Point Models will be held on May 1–2 at the Massachusetts Institute of Technology in Cambridge, Massachusetts, and is co-hosted by the University of Wisconsin–Madison and the University of Texas at Austin. The conference received funding from the Society for Political Methodology and the National Science Foundation as well as from the Massachusetts Institute of Technology. All reasonable travel costs for participants will be paid by the conference.

The conference is a thematic methodology meeting of the Society for Political Methodology and is intended to ​provide ​a ​forum ​for ​scholars ​to ​discuss ​recent ​advances in ideal point models, ​as ​well ​as ​continuing ​methodological ​challenges, ​in ​the ​development ​of ​IRT ​models ​for political science ​questions. ​The goal of thematic methodology meetings is to provide a smaller and more focused forum for scholarly exchange on a scale reminiscent of the original annual meetings of 25 years ago.


Spatial models have long been fundamental, even “iconic” (Brady 2011) tools of political science research, and are essential to many of the most important questions and theories in political science. New statistical techniques and data sources have enabled researchers to advance the frontiers of spatial modeling on a number of fronts. However, there are also many questions about how the identifying and bridging assumptions used in IRT models affect substantive inferences.

Techniques ​to ​bridge ​spatial ​models ​across ​time, ​institutions, ​and ​data ​types

While ​most ​ideal-point models ​are ​focused ​on ​a ​single ​institution ​or ​group ​of ​voters, ​some ​of ​the ​most ​important ​recent applications ​of ​spatial ​models ​have ​involved ​“bridging” ​the ​preferences ​of ​political ​actors ​in ​different contexts ​onto ​a ​common ​scale. ​Examples ​include ​producing ​estimates ​that ​are ​comparable ​across ​time, ​across institutions, ​or ​between ​elites ​and ​the ​mass ​public. ​Recently, ​however, ​some ​scholars ​have ​expressed concern ​that ​our ​ability ​to ​estimate ​models ​that ​bridge ​across ​time ​or ​context ​has ​outstripped ​our understanding ​of ​the ​assumptions ​that ​undergird ​these ​models. ​Given ​the ​importance ​of ​the ​research conclusions ​that ​hinge ​on ​these ​assumptions, ​it ​is ​crucial ​that ​scholars ​develop ​a ​shared ​set ​of standards ​for ​describing ​and ​evaluating ​assumptions ​for ​bridging ​spatial ​models.

Hierarchical ​models

Although ​early ​work ​on ​ideal-point ​estimation ​treated ​both ​bill ​and ​ideal-point parameters ​as ​non-random, ​subsequent ​work ​suggested ​a ​random ​effects ​approach ​to ​ideal-point ​estimation. ​A more ​recent ​line ​of ​research ​has ​sought ​to ​augment ​the ​standard ​ideal ​point ​model ​with ​additional information. ​For ​instance, ​various ​models ​have ​added ​hierarchies ​formed ​from ​partisanship, ​issue ​area, ​and topic ​models ​of ​bill ​text.

Nonparametric ​and ​flexible ​parametric ​estimation

Most ​approaches ​to ​ideal-point ​estimation ​assume ​a particular ​functional ​form ​of ​the ​utility ​curves ​and ​a ​particular ​distribution ​for ​the ​error ​terms. ​In recent ​years, ​some ​scholars ​have ​moved ​toward ​nonparametric ​approaches ​to ​ideal-point ​estimation. ​Another literature ​has ​retained ​a ​parametric ​framework ​but ​attempted ​to ​relax ​the ​parametric ​assumptions ​or distinguish ​between ​common ​choices. ​Both ​of ​these ​literatures ​have ​important ​implications ​for ​work ​on spatial ​models ​in ​political ​science.

Dimensionality ​of ​IRT ​Models

Dimensionality ​of ​IRT ​Models: ​There ​is ​no ​consensus ​in ​the ​literature ​on ​the ​best ​way ​to ​estimate the ​number ​of ​dimensions ​in ​a ​spatial ​model. ​The ​most ​common ​approach ​is ​to ​assess ​whether ​adding dimensions ​increases ​the ​percentage ​of ​items ​correctly ​classified ​by ​the ​model. ​But ​recent ​work ​in ​the congressional ​literature ​has ​shown ​that ​polarization ​of ​preferences ​can ​produce ​misleading ​evidence ​of low-dimensionality. ​We ​hope ​that ​this ​conference ​will ​help ​scholars ​discuss ​new ​ways ​to ​assess ​the dimensionality ​of ​spatial ​models.