Boundless but Bundled: Modelling Quasi-infinite Dimensions in Ideological Space

Ideological summary scales, derived from policy position items, are prevalent in political psychologyand behavioral research. However, past scholarly practice shows little to no consensus as to howmany such scales (i.e., ideological dimensions) researchers should consider in order to adequatelycapture the main political dividing lines among mass publics. Our comprehensive literature reviewsuggests — and extensive statistical simulations using ANES and CCES data confirm — that theoptimal number of latent ideological dimensions grows indefinitely as researchers include additionalissue position items in their models. Instead of increasing measurement precision, additional issueposition questions thus increase uncertainty about what underlying construct they are supposed tomeasure in the first place. At the same time, nearly all latent ideological factors detectable withinpolicy position data are sizably and positively correlated with one another, suggesting that ideologicalsub-dimensions ultimately stem from a more abstract, unifying parent dimension. We propose aBayesian hierarchical latent variable modeling framework which seeks to reconcile the boundless, yetcorrelated dimensional nature of mass ideology. The proposed model estimates ideology as a higherlevelexpression of correlated, lower-level building blocks, allowing researchers to simultaneouslyevaluate whether external predictors, such as income or gender are consistently related to specificideological sub-dimensions (e.g., economic or socio-cultural ideology) or, instead, a generalized,uni-dimensional representation thereof. Our results underscore the potential of this approach, offeringinsights into the unique characteristics of different ideological sub-factors and their overarching parentdimension.