Title: Probabilistic feature models with individual differences in feature selection
Authors: Meulders, Michel
Vermunt, Jeroen
Van Mechelen, Iven
Issue Date: Sep-2014
Publisher: KU Leuven@campus Brussel
Series Title: Research Papers FEB@campus Brussel
Abstract: Probabilistic feature models (PFMs) can be used to explain judgements of persons about
binary object-attribute associations on the basis of latent features. More specifically, PFMs
assume that persons classify both objects and attributes in terms of binary latent features
and that the observed judgement is derived as a non-compensatory (e.g. disjunctive or
conjunctive) mapping of the object- and attribute classifications. In this paper we develop
multilevel latent class extensions of the PFM that allow to model heterogeneity in the
object-attribute association probabilities across persons by assuming that persons select
each of the latent features with a class-specific probability when making object-attribute
judgements. In addition, statistical dependencies between object-attribute associations
with a common element are modelled by assuming that a person relies on the same object
classifications for all attribute judgements with regard to that object (or, alternatively, on
the same attribute classifications for all object judgements with regard to that attribute).
Compared to existing PFM extensions, the model proposed in this paper has several
advantages. First, it allows the user to independently specify the number of features and
the number of latent person classes, leading to a more flexible modelling. Second, unlike
models with class-specific object- or attribute parameters the models presented in this
paper use a small set of parameters to model heterogeneity, leading to more stable
parameter estimates and models that are easier to interpret. As an illustration, the models
are used to analyze data on hostile behavior and psychiatric diagnosis.
Publication status: published
KU Leuven publication type: IR
Appears in Collections:Research Centre for Quantitative Business Processes, Campus Brussels (-)
Faculty of Economics and Business (FEB) - miscellaneous
Faculty of Social Sciences - miscellaneous
Quantitative Psychology and Individual Differences

Files in This Item:
File Description Status SizeFormat
14HRP02.pdf.pdfHUB Research Paper 2014/02 Published 338KbAdobe PDFView/Open


All items in Lirias are protected by copyright, with all rights reserved.