Scaling
-
Prof. Cees van der Eijk (Methods and Data Institute)
Content
The scaling clinic continues where the measurement clinic ended, but it can, in principle, be taken without having participated in the Measurement clinic.
Scaling is the over-arching term for a number of measurement models that differ from the factor analytic tradition, generally by making fewer or different assumptions than in the case of factor analysis about the level of measurement of the items.
This clinic covers the following topics:
The item-response theory (IRT) framework
IRT is a set of models that relates the probability of particular responses (e.g. answers to survey items, scores on test-items, etc.) to the position of respondents and of items on ‘underlying’ dimensions, which are often referred to as ‘latent traits’, hence these models are also known as latent-trait models.
The general structure of these models is explained and compared to the factor-analytic approach.
Two particular classes of IRT models are subsequently introduced:
a) monotone (also known as cumulative) models, such as the (parametricised) Rasch model and the (non-parametric) Mokken model. The Rasch model is particularly popular in education and other testing contexts where large pools of items are available; the Mokken model is particularly popular in the context of attitude measurement via survey research where the number of relevant items is generally smaller.
b) single-peaked models, particularly relevant in the analysis of indicators that reflect preferences.
IRT-applicability issues
This section discusses the question how to diagnose the applicability of different models for the data at hand, and includes in this a comparison with factor analysis.
Multi-dimensional scaling (MDS)
MDS is a set of approaches for the measurement of latent characteristics of a set of items, based on observations about the similarity or dissimilarity of those items. It results in a ‘map’ of the items in a multi-dimensional space; the coordinates of the items on the dimensions of the space are then the measures of the items on the latent dimensions.
MDS-applicability issues
This section discusses how to diagnose one’s data in order to assess the applicability of different MDS models.
This clinic will be supplemented with online learning materials which can be accessed after the event.
Prerequisites
Introductory understanding of statistics, e.g. the mean, standard deviation and standard error of a variable; correlation and covariance. If you are attending the clinic as part of our Researcher Development Initiative then you automatically fulfil the prerequisites.