SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Imprint / Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University

Contributions: Abstract

Factor analysis of non-normal variables

Kees van Montfort
Free University Amsterdam
The Netherlands

This presentation deals with a factor analysis model. Usually one assumes that the model has normally distributed variables. If the variables are not normally distributed, the well-known estimators using only variances and covariances of the observed variables can generate estimates of the factor loadings which are not correct (i.e. biased). This presentation uses characteristic functions to derive an approximately consistent BGLS-estimator of the factor loadings in a model with both normal and non-normal variables.

An example with generated data and an empirical example will evaluate the estimator. Furthermore, the estimator will be compared with well-known estimators using only second order moments, consistent estimators using third order moments and estimators using normalizing transformations.

It turns out that our method gives good estimates of the unknown model parameters for sample sizes larger than 50. Besides, our estimation method does not have the disadvantage of the abovementioned estimation methods.