Nonparametric Estimation of a Latent Trait Density.
Abstract
A method for estimating the density of the latent trait (or ability) in an item response model is given based on the logspline density estimator of Stone and Koo. In the implementation presented here, families of densities whose logarithms are quadratic splines are used to estimate the unknown latent trait density. The number of knots in the spline is variable permitting arbitrary densities to be well approximated from the logspline family. Because the family is exponential and contains all normal distributions, the likelihood ratio test can be used to test for normality. An ad hoc method is proposed for choosing the number of knots, and the method is illustrated with two simulated data sets. (KR)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 31, 1988
- Accession Number
- ADA198128
Entities
People
- Paul L. Speckman