Partial Likelihood Estimation of IRT Models with Censored Lifetime Data: An Application to Mental Disorders in the ESEMeD Surveys

Developmental studies of mental disorders based on epidemiological data are often based on cross-sectional retrospective surveys. Under such designs, observations are right-censored, causing underestimation of lifetime prevalences and correlations, and inducing bias in latent trait models on the obs...

Full description

Saved in:
Bibliographic Details
Published in:Psychometrika 2014-07, Vol.79 (3), p.470-488
Main Authors: Forero, Carlos G., Almansa, Josué, Adroher, Núria D., Vermunt, Jeroen K., Vilagut, Gemma, De Graaf, Ron, Haro, Josep-Maria, Alonso Caballero, Jordi
Format: Article
Language:English
Subjects:
Assessment
Behavioral Science and Psychology
Bias
Censorship
Epidemiologic Research Design
Epidemiology
Estimates
Health services
Hospitals
Humanities
Humans
Item response theory
Law
Likelihood Functions
Mental disorders
Mental Disorders - epidemiology
Mental health
Models, Theoretical
Observation
Parameter estimation
Psychiatry
Psychology
Psychometrics
Psychometrics - methods
Psychopathology
Quantitative psychology
Simulation
Statistical Theory and Methods
Statistics for Social Sciences
Survival analysis
Testing and Evaluation
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Developmental studies of mental disorders based on epidemiological data are often based on cross-sectional retrospective surveys. Under such designs, observations are right-censored, causing underestimation of lifetime prevalences and correlations, and inducing bias in latent trait models on the observations. In this paper we propose a Partial Likelihood (PL) method to estimate unbiased IRT models of lifetime predisposition to develop a certain outcome. A two-step estimation procedure corrects the IRT likelihood of outcome appearance with a function depending on (a) projected outcome frequencies at the end of the risk period, and (b) outcome censoring status at the time of the observation. Simulation results showed that the PL method yielded good recovery of true frequencies and intercepts. Slopes were best estimated when events were sufficiently correlated. When PL is applied to lifetime mental health disorders (assessed in the ESEMeD project surveys), estimated univariate prevalences were, on average, 1.4 times above raw estimates, and 2.06 higher in the case of bivariate prevalences.
ISSN:0033-3123
1860-0980
DOI:10.1007/s11336-013-9400-y