Autori
Dai, Xiaowen Jin, LibinShi, LeiTitolo
Quantile regression in random effects meta-analysis modelPeriodico
Statistical methods & applications : Journal of the Italian Statistical SocietyAnno:
2023 - Volume:
32 - Fascicolo:
2 - Pagina iniziale:
469 - Pagina finale:
492In meta-analysis model, due to the appearance of publication bias or outliers, as well as the small sample size, the normal assumption is usually unreliable. Therefore, the exploration of more robust estimation, such quantile regression (QR) method, is extremely important in meta-analysis area. This paper studies the QR estimation method in random-effects meta-analysis model based on the reformulation by asymmetric Laplace distribution (ALD). The maximum likelihood estimation using Monte Carlo Expectation Maximization algorithm and the Bayesian estimation using Markov chain Monte Carlo (MCMC) algorithm are proposed for computation of the QR estimates. The significance tests of regression coefficients are suggested using likelihood ratio statistics. For MCMC algorithm, a simple and efficient Gibbs sampling algorithm is employed based on a location-scale mixture representation of the ALD, and information criterions are considered for choosing the hyper-parameters. Monte Carlo simulations are conducted to study the finite sample performance of the proposed methodology and analysis of two real data sets are presented for illustrations. Our results show that QR estimation methods perform very well, especially in case of non-normal assumption in meta-regression models. The detailed algorithms and software code are available for easy use in applications.
SICI: 1618-2510(2023)32:2<469:QRIREM>2.0.ZU;2-Y
Esportazione dati in Refworks (solo per utenti abilitati)
Record salvabile in Zotero
Biblioteche ACNP che possiedono il periodico