Censored Probit Estimation with Correlation near the Boundary: A Useful Reparameteriztion

The conventional computational algorithms for full information maximum likelihood (FIML) estimation of the censored probit model (see Farber, 1983), will sometimes fail to converge when the estimated value of the correlation coefficient (ñ) approaches ±1; even when the true value of ñ is not at a boundary. We show that a simple reparametrization of the censored probit model may afford straightforward Newton-Raphson convergence to the true FIML estimate for cases in which likelihood maximization under the conventional censored probit parameterization would have failed. Moreover, our method avoids the computational and inferential complications that arise in alternative methods that, based on a specified criterion, suggest fixing the estimated value of ñ at -1 or +1. For the purpose of illustration the method is used to estimate the determinants of elderly parents’ receipt of informal care from their children.

Issue Date:
Publication Type:
Journal Article
PURL Identifier:
Published in:
Review of Applied Economics, Volume 02, Number 1
Page range:
Total Pages:
JEL Codes:
I12; C24; C63

 Record created 2017-04-01, last modified 2017-08-25

Download fulltext

Rate this document:

Rate this document:
(Not yet reviewed)