Fitting mixed-effects models when data are left truncated |
| |
Authors: | Jostein Paulsen Astrid Lunde Hans Julius Skaug |
| |
Institution: | aDepartment of Mathematics, University of Bergen, Johs. Brunsgt. 12, 5008 Bergen, Norway;bDepartment of Public Health and Primary Health Care, Epidemiology and Medical Statistics, University of Bergen, Kalfarveien 31, 5018 Bergen, Norway |
| |
Abstract: | Damage sizes, i.e. all damages occurring to a policy and not only those that are reported to an insurance company, are modelled as a linear mixed model. Only those damages that are larger than their deductibles are reported to the company, and this fact should be taken into account when analyzing such data. In statistical terms, the problem is to make inference in a linear mixed model with left truncated data. Estimation methods based on a Monte Carlo simulation of the likelihood are proposed, and extensive simulations to evaluate the quality of the methods are reported. The proposed methods are then used to analyze claimsizes for some marine insurance data, where shipowners represent random effects and technical data about the ships represent fixed effects. |
| |
Keywords: | Deductibles Random effects Likelihood estimation Monte Carlo techniques Prediction |
本文献已被 ScienceDirect 等数据库收录! |
|