• Adaptive-Lasso for Cox's proportional hazards model (based on penalized partial likelihood)

The computation codes, written in the R language, conduct variable selection and model fitting simultaneously in the proportional hazards model (Cox 1972). The adaptive-lasso estimate is obtained using the penalized partial likelihood method with weighted L-1 penalty. The algorithm iteratively solves least square problems with weighted L-1 penalty via a modified shooting algorithm of Fu (1998). The codes were developed by Wenbin Lu and Hao Helen Zhang. The original paper is published in Biometrika (2007, 94, 1-13).

Download the R code here for a simulation study (without tuning). PBC liver data example: R code (with tuning using GCV).

• Adaptive-Lasso for proportional odds model (based on penalized marginal likelihood)

The computation codes, written in the R language, conduct variable selection and model fitting simultaneously in the proportional odds model (Bennett 1983). The adaptive-lasso estimate is obtained using the penalized marginal likelihood method with weighted L-1 penalty. The algorithm iteratively solves least square problems with weighted L-1 penalty via a modified shooting algorithm of Fu (1998). The codes were developed by Wenbin Lu and Hao Helen Zhang. The original paper is published in Statistics in Medicine (2007, 26, 3771-3781).

Download the R code here for a simulation study (without tuning).

• The marginal likelihood based boosting algorithm for nonlinear transformation models

The computation codes, written in the R language, fit the nonlinear transformation models using a smoothing spline based boosting learning algorithm (e.g. Buhlmann and Yu, 2003; Li and Luan, 2005). We first compute the log marginal likelihood function and its gradient using importance sampling. Then we use the negative gradients as responses and fit the model using component-wise cubic smoothing spline as the base learner. The codes were developed by Wenbin Lu and Lexin Li. The original paper is published in Biostatistics (2008, 9, 658-667).

Download the R code here.

• Efficient kernel estimation for accelerated failure time cure model

The computation codes, written in the R language, fit the accelerated failure time cure model via kernel-based nonparametric maximum likelihood estimation. An EM algorithm is developed to calculate the estimates for both the regression parameters and the unknown error density, in which a kernel-smoothed conditional profile likelihood is maximized in the M-step. The variance of the resulting estimators are obtained using an EM-aided numerical differentiation method. The codes were developed by Wenbin Lu. The original paper is published in Statistica Sinica (2010, 20, 661-674).

Download the R code here for analysis of a breast cancer data (Farewell, 1986).        

• On estimation of partially linear transformation model

The computation codes, written in the C language, solve the martingale-based global and local estimating equations for a general class of partially linear transformation models. A resampling method is used to estimate the variance of the proposed estimators. The codes were developed by Wenbin Lu and Hao Helen Zhang. The original paper is published in Journal of American Statistical Association (2010, 105, 683-691).

Download the C code here for analysis of a lung cancer data (Kalbfleisch and Prentice, 2002). In the provided codes, gama = 1, which gives the partial linear proportional odds model. But setting gama = 0 gives the partial linear proportional hazards model. In general, for gama >= 0, it gives a family of survival models considered by Dabrowska and Doksum (1988).        

• On estimation of linear transformation models with nested case-control sampling

The computation codes, written in the C language, solve the inverse selected probability weighted (ISPW) martingale-based estimating equations for a general class of linear transformation models with nested case-control sampling, which generalizes Samuelsen's maximum pseudo-likelihood estimation method (Samuelsen, 1997) for the proportional hazards model. The codes were developed by Wenbin Lu and Mengling Liu. The original paper is published in Lifetime Data Analysis (2011, in press).

Download the C code here for analysis of a data from the Wilms’s Tumor Study (D’Angio et al. 1989; Green et al. 1998). We conducted nested case-control sampling within the full cohort with the control size = 1 or 2. In the provided codes, size =2 and gama = 2, which gives a model from the family of survival models considered by Dabrowska and Doksum (1988). In this family, gama = 0 gives the partial linear proportional hazards model while gama = 1 gives the proportional odds model.