A profile conditional likelihood approach for the semiparametric transformation regression model with missing...

Area of Science:

• Statistics
• Biostatistics
• Survival Analysis

Background:

• Missing covariate data is a common challenge in semiparametric regression models.
• Existing methods like complete-case analysis can lead to biased results or reduced efficiency.
• Accurate handling of missing data is crucial for reliable statistical inference.

Purpose of the Study:

• To develop a novel statistical approach for handling missing covariates in general semiparametric transformation regression models.
• To provide a simpler and more efficient alternative to full maximum likelihood methods.
• To improve the accuracy and reliability of regression parameter estimation in the presence of missing data.

Main Methods:

• A profile conditional likelihood approach is proposed.
• The Kaplan-Meier estimator is used to estimate the marginal survival function.
• Covariate distribution and model parameters are estimated from a conditional likelihood, incorporating the Kaplan-Meier estimator.

Main Results:

• The proposed method yields consistent and asymptotically normally distributed estimators for regression parameters.
• Simulations show very high relative efficiency compared to other methods.
• The estimator can be more efficient than complete-case analysis and corrects bias when data are missing at random.

Conclusions:

• The profile conditional likelihood approach effectively handles missing covariates in semiparametric transformation regression.
• This method offers a simpler, more efficient, and less biased alternative to existing techniques.
• The approach has potential applications in models like the generalized probit model with missing continuous covariates.