Residual plots for the censored data linear regression model

Area of Science:

• Statistics
• Biostatistics
• Survival Analysis

Background:

• Linear regression with censored data requires specific assumptions for accurate estimation.
• Traditional residual plots are effective for uncensored data but not directly applicable to censored data.
• Assessing model fit is crucial for reliable statistical inference.

Purpose of the Study:

• To propose novel graphical methods for assessing linear regression model assumptions with censored data.
• To provide tools analogous to residual plots used for uncensored data.
• To demonstrate the utility of these plots in practical data analysis.

Main Methods:

• Development of new residual plot types specifically designed for censored data.
• Adaptation of established residual plotting techniques for use with survival data.
• Application and evaluation of the proposed plots using the Stanford heart transplant dataset.

Main Results:

• The proposed plots effectively reveal potential violations of linear regression model assumptions for censored data.
• Visual assessment using the new plots aids in identifying appropriate model specifications.
• The plots demonstrated utility in the context of the Stanford heart transplant data analysis.

Conclusions:

• The novel residual plots offer a valuable diagnostic tool for censored data linear regression.
• These graphical methods enhance the ability to check model assumptions and select appropriate models.
• The proposed techniques improve the reliability of statistical modeling in the presence of censored observations.