Accounting for measurement error in log regression models with applications to accelerated testing.

Area of Science:

• Statistics
• Reliability Engineering

Background:

• Parameter estimates in regression models are prone to bias when explanatory variables contain measurement errors.
• This bias is particularly problematic in accelerated lifetime testing, where extrapolation amplifies prediction errors.
• Additive stochastic components in log regression models can also introduce bias if incorrectly modeled as multiplicative.

Purpose of the Study:

• To develop and evaluate a robust log regression model that accounts for both measurement error and additive error.
• To improve the accuracy of parameter estimation and prediction in accelerated lifetime testing.
• To provide a more reliable method for analyzing accelerated testing data.

Main Methods:

• Approximation of a log regression model with measurement error and additive error using a weighted regression model.
• Estimation of the weighted regression model via Iteratively Re-weighted Least Squares (IRLS).
• Application and comparison of the proposed model to existing methods using the reduced Eyring equation in accelerated testing scenarios.

Main Results:

• The proposed weighted regression approach effectively mitigates bias caused by measurement and additive errors.
• Simulations and real-world data analysis demonstrate the superiority of the new model over previously accepted methods.
• Improved accuracy in parameter estimates and extrapolated predictions for accelerated lifetime testing data.

Conclusions:

• The developed weighted regression model offers a significant advancement for analyzing accelerated testing data.
• Accounting for both measurement and additive errors is crucial for reliable modeling and prediction in reliability engineering.
• This method provides a more accurate and robust alternative for practitioners in accelerated lifetime testing.