Regression analysis method for mixed data

In life test, complete data are often derived at high stress level and censored data often arise at low stress level. Aiming at the mixed data which are composed of complete data, type I censored data and type II censored data, a regression analysis method for mixed data is proposed. The best linear unbiased estimations of regression coefficients and the standard deviation, and the point estimation and confidence limits of percentiles are given under location-scale distribution. The regression methods for mixed data from normal distribution, Weibull distribution and extreme value distribution are discussed in detail. Thus, the traditional regression analysis method which is strictly applied to the complete data is extended to the mixed data composed of complete data and censored data. Compared with the traditional method, the presented method can reach a high precision and a great number of specimens can be saved.