| Robust statistics is a key branch of statistics, which not only closely related to statistics, but also plays an important role in the development of statistics. Therefore, the study of the history of robust statistics has a very important theoretical value and practical significance. Based on an extensive investigation of original sources and research papers on this subject, provided the history of the development of robust statistics through the analysis of three questions:"What is robust statistics","Why is robust statistics" and "How robust statistics was created and developed". By (he combination of historical literature analysis and the theory of modern robust statistics, the combination of different ways of history development and different subjects, the main contributions of this study are as follows:1. Discussed the early works of error theory, studied the works of Euler, Gauss and Laplace in error theory, and analyzed the generating process of the basic concepts in error theory.2. Discussed the method Statistics used for deal with the outliers before the creation of the Robust Statistics. The main way is to analyze the background of the person and the method together, then concluded the reason of using this method. The main part is about Simon Newcomb’s mixed normal distribution and Daniell’s probable integral method. Then discussed the way how Box give the concept of the Robust3. Studied Turkey’s important contributions to robust statistics. Investigated the theory of contamination distribution, including the trimmed mean method, Winsorizing method and averaged absolute deviation method. Some investigations about the double weights of position estimation are conducted through analyzed the works of Turkey in Princeton Robust Year. Moreover, analyzed Turkey’s robust ideas, the contributions of Time Series, as well as how he brought the robust ideas in area of Time Series. 4. Studied Huber’s important contributions to robust statistics. According to the paper’ Robust estimation of location’, which published in1964, discussed the work in the paper, studied the important background and hot problem in that period. Then Huber gave other estimator based on the M-estimator, for example, R-estimator and L-estimator. They are very useful tool for the development of the robust statistics. Last analyzed the impact and contribution of these theories.5. Studied the work of Hampel, analyzed the meaning of it in the development of the Robust Statistics. Focus on the two important concepts which are influence function and break point, point out the effect of that two concept in development of the Robust Statistics, and also discussed the relation between them. Last compared the Minimax approach and the influence function, pointed out the difference and relationship. The result is, when both of them are in normal distributions, they can get a group of very similar result.6. Summarized the history and the future of the Robust Statistics. Go through the basic idea and theory of the Robust Statistics, then summarized the history of the development. Then analyzed the main theories in Modern Robust Statistics, which included the Winsorized Method, Bootstrap Method and Rank Static Method. According to the main methods in Modern Robust statistics, then overlooked the development in the future. |
