Method And Application Of Statistical Data About Outlier-Diagnosis

Using statistical data to analyse and forecast economic traits and trends is the usual method for all scholars and even amateurs. However, with further development of information technology, people have higher and higher requirements of data quality, new theory of data-diagnosis come out and motheds of it are more perfect, especially on the identification and treatment of outliers, which can not be neglected in the statistical diagnosis process. In this paper, we discuss the identification problem of outliers both from regression model and residuals distribution, An optimal method of diagnosing outliers under supposed condition is obtained by summarizing and comparing various test statistics under the distribution or model, then, we use the method to test our country's macro-statistical data.For common regression model, Outlier is the data point that is maximum deviation from the eastablished model. Analyzing data point's impact on the regression by case-deletion model, if the impact on estimator of the regression equation exceeded the critical value, then the point is outlier. With the increasingly complex relation ship between economic phenomenon, there is almost no longer a simple linear relationship between them. On the base of studying outliers under the non-parametric and parameters model by predecessor, we expand this classical methods and analyse practical application.For statistical sample data, Outlier is the data that is not from the same distribution with data set, In other words, outlier and most of the data are not from the same distribution. Although scholars from various countries have made outstanding contributions of outlier identification under the normal distribution, exponential distribution, extreme value distribution, Weibull distribution Distribution and so on. On the base of studying outliers under non-normal distribution, we propose a new statistical quantity-type statistic under I-type extreme value distribution that can be avoided shielding effect in any case; Secondly, we summarze various diagnostic method about outlier and use "comparative statistics" to compare the power of three commonly used test statistic under normal distribution.Finally, Based on summarizing and evaluating the main conclusions and contributions, we propect the future development of outlier diagnosis.