术语“离群值”的详细信息

英译术语:

outlier

汉语语境:

发现离群值也可以通过观察值的频数表或直方图来初步判断，也可通过统计软件作观察值的箱式图来判断，如果观测值距箱式图底线Q1（第25百分位数）或顶线Q3（第75百分位数）过远，如超出箱体高度（四分位数间距）的两倍以上，则可视该观测值为离群值。当数据近似正态分布时，有一种较为简单的方法，可用均数加减2.5s来判断，如观测值在此范围以外，可视为离群值。在统计学上也可用线性回归的方法来对离群值进行判断。当出现离群值的时候，要慎重处理，要将专业知识和统计学方法结合起来，首先应认真检查原始数据，看能否从专业上加以合理的解释，如数据存在逻辑错误而原始记录又确实如此，又无法在找到该观察对象进行核实，则只能将该观测值删除。如果数据间无明显的逻辑错误，则可将离群值删除前后各做一次统计分析，若前后结果不矛盾，则该例观测值可予以保留。

英语语境:

Finding outliers can also be through the observation value of the frequency table or histogram to determine the initial, and it can also be through statistical software for box plot of the observed values to judge, if the observations from the box bottom line Q1 (25th percentile) or top line Q3 (75th percentile) too far, such as in excess of more than two times of the box body height (interquartile range), visible at the time of the observation value as outliers. When the data is approximate normal distribution, there is a relatively simple method, can be used to count and subtract 2.5s to judge, such as the observation value in the range, can be regarded as outliers. In statistics, the linear regression method is also used to estimate the outliers. When there are outliers and should be handled carefully, to combine the professional knowledge and statistical method, first of all, we should carefully check the original data, whether from the professional to be a reasonable explanation, such as data logic errors and original records and rightly so, and not in the observed objects to find the verification can only be the observation value is deleted. If there is no obvious logical error between the data, then the outlier can be deleted before and after the statistical analysis. If the results are not contradictory, the observation value can be reserved.