统计问题(22):非参数检验知多少(多选)
Question
Which if any of the following statements are false?
a) Non-parametric statistics may be used for normally distributed data
b)Correlation coefficients are inappropriate for non-normally distributed data
c)Medians are always a more valid way than means are to compare non-normally distributed data
d) The significance level of a test should be increased when multiple tests are made in a study
Answer
Parametricstatistics (like t tests)work well for data from a normally distributed population but non-parametricstatistics may also be used. Non-parametric statistics do not requireassumptions about the distribution of the data, but they will be a little lesssensitive at picking up real differences if used when a parametric test wasappropriate. Also, 95% confidence intervals are harder to compute and will be widerwhen based on non-parametric statistics.
Pearson’s rank correlation coefficient makes no assumption about the distribution of thevariables, only that there is an order to the values.
A dichotomous variable has only two values. An example of a dichotomous variable is a person’s vital status—either dead or alive. This variable is certainly not normally distributed, but comparing two treatments on the basis of the median value would be unhelpful. On the other hand if the variable were coded as 0 for dead and 1 for alive, then the mean value would be the proportion surviving ineach group. So the mean would be more informative than the median.
Multiple testing increases the chance of a false positive finding so, if anything, the significance level should be reduced to guard against this.
中文解释:
参数检验(如t检验)对于来自正态分布总体的数据非常有效,但也可以使用非参数检验。非参数检验不需要关于数据分布的假设,但是如果在合适的参数检验中使用的话,它们在获取实际差异方面将不那么敏感。而且,基于非参数检验时,95%的置信区间很难计算,并且会更宽。
所以答案是选择 BCD
更多统计问题(最近6期)
联系客服