I have a question about t distribution. I have a set of data and I want to know if they are drawn from a t distribution or not! Can anybody help me how I can do this in Matlab? Thanks

statistical test for data drawn from t distribution

dpb 2014-8-6

First, how many samples? If n>= ~ 30, the t distribution approaches normal so testing for normality would be essentially the same. There's at least one test for such packaged in Matlab Statistics Toolbox.

MINA 2014-8-6

n>30. The thing is I'm modeling my data with ARIMA and there I have to choose whether the innovations have Gaussian distribution or t distribution. when I use AIC/BIC criteria to select which model is better, I get a model with innovations drawn from t distribution. Now When I compute the residuals I want to see if they are coming from t-distribution or not. I know that they are not normally distributed- by using kstest.

dpb 2014-8-6

编辑：dpb 2014-8-6

Then w/ n>30 they won't be t, either, because the two distributions are asymptotically identical. The higher N, the closer and the lesser chances of showing anything different.

MINA 2014-8-6

@dbp ok I understood. Thanks.

dpb 2014-8-6

在 MATLAB Online 中打开

BTW, you can use kstest on the data with a cdf from tcdf; see

doc kstest

for the details. There's lillietest that's more powerful specifically for normality.

But, as noted, for N of 30 or greater the likelihood of there being sufficient difference between T and N is minimal at best to be able for statistical hypothesis test to tell the difference.

statistical test for data drawn from t distribution

5 个评论
显示 3更早的评论隐藏 3更早的评论

回答（0 个）

类别

标签

Community Treasure Hunt

statistical test for data drawn from t distribution

5 个评论 显示 3更早的评论 隐藏 3更早的评论

回答（0 个）

类别

标签

另请参阅

Community Treasure Hunt

5 个评论
显示 3更早的评论隐藏 3更早的评论