What is the name of subject for testing which distribution the set of data follows?
I become frustrated since I am a beginner in this area.
Asked
Active
Viewed 38 times
0
-
I've added the relevant tag that seems to answer your question, in order to help you find related posts. I note that in answers to an [earlier question](http://stats.stackexchange.com/questions/159841/what-is-the-subject-of-statistics-called-if-the-subject-is-used-to-studying-if-t) you were already linked to the Wikipedia article on goodness of fit (which I link to a subsection of in my answer here). – Glen_b Jul 04 '15 at 02:52
2 Answers
1
This would fall under hypothesis testing. You hypothesize a distribution and test whether it is plausible that the data follows that distribution.
shadowtalker
- 11,395
- 3
- 49
- 109
-
-
@Victor Your question uses the word *test*; and ssdecontrol has given an answer about testing. it's hard to imagine what other topics you seek that follow the condition of your question (are tests) but do not involve hypothesis testing. – Glen_b Jul 04 '15 at 02:45
1
In most situations, you can't find out which distribution a set of data has been drawn from; you can sometimes be quite confident that it hasn't been drawn from some distributions (e.g. if it's clearly bimodal, you might be able to be pretty sure it won't be a random sample from a unimodal distribution).
So if you're testing whether your data come from a specific distribution, say a normal distribution (using something such as a Shapiro-Wilk test), then a rejection might lead you to be confident that your data are not drawn at random from a normal population -- but a failure to reject doesn't imply the data are drawn from a normal population, only that you didn't have a large enough sample to detect deviations from it.
There are some tests for a specific null hypothesis (i.e. with all parameters specified, such as the Kolmogorov-Smirnov, or the Anderson-Darling), and there are other tests for a specific distributional form but with some parameters unspecified (e.g. the Shapiro-Wilk test for normality).
This kind of testing is called goodness of fit testing.
There are many posts on site about this. One that it might be worth starting with is Is normality testing 'essentially useless'?
