algorithm for testing a variable

Asked Jul 24 '17 at 13:10

Active Jul 24 '17 at 13:10

Viewed 244 times

I want to test if a quantitative variable is different in 2 groups.

As for what I understood so far, here is the algorithm I should use :

if the length of my 2 subgroups are >30
- --> z-test
else, if the variable is normal (shapiro.test() returns a p-value>0.05) and the variances are mostly equal (var.test() returns a p-value>0.05)
- --> t-test
else
- --> wilcoxon-MW test

This is not stated clearly anywhere but just some kind of a grouping of my classes, am I doing it right ?

Where can I find clear resources on this kind of question ?

asked Jul 24 '17 at 13:10

Dan Chaltiel

Your algorithm doesn't make any sense to me. – SmallChess Jul 24 '17 at 13:19
Your comment is not really helpful. Can you elaborate maybe ? – Dan Chaltiel Jul 24 '17 at 13:20
https://stats.stackexchange.com/questions/85804/choosing-between-z-test-and-t-test – SmallChess Jul 24 '17 at 13:21
You don't need to switch between z-test and t-test. – SmallChess Jul 24 '17 at 13:21
Could you post a good version of this algorithm ? Or provide a link where I could learn ? And the z-test is not the only part I'm concerned about, are the conditions badly tested too ? – Dan Chaltiel Jul 24 '17 at 13:25
You'll need to fix the algorithm yourself because it's unclear exactly what you want to do just from the question. For example, you want to check "different". Different in what? Different in means? In variance? Anyway, you don't need to switch t-test and z-test. – SmallChess Jul 24 '17 at 13:26
Different in means. I think these are the conditions of the Student's t-test. – Dan Chaltiel Jul 24 '17 at 13:27

0 Answers0