2
$\begingroup$

I have rainfall time series of 85 points from 1901-2013. I want to check whether the mean rainfall during 1901-1970 is significantly different than that of the mean rainfall during 1971-2013.

Here, the data would not 'necessarily' follow the normal distribution. So, one of the main assumption for applying t-test may not be fulfilled.

Now as the length of the dataset for two samples would be 70 and 43, can I use the paired t-test or Welch's t-test assuming that the length is sufficient to circumvent the assumption of normality.

What should be the minimum length of data to overcome the normality condition in t-test, is 43 good enough?

In case, it's not then what test should be applied?

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ I doubt there's a good reason to think you'll have near-independence across time, so I wouldn't suggest either. $\endgroup$
    – Glen_b
    Commented Mar 29, 2016 at 5:37

1 Answer 1

Reset to default
2
$\begingroup$

I would use the Mann-Whitney U test, which is quite robust and reasonably powerful. Otherwise you can first check for normality (eg with the Shapiro-Wilks test) and then use the t test if the hypothesis of normality is not rejected.

Improve this answer
$\endgroup$
1
  • $\begingroup$ How do you answer to the possible serial correlation? $\endgroup$
    – kjetil b halvorsen ♦
    Commented Nov 24 at 17:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.