0

I have a data set comprising of responses from approximately 385 people on several variables such as their shopping habits, mode of transport to the shopping centre etc. I want to do Pearson correlation on the variables (by age, gender, income level etc. of the respondents).

Is there any rule that says what would be the maximum number of tests that can be reliably performed on this data set? I am unsure if "reliable" is the correct term; I mean "valid" or "sound" in this case.

My alpha is 0.05 and I have corrected it using Sidak's method, so the experiment wise error is 0.05 for the 80 tests that I want to perform on this data set.

Adhesh Josh
  • 2,935
  • 16
  • 50
  • 67
  • It looks like there is an error in the reported $\alpha$ value. Setting your FWER at 5% means that nominal $\alpha$'s are about 0.000641. – chl Sep 07 '11 at 13:40
  • 2
    I wonder whether looking at correlation coefficients would be useful. What would the test results tell you? – whuber Sep 07 '11 at 14:04
  • Your question is closely related to this one: http://stats.stackexchange.com/questions/5750/look-and-you-shall-find-a-correlation – Jeromy Anglim Sep 08 '11 at 03:40
  • @chi It is 0.05 for all 80 tests which means, as you correctly pointed out, that the alpha is lower for the individual tests. – Adhesh Josh Sep 08 '11 at 15:43

0 Answers0