2

I am trying to interpret the results of a null hypothesis A/B testing by conducting a two-sample t-test and I am using scipy's stats.ttest_ind function.

ttest_ind function has a parameter called "equal_var" of a boolean type. When this is true (default), this seems to assume that A and B follow the same variance.

In the case of a null hypothesis A/B testing, should this be set to True or False? Also, how does this parameter change the formula? Not so sure this is the right forum for this type of question.

halfer
  • 19,824
  • 17
  • 99
  • 186
kee
  • 10,969
  • 24
  • 107
  • 168

1 Answers1

1

TL;DR: if you want to be safe use equal_var=False

Here you can find elaborate answers to the question why not always using the Welch's t-test (equal_var=False). Admittedly, there you can find arguments for both side but only academic references in favor of using Welch's t-test. On Wikipedia it is argued in the same direction: if variances are unequal, you want to use Welch's t-test, if they are not you are still safe.

Sebastian
  • 101
  • 1
  • 5