Simpson's paradox in action
Generalizing our A/B test results to different segments of the population can be of utmost importance to the business. Sometimes we want to save the cost of running other tests in different cities, by different devices, etc. Making sure that our results are consistent by subpopulations can increase our confidence to make such generalizations.
Examine the simp_balanced
and simp_imbalanced
datasets for Simpson's paradox to gain a good sense for how this phenomena can occur in A/B testing scenarios.
This exercise is part of the course
A/B Testing in Python
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
# Calculate the conversion rate per variant and then browser
imbalanced_variant_rate = simp_imbalanced.____('____')['____'].____()
imbalanced_variant_browser_rate = simp_imbalanced.____(['____','____'])['____'].____()
print(imbalanced_variant_rate)
print(imbalanced_variant_browser_rate)