Simpson’s paradox happens when behaviour at the group level is different from the behaviour of its subgroups.
For example, consider a mobile app with 10,000 Android users and 5,000 on iOS. If we make the assumption that all users spend equally, then it may make more sense to prioritize development of features on Android.
However, this decision may be misguided when we drill down to the subgroup level. Spending broken down by tablet and phone may reveal better percentage of monetisation for both subgroups on iOS.
The following article provides a good example of Simpson’s paradox at work in mobile app monetization. It also suggests how to avoid Simpson’s paradox through smart user segmentation.
In conjunction with Big Data Week, we compiled a list of the best Big Data articles here.
(Image credit: Fotopedia)