Minimum Viable Experimentation
The Analytics Engineering Podcast
00:00
The Failure of Interpretation in Social Platforms
A big assumption of experiments is that each individual observation doesn't affect the other ones. At Lyft, we would run these like driver randomizations. Some drivers would get some incentives and some wouldn't. And we'd say, hey, they drove more hours because we gave them these incentives. But they were just stealing rides from other drivers. So actually the net effect of that on the whole system was actually zero.
Transcript
Play full episode
