Online Program

Return to main conference page
Thursday, October 19
Thu, Oct 19, 3:50 PM - 5:00 PM
Aventine Ballroom E
Speed Session 2

How to Effectively Communicate Commonly Misinterpreted Statistical Concepts (303865)

Hoi Yi Ng , Amazon 
*Paavni Rattan, Amazon  

Keywords: statistics, experiments, randomized control trials, AB testing, amazon

Amazon's supply chain is innovating at a tremendous pace. Due to our obsession with data driven decisions, the impact of every new feature needs to be measured with rigor. It is an exciting time for us at IPC (Inventory Planning & Control) LAB, a team that provides an experimentation platform, as the problems we face become more novel and challenging. Our automated platform and offline studies design and perform randomized control trials, then provide p-values, point estimates of the treatment effects and their corresponding confidence intervals. As common as these concepts are within the statistics community, it is challenging for our customers to interpret these terms and as a result, the results of the experiments can sometimes be misinterpreted. Another challenge lies in the constant need from our customers to quantify the results with a single number without measures of uncertainty. This presentation illustrates with examples how our team of statisticians enable our customers to draw inferences using statistics. We will discuss the challenges we have faced in doing so and our proposed solutions with our continuously evolving experience in this area.