[MUSIC] Let's look at the next follow up question that helps elicit why groups may not be probabilistically equivalent. The question is. Is there reverse causality? Or, did managerial decisions cause outcomes to change? Or was it outcomes that drove managerial decisions? In the Pentathlon case, we already know that there are pre-existing differences between users who receive different amounts of email. Let's see whether the reason might be that there is reverse causality. We want to ask, did more emails cause better business outcomes, for example, higher revenues? Or is it possible that higher revenues cause more emails to be sent? From our discussion from this case we know that pentathlon, because email promotions are de-centralized. Tends to send more emails to consumers who purchase from more departments. These consumers also tend to have higher revenue. As a result, indeed, there is reverse causality. Revenues drive email frequency, instead of email frequency driving revenue. In interpreting this data, Francois Gabriel mistook the direction of causality. Because email frequency is determined by revenues, the three groups of consumers are not probabilistically equivalent. They differ by more than just how many emails they got. Namely how good a customer they offer pentathlon. Let's think about another example to illustrate. I would like you to read the Microsoft social engagement case. This case describes a situation at Microsoft. Where a digital marketer discovers evidence that consumers will actively share content from Microsoft websites generate much more revenues than those who do not. The case shows what happens when as a result of this evidence, the company invests in social engagement. Your job is to make sense of what happened to social engagement and to revenue. Pause the video, read the case, try to answer the case questions and then come back for debrief. The starting point of the case is that social visitors spent much more than other visitors. With social we mean visitors who share content from Microsoft websites for example by posting to Twitter, or Facebook, or by sharing something by email. As you can see, while only 0.8 percent of visitors are social they produce 5.1 percent of the revenue. As a result their spending is six and a half times higher. To Soneil Kosla the message is clear. If they could increase social engagement with Microsoft, revenues for the companies would go up. And so, he advocates having Facebook and Twitter links in prominent positions on every page of their core website. What happens is this, when Kosla compares the 90 days before the change to the 90 days after the change. He finds that overall visits and revenues are down a little, but not out of line with seasonal variation. The real surprise comes from looking at the sharing data. While social sharing is up, from 3.4 to 4.4 million, as Kosla had hoped, the percent of visitors who shares unchanged at 0.8%. And most disturbingly, the revenue generated for such visitors is also unchanged. The case asks you to consider why social sharing is up, but revenues are flat. The answer is this, Kosla assumed that giving people opportunities to share would increase social engagement. Which in turn would make people more likely to buy because it increases the loyalty to MIcrosoft. Ask yourself this. Could reverse causality be at play? What is more plausible, that the act of sharing leads to more loyalty? Or that loyalty leads to more sharing? The answer, of course, is the latter. When Kosla gave visitors more opportunities to share, it did not make customers more loyal. Instead, it gave customers who were already loyal to Microsoft. Meaning that they spent more money on average. It gave these customers more opportunities to share. This explains why the number of social visitors did not go up, but their sharing activity did. So, is there reverse causality, is the second follow-up question when we have trouble answering directly. Whether there are any preexisting differences between groups. [MUSIC]
