Case definition of conversion

- [Narrator] In this video, we are going to cover what I call the case definition of conversion. Which means getting into the details of exactly how to measure conversions. I'm getting into the weeds with you because measuring conversions can be complicated. Some conversions are easier to measure than others and that may actually impact your experimental design. For example, I talked about trying to measure my book sales directly, but realizing I only get biannual reports of my book sales, it might be easier for me to measure something else related to my book sales that gauges interest in my book. Like how many new reviews I get on Amazon. And use that as a proxy for a conversion. The fact that I only get biannual reports limits my ability to measure my book sales in smaller time increments, and there are other limitations. If I'm using a conversion other than a sale, like a LinkedIn connection, what does that even mean? Maybe that connection really doesn't get me closer to a conversion like a conference invite, or book sale. And worse, you can measure the same conversions in different ways. My friend did a public health campaign on Twitter and boy, is that a good example of multiple ways to measure the same thing. Some people use the app Twitter analytics to get the metrics related to their tweets. If you do that, then you compare those metrics to what you see when you log into your actual Twitter account, you will see that the 2 counts don't exactly match. Twitter knows this and has explained it in their documentation. And if you use third party tweeting software, such as Buffer, you will find that it might count tweets differently than Twitter and Twitter analytics. So you have to think about which one you are going to choose to use for your experiment if you have Twitter related conversions. So to help you think practically about the details of counting conversions, I want to introduce you to a concept from my field, Epidemiology, where we study disease outbreaks. Imagine there is a possible Influenza outbreak. Epidemiologists have to go into the community and count the cases of Influenza. But how do they count a case? Is someone with a fever a case? Just because they have that symptom? Or do they need laboratory confirmation of actually having the virus? In real life, we will set up a number of different case definitions, as we call them. One might be a symptom only definition, and one might be more strict, and require symptoms and laboratory confirmation. For each of these, we start out with a working case definition. Which is a list of criteria the patient must meet to qualify as a case, by definition. It is helpful to have real cases to guide us in developing these criteria. Let's try to come up with a working case definition for a conversion together. Let's pretend I have the objective to increase my profile in the data science community through articles and posts on LinkedIn. Then, whatever I choose to measure as a conversion should be measurement of how successful I am at actually increasing my profile in the data science community on LinkedIn. So, one way I could do it is count how many likes I get on my articles and posts on LinkedIn. But then I started thinking, articles are a lot more work than posts, and does it matter who is actually liking my articles and posts? I do a lot of working in the nursing community, as well as the data science community. Do I want to reject the love of the nurses? And maybe likes aren't even the right thing to count. I have to consider all these points in developing my working case definition of a conversion. Okay, we admit it, there are some challenges in getting into the details. But let's try to come up with some working qualification criteria anyway to measure this conversion. Here is what I came up with. I want to measure likes on LinkedIn from anyone, including the nurses, but only on articles. Because those are the ones I'm really pouring my heart into to try to increase my profile in the data science community. But, what I'll do, is only count these on the actual data science articles I post. Not the ones about other topics, which do not touch data science specifically. Like when I talk about healthcare and nursing education. This is our working case definition. This is what I think I will measure as a conversion to meet my objective. The term I am using is working case definition. That's the case definition you develop based on what you know conceptually. But will it actually work? Can I actually measure this? In order to answer this question, I will need to test the definition on real data. That is why it is called a working case definition. You can use it on real data to see if it will work, and that will give you an opportunity to improve it if you can see a better way.