Now I want to give you an example of exploratory and confirmatory analysis. In a way the examples that I've shown you before, the one where I'm using a software, an interactive visualization to analyze data is really an example of exploratory analysis. But let's go to another example. I'm using once again the same dataset. So this is a dataset describing vehicle collisions in New York City. And let's say that now I have a new question. So having a question, is the idea of exploratory analysis. I have a series of questions that I want to be able to answer. So here my question now is, how does the volume or frequency of injuries change over time? Let's see. How do I do that? Well, I select date, I select the total number of people injured, and now, I have a line graph showing me how these volume changes over time. So, very interesting. We can see a lot of different patterns. So, on the x-axis you can see that we have a number of years. So, 2012, 2013, 2014, up to 2017. And in between we have data about every single month. Okay? So as you can see, one of the first interesting patterns that we can detect here is that there is some seasonal changes. Seasonal patterns. So there are some deeps, and then the volume goes up and then it goes down again, and then up and down again, and then up and down again. So, when does it go down? Well, it's always in February and January. So this is December, January, February. December, January, and February, and so on. So, what does it mean? How do we interpret this pattern? Well, one idea here is that we may imagine that in winter, this is winter time, the number of collisions go down. So now we have a question, why does it go down? So, that's a very interesting aspect of exploratory analysis. Very often, the output of exploratory analysis, is the generation of new questions. This is typical of exploratory analysis. We start with some answers, and, sorry, we start with some questions, and what we produce is not only answers to these questions, but also new questions or new hypotheses. What else do we see here? Well, we see a big, big increase in August 2016. Why do we have this big increase? Either no. That's another question. And why do we have this very, very deep reduction in number of collisions here? Either no. So once again, the output of exploratory analysis very often, is new questions. But now let me give you an example of confirmatory analysis. So, let's start from the same data in the same chart. And let's go back to the observation that, the pattern that we observe over time tends to have a repetition of some sort. Right? In winter time, the number of accidents of collisions go down. So, one hypothesis I have is that this pattern is going to be more pronounced when we analyze data about different kind of collisions. In particular in the dataset, we have information about the number of pedestrians who have been injured, the number of cyclists, the number of motorists, and so on. And now you have a hypothesis that this pattern is going to be much more pronounced for cyclists and motorists. So let's see if this is true. This is an example of confirmatory analysis. I have an hypothesis, and I want to use visualization to see whether my hypothesis is true or not. How do I do that? Well, I have to change the chart a little bit. Rather than having the total number of people injured, I have to look at the cyclists, followed by the motorists, followed by the pedestrians. What do we see here? Now we have three charts, cyclists, motorists, and pedestrians. And when we look at the patterns, we can see that the pattern of the cyclist is much, much more pronounced than the patterns of the motorist and pedestrians. Note that my hypothesis was that the pattern would be more pronounced, both for cyclists, and motorists. But when I look at this chart, it looks like my hypothesis is only partially confirmed. Yes. We do have a much more pronounced pattern for cyclists, but not necessarily for motorists.
