Join Barton Poulson for an in-depth discussion in this video Actionable insights, part of Data Science Foundations: Fundamentals.
- [Voiceover] When reporting the results of your data science project, you need to be be able to provide actionable insights to your client. I think of it this way. Data and data science is for doing. And that's a paraphrase of one of my intellectual heroes who said, "My thinking is first and last and always for the sake of my doing". That's from American philosopher and psychologist, William James. You don't want to get in a situation of working really hard without knowing what you're doing and without having anything useful. Reminds me of another line.
"We're lost, but we're making good time", from Yankees catcher, Yogi Berra. The idea here is you want to be able to point the way. Remember, the project was conducted for a reason. Why was the project conducted? The goal of the project is usually to direct some future action. They're trying to make a decision, one way or another. And the analysis that you conduct should be able to guide that action in a meaningful and well-informed way. One of the things that you need to do is you need to be able to give your client the next steps.
Tell them what comes next, and justify your recommendations with the data that you analyze. Also, be as specific as you can be and make sure that the recommendations that you give are in fact doable by your client and that each recommendation you give should build on the other step by step by step. Now, there is one major complication in all of this and it has to do with correlation versus causation. Correlation, of course means that two variables are simply associated, whereas causation says, where one produces directly changes in the other.
Here's the deal though. You have correlation, all data is inherently correlational but your client wants causation because they want to know if I do this will it produce this particular outcome. So there's a disconnect. And the question then is, how do you get from correlation, which is what you have, to causation, which is what your client wants? I'm not asking a philosophical question about anthology or epistemology, rather, it's a practical concern. And it's one that we can give a practical answer to.
Now there's a few things you can do to get causality. Number one is you can do an experimental study, a randomized controlled trial. That's theoretically the simplest path to causality. Sometimes, very hard to do in practice and something that needs to be done before the data's gathered. You can also do what are called Quasi-Experiments. These are methods that use non-randomized or correlational data for causal inference. It's a whole class of studies and they're frequently used in education and economics and epidemiology, but they can also be difficult to carry out.
And then finally theory and experience. We're talking about research based theory and domain-specific experience. Think of it as an implicit form of betting calculations that people can do in a way that draws on their experience and makes sense in their context. Next, think about the social and contextual factors of your recommendations. You may recall the data science Venn Diagram, that says data science is composed of three particular practices. That there's coding, that there's statistics and mathematics and that there's domain expertise, however; some people have proposed a fourth circle to be added to this diagram.
And that's social awareness and understanding of how things play out between people. Let me give you a few ideas of how that might work. Your recommendations need to be consistent with your client's mission. Most organizations have mission statements. This can help you place your recommendations within that explicitly defined context. You also need to make recommendations that are consistent with your client's identity. Certain recommendations, even if they appear to match the client's mission and maximize outcomes, are inconsistent with their identity, like charging admission to a religious services.
You need to be aware of the business environment that they're operating in. Know about related companies and government regulations so you can give them a competitive advantage and comply with legal and policy requirements. And then finally, be aware of the social environment. How will your recommendations effect relationships within the client's organization? I mean, that's office politics. You may not know about that, but if you can, try to be aware of it when making your recommendations. And so our conclusions are these. Your analysis needs to address the goals of the client.
You need to give specific next steps based on your analysis. And in doing so, you should be aware of the social and even the cultural context that they are going to be implemented in.
- The demand for data science
- Roles and careers
- Ethical issues in data science
- Sourcing data
- Exploring data through graphs and statistics
- Programming with R, Python, and SQL
- Data science in math and statistics
- Data science and machine learning
- Communicating with data