Welcome!

1. Welcome!

Welcome to this course on data-driven decision making. We will cover the many ways data, models and visuals can aid your decision making. Let's get going!

2. Types of data

Data can take many forms including numeric, factor, Boolean and strings. In this course we expand to include information gathered from models, conclusions reached through visualizations, or by interacting with dashboards.

3. Exploring patterns without an outcome

Often, basic analyzes are exploratory or explanatory. Exploratory has no predefined outcome. You're looking for insightful patterns. Exploratory projects can be useful to understand data availability, or integrity issues such as faulty information. Suppose you were a hospital administrator trying to learn about your patients. With exploratory analysis you could identify common characteristics among all patients such as age, and gender averages. You aren't looking for anything in particular, instead learning a general sense of the data.

4. Explaining a specific outcome

In contrast explanatory analysis seeks to understand a specific situation. Let's continue with the previous example. During the exploratory analysis you may identify an odd cohort in the data. For example, elderly diabetic patients discharged on Fridays may be readmitted often. If that pattern is recognized, your efforts may shift to an explanatory analysis concerning these specific patients. Now you have a specific outcome you are looking to explain.

5. Predicting a future outcome

The last high-level category moves from exploratory, and explanatory to predictive. Previously, elderly diabetic patients were found to be readmitted more often than others. An explanatory analysis may indicate that the nurses on Friday's shift need additional training. During predictive analysis one builds a model predicting the probability of a patient's readmission. Predictions are future-oriented.

6. Who is doing the analysis and why is the analysis being performed?

Often, the level of detail and methods employed depends on the audience. Exploring patterns helps business leaders identify new markets or learn about their business holistically. If you are in charge of an operation, you are likely tactical, looking to explain,and optimize your operation. Lastly, predictive models help understand specific interactions. Keep in mind, a single prediction is a tactical outcome. Thus, senior leaders focus more on a predictive model's overall impact rather than a single outcome.

7. How was the data collected and when?

Don't overlook having the right data available. Next, match the data to the problem. Don't assume data is error-free. Further, there can be issues in collecting or reporting data.

8. What method is being applied?

Next ask about the method utilized. Sometimes a simpler method has similar results, particularly in machine learning. Don't assume complexity is superior. For example, a deep neural-net may be the most accurate but a simpler model like a decision tree may be acceptable given the use case. It's a good idea to learn what methods were applied, any alternatives explored and in the case of machine learning to explicitly ask about the trade-off in accuracy versus complexity.

9. What was the outcome?

Asking about results is an important way people demonstrate fluency and instill trust. Ensure you understand the results of a technical project. Trust the practitioner to operate with technical fluency. As the decision maker you need to ensure the output aligns to expectations. If it doesn't then probe why and ultimately ensure the analysis adds value.

10. Who is being impacted?

Lastly, discuss the potential impact. For predictions, the impact assessment is knowing good and bad. Consider building a credit prediction to increase your company's revenue. That's a great impact. However, if the model adversely impacts women or people of color your company could be in violation of the law let alone ethics. Mitigating negative impacts is an important step in probing a data-driven decision.

11. Let's practice!

In the next exercises you will classify data-driven examples as exploratory, explanatory, or predictive analyses.

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.