"Talk Data to Me - the art of storytelling" by Diana Pholo

Collecting data is now seen as a very important aspect of business. Many companies therefore invest in solutions such as Business Intelligence tools, spreadsheets and dashboards in an attempt to extract useful information from their data. However, these tools still fail to present what is hidden in the data because they do not bring out their underlying significance. This is where data storytelling comes in. It helps to visualise information, understand the narrative behind it and share it.

This talk is aimed at anyone who wishes to get started with storytelling using data. What do you look for? How do you present your findings to stakeholders? Together, we will have a look at common visualisation tools provided Matplotlib, Pandas and Seaborn (accompanied with code snippets and sample outputs). We will then look at the story behind the Harambee youth employment journey data. (Data provided by Harambee Youth Employment Accelerator)

Transcript

1. Talk Data to Me: The Art of Storytelling Diana Pholo Predictive Insights
2. Introduction • Data is just data if we do not effectively visualize and communicate it.
3. Story components • Situation • Problem • Insights • Next Step
4. The Background
5. Background • harambee.co.za
6. The Problem
7. Problem Definition •
8. Finding the story
9. Exploring the data
10. Exploring the data (cont.) • Nice tool: pandas ◦ head() ◦ tail() ◦ describe() ◦ isnull() ◦ etc.
11. Exploring the data (cont.) •
12. Exploring the data (cont.) •
13. Exploring the data (cont.) •
14. Exploring the data (cont.) •
15. Data Wrangling • Examples: ◦ Convert data types ◦ Dealing with NAs ◦ Engineer features
16. Data Wrangling (cont.) • E.g.: Type conversion
17. Data Wrangling (cont.) • E.g.: Feature Engineering
18. Feature Distributions • Example: boxplot with Pandas
19. Feature Distributions (cont.) • Example: Histograms
20. Feature Distributions (cont.) • Histogram with Pandas
21. Feature-Feature Relationships •
22. Feature-Feature Relationships (cont.) • Example: boxplot with Pandas
23. Feature-Feature Relationships • Correlation
24. Feature-Feature Relationships • Heatmaps
25. Feature-Feature Relationships • Barplot
26. Feature-Feature Relationships (cont). • Scatter plots
27. Real-life example
28. Background • South African labour market more favourable to men • Harambee takes in more women than men
29. Problem • Percentage of women finding employment is still lower
30. Insights • Example: Women generally have more responsibilities
31. What’s next? • Provide child care • Promote inclusive workplace culture • Teach shared responsibilities
32. In Conclusion
33. Recap •
34. Best practices • Create interesting presentation • Avoid unnecessary details • Define & Understand Audience • Humanize data
35. Don’t • Determine narrative beforehand • Spend more time on presentation than questioning results validity • Analyse data without big picture • Use storytelling as a shortcut