A Cookbook for Exploratory Visual Analysis

Initial writeup: Sept 2022. Last update: Sept 2022.

When I worked in a news media, I found out 90% stories can be discovered with a few simple steps. Here's the cookbook of visual analysis methodology, despite which visualization tool one uses.

Methodology

A common exploratory visual analysis process:

1. Add a flat table to view actual data. Pick a few dimensions/ metrics to understand the data points.
2. Apply 1-dimensional analysis on all (all interested) metrics.
   1. MIN, MAX, MEDIAN can already tell many stories
   2. One can make histogram to view the distribution
   3. You may want to discretise continuous variable sometimes
   4. You may want to turn textual but ordinal variable into numeric variable sometimes
3. Leverage time series if there is a date time dimension
   1. Pick one metric and check its statistics conditioned on a time part can usually yield insights
   2. Plot a metric over time dimension to observe the trend.
      1. Use breakdown dimension to compare multiple values.
4. Leverage bar/ pie for categorical data
   1. You may want to sort by the metric for most of the time to make the presentation meaningful.
   2. Stacked bar and 100% stacked bar can emphasise different information
5. Map is good to uplevel the presentation but it is same as a bar chart under the hood (one categorical dimension + one metric)
   1. Map adds value only when the metric distribution is geo related
6. Apply 2-dimensional analysis.
   1. Add controls (filters) as many as possible
   2. Use controls to filter the view and answer the question: what is A statistics of B metric when C column equals value D
7. Further more, leverage scatter plot (2-D) and bubble chart (3-D to 4-D) to visualise more variables in one go.