Introduction
Today's business decisions are closely connected with data analysis. While most analysis of business data involves comparing single quantitative variables (such as sales across different regions), there is often a need to analyze data to find similarities and differences among groups of objects. For example, finding the best performing products across different regions by comparing revenue, sales, and marketing budget, analyzing multiple instances of several variables at once.
Data visualizations provide an effective way to make sense of large datasets by stimulating visual thinking. This article explores visualization techniques that can decode business data and make product performance analysis comprehensible.
Example: Telco Mobile Usage Plans
Consider a telecommunications company with various mobile-usage-plans to offer. Plan comparison requires multiple characteristics:
- Price
- Duration
- Revenue
- Units Sold
- Marketing Budget
The values of all these variables for each plan combine to form the plan's multivariate profile. The goal is to compare such profiles to answer questions like:
- Which mobile-usage-plans are alike?
- Which mobile-usage-plans are most exceptional?
Visual Representation Approaches
Initial data visualization goals should focus on gaining insights rather than jumping directly into numerical analysis. Once users understand the overall dataset, they can drill-down for exact numbers and deeper analysis. Spotting similarities or differences across datasets requires thoughtful use of visuals with uniformity.
Visualization Techniques for Multidimensional Data
- Trellies or small multiples
- Multiple concurrent views with brushing and filtering
- Spider charts
- Heatmaps
- Parallel coordinates
Spider Charts (Radar Charts)
Spider charts, also called "Whiskers" or "Star" charts, are advanced forms of Chernoff faces (a 1973 concept for presenting multivariate data in face shapes). In spider charts, multiple lines radiate from a center point, with each line representing a different variable and its length encoding the variable's value.
Heatmaps
Heatmaps effectively encode quantitative values as color or hue variations. In a heatmap representing mobile-usage-plans, each row displays the plan's profile:
- Higher-than-average values appear in blue
- Lower-than-average values appear in red
Heatmaps are excellent for finding exceptions. For example, if plan-4 has all values lower than normal, the heatmap immediately reveals this. Similarly, plan-7 shows shorter duration like plan-4, but better marketing budget suggests plan-7's poor performance may be due to other factors.
Using gray for average values improves visual attention for extremes.
Parallel Coordinates Plots
Parallel coordinates can display hundreds of lines representing data profiles (such as mobile-usage-plans) and provide excellent insights when used thoughtfully. This technique helps:
- Find similarities between product performances
- Cluster products based on similarity criteria
All evaluated dimensions must be on the same percentage scale for meaningful comparisons.
Interactive Features
Advanced interactive visualization techniques like highlighting emphasize specific lines of interest. For example, you can focus on plans with the lowest marketing budget and analyze their revenue.
Further techniques like brushing and filtering enable detailed insights. Rather than relying on complex statistical clustering algorithms, modern visualization tools allow transforming business data into visually comprehensible objects for intuitive analysis.
Conclusion
Visual patterns provide simple ways to identify similarities in product performance analysis compared to scrolling through tables of rows and columns. Statistical clustering algorithms are complex, but current visualization advancements enable turning business data into comprehensible visual formats for effective insight generation.