Power BI Visualisation Best Practices

It is said that your data is only as good as your ability to communicate it, which is why choosing the right visualisation and presenting it well is so imperative.

Today we’ll look at several common visualisations available in Power BI and their recommended design practices.

* All samples on this blog are created using standard visualisation elements in Power BI Desktop (May 2016 release) *

Visualisations in A Nut Shell

Before creating any visualisation, it is useful to understand the types of data that are visualised and their relationships to each other. Here are some of the most commonly used ones from Power BI :

* Courtesy of Data Visualization 101 published by Hubspot

NOMINAL COMPARISON

A simple comparison of the quantitative values of subcategories. Example: Total sales in different regions

TIME-SERIES

This tracks changes in values of several consistent metric over time. Example: Monthly sales Vs. targets

CORRELATION

Data with two or more variables that may demonstrate a positive or negative correlation to each other. Example: Income by hours of work

RANKING

This shows how two or more values compare to each other in relative magnitude. If plausible, use of logarithmic scale is recommended.

PART-TO-WHOLE

This shows a subset of data compared to the larger whole. Example: Percentage of customers purchasing specific products

TREND & OUTLIERS

This shows the trend of a metric over time. Alos useful for highlighting data outliers. For example, ice cream sales in a winter season region

Bar & Column Charts

Bar charts are very versatile. They are best used to show change over time, compare different categories, or compare parts of a whole

Do’s & Don’ts

USE HORIZONTAL LABELS

Avoid steep diagonal or vertical type, as it can be difficult to read

USE CONSISTENT COLORS

Use one series of colors for bar / column charts. You may use an accent color to highlight a significant data point but try to avoid using too many color variations

ORDER DATA LABEL

Order categories alphabetically, sequentially, or by value

START FROM 0, IF POSSIBLE

Starting at a value above zero truncates the bars and doesn’t accurately reflect the full value

Pie & Donut Charts

Pie chart is arguably one of the most popular (and controversial) charts of all time. They are recommended for making part-to-whole comparisons with discrete data.  The visualistion works most effectively when there are only a few discrete values.

Do’s & Don’ts

KEEP UNDER 5 CATEGORIES

It is hard to differentiate smaller values on a pie chart. Depicting too many slices diminishes the impact of the visualization. If possible, group smaller slices into an “other” category

DONT COMPARE PIES

Slice sizes are difficult to compare even when lined up side-by-side. Use a 100% stacked bar chart instead.

ENSURE IT TALLIES UP

If you are displaying percentage on a pie chart, verify that values sums to 100% and that pie slices are sized proportionally to their corresponding value.

About Order of Pie Slice

Although itis not configurable in the current release of Power BI (not yet), but I generally recommended positioning the largest pie slice at 12 o’clock position, then place remaining slices in descending order, going clockwise direction.

To Be Continued (Part 2)

In the next session I will talk about design practices for the following visualisations:

  • Line Chart
  • Area Chart
  • Scattered Plot

In Part 3, I will also go over the following visualisations:

  • Bubble chart and visualisation using map
  • General advice for a better visualisation design

Stay tuned !


The following two tabs change content below.

Jixin Jia

Jixin 'Gin' Jia (MBA, MCSE, MCP, ASA) is a certified Azure Solutions Architect. He is a MCSE in Business Intelligence and Data Management & Analytics. With a Mathematics, MBA and IT background, Jixin combines advanced analytics and modern BI technologies in building data-driven intelligence. He is a regular speaker about machine learning, quantitative analysis and Microsoft BI in many public forums. He is now an active Data Science program candidate at Harvard University.