The 3 Keys to Effective Dashboard Design
Good dashboard design boils down to 3 factors:
We need to include the right metrics for the right audience. A lot of statisticians and data scientists go wrong here by trying to get too fancy with the metrics used. Try to use universally understandable metrics (such as averages, proportions and counts) rather than less commonly used measures (such as standard deviations and interquartile ranges).
We need to know when to use a chart versus a table versus ‘impact metrics’. If we have decided to use charts, we need to choose the right charts for the job. Chart selection should be based on the science of vision and perception (see e.g. Cleveland and McGill, 1984), and not based on what looks aesthetically pleasing at the time. Once we have selected our charts based on what we know about human vision, we need to design these charts well, free of chart junk and with high data-ink ratios (see Edward Tufte’s work for more).
Once we have decided on our metrics and charts, it is time to arrange these items appropriately in an appropriate spatial manner. Eye tracking studies by Tableau suggest that humans read dashboards in a “Z” pattern, reading across the top of the dashboard from left to right first. Our most important metrics or graphs should be positioned across the top of the dashboard, with the most important numbers in the upper leftmost corner. We should then implement appropriate interactivity into our dashboard via tooltips, filters and highlighting effects (you will be out of luck here if you are working in Excel. Think about upgrading to Power BI or Tableau). Where possible, the dashboard should be dynamically linked between the various graphical/tabular components, so that when we filter to a subset in one graph/table, the other graphs/tables will dynamically adjust also.