Visualization Best Practices for Einstein Analytics

 

3 Simple tricks to make your Einstein Analytics Visualizations more meaningful.

 

The old adage “a picture is worth more than a thousand words” is true but if the right visual representation is not used, it becomes misleading and difficult to interpret. The following use cases explain some misrepresentation scenarios and then suggests solutions to fix them in Einstein Analytics.

Displaying missing values and values for incomplete time periods in a Time-series chart

Most visualization tools just ignore the missing value by joining the data points corresponding to the neighbors of the missing value. But this gives an impression that nothing is wrong with the data which can actually be false.

In the example below, data for the year 2012 is missing. The default Timeline chart joins the data points for 2011 and 2013. [Image 1] The dashboard viewer can easily miss the absence of the data point for 2012. And if “Show Points” option for Timeline Chart is disabled, then it would make it almost impossible for the viewer to realize that there is a missing value. [Image 2]

Image 1: Timeline Chart with missing value for 2012. Data points for 2011 and 2013 are joined.

Image 2: Timeline Chart with missing value for 2012. Timeline Chart Properties: Show points = Disabled

Best practice to help with this situation would be to show missing value as a gap in the timeline chart. Doing so makes it clear that data for 2012 is missing. [Image 3]

Image 3: Timeline Chart with missing value for 2012. Timeline Chart Properties: Handling Missing Values=Gaps in Graph

Displaying Values for Incomplete Time Periods

When it comes to incomplete time periods, the values are generally lesser compared to the values of previous time periods. This can give an incorrect impression to the dashboard viewer and it is very important to pass on the fact to the viewer that the data value is for an incomplete period. Here’s a good way in Einstein Analytics to do the same. [Image 4]

Image 4: Timeline Chart indicating a data value for an incomplete time period.

Visualizing Wide Variations in Data

While visualizing data, many times we come across situations when the difference between the highest and lowest value of a grouping is very high. For example: In the column chart below, [Image 5] the data values for ‘Water’, ‘Buttermilk’ and ‘Wine’ are very small compared to ‘Coffee’. This makes the size of the bar for the former 3 very small which in turn makes it difficult to compare among them and also very hard to click on the bar for the dashboard to facet.

 

Image 5: Column Chart showing wide variations in highest and lowest data values.

A good solution to this problem is to show 2 graphs. The first one would be the same as the above image and would show the complete picture. The second graph would just zoom in on the smaller data values. Both these graphs could be placed adjacent to each other to showcase the data more effectively.

Using this technique would make it easier for the dashboard viewer to compare smaller data values and also click on any of them in the dashboard to facet. Below is a demonstration of the solution explained. [Image 6]

Image 6: Two Graph solution for a Column Chart with wide data variations.

Width of a Bar in a Bar Chart – How does it matter?

A bar graph encodes values by showing variations in its length. The width of the bar or ratio of length to width should be such that it makes it easier to look at the length of the bars, it’s values, and compare them. Excessive width distracts attention from the lengths of bars, wastes space, and simply does not look good. Two graphs with bars of the same length do not seem the same if the bars in one are much wider than the bars in the other.

The below visualizations showcase these differences: [Image 7] [Image 8]

Note: In the visualizations below, width refers to the vertical thickness of the bar (on Y-axis) and length is shown on the X-axis.

Image 7: Bar Chart Properties: Auto Fit = Fit leading to thicker bars which seem to distract attention from length of the bar.

Image 8: Bar Chart Properties: Auto Fit = None

Einstein Analytics Considerations for Bar Width:

None: The width of the bar remains fixed irrespective of the number of bars and the space assigned to the widget.

Fit: All the bars will fit in the allocated widget space. If there are many values for the dimension, each bar would become very thin but would still fit. If there are very few values for the dimension, the width of the bar would increase and so will the spacing between them.

Preserve Labels: The name of the grouping value will always be shown. If there are few values for the available space, the bar width would increase and so will the spacing between bars. If there are many values for the dimension, a scroll bar will be added after reaching a point beyond which the bar width cannot be decreased without affecting the grouping value names.

Conclusion:

Using the above discussed tricks will help send out the right message to your dashboard viewer and can add a lot of gravitas to an analysis. Clearly a nice arrow to have in a Data professional’s quiver! If you found this post interesting or helpful, please share it with your network. Happy Reading.

References:

Few, Stephen. (Q1′ 15, Q1 ’16 and Q3 ’16). Perceptual Edge Visual Business Intelligence Newsletter. Retrieved May, 2017