The two most common ordered categorical scales that a product manager is likely to come across are the Net Promoter Scale and Likert Scale. This article will focus on graphing Likert scale data. Likert data are very common in market research, user satisfaction, opinion, and and other quantitative research. The most authoritative guidance for the best way to graph Likert data is probably from Richard Heiberger and Naomi Robbins' paper in The Journal of Statistical Software, "Design of Diverging Stacked Bar Charts for Likert Scales and Other Applications."
The diverging stacked bar chart allows the viewer to easily compare divergence in response from the "zero" or "neutral" line. Divergence between "Strongly Agree" and "Agree" is secondary in most scenarios where Likert scales are used, but divergence from "No Opinion" is paramount. For example, in the figure below, from Heiberger and Robbin's example generously published to a public Tableau gallery, respondents were asked to answer on one of five levels of agreement or disagreement to the question “Is your job professionally challenging?” It is more valuable to the viewer to discern whether or not respondents agree or disagree that their job is challenging. Discerning the difference between "Strongly Disagree" and "Disagree" is secondary. "No Opinion" is the "zero" or "neutral" line in this example.
Diverging stacked bar charts aren't the only way to graph Likert data. The reader should note that there are dissenting opinions on the subject. However, it is difficult to equivocate their usefulness when divergence from the center line is the goal of one's chart, as the above example clearly illustrates.
Comments