“A picture is worth a thousand words.”
Similarly, a good figure or graph displays interesting data or relationships therein in a way that is informative, accessible, non-redundant, possibly thought-provoking, and visually pleasing. There is any number of ways to render a graph or figure ineffective, for example, through a lack of labels or annotations, stacking information that should not be stacked (e.g., stacked line or bar charts), ambiguous or obscure variables (e.g., a ratio of two variables that obscures whether differences among observations stem from differences in the nominator, denominator, or both), redundancy with the accompanying text or a table (e.g., a pie chart that adds nothing to what is already in the text or a table), or visual clutter.
To obtain a visually pleasing figure one should aim for a minimalist quality. The data and its scaffolding (e.g., the grid) should be displayed with as little ink as possible without impairing the information content and its accessibility. More detailed advice and many examples can be found in Tufte’s (2001) The visual display of quantitative information.
One figure of my own making that I consider successful in these respects is an enhanced version of a figure published in Malter (2014) On the causality and cause of returns to organizational status: evidence from the grands crus classés of the Médoc, Administrative Science Quarterly. The figure below, handcoded in R, integrates and enhances figure 1 and table 1 as they appear in the article.
Chateau quality, 1991-2008: The figure shows the mean and confidence intervals of wine quality for the grands crus classés of the Médoc over the vintages 1991 to 2008. The grands crus classés of the Médoc are the 61 (of about 400) chateaux in the Médoc, Bordeaux, France, that were classified as grand crus (“great growths”) based on historic prices in 1855. The classification sorted the 61 producers into five grands cru classés. In the figure. dotted lines separate the five classes, with the first class at the top and the fifth class at the bottom.The y-axis names the chateau whose data is displayed to the right. The x-axis indicates wine quality as measured by the ratings a producer received in the Wine Spectator (a wine magazine). The data for each chateau show the point estimate of its average quality over the vintages 1991 to 2008 and its 95% confidence interval. Within each class, producers are sorted by their average quality, with the highest average quality at the top and the lowest average quality at the bottom of each class. The vertical white line dividing a grey bar shows the average quality for an entire class; the grey bar itself shows the 95% confidence interval around the class average. With this setup the figure allows an assessment of
the quality hierarchy of the producers today and whether it aligns with the classification (i.e., with the historic price hierarchy before 1855)
whether and which pairs of producers produce at a similar or different level of quality from each other,
whether the average quality of an entire class is different from the average quality of another class,
whether a producer out- or underperforms its class average and heterogeneity of quality within classes
whether a producer produces on the level of quality that would merit up- or downgrading the producer in the classification if the classification were flexible.
Overall, the graph conveys a wealth of information in an accessible and visually compelling way. Virtually every inked pixel contributes to conveying part of the information. Nonetheless, one might argue that more ink spent on the scaffolding could benefit accessibility. A second x-axis at the top, a second y-axis on the right with tick marks only, and horizontal gridlines that facilitate connecting a chateau’s data with its name would all be consistent with the minimalist approach as long as it improves the accessibility of the information contained in the figure. As with all good things, it may take experimentation to arrive at the best solution, but starting from a set of principles that give clear guidelines might help us get there more effectively and efficiently.