Consistency.

Consistency: Each series must maintain its colour, texture or, in the end, its codification. Click on the image to enlarge it. Source: graph made by the author

Consistency is important in whatever type of graphic, especially if it takes part in a presentation or a set of graphics appearing together in a panel or on the same page. It implies compliance with a rule, essentially of codification. The same colours or textures should mean the same across the representation.

• Code the bars of the same data series or corresponding elements always in the same way in what regards to scale, colours and textures. This is something that most graphic programs generating bar charts do, but remembering it doesn't harm.
  
  

• Place the bars corresponding to the same concept always in the same manner.
  
  
• Don’t emphasize some bars in detriment to others if they are of the same data series.

Spatial Contention.

Well defined groups favour the comparison between data series. Click on the image to enlarge it. Source: graph made by the author	Unorganised groups where the series of different categories seem to group Click on the image to enlarge it. Source: graph made by the author

The physical space a graphic occupies and that of the elements that make it up is important for the comprehension of the message it conveys.

• In grouped bar charts, every group has to be separated from the others to emphasise its condition of group. Here the Gestalt rule of proximity applies (see number 19 “Gestalt and Visual Momentum”).
  
  
• The bars should not exceed the extension of the scale.
  
  
• The bars of an overlapped graph should not overlap so much that they make us think that it is a stacked bar chart.

Layout.

Depending on how the elements of a graph are laid out the interpretation of the same can be enhanced or hampered.

• In stacked bar charts it’s convenient to place in the lower part the values that have a lower variation in order to obtain a more or less uniform structure that enhances the differences between the others.
  
  
• When there are several independent variables to show, chose the most important as categorical or temporal axis and treat the remaining ones as data series.

Stacked bar graphs. When placing the most uniform series in the lower part it's easier to gain insight Click on the image to enlarge it. Source: graph made by the author

Use.

• Use bar graphs to allow the user comparing specific values.
  
  
• If you have a categorical axis, then bar graphs are very web suited to represent it.
  
  
• If you have a temporal axis and you would like to see the trends of change over time don’t use a bar graph. In this case a line chart can show the trend in a better way. From a cognitive psychology standpoint, a line, even though it could be zigzagging, better illustrates the continuity of its values.
  
  
• If, on the contrary, you have a temporal axis and you are interested in comparing specific values, then the bar graph is the appropriate one. Bars convey continuity in a much worse way than lines, nevertheless they are very useful to allow the user to compare particular values.
  
  
• Vertical or horizontal bars?
  
  According to Koslyn, as a general principle, the properties of the visual pattern shouldn’t be in contradiction with the real properties of what is being represented. For example, heights are better represented with vertical bars than with horizontal ones, while lengths are better shown with horizontal ones.
  
  

  The properties of the visual pattern shouldn’t be in contradiction with the real properties of what is being represented. Click on the image to enlarge it. Source: graph made by the author

  • On the other hand it’s a widely spread convention in the western world that top is in the upper or right part and bottom is on the lower or left part of the coordinate axis.
    
    
  • In absence of other restrictions, if labels are long, it’s advisable to use vertical bars.
    
    
  • When in doubt use vertical bars for the simple reason that most cultures recognise height as a bigger value, while this does not apply for horizontal extension.
    
    
• Respect the cultural context of the target audience of your graphical representations.
  
  

• Paired Graph, age pyramid. Click on the image to enlarge it. Source: graph made by the author with data from the "Instituto Nacional de Estadistica (INE)"

  Use paired or bi-directional bar graphs to show differentiated trends amongst levels of a single variable. The classic age pyramid distinguishes the survival rate of male and female humans across the age of the individual. The independent variable is the amount of alive people in an age group. There’s a clear difference between males and females.
  
  
• Use paired graphs if the comparison between pairs of values makes sense.
  
  
• Don’t use paired graphs for more than two independent variables (grouped and paired bar graphs). The patterns that appear when you build paired graphs of several data series become quite complex and difficult to interpret.
  
  
• Use stacked bar charts to represent components of a whole in a categorical scale.

Up to here we have tried to review the most effective practices to apply when building a bar graph. We shouldn’t forget that recommendations must not constitute a corset that inhibits the creation of an excellent graphic if the application, the nature of data or the target audience make it advisable.

For this reason if the result is a graphic that makes the nature and relations among data more comprehensible, abandoning all the previous advice is the most advisable.

Most of the above mentioned best practices come from the books:

• Information Graphics, A comprehensive Illustrated Reference. Robert L.Harris Oxford University Press.
• Elements of graph design. Stephen M. Koslyn . W.H Freeman and Company NY.

Links of this issue: