I haven’t had time to respond to this discussion until now (I'm currently on my way home from South Africa), but I’ve found it interesting to follow its development without intervening. In my work, I don’t establish and follow strict rules, but instead advocate best practices that are rooted in human perception and cognition, and try to help others understand these practices deeply enough to know when to break the rules. I teach people to always start the scale of regular bar graphs (not range bar graphs) at zero because only then do bar lengths accurately reflect the values that they represent. In the case of range bars as I’ve used them to display change between two points in time, however, their lengths represent the amount of change and the positions of their ends represent the two time-series values between which the change occurs. In other words, it is the positions of the two ends of the bar that correspond to the quantitative scale. The amount of the change between these two points, represented by bar lengths, can only be interpreted in relation to the scale by subtracting one value from the other, which is not easy to do. I would never use a range bar graph to primarily feature amounts of change, but only to provide an approximate sense of those amounts.
Andrej—As a Rolf Hichert Certified Consultant, you follow a strict set of rules that are sometimes too strict in that they don’t apply to all situations. Insisting that the scales of all graphs begin at zero forbids many graphs that are not only useful but are often necessary. While it is true, as you said, that one story in a graph that has a series of values that are all close to one another in distance from a zero-based scale is that the values differ only to a small degree, it is not the only useful story. For example, in a series of tests that measure temperatures all varying between 30 and 31 degrees Celsius, a difference of a tenth of a degree might be significant, but a graph with a zero-based scale would not reveal this. Whenever we design graphs, either for our own use or to present data to others, we must first understand what must be accomplished with the data and then find a form of visualization that supports this use. When rules get in the way of useful solutions, the rules should be abandoned.
Michael—As you acknowledged in your recent posts, you’re finding it difficult to see bars used in ways that don’t conform to a standard bar graph. Bars are just rectangles that we use to encode values in graphs for various purposes. We shouldn’t limit their use to one form of graph any more than we should limit the use of discrete data points to a scatter plot.