To study importance of “right scale” let’s see the following graph which is from popular currency exchange website.
Source: www.xe.com
Now suppose you want to know the actual numeric value of right most point, we can see it is little less than half most point between 1.25 and 1.4 (i.e. little half of 1.4-1.25= 0.15), so about 0.6, now adding this to 1.25 it becomes 1.31.
The point I want to convey here is with wrong scaling techniques, it requires more of a mental work than one should actually perform to gain insights from the visualization. One common source of this problem is algorithm used by common graph rendering software to create these scales. As a designer, one should be aware of this common problem and should consider the following points so that it is easy to perceive values from the graph.
1. All intervals should be equal: This means that the quantitative distance between 2 labels should be equal because if intervals are not equal, it becomes difficult to perceive the values in the graph.
2. Scale should be power of 10 or power of 10 multiplied by 2 or 5: Power of 10 include 10 itself, 10 multiplied by itself any number of times (10*10 or 10*10*10) or 10 divided by itself any number of time (10/10 = 1, 10/100 = 0.1 etc).
Also, it is important to note that 10 multiplied by 2 or 5 is not a constraint in cases where audience thinks of the measure as occurring in groups of any particular size. For example, months (3 or 12), RAM in Gigabytes (4 or 16) etc. A scale of month in form of (0, 5, 10, 15, 20..) is less cognitively fluent than the scale (0, 3, 6, 9, 12..)
3. Scale should be anchored to zero: This does not mean that scale should include zero, instead it means that if scale was to be extended to zero, it should have one of the labels as zero. For instance if we were supposed to extend the above graph the scales in decreasing order would be (0.80, 0.65……….0.20, 0.05, -0.10, -0.25) i.e. this scale has no place for ‘zero’ label hence it is an example of bad scaling.
4. Number of intervals: There is no general rule for this but the scale should provide as many intervals needed for the precision that audience requires but not so many that the scale gets cluttered.
5. Upper and lower bounds of the scale: The general rule is that the scale should extend as little as possible above the highest value and below the lowest value while still respecting the first 3 constraints defined above.
Exceptions to rule 5: a)When using bars, the scale must always include zero, even if it results to an extended scale. b)If zero is within 2 intervals in the data, the scale should include zero.
So next time, it is better to evaluate your scale on these five points before finalizing your graph.
Caution: Above rules apply to only linear quantitative scales.
References:
http://www.perceptualedge.com/blog/?p=2378
http://www.xe.com/currencycharts/?from=USD&to=CAD&view=1D