When displaying a single variable, you should use a single dimension. That is, a point on an axis. Among multiple points, the distance of the point from the origin is what is being compared. To make this explicit, a line can be drawn from the origin to the point. With a linear scale, a line with half the length represents half the value.
To make these lines more visually salient, they are often made into bars. As long as the bars have equal width, the areas of the bars are still in the same proportion as the simple lines.
Many chart engines allow for 3D Bars for their visual appeal. Since each rectangular solid has equal depth, the volumes are in the same proportion as the bars and lines.
The same idea can be used in bubble charts. A variable can be represented by the size of the point. In visual comparison, this variable should be the area of the point. Some engines mistakenly tie the variable to the radius of the circle. It is hard to tell if a point is exactly half the area of another point, but in terms of visual salience, it works.
Now, look at the bubbles used by Advizor Analyst/X in:
Multivariate analysis using parallel coordinates.
The "bubbles" are shaded to look 3D, that is like spheres. Should we compare the volumes of the spheres? Unfortunately, the volume of a sphere is not linearly proportional to the area of its 2D projection (circle). In fact the volume of a sphere is r*4/3 times the area of the circle. So, comparison of relative value is skewed.
Visual appeal in a chart is nice to have, but not at the expense of the information it represents.
I think the original graph of postage stamp prices is fine. The x-axis uses regular intervals, and the points demarcate the actual known values. I agree that a step/bar graph might be preferable for some purposes, but if you want to see if the rise in stamp prices is in line with inflation, the line is better. If a step graph is used, the trend line should connect the midpoints of the bars (see my version, which includes a CPI line). In effect this spreads out the changes as though they were more continuous, and the total area under the graphs would be about the same.
However, the example of households with computers and internet access has many problems. The original is missing data, but the x-axis it uses is categorical. Instead, it should use a continuous scale so that the gaps are apparent. Like the first one, it uses both points and lines. The points indicate the known data, and the lines help you to interpolate what the missing values might be. Another problem with it is that it seems to have a different purpose. The title of this chart says, "In 2003, more than 88% of households owning a computer were online, up 40% from 1997." To arrive at this fact requires dividing the Internet Access number by the Presence of Computer number. Instead, why not just graph this ratio on the chart? That's what I've done in the second chart below. Notice how this allows you to also see that a greater percentage of computer households had internet access in 2001 than in 2003.
Drawing points and lines. This is the most common and, in my opinion, the most intuitive. Line graphs always do some amount of interpolation, otherwise it's really a point graph.
Posts
