It’s not poor scanning quality or Web resolution that makes the data in this figure invisible. The actual figure in the paper is this unreadable. This figure was designed in color, but was printed in black and white, rendering the thin lines between data points invisible. In addition, the figure lacks units, comprehensible labels, and a caption. At the very least, commas between the zeroes in the Y axis would have been helpful.

Lessons learned:

• Avoid background color or shading in graphs.
• Follow Tufte’s Rule: Minimize the ratio of ink to data.
• Choose high contrast elements to make lines or other relationships between different subsets of data distinguishable in black and white.
• Clearly label important data points, axes, and other elements of the figure.

Graphics programs make it easy to create all sorts of cool graphs. But the 3D perspective in this graph not only irritates the eyes, but also makes the data difficult to interpret. More importantly, the graph itself does little to help illuminate the near-zero time data, which comprises a large portion of the results reported. A log-plot would have helped resolve the full range of the data obtained.

Lessons learned:

• Pick the right graph for the job.
• Every graph should have a purpose. Think about what your reader needs to see in order to understand what you observed and what that led you to conclude.
• Avoid 3D effects, fill colors, and other bells and whistles, which may ultimately confuse or distract the reader.

Many of the algorithms that you will study in these reports accomplish their tasks very quickly. A log-scale plot can help you determine near-zero differences in performance. The two figures shown here are taken from the same paper. The team first shows the execution time vs the number of edges on a regular scale, which reveals general performance of the algorithms. The team then shows a log-scale plot, which resolves the data clustered uselessly at the bottom-left of the first figure.

This team incorporated informative axis labels, a comprehensive key to the lines displayed, and descriptive captions (not shown). While the use of color is pleasing, many of the colors have similar hues that can be difficult to distinguish. And were this figure to be presented in black and white, it would be incomprehensible.

Lessons learned:

• Log-scale plot can resolve clusters of near-zero data.
• Labels help readers see your point by simply scanning the graph.
• Color can be helpful, but use with caution.
  

If you think this page of charts looks difficult to read here, it isn’t any easier to read on an 8 1/2 x 11 sheet. The axis labels are small, the volume of zeroes on the Y axis makes the numbers hard to decipher, and no captions are included to help guide the reader. Moreover, most of the pages in this report contain four or more figures. Your reader isn’t interested everything you observed, just a representative sample.

Lessons learned:

• Less is more. Show only the most relevant, representative data to support the points you need to make.
• Make your figures large enough that they can be easily scanned.
• Always include captions and refer to your figures in the text.

Ea - BEFORE

Sometimes, a little more thought to the point you are trying to communicate can improve your data display. In Exhibit Ea, the original table from the student report is on the left. In the discussion about this table, the students point out the Floyd-Warshall algorithm varies only with the number of vertices, while Bellman-Ford varies with both vertices and edges. This variation is difficult to see in their table, though.

The example above (right) shows one way to reorganize the table structure to make the variation easier to spot. The reader only needs to process two rows of data. The column organization makes it easier to see what’s happening as edges and vertices change. Aligning decimal points also aids comparison.

Eb