Chris Mack
Gentleman Scientist

I don’t know if things have gotten worse, or if I have just become less tolerant. In either case, I have seen far too many examples lately of a serious problem in lithography presentations and publications: the graph with no axis labels.

At first blush, one might conclude that a graph that doesn’t put any numbers on one or both axes would not be effective at conveying information. First impressions, in this case, are exactly correct. So why would an author do such a thing? Even if you haven’t been there yourself, it is very easy to imagine the situation: When sending the paper through internal approval channels, your boss, the marketing manager, or someone from legal objected to the graph. It discloses “company confidential” information. It cannot be published.

So, what to do? Leaving out the graph and its discussion will eviscerate your paper, making what is left weak, if not completely unpublishable. The “easy” solution is simply to erase the numbers from one (or both) axes, thus satisfying the demands for corporate secrecy. But is this really a solution? While the graph may appear to be informative, it probably isn’t. At best, it says “look how much I know; too bad for you that I’m not telling you what it is.” At worst, it is a deceptive attempt to pass off shoddy work as real science, or to draw conclusions that the data does not support.

There is no place for the label-less graph in a scientific publication, in lithography, or in any other field. The reason is simple: Scientific knowledge must be reproducible. Thus, a publication that purports to present scientific (or engineering) knowledge must provide sufficient information so that the results of the paper can be reproduced by others. A graph with no labels cannot be reproduced and thus, is not admissible.

Most refereed journals have adequate review processes to weed out such flimflam. But the conference presentation and associated non-refereed proceedings rely on the integrity of the author to avoid such ethical dilemmas. There are two ethical ways out of this predicament. First, the author could simply remove the graph completely from the paper. If the resulting paper is too weak to be worth publishing, it should be withdrawn. While this option is often embarrassing to the author, it shouldn’t be: There is no shame in it when your best efforts can’t prevent it.

Often, however, there is an alternative to scrapping the graph. Since the ethical obligation is to ensure the results of the paper are reproducible, changing the scales of the graph from absolute to relative values will sometimes meet the competing goals of secrecy and reproducibility. By picking and clearly identifying a baseline case, data values can then be made relative to this baseline case. If the baseline case is described sufficiently so that others can generate data for it, use of relative values becomes acceptable. (Note that this is not the same as removing the labels and claiming the axis is in “arbitrary units.”)

Of course, a graph with relative values provides less information than one with absolute values. Thus, one must still face the issue of providing a paper of sufficient quality and information content to be useful to the reader or listener. But if the principle of reproducibility
is rigorously followed, one can at least avoid the ethical dilemma of publishing a paper that only appears to be scientific.

If you have comments on this editorial–or anything else–do not hesitate to e-mail Dr. Levenson at marcl@pennwell.com or to write him at 201 San Antonio Circle, Suite 225, Mountain View, CA 94040.