I am writing a paper in computer science literature. I have a method that has two main parameters. It turns out that one of these parameters does not have a significant impact on the quality of the results, but it has a big impact on the efficiency of the method. I have a question:

  • How should I explain this in my experimental results section? Should I draw some plots and show that for different variation of this parameter the result is the same (this seems numb to draw some plots that all overlap) or I should write a paragraph and explain this phenomenon?
  • This is, afaik, known as sensitivity analysis. Surely, other papers before have had to deal with the issue of how to present the fact that the model isn't sensitive to a specific parameter? – xLeitix Nov 7 '14 at 9:58

Actually, I had a programming part in my masters thesis in which I wanted to show that some parameters do not have significant impact on the results. This is what I did which seems to be a combination of your methods:

I drew a plot with all the parameter remaining the same except one parameter that I want to study on. Even if the variation of the parameter does not show any significant change in the results, it does worth to show the results in a plot. It is not important that all the lines overlap. At least the reader sees that the results are the same. But, in a paragraph or at least one sentence following the graph I would talk about this. You may write that all the plots overlap, because of this and because of that; and talk about why such results happen. Also, in the caption of the plot, write something that directly says something about such insignificance.

  • 1
    I absolutely agree. I love seeing plots with a whole bunch of overlapping flat lines that essentially say "We hit our method with all sorts of hammers, and it didn't budge!" – jakebeal Nov 7 '14 at 13:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.