0

Tasked with showing the distribution of a certain data set in a different way, I wanted to try to plot a kernel density.

After seeing it however, my co-worker advised against it saying that because it's smooth: "In a histogram the observations are placed in baskets, but in the smooth kernel plot we could pick a value between the two baskets, so we're assuming values that are there that we don't have raw data on."

At the time I couldn't think of a good response (my fundamentals are shaky), but I have seen a handful of kernel density plots in the past and figured there must be a good reason for making them. After doing some research on it and reading Interpretation/use of kernel density, it seems that one does not look at any one "point" but rather the distance between to points of interest and that gives the probability.

Question

If I could go back in time, what would have been a good response? Or are there actually caveats (other than point vs AUC?)?

Further Clarifications:

  • utility is not quite as important as aesthetics for this task
  • user statistical literacy can assumed to be advanced
  • kernel type: Epanechnikov
Arash Howaida
  • 741
  • 7
  • 19
  • Could you elaborate on the intended purpose of displaying the data in this way? That would suggest criteria for evaluating the relative merits and disadvantages of the two visualization techniques. – whuber Jun 03 '19 at 13:31
  • @whuber We simply want to show the spread of the data, but since histograms look like bar charts the visual is prone to looking boring. The aim is to show the spread of the data in a more interesting way -- so that is why I suggested kernel. We are just not in consensus about how to read it. – Arash Howaida Jun 04 '19 at 00:03
  • 2
    Thank you for explaining. I am troubled by the implicit dichotomization of utility and aesthetics, especially for a statistically literate audience. Such an audience is likely to find an *insightful, effective* graphic to be stimulating rather than boring. Might I suggest first selecting an effective visualization, and then decorating it afterwards? For more about this point of view, see (especially) Tufte's discussion of chartjunk in his *Visual Display of Quantitative Information.* – whuber Jun 04 '19 at 12:00

0 Answers0