This is related to an earlier question on code review at CV.

While 'code review' itself is understandably out of scope, I wonder about questions like this one which I asked on SO, and this question on plotting nicely.

I imagine that there is a methodological gradient that goes from theory to programming mechanics, but I'm curious where CV stops. This answer seems particularly salient.

Here are some examples of possible edge cases I thought up and the site they border on:

  1. Plotting results effectively ("data-to-ink" ratio) -- SO, TeX?
  2. Keeping data organized throughout scaling and transformations (for example, keeping track of categorical variables so they still match the transformed features) -- SO
  3. Extending existing statistical packages without creating kludge (both etiquette and methodology here) -- SO
  4. Implementation of a stochastic algorithm (is this the right way to code this, or is there a better algebraic expression for X?) -- MSE
  5. Reporting results in limited space; how to prioritize? (i.e., papers etc.) -- Academia
  6. Preparing chains of statistical analysis elements so that reports can be made mostly agnostic to input data -- SO
  7. What's the best way to store data you're going to start/continue analysis work on in <this language>? -- SO
  8. Basic checklist of "things to try" on data when exploring/classifying (CW) -- SO
  9. If I want to do X, when should I format my data like A? If I'm trying to do Y, should I transform my data to B? What's the best way to go about that in <this language>? -- SO

These are all methodological questions, and I'd like to ask some of them, in fact. I'd like these questions to be allowable in the context of review specifically, because that's really the best way to learn.

  • 3
    $\begingroup$ (1.) strikes me as clearly on-topic here. (Regarding your SO post on scatterplots w/ many points, you may be interested in this CV thread: More efficient plot functions in R when millions of points are present.) $\endgroup$ Commented Feb 2, 2014 at 4:35
  • 6
    $\begingroup$ Many of these edge cases would be acceptable here, especially if couched in a way that exhibited their general interest to a community of statisticians and data miners. The ones that definitely are not on topic here have nicely been flagged already by "<this language>": when a question is language-specific, it almost surely ought to be on SO instead of here. $\endgroup$
    – whuber Mod
    Commented Feb 2, 2014 at 4:58
  • $\begingroup$ Maybe, but I feel like if you're a statistician you're probably used to 'flat' 2D tables for most work, whereas algebraists and engineers do a lot with variable dimension 'cubic' matrices; why and what merits those approaches provide was not really addressed in the question I asked on SO (due to SO's scope). $\endgroup$ Commented Feb 2, 2014 at 11:44
  • 1
    $\begingroup$ I think the key is how much statistical knowledge is needed. I think that, among moderators, I am for a more inclusive approach here. But the guidelines apply to all. $\endgroup$
    – Peter Flom
    Commented Feb 2, 2014 at 13:29


You must log in to answer this question.

Browse other questions tagged .