I see no other forum to raise this, so I'm raising it here.

Question: Is there any consensus or literature on the proportion of font parameters within plots?

I feel that this is a reasonable thing to expect there to be literature on, even though it's not an exact question such as solve for x : 2x + 3 = 8 or whatever. Highlighting literature on the matter was specifically said to be a suitable answer - even if said literature is subjective (I'm happy to follow the subjective advice of someone who's an expert at something).

  • 9
    $\begingroup$ Assuming this policy still stands, then literature/reference requests are on-topic. However, is "the proportion of font parameters within plots" really related to statistics, compared to, let's say... User Experience, or maybe Graphic Design? $\endgroup$
    – Andrew T.
    Commented Jun 21, 2020 at 12:51
  • 3
    $\begingroup$ @Andrew Your point is a good one: this is an "edge case" of a topic that straddles several disciplines. But isn't the relevant issue whether the topic is appropriate on CV, rather than whether it is appropriate (or even most appropriate) on some other site, given the OP has elected to post it here? $\endgroup$
    – whuber Mod
    Commented Jun 21, 2020 at 13:25
  • 2
    $\begingroup$ @whuber fair point, I only gave alternative sites in case it's really off-topic since I don't know if it's on-topic or not on here. $\endgroup$
    – Andrew T.
    Commented Jun 21, 2020 at 13:36
  • 2
    $\begingroup$ @AndrewT. yes - I would say that it is within statistics given things such as tufte's ink ratio. Basically - it's something that I think is reasonable to expect someone to have considered and written about within the context of statistics. $\endgroup$
    – baxx
    Commented Jun 21, 2020 at 15:40
  • 2
    $\begingroup$ @baxx I won't argue if the question is on-topic or not since I'm not a regular here. I just have a feeling the question might fall into the boat programming issue (in this case: just because it's used in statistics doesn't make it a question about statistics), but the mod has spoken, so... $\endgroup$
    – Andrew T.
    Commented Jun 21, 2020 at 15:46
  • 1
    $\begingroup$ @AndrewT. I'd completely disagree that this is a "boat programming" question. Unless you'd also argue for material such as Tufte to be in graphic design, I'd also disagree with that though. $\endgroup$
    – baxx
    Commented Jun 21, 2020 at 17:21
  • 4
    $\begingroup$ I can understand why you are puzzled, indeed annoyed. but the tone of this and indeed the question itself is a little disconcerting. In the original you state "Suitable Answer The answer does not have to be in Python or use the above code, that is just to illustrate the point." Indeed, the stance on CV is that code is never more than by way of example to give substance to a statistical question. A secondary objection to your question is that it is dominated by your code. I am not clear it is off-topic, despite @Tim's well argued rationale. but I doubt there is a really useful answer. $\endgroup$
    – Nick Cox
    Commented Jun 22, 2020 at 7:17
  • 1
    $\begingroup$ It's open again. $\endgroup$
    – Nick Cox
    Commented Jun 22, 2020 at 16:15
  • $\begingroup$ @NickCox i appreciate your points there - but Python is what I use and I wanted to make it as straightforward as possible for others to quickly reproduce what was in the question. I think it's reasonable to do so, given that I explicitly state that the question isn't about the code it seems to fit in with the CV stance? Personally I feel as though it's better with the code than without, as it enables things to be reproduced, removing that for the sake of people who aren't going to read doesn't seem worth it. $\endgroup$
    – baxx
    Commented Jun 23, 2020 at 14:33
  • 2
    $\begingroup$ The code does no harm and certainly it is available for anyone wishing to use it. My own feeling is that the question would be perfectly clear without the code. Even with free software that one uses (my text editor and my TeX implementation fall under that heading) it's easy to overestimate (1) how far everyone else uses what you do (2) how easy it is to run code even if it's reproducible (there are usually key details obvious to the experienced but not the learner). (All software one doesn't use is plausibly wonderful but overrated by its fans,) $\endgroup$
    – Nick Cox
    Commented Jun 23, 2020 at 15:31
  • $\begingroup$ good points, I suppose I'm so used to trying to produce a MWE that I didn't consider it actually obfuscating things wrt the core question $\endgroup$
    – baxx
    Commented Jun 23, 2020 at 18:18
  • 1
    $\begingroup$ Cross Validated != Stack Overflow. $\endgroup$
    – Nick Cox
    Commented Jun 23, 2020 at 22:15
  • $\begingroup$ @NickCox i'm aware, reproducibility is usually a pretty valuable thing though, so I'm not saying that I don't think the code should be in the post I made (or posts in general where available). I'm just saying that I didn't consider it being something that would throw someone reading it, in future I would provide more clear disclaimers. $\endgroup$
    – baxx
    Commented Jun 23, 2020 at 22:21
  • 2
    $\begingroup$ I naturally agree: reproducibility is important for coding questions. I am active on Stack Overflow, although only on Stata questions. Lack of a MWE is a major issue limiting the value of many questions. Other way round, many people post here expecting we will debug or write code for them. If a statistical question is at the heart of their question, it may be on-topic here, but usually it isn't. $\endgroup$
    – Nick Cox
    Commented Jun 23, 2020 at 22:27
  • $\begingroup$ "tufte's ink ratio" is a design and communication prinicple articulated by Ed Tufte, but which often suffers when put to the empirical test: e.g., his dot-dash plots are less legible to educated audiences than box and whisker plots used to communicate the same information. $\endgroup$
    – Alexis
    Commented Jul 16, 2020 at 17:23

2 Answers 2


Part of statistics is

The art of telling a story with [numerical] data.

The World of Statistics.

The techniques and principles of telling such stories are primarily graphical. Those who use two-dimensional graphics to represent and communicate quantitative data usually identify themselves as cartographers. Statisticians, although we might not aspire to be cartographers, must be intelligent, well-informed users of cartographic methods.

The cartography literature includes experimental results and recommendations concerning elements outside the immediate plot area. This includes titles, axis labels, comments, legends, footnotes, captions, and anything else that is integrated with the graphic and is important for interpreting it. For example, Borden Dent (Cartography, Fifth Edition) wrote

... titles, legends, and other explanatory information, like any other graphic elements, need to fit into the whole plan for the [graphic].

(Dent, Chapter 14.)

He proceeds to recommend a gradation of type sizes for various such elements. Then he turns to considerations of spacing of titles and (rather vaguely) to the placement of the elements in the overall graphic. It's all rather obvious when you read it--but then when you examine many statistical graphics, especially those produced only by statisticians, it becomes apparent that these obvious principles are not frequently observed. (Many people asking questions on CV post graphics that are so deficient in this regard that they are unreadable.)

Unlike moderator @Tim, whose opinions I respect, I do not construe this site to have a "pretty narrow scope," but view it more expansively as covering methods, principles, procedures, and concepts needed for thinking about and applying statistics. In this context the question we are discussing is not a case of the "boat programming" meme (wherein the involvement of statistics is only incidental to the question, and trivially so).

After reviewing Dent (and a few other cartographic classics) I felt that reopening the question would be appropriate and did so. Please do not read any criticism of the original closure into this action, though: the subject matter clearly is on the edge of what most of us would consider on topic on CV. That, and the apparent triviality of the question, were sufficient justification for the initial closure decision.


Since it was me who closed the question, let me share my thoughts.

First of all, my concerns were mostly pointed out by @Andrew T., i.e. choosing font parameters seems to be more related to graphic design. The appearance of the question on CrossValidated.com seems to be a case of boat programming (although I didn't know this term before). Asking about relative font size does not really differ from asking about choosing font family, or colours, for the graph, how to align it on a page, etc., so this is in a graphic design domain. In many cases if figures are submitted with a publication, a graphic designer is the person who makes appropriate adjustments so that they look nice.

Moreover, it is a broad question, where the answer would to great extent depend on personal taste ("does this look better, or that"), but also on many different factors. A figure, or any other element of a publication, or graphic design, would look different on screen, than when posted on a blog, or in a journal article, different if it were printed in a book, or in a PowerPoint presentation, different depending on whether they are printed in colour, greyscale, black and white, or depending on what would be their size (e.g. a small plot in paper vs poster at a conference), using different fonts may need adjustments in their size, etc. Because of all those factors, there will be no single answer to such question.

The question has the following acceptance criteria for the answer

The answer does not have to be in Python or use the above code, that is just to illustrate the point. If there's a way that the fontsizes could me made a function of something such as the title, and literature to support the sizing of them, that would be appreciated.

Notice that this is already handled by the defaults of the plotting software you are using. Whatever you are using, matplotlib, base R graphics, ggplot2, Excel, etc. they all have some predefined defaults that are based on what their authors considered as best practice, or what was most consistent with their aesthetic taste. The fact alone that the OP considers changing those defaults shows that there is no "one size fits all" solutions for this problem. Even more, ggplot2 alone has nine built-in themes with different visual styles, since different users may prefer different appearances of the figures.

Making readable plots should be a concern of any statistician or data scientist, since to communicate our findings we often need to present them. The reason why I closed the question was that it simply does not seem to fall into the pretty narrow scope of this site. The boundary in such cases is often hard to draw, so if the community disagrees, I won't insist on this decision.

  • $\begingroup$ I interpreted the question as asking whether there were any guidelines beyond personal preference and "looks nice." There is literature indicating that people have a harder time extracting information from "area" cues (e.g., a pie chart) vs length ones. Similarly, certain colormaps are seem worse, in terms of perceptual consistency as well as accessibility, than others. Jet, which was the Matlab default for ages, is particularly bad in that regard, so "trust the defaults" also isnt a great solution IMO. $\endgroup$ Commented Jun 25, 2020 at 13:18
  • $\begingroup$ @MattKrause I'd agree if the question was more general, but here it asked about font size to be used in title, something very specific and totally related to graphical side, rather then accessibility (unless the answer would be "use font not smaller then..."). $\endgroup$
    – Tim
    Commented Jun 25, 2020 at 13:23
  • $\begingroup$ Eh...the relative "badness" of jet isn't just related to accessibility: it's not perceptually uniform, so some 1 unit changes look bigger than others. Ditto for areas vs lengths. Other than some vague notion of "crowding", I can't think of a psychophysical/perceptual principle that would give you much guidance about the relative size of the titles vs axes, but...that's an actual answer to the question. $\endgroup$ Commented Jun 25, 2020 at 13:31

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .