19
$\begingroup$

I hardly see what kind of advantage a vertical bar chart could have over a horizontal one (edit: by "horizontal bar chart", I mean a chart with bars oriented horizontally). On the other hand, vertical bar charts are so prevailing that it makes me wonder if I'm just lacking imagination or experience.

enter image description here

In particular, I'm wondering:

  • If you have many categories to plot, the labels for each category on a horizontal bar chart are easier to read (e.g. you don't have to move your head to read a vertical or rotated label). Are there situations where this isn't true?

  • If you have long labels for the categories, an horizontal bar chart requires less line breaks than a vertical bar chart. Thus, a horizontal bar plot is probably easier to read. Again, are there situations where this isn't true?

These are two examples where horizontal bar charts are likely superior, but maybe there are other situations I'm not thinking of, and where vertical bar plots have an advantage. I'd be interested to learn about that. (This blog post talks about the issue, if you're interested in some visual examples and my previous examples are not clear - note that I'm not a SAS user, by the way).

A practical stake for me is work-related, i.e. automation of graph generation. Recently I've been wondering if I should make horizontal bar charts the default (as it's easier to deal with labels overlap and so on, it would save me some programming time). On the other hand, I don't want to generate incorrect or misleading data visualizations, or run into other problems I did not anticipate. Hence my question.

I'm definitely not looking for answers similar to "I find vertical bars ugly" or other opinion-based answers. I'm looking for practical, objective issues that members of an audience may encounter with horizontal bar charts, but not with vertical bars. Which makes me think that I'd also be interested in references on the subject - peer-reviewed publications based on experiments would be perfect.

Edit:

I realize that my question assumes we're dealing with graphs in English or other languages where vertical writing is not an option. It is because the practical issue that caused my question does not involve languages like Japanese or other East Asian languages where vertical writing is possible. However if someone has information about graphs in those languages, and how the problem possibly differ compared to other languages, I'd be curious to hear about it (I guess that the option to write vertically simply gives more flexibility and makes things easier relative to vertical bar plots, but as I'm not familiar at all with those languages I may be wrong). A word or two about those languages might be also directly helpful to other people than me.

By the way, thanks to all the people who answered so far, it has been interesting and helpful.

$\endgroup$
19
  • 4
    $\begingroup$ Note that "vertical" and "horizontal" refers to the orientation of the bars. What SAS calls a "vertical" bar chart could easily be called a "horizontal" one, based on it being wider than tall, and categories arranged horizontally. Can you please edit this clarification into your question? Without it, this may be misunderstood majorly unless one clicks through to the SAS blog post. $\endgroup$ Commented Mar 13 at 12:26
  • 1
    $\begingroup$ I think this varies a bit by field. But, with a small number of categories, I think it's just "this is what we're used to" along with "up is bigger" being, perhaps, a bit more intuitive than "to the right is bigger". $\endgroup$
    – Peter Flom
    Commented Mar 13 at 12:28
  • 1
    $\begingroup$ I took the freedom to edit the post and add a picture, feel free to revert (+1 by the way). $\endgroup$
    – dariober
    Commented Mar 13 at 13:01
  • 6
    $\begingroup$ Because there are objective theories about the design and perception of quantitative graphics (see Bertin, Tufte, et al.), this is not necessarily an opinion-based question, so I have reopened it. $\endgroup$
    – whuber
    Commented Mar 13 at 13:28
  • 2
    $\begingroup$ @RobbieGoodwin No, as I got interesting answers, very helpful to solve the practical problem I have described in the body text. From your comment (tell me if I misinterpreted it), I understand that you're annoyed because you find it would require a textbook to answer it, or that the question is too basic to be interesting to you. However no one forces you to answer it, and even if this question was high-school level, I wouldn't find anything wrong with asking about it. Everyone does not necessarily had the education you had access to. Hopefully this website is meant to correct this gap. $\endgroup$
    – Coris
    Commented Mar 16 at 7:56

4 Answers 4

15
$\begingroup$

If you have "many" categories, then a vertical chart will be wider than it is tall. A horizontal chart will be taller than it is wide. (Note the potential for confusion what "vertical" and "horizontal" mean, per my comment.)

Which one makes more sense depends on your situation. Mobile phone screens are usually in portrait mode and would work better with horizontal charts. Laptops or desktop computer screens are usually in landscape orientation and may work better with vertical charts. If you are writing a paper, which one makes best use of your page real estate depends very much on the number of categories.

I would assume that humans are better at comparing the length of horizontal bars than the height of vertical ones, but do not know of any research on this. I am sure someone has already looked at this.

$\endgroup$
13
  • 9
    $\begingroup$ See William Cleveland's work. As I recall, his studies did not find significant differences in perception of vertical vs. horizontal lengths. $\endgroup$
    – whuber
    Commented Mar 13 at 13:29
  • 2
    $\begingroup$ @whuber, this is really interesting (came up recently), but I couldn't easily find where he did these comparisons. Can you point me to a paper or to page/chapter references in a book? $\endgroup$
    – Ben Bolker
    Commented Mar 14 at 21:54
  • 2
    $\begingroup$ @BenBolker I'm wondering the same question . If whuber does not remember the reference, maybe that could be a new question, in case someone else has this information? $\endgroup$
    – Coris
    Commented Mar 15 at 4:20
  • 1
    $\begingroup$ @Ben William S. Cleveland (1985). The Elements of Graphing Data. See especially chapter 4, "Graphical Perception," and section 4.3, "Theory and Experimentation," summarizing Cleveland's work at Bell Labs. (I'm looking at the older first edition.) It's worth reading if only to understand the nuances of assessing graphical perception. For a more comprehensive account, MacEachren's book will teach more than you ever wanted to know: MacEachren, Alan M. 2004. How Maps Work: Representation, Visualization, and Design. Guilford Press. $\endgroup$
    – whuber
    Commented Mar 15 at 13:23
  • 1
    $\begingroup$ Hmm. Looking at the revised edition on the Internet Archive here, it doesn't look like this information is there (Chapter 4 doesn't seem to have anything at all about his experimental work). I would have to get the first edition by interlibrary loan ... I may ask another question (or @Coris might?) to get a summary of any experimental work on horizontal vs vertical layouts ... $\endgroup$
    – Ben Bolker
    Commented Mar 15 at 13:53
24
$\begingroup$

The accepted answer from @Stephan Kolassa is fine by me, but unsurprisingly more can be said.

First, let us clear out of the way a small question of definition. To some authors, bar charts are those with horizontal bars, while vertical bars define a column chart. If you think this distinction is somewhere between unnecessary and absurd, I agree with you. But it's far from dead, although I sense its use is declining. In what I see, the more common distinction is -- as in the question title -- between horizontal bar charts (bars go from left to right by default) and vertical bar charts (bars go from bottom to top by default). Neither term rules out (e.g.) pairing of bars, as in so-called population pyramids showing age distribution of population with bars for numbers of males and females laid back to back.

The leading point in my experience over whether to use horizontal or vertical bars is that designers should show text labels legibly, as mentioned in the question. All bar charts need clear explanations of the categories shown. This is usually easier with horizontal layout. With a vertical layout, some people use a legend instead of text by each bar. At best that is an awkward solution, as are some other work-arounds in common use, such as very small font sizes, text running at an angle (even vertically), or abbreviations verging on the cryptic. This comment, however, bites most, or even only, if text in the language used is customarily shown horizontally.

Conversely, some very concise abbreviations have become so widely accepted that they suit vertical bars, including chemical element symbols such as Cu and Zn and two-letter codes such as AL and AK for states of the United States.

Yet again, as @matt_black flags in his answer, some bar charts don't need labels for every bar, as is often true if bars are for a series of dates and mental interpolation is easy enough.

In addition, two points arise quite commonly.

In statistical graphics, vertical bars match a common convention that whatever is the outcome or response of interest is usually plotted vertically unless there is good reason to do otherwise.

Occasionally, plotting vertically or horizontally may seem natural for other reasons, even if pure convention or just a matter of psychology. Thus a bar chart for heights of buildings would seem very odd if presented horizontally, while durations (lengths of time) might seem better suited to horizontal layout.

$\endgroup$
4
  • 3
    $\begingroup$ +1, IMO "whatever is the outcome or response of interest is usually plotted vertically" is the main point: convention. Research would be much more painful if authors were randomly choosing axes for their independent and dependent variables. $\endgroup$
    – Sam
    Commented Mar 13 at 21:13
  • $\begingroup$ @Sam Thanks for the upvote, but I don’t fully share your concern. In the Earth and environmental sciences, and in archaeology, height above or depth below the surface often are plotted vertically, even though it is usually not an outcome. People don’t seem much troubled by this different convention. $\endgroup$
    – Nick Cox
    Commented Mar 13 at 21:19
  • $\begingroup$ @Sam I'd say that convention is a bit weaker for bar charts which by their nature cannot have continuous bins. A graph often shows a function mapping an independent variable to a dependent one, and the non-redundancy of groups in a bar chart means you can always define a function mapping a bar to value, but not necessarily the other way around. I usually read the discrete bar categories as the independent variable whether they're on the x- or y-axis, since the continuous values are always mathematically a function of them. $\endgroup$ Commented Mar 14 at 12:27
  • 3
    $\begingroup$ @NickCox If the independent variable is literally "height", that is a more than reasonable justification for a change from convention. As are arguments about the medium, number of categories, etc. In my experience, if there is no strong argument for these exceptions, one is expected to present data in a format adhering closely to the formats of previous (especially cited) research in that field. $\endgroup$
    – Sam
    Commented Mar 14 at 14:16
10
$\begingroup$

One issue is to synchronize intuition with the convention of graphing mathematical functions with the dependent variable on the y-axis, and the independent variable on the x-axis (i.e., matching a vertical bar chart). For instance, if the data is quantitative, it's easier to eyeball whether it matches some known probability distribution previously studied. Here's an example from the Weiss Introductory Statistics textbook:

Vertical bar chart with normal curve superimposed

$\endgroup$
3
  • 5
    $\begingroup$ I don’t think the task would be harder if the axes were swapped. I have published papers with altitude histograms and so a vertical magnitude axis with no dissent from anyone evident. $\endgroup$
    – Nick Cox
    Commented Mar 13 at 22:08
  • 1
    $\begingroup$ Violin plots are popular visualizations that (typically) maps the shape of the distribution to y-axis instead of x-axis, as on your image. On the other hand, ridgeline does the opposite. en.wikipedia.org/wiki/Violin_plot en.wikipedia.org/wiki/Ridgeline_plot $\endgroup$
    – Colombo
    Commented Mar 14 at 22:25
  • 1
    $\begingroup$ Violin plots or ridgelines don't in my experience show comparisons with theoretical or expected distributions, which is the point being made here, as I understand it. $\endgroup$
    – Nick Cox
    Commented Mar 15 at 9:11
3
$\begingroup$

The conventional wisdom here is based on the readability of categorical labels which are often hard to read if the text is not horizontal.

But this can sometimes be unhelpful in context where the categories from part of a regular series. For example, if the plot is of monthly performance over several years. Here, not every bar needs to be labelled with the full name of the month as the reader of the chart knows the sequence of months and the reader can interpolate the specific months easily without the labels (eg when year boundaries are labelled). Plus, for data like time series, the common charts show them from left to right so the bars may be interpreted more naturally in that orientation.

So the penalty of readability for a chart oriented this way is not a notable issue.

$\endgroup$
4
  • $\begingroup$ Indeed. Sometimes time series are shown with bars and if there are many times and so many bars, then there is usually neither space nor need to label every time point. So agreed, but that said wouldn't a line chart work better in most circumstances? People often don't feel constrained by the difference between measurements at points and over intervals. However, your point is conditional on bar charts being the chosen method. $\endgroup$
    – Nick Cox
    Commented Mar 15 at 10:28
  • $\begingroup$ @NickCox The issue for using line charts over bar charts is subtle. Bar charts make it visually clear that the measurement is not continuous. Line charts hint to the user that measurements are. Whether this distinction matters is dependent on context. Sometimes,for example, even when data could be measured daily it is only available in weekly or monthly aggregates. $\endgroup$
    – matt_black
    Commented Mar 15 at 10:39
  • $\begingroup$ I agree. Are you thinking I don't? Sorry if I am unclear. $\endgroup$
    – Nick Cox
    Commented Mar 15 at 10:44
  • $\begingroup$ @NickCox Not disagreeing, just clarifying the rationale. $\endgroup$
    – matt_black
    Commented Mar 15 at 10:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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