Graphs and tables, tables and graphs

Jesse Wolfhagen writes:

I was surprised to see a reference to you in a Quartz opinion piece entitled “Stop making charts when a table is better”. While the piece itself makes that case that there are many kinds of charts that are simply restatements of tabular data, I was surprised that you came up as an advocate of tables being a “more honest” way to present information. It seems hard to see downstream effects by looking solely at tables.
So then I looked at the link, which led me to your blog post from 2009. Specifically, from April 1, 2009. Yes, like any good satire, your post was taken at face value!
So while yes, there are reasons for tables and reasons for charts and misuses of both formats, I might humbly suggest you put a tag on your future April 1 posts (on, say April 2), because it’s the internet age: satire and close inspection of dates are dead, but text searching and confirmation bias are alive and well.

Yup.  For anyone who has further interest in the particular topic of tables and graphs, I recommend this paper from 2011, which begins:

The statistical community is divided when it comes to graphical methods and models. Graphics researchers tend to disparage models and to focus on direct representations of data, mediated perhaps by research on perceptions but certainly not by probability distributions. From the other side, modelers tend to think of graphics as a cute toy for exploring raw data but not much help when it comes to the serious business of modeling. In order to better understand the benefits and limitations of graphs in statistical analysis, this article presents a series of criticisms of graphical methods in the voice of a hypothetical old-school analytical statistician or social scientist. We hope to elicit elaborations and extensions of these and other arguments on the limitations of graphics, along with responses from graphical researchers who might have different perceptions of these issues.

Still relevant seven years later, I think.

14 thoughts on “Graphs and tables, tables and graphs

  1. Tables have the advantage of being *much* more easily machine-readable and parsable, which creates possibilities for both massive meta-analysis, databasing, and even machine-driven inference that probably — especially going forward — dwarf the relevance of any particular human getting a little bit more out of the figure than the table, in one study (especially since that one study is probably a type M error).

    • We live in the 21st century, this shouldn’t be either/or: every paper should have fully machine-readable data/results tables provided *in addition* to figures (but humans shouldn’t be forced to look at them!)

    • Rick:

      Raw data should be published, and all code should be runnable so all graphs can be reproduced. I don’t see the point of tables, given that the machines should be reading the raw data, not various summaries.

  2. There may also be individual differences in preferences for graphs or tables — possibly just what one is used to, or possibly more inherent individual differences in how readily different people’s brains take in information in different forms.

    • Today, with more and more graphs consumed online instead of print there’s also possibilities of integration. i.e. What looks like a graph but with the ability to drill down on data points by selection that leads to the table.

      I feel this is the best of both worlds. Eventually, once we are rid of the shackles of print, lot of unconventional options open up.

  3. @Martha: I’ve found A.S.C. Ehrenberg’s ideas on table design useful, for those times and people that seem to call for tables. You’re probably familiar with them (e.g., “Rudiments of Numeracy at http://www1.maths.leeds.ac.uk/~sta6ajb/math1910/p4.pdf).

    @Rick G.: I agree with the “both/and” sentiment. I see two general problems with a tables-only approach: data sets seem to be getting generally larger, and the prospect of pulling a few tens or hundreds of thousands of rows out of a printed or PDF publication doesn’t sound like fun, and following something like Ehrenberg’s guidelines for tables (e.g., round to two significant digits) renders tables less than useful for meta-analysis. In the above article, Ehrenberg notes that “… no information need be completely lost by rounding. The two-digit
    rule is a guideline for statistical working tables and the final presentation of results, not necessarily for basic data records. One can put the more precise data in an appendix or, better still, in a filing cabinet or other data bank just in case somebody should want them sometime.”

  4. Oh no!!!! I am the author of that Quartz article. I stand by the majority of that article, but I screwed in my use of Andrew’s comments. What a dunce I am! Will have to figure out how to appropriately edit.

  5. I have recently been *made* (by an insisting editor) putting the table into the main text alongside the graph that had the same information. I tried insisting on the graph in the text (for human readers) and the table in the appendix (for the meta-analyses etc.). The editor and one of the reviewer insisted on being more accustomed to tables and thus finding them easier to read (sic.) than tables. It might still take a while…

  6. Graphs are definitely better when feasible. But I think that tables are still useful sometimes, for situations where data visualization is in more than 3 dimensions. Multiple regression comes to mind.

Leave a Reply to Rick G Cancel reply

Your email address will not be published. Required fields are marked *