3 ms·

The submitted article includes one section that begins, "Ever heard of the data-to-ink ratio? If so, skip this." But please don't skip that. That is an excellen

The submitted article includes one section that begins, "Ever heard of the data-to-ink ratio? If so, skip this." But please don't skip that. That is an excellent visual illustration of how to remove "chartjunk" (Tufte's term) from a data graphic, vividly showing how much a chart can be improved by being simplified. Cute captioned photos are a dime a dozen, especially since websites for producing them for free popped up to feed some people's Facebook walls. Don't bother. Use a diagram when it really illustrates something, and write clearly the rest of the time. One of the strengths of the discussion on Hacker News is paragraphs of actual thoughtful text.

hsuresh 13y ago
Yup, i thought the animation was brilliant. Pretty much summed the blog post. The equivalent in HN is the wall of text - breaking down the message into paragraphs, and keeping the content crisp makes posts/comments a lot more readable. In fact, wouldn't data-to-ink ratio apply even for text?
jakobe 13y ago
I think that this animated GIF is actually very bad advice. 1) They remove axis labels, forcing the reader to infer the axis from context. 2) They remove the horizontal lines, making it harder to compare different values visually. 3) They remove tickmarks, so the reader no longer knows if the origin is at zero or not. 4) They add labels with 3 significant digits, which gives the reader a false sense of precision. Overall they arrive at a graphic that may be nicer to look at, but is functionally much worse than the original.
- notacoward 13y ago
  I had pretty much the same reaction. Yeah, definitely remove backgrounds and shadows and anything that's redundant. That's great, the original was awful, but they took the minimization too far. What they ended up with was missing information and IMO wasn't as nice to look at as some of the intermediate forms had been either. The one part that's really tricky is color. I actually liked the colored version a bit better, even though it provided no extra information content. Sure, if you want to use color for only one value (bacon) as a highlight that makes sense, but more generally I think a little bit of appropriate color can help maintain attention.
- ctdonath 13y ago
  Much worse? or more in line with what most readers want: simplicity? The 3 significant digits is the raw full-precision data, if you want to do more with the graph than get a brief impression. This in contrast to having to infer an approximation by following lines to a coarse axis and visually interpolating the desired values. Providing this data obviates objections #2 & #3. Don't let the tickmarks, there or not, distract you from the zero. A favored lie in statistics is to conceal the lack (or off-the-chart) of zero, betting that the reader won't notice it having been dazzled by the dramatic angles on what (in many cases) is a rather flat data set. Default assumption should be the origin IS zero; if you're hiding the zero, ask yourself if you're really trying to lie to the reader. Marking the origin may provide plausible deniability, but most readers won't notice it - so act on truth, rather than concealing the lie in plain sight. Comparison should be as natural as the context & form of the graph makes appropriate. If it's not suitably obvious, reconsider the graph type. If the reader wants more accuracy, a graph may not be the best medium - and golly there's the raw data. The context should make the axis labels obvious and redundant. If it's not already obvious, reconsider whether the title or other commentary can be reworded to make it obvious. Of course, all this depends on what you're really trying to tell the reader, and what form follows that function. If tick marks or labels or origin value is necessary, include it ... but whatever isn't necessary, drop it.

hsuresh 13y ago

Yup, i thought the animation was brilliant. Pretty much summed the blog post. The equivalent in HN is the wall of text - breaking down the message into paragraphs, and keeping the content crisp makes posts/comments a lot more readable. In fact, wouldn't data-to-ink ratio apply even for text?

jakobe 13y ago

I think that this animated GIF is actually very bad advice. 1) They remove axis labels, forcing the reader to infer the axis from context. 2) They remove the horizontal lines, making it harder to compare different values visually. 3) They remove tickmarks, so the reader no longer knows if the origin is at zero or not. 4) They add labels with 3 significant digits, which gives the reader a false sense of precision. Overall they arrive at a graphic that may be nicer to look at, but is functionally much worse than the original.

notacoward 13y ago
I had pretty much the same reaction. Yeah, definitely remove backgrounds and shadows and anything that's redundant. That's great, the original was awful, but they took the minimization too far. What they ended up with was missing information and IMO wasn't as nice to look at as some of the intermediate forms had been either. The one part that's really tricky is color. I actually liked the colored version a bit better, even though it provided no extra information content. Sure, if you want to use color for only one value (bacon) as a highlight that makes sense, but more generally I think a little bit of appropriate color can help maintain attention.
ctdonath 13y ago
Much worse? or more in line with what most readers want: simplicity? The 3 significant digits is the raw full-precision data, if you want to do more with the graph than get a brief impression. This in contrast to having to infer an approximation by following lines to a coarse axis and visually interpolating the desired values. Providing this data obviates objections #2 & #3. Don't let the tickmarks, there or not, distract you from the zero. A favored lie in statistics is to conceal the lack (or off-the-chart) of zero, betting that the reader won't notice it having been dazzled by the dramatic angles on what (in many cases) is a rather flat data set. Default assumption should be the origin IS zero; if you're hiding the zero, ask yourself if you're really trying to lie to the reader. Marking the origin may provide plausible deniability, but most readers won't notice it - so act on truth, rather than concealing the lie in plain sight. Comparison should be as natural as the context & form of the graph makes appropriate. If it's not suitably obvious, reconsider the graph type. If the reader wants more accuracy, a graph may not be the best medium - and golly there's the raw data. The context should make the axis labels obvious and redundant. If it's not already obvious, reconsider whether the title or other commentary can be reworded to make it obvious. Of course, all this depends on what you're really trying to tell the reader, and what form follows that function. If tick marks or labels or origin value is necessary, include it ... but whatever isn't necessary, drop it.