Data looks better naked | Hacker News Reader

Data looks better naked

darkhorseanalytics.com

172 pointschethiya48 comments

Open articleView on HN

Agreed until it removed the numbers in y-axis. I want to see the scale, minimum, and maximum points in evenly spaced lines, whether logarithmic or linear space is used. This way, these details slightly disappear, or become harder to interpret.

The horizontal lines and y-axis labels were particularly helpful for comparing non-adjacent bars.

And with those numbers gone, there's more room to mislead.

The original source with blog post: http://darkhorseanalytics.com/blog/data-looks-better-naked/

For every chart you could ever want with a very easy-to-use API, highcharts (http://www.highcharts.com/) is your best friend. Free for personal use, inexpensive for commercial license. Highcharts will let you show as much or as little as you want. It's funny the author chose calorie per food portion for the example, I'm actually developing a web app that helps plan diets and I did exactly the same thing - remove labels, special effects, legends, etc. The result is magnificent, it really helps with both design and the usefulness of your data.

Agreed. After evaluating a number of charting libraries, we purchased the "web app" license of highcharts.js and we've been pleased with its looks and ease-of-use. (in case it's not clear, we are a customer & not otherwise affiliated).

Out of topic, but counting calories don't mean squad for a diet. It is a myth. Read http://garytaubes.com/ for a good research on this.

Counting carbs is what is important. Low Carb, High Fat is what science tells you should do to loose wait. Using thermodynamics to understand our biology is useless.

EDIT for on topic add: Useless data is useless, naked or not. They should have chosen a better example.

Nutrition science is highly contested; for every report or paper that claims one thing, another will completely refute it.

That's just following Edward Tufte's maxim to minimize chart junk.

http://en.wikipedia.org/wiki/Chartjunk

Any of you on HN that aren't familiar with "The Visual Display of Quantitative Information" should give it a read (and the rest of his books).

> and the rest of his books

I find his books to be a bit too light on the content. All of them combined would've made a great book that would still be of a reasonable thickness, but he is self-published and he makes his living from lectures and book sales, so he's probably not too interested in that.

His books were written over 24 years so they idea of writing one great book is just not a practical one. He made his living as a professor not simply self-publishing, lectures and book sales. For someone who evangelizes effective presentation of information, creating a CliffsNotes or Nutshell version just completely misses the point of his work.

I can't agree. The rest of those books contain roughly the same amount of actual information as his first. I went to his workshop, in which he seemed to be showing as much restraint as he could to not talk exclusively about his lawn sculptures.

Tufte has some great ideas, but would benefit from a ruthless editor. As a guy who's self-published, he does not get the benefit of that kind of feedback.

"The Visual Display of Quantitative Information" is not a howto guide. It's not a list of 101 tips and tricks for drawing charts, where one would wish for 200 tips instead. It's a framework, a theory, of how to think and reason about visualizing information. It is something you can take and apply in many different situations. Then the question really is, is there anything wrong with the theory it conveys?

I own all of his books and have read through them several times. I see them as filling a role somewhere between a guide and a listing taken together. It's easy enough to get 95% of the practical value by just reading one of them, but the remaining 3 provide a number of examples which can help to provide a broad perspective on Tufte's ideas while inspiring your own interpretation of them.

If you just want to make better charts, buy, a from memory guess, Envisioning Information. It provides most of the basics of The Visual Display of Quantitative Information but develops the ideas a little further and provides better, more modern examples. If you think it's a cool topic then buy them all because they're all pretty nice books.

It provides steps to get from the junk to something relatively clean. Many people could benefit from following that, but software that doesn't spit out the junk in the first place would be better

They could have used horizontal bars instead of vertical. It takes up much less space and easier to place the labels. This GIF shows a similar animation, but with a 90 degree rotation and a little more information.

http://blog.bissantz.com/images/2008/01/tod_der_businessgraf...

I wonder why they add the table grid at the end. A grid could have been useful if there were a lot of columns, but certainly not in this case.

This is a similar idea I was working on http://vpj.svbtle.com/variable-length-underlining-to-help-se...

Stop using gray text on a white background.

Not sure about lightening the labels and removing bolding. Contrast and readability goes down.

The idea is to keep focus on the data. The labels are reference when exact numbers are needed, but the data trend is king.

The trend is meaningless without a reference. For example:

http://i.imgur.com/zXkkJw4.png

vs.

http://i.imgur.com/f6A84ds.png

It's the same exact data (which I made up, by the way) but the first one tells a very different story from the second.

This has nothing to do with putting labels in bold or not, unless I misunderstood?

Readability may go down for the lightened elements but the point I took is that you can actually increase contrast by recognizing it as a two-way street. I.e., de-emphasizing less important details is perhaps even more effective at making the main emphasis (in this case the relative heights of the bars) stand out than trying to highlight it directly.

Don't get me wrong--I agree that the units/what-are-we-actually-measuring, 'Calories per 100g', should be darker in the final graph (though I don't think it needed to be bold). I also think the direct-labelling step, which obviates the lightening of the y-axis and over-shoots attenuation in favor of full, bright whiteness is an improvement.

Lightening may not be the best independent step but I think it's at least illustrative as an intermediate one.

Anybody has good resources (preferably free) on creating online visualizations. Not links to the d3.js documentation but good tutorials creating visualizations you often use on the internet. Like creating your own cool dashboard and stuff like that.

Here: http://hackershelf.com/topic/visualization/. I'm sure you can find more if you fish around the "topics" section.

http://flowingdata.com/category/tutorials/

not web-centric but the best source I know of offhand

(edit to add ..)

http://chimera.labs.oreilly.com/books/1230000000345/index.ht... is a viz book that uses d3. I haven't read it but the guy who wrote it does good work

Forget documentation, look at tutorials (list here: https://github.com/mbostock/d3/wiki/Tutorials, for example look at this one: http://vogievetsky.github.io/IntroD3/).

And later - look at examples (https://github.com/mbostock/d3/wiki/Gallery or http://christopheviau.com/d3list/gallery.html). IMHO it is the best way to learn D3.js and data vis in general.

Some other links: https://delicious.com/stared/d3js

Good analysis here: http://betterposters.blogspot.co.uk/2013/09/link-roundup-for...

If you don't have the y-axis, then a) there's no point in having it as a graph and b) there's the possibility for manipulative display.

This is kind of neat from a graphical design perspective, but it doesn't actually relate minimalistic chart design to more informative analytics in any meaningful manner. A concise, well-written blog post would do much more (than a short, hasty animation) to inform people.

Actually pretty sure these are all images taken from a blog posted earlier on HN.

edit: Chaz pointed the original out.

Regrettably the blog post is just a link to a slideshow with no relevant additional information.

Got the gif from a friend. Thanks for point out the original blog post. Just wanted to share the difference between having high redundancy vs less redundancy in visualizations.

Adding words to this concise, clear, and convincing visual argument would be the equivalent of adding the chartjunk back to the graph.

Turn it sideways. Remove the bars.

Does anyone have any examples of great analytics dashboards(of any kind)? I've always been a bit unimpressed with most of them. Making the base presentation simple and clean while still allowing powerful filtering can be tricky.

My company makes analytics dashboards for Statistical Quality Control and Flavor profiling of artisan products... I think it's pretty good with its filtering options and the like. Link to our first tutorial here: https://www.youtube.com/watch?v=MvHbNcrobwc

This is old and does not really demonstrate the importance like a study would.

I wish I hadn't seen that potato chips have that much more calories than pizza.

Edward Tufte to the rescue!

"You get used to it. I don’t even see the code anymore, all I see is blonde, redhead, brunette..." -That guy who betrayed Neo for a Juicy Steak.

Takeaway: I should eat more chili dogs.

There was nothing wrong with that chart in the first place. What a pointless optimization.

Well it was butt ugly. Personally I think they went too far in minimizing the look but that's just me.

But it did start off seriously ugly, but then again, it was made to be. I guess it was something of a straw graph, no one would really make a graphic like that.

somebody got good websites on best practice visualization?

http://www.edwardtufte.com/tufte/

http://www.perceptualedge.com/

I've read books by both these authors and they were very good. Their websites cover most of the important content found in the books.

http://www.nbr-graphs.com/ got some example. Her book was also a good and easy read on understanding more into the topic.

http://www.amazon.com/Graph-Design-Mind-Stephen-Kosslyn/dp/0...

This (rather old) paper is still worth reading: Cleveland, W. S. (1984), "Graphical Perception and Graphical Methods for Analyzing Scientific Data"

http://courses.ischool.berkeley.edu/i247/f05/readings/Clevel...

See the codes or products of javascript library d3 http://d3js.org/ I learned a lot from here

Reply on news.ycombinator.com