Seth's Blog: How to make graphs that work | Hacker News Reader

Seth's Blog: How to make graphs that work

sethgodin.typepad.com

38 pointsGVRV18 comments

"First, programmers don't often have a lot of taste. The fonts [in Excel and Powerpoint] are flaccid, the defaults are wan and uninspiring. There's no sophistication."

The programmers working on the latest version of Excel are about as likely to choose the fonts as they are to design the advertising campaign for it.

Godin didn't mean "programmers", he did mean "corporate drones". But that's his target audience. So he writes "programmers" to give his audience somebody to distinguish themselves a bit, like "if you make awesome presentations you are not so un-hip like those".

"Godin didn't mean "programmers", he did mean "corporate drones"."

Well, I think you're giving him too much credit, I took his words at face value and I think that he meant exactly what he said.

If you're right, though, what exactly is his point? That you should draw your graphs by hand? Use an unpopular spreadsheet? That you should change the font etc. from the default, which is a deliberately safe choice? What makes him think that a 'marketer' can produce a more appealing graph than a 'corporate drone' graphic designer?

"If you're right, though, what exactly is his point?"

How does the overall point of the paragraph / article change when we change the job title of the people who create the templates?

Whoa - he didn't put a time scale or units on his graph are we looking at one month, one year, or 50 years?

His graph definitely looks more attractive but I would add that it's very important to put scales and units on your graph. And don't trick people by playing with the units. Make sure that they are meaningful.

This is not one of those rules you want to break.

I can think of two reasons for doing that:

1) You're trying to deceive your audience. It's one of the many ways covered in the classic, "How to lie with statistics"

2) You're presenting secret data. You're allowed to show the trends, but not any of the actual units.

Also, I thought the title on this graph was misleading. it said "Rapid Growth", and certainly, it was going up, but rapid compared to what? You always want to be asking yourself, "compared to what?" when you're looking at visualizations.

Off the top of my head, I can think of way more than "four reasons I can imagine [I] would want to show someone a graph." How about "look at this interesting pattern in the data," or "look how data set foo correlates with data set bar?" Neither of those fall under "help me figure it out", and they're both telling at least as much of a story as "everything is going great".

Also, while his graph is somewhat more pleasing to my eyes than the original, it still doesn't tell me what I really want to know about that data. I have two questions about that line: what happened with that first really big spike, and does he think that the two similarly-shaped regions that follow it represent the beginning of a cycle that we might see again. His graph is called "Rapid Traffic Growth", but I think there's a more interesting story in there.

It might be worth breaking the graph into 3 sections or somehow highlighting those interesting points.

One of his rules says "Good results should go up on the Y axis. This means that if you're charting weight loss, don't chart "how much I weigh" because good results would go down. Instead, chart "percentage of goal" or "how much I lost." "

A graph that needs to track weight loss should chart weight loss, but imo it isn't a good idea to say that good results should always go up on the y axis.

If I were interested in tracking weight, I'd rather have a graph track the actual "weight" instead of some other convoluted metric like "how much I lost". Weight going up may be a "bad" result for some users and it may be a "good" result for other users. So it wouldn't make sense for a weight-tracking for a graph to categorize it as good or bad

http://blog.CascadeSoft.net/2009/07/05/product-ui-design-why... also addresses this topic.

He left out "Read Tufte."

Or at least his disciple, Stephen Few.

I worked on some dashboard projects and have found that following some basic good guidelines from Few really improved the end results.

As a plus, the Few books were short and simple, so a minimal time investment was well-rewarded.

Probably because what he is saying in this post doesn't align well with Tufte.

The premise of Tufte, or at least his aesthetic taste, is that graphs should be as information-dense as possible. The premise of Seth's post is that the only point of a graph is to get someone to understand the emotional impact of an idea. So instead of sending someone an email telling them that sales are down 12% this quarter, you show everyone a graph with a thick red line sloping down and to the right.

That isn't a graph, it is a graphic.

Or a "chartoon", as Tufte calls them. http://www.youtube.com/watch?v=YslQ2625TR4

There's a similar emphasis on removing "chart junk." I don't have the books handy to confirm, but I don't think Tufte is entirely about information density, especially because in some cases, there just isn't density available. The case study about the shuttle disaster in one of the books was very much about presenting just the right information and communicating the "emotional impact of an idea."

These suggestions aren't very good suggestions for making charts in general. Instead they seem to be biased for an audience which doesn't want to parse and understand your data but simply be wowed by hand-waving and feel quantitative reassurance in that belief.

John Tukey (I believe) suggested that there were two broad classes of charts: exploratory and descriptive. Exploratory (Tufte style) seek to visually describe as many variances in your data and their relations both to one another and the marginal totals. With all this at hand, a careful analysis may reveal patterns and idiosyncrasies in your data. Descriptive comes later, once you know what you want to present, and seeks to maximize the visibility of whatever detail you're highlighting while providing appropriate context. It's the kind of statistic that "lies" via omission (or, rather, selective attention) but it also leaves the hints that let a careful reader suspect a larger picture.

Godin's suggestions don't really produce either of these though. Without contextualization his graphs are simply ethos arguments wrapped up in a white lab coat.

Not to rag on him: I'm sure he knows this, too. These aren't really suggestions for good graphs but instead for good presentation of technical matters to non-technical audiences: an endeavor which is probably doomed to live as a half-lie.

He neglected to say what program to use to make the graphs. Any suggestions?

You can use all the traditional tools. Just take the time to tweak the settings to achieve something that is both unique and effective. Office 2007 has much better defaults and you can accomplish a more original look with only a few clicks, so at least do that. But if you want a graph that really rocks, you need to really tweak it.

Reply on news.ycombinator.com