Tufte, you need to give me a little bit more...

March 1, 2011 4:49 PM

How should I display this data? Chlamydia screening rates among 20 clinics, cut a few different ways. Big picture question: resources for really good charts and graphs?

Yes, I went to a Tufte class today.

So it got me thinking: how can I best visualize this data? I am looking at chlamydia screening of sexually active young women: twenty clinics, patients split between internal medicine and pediatrics PCPs, and patients included either by having a hormonal birth control prescription or some other reason. I want to get my audience discussing the variation and thinking about ways to improve.

I know I could do this with a bunch of different displays (a graph comparing the rates of the different clinics, a page comparing the clinics pediatrics only, etc etc) but I wonder if there is something more clever.

Big picture, any books or resources for excellent charts and graphs? I have the Tufte books right here, but I want more examples. Bonus points if they display healthcare data. Even suggestions on terms to search on Google Images would be much appreciated.

posted by teragram to health & fitness (7 answers total)

A good first place to start is Nathan Yau's Flowing Data. The links there and the discussions there are right up your alley. The search terms you're looking for are "data visualization" and "visual display of quantitative data" or "VDQI." There is a programminbg language called Processing though I do not have personal experience with it. O'Reilly also publishes some books on this topic. Finally, you might look around the sites of GGobi, ggplot, and Colin Ware.
posted by proj at 5:12 PM on March 1, 2011

Tableau Software is a popular way of visualizing data, it was formed by a bunch of folk from Stanford working on visualizations a few years ago, plus a really well-known visualization researcher (Jock Mackinlay).

If you're able to do some programming, prefuse and Flare are toolkits designed for visualizations. I've seen students use processing to create some slick visualizations too.

This paper on the Visualization Zoo is probably the best intro to interactive visualizations that I've seen.

Lastly, keep in mind that Tufte is a good start, but his approach is by no means comprehensive or even fully agreed-upon in the community. He focuses a lot on designing from first principles (which is good) but he doesn't have a lot to say about interactive visualizations, nor are his principles fully derived from empirical approaches.

Good luck!
posted by jasonhong at 5:13 PM on March 1, 2011

One interesting (free) visualization tool I've seen for categorical data is Parallel Sets. It would allow you to explore the various combinations of categories relative to one another.
posted by i love cheese at 5:15 PM on March 1, 2011

Informationisbeautiful.net
posted by Sparx at 5:26 PM on March 1, 2011

Just one word: Scatterplots.

You can use color to display a third dimension.
posted by demiurge at 6:17 PM on March 1, 2011

I think the most important part of data visualization (coming from someone who's been known to do some, e.g.) is that you're telling a story. You need to look at your data and find the way that best represents the story you're trying to tell.

I want to get my audience discussing the variation and thinking about ways to improve.

It really really is going to depend on what your data itself looks like. I think the more specific of a point you're trying to make, the easier this is going to be on you, because you won't be left with so many options about what to do.

First off, you're definitely going to want to group your clinics. The fewer the groups the better, always simplify simplify simplify. Along with that, showing fewer dimensions at a time is also easier to understand. Break it into a couple charts if you have to - your aim is to have someone understand what the story is at a very brief glance, then have the numbers act as backups if you can hold their eyeballs long enough.

Secret tip: when you group things, look at both medians and averages before figuring out what you want to display.

And since scatterplots were mentioned - I don't think they're very easily understood at a glance. I think bars are #1 in terms of ease of understanding.

I have a few decent books about data viz, but in the end Tufte is your go-to guy. Surfing Flowing Data is helpful for inspiration but can also be a little demoralizing because of how badassedly awesome a lot of those folks are. Feel free to memail me if you want to run ideas or drafts by someone.
posted by soma lkzx at 8:56 PM on March 1, 2011

I like the figures and charts that the Dartmouth Atlas of Healthcare pulls together (in the PDF atlas, as well as the website itself); that's pretty much all about variation in health care use and what is driving it. I think one thing they're particularly good at is showing the relationship between different units of analysis - in other words, some states have higher Medicare spending than other states, but that's due to the choices made by individual hospitals and physicians within those states.

See, for instance, the plot on page 56 of the PDF (Ratios of Hospitalization Rates for Hip, Ankle, and Forearm Fractures)--it's very effective at telling the story about variation in standard care for different types of broken bones, even as it's not a terribly flashy chart. I could see something similar to that (although not standardized and centered to 1.0 for the average) as being very effective for showing how the average screening rates are affected by individual practitioner behavior. If you see one clinic with a 50% rate and another with a 70% rate, is it that practitioners in each clinic are tightly clustered at those means, or is it that the clinic with the higher rate just has a few high-performing docs? (You could color-code the dots representing clinician rates into red for IM and blue for pediatrics to add in another layer of the analysis.)
posted by iminurmefi at 9:11 AM on March 2, 2011