I did find this viz to be very compelling, but when I opened it up to understand how Joe did it, I began to wonder if the horizon graph was the best way to convey the information, especially given how much effort appeared to go into the construction of the viz. Granted, Joe is a master at this stuff and so the table calculations probably took him a couple minutes to throw together, but for mere mortals the layers of calculations might be a bit daunting. :-) That, and I don't fully understand how to read horizon charts (although I think I understand their intent).
So I took a stab at an alternative view using a simple bar chart using the absolute value of the delta (for the size) and the the native delta (for the color). From my perspective, the same insights can be gleaned from the bar chart (e.g. which states deviated most from the average and when). In addition, I added a line chart at the top that shows the overall unemployment rate (for context) as well as adding the monthly unemployment rate and average rate into the tool-tip of the bar chart so that the user has those numbers handy.
Here's my alternative viz: http://public.tableausoftware.com/views/AlternativetoUnemploymentHorizionChart/AlternativeView
I'd be interested in your thoughts.
Hey Mtmixon, your comments made me realize that I maybe did something that's not so cool. I posted a link to Joe's viz because I found it interesting, but please realize I, Shawn posted this viz without Joe's permission, and I don't think it's fair to critique a viz here unless the author has asks for the abuse/criticism. Sorry about that Joe, I'll be more careful in the future.
I didn't intend any abuse. Joe does fantastic work, and when I have the mental capacity I often like to deconstruct his vizzes so that I can learn from them. In this case, it isn't Joe's viz that I was questioning (he created a perfect horizon graph) but whether the horizon graph conveyed the intended metric clearer/better than a simpler-to-construct bar chart. Again, I'm new to the horizon graph and so may not understand how to interpret it correctly, but I got the same insights from the bar graph as I got from the horizon graph, namely which states deviated the most from the average unemployment rate and when. So I am just curious as to what I might be missing. I suspect others who are similarly unfamiliar with the horizon graph may be wondering the same thing. Or perhaps I'm alone in my confusion. :-)
Shawn, no worries, no need to ask for permission to dive deeper into an analysis, we are all in this community together , and I think you did a great job!
Mike, a few things, as for the table calc use, I created this workbook as an example of an interesting analysis that can be done with them without ETL in Tableau.
As for horizon charts being an effective chart type, see the research and comments at:
Stephen Few says: "Though it takes some getting used to at first, once you’ve learned how to read it, it works quite well."
Thank you for the complements
Thanks for the links, Joe. I had read the Few article a while ago, which is why I was familiar with horizon graphs, but I had forgotten some of the details. In re-reading it, along with the study from Berkeley, I can better appreciate the purpose of the horizon graph, but find that some of my original questions/reservations remain. In particular, I don't see anyone using a horizon graph for any real precision analysis, outside of perhaps spotting an interesting trend and then zooming in (or viewing a more detailed companion chart) to see the actual numbers. The immediate benefit of the horizon graph is the ability to view a lot of data at the same scale and quickly identify two things:
1. The extremes (e.g. which states deviated the most (either higher or lower) from the average unemployment rate).
2. The general direction of the metric for all of the items (e.g. which states tended to be consistently above or below the average unemployment rate)
I think these two primary takeaways are also possible with the bar chart I put together, even though aesthetically the horizon chart is more pleasant to look at (the shapes are smoother and the colors more muted). Also, the layering of the horizon chart allows you to more easily hover over those states/months that hug the average, whereas on the bar chart they barely register. Both of those items can be addressed quite easily, though (via color palette choices and playing with the scale to make the smaller values more visible).
So, I'm still on the fence as to whether the horizon graph is a significantly better choice for this type of data set, given the extra effort required to put it together. Still, this has been a fun exercise, and has definitely gotten me thinking about how/if I would want to employ them (or some variant) in my visualizations.
Thanks for putting this together.