Graphing Levels and Rates
Posted: Tue Nov 22, 2005 11:28 am
Posted by Dan Goldner <dan@ventanasystems.com>
It is well established that most people cannot easily perceive the relationships among an inflow, an outflow, and a stock by looking at their time series. Thus, exercises in graphical integration (e.g. Sterman, Business Dynamics, chapter 7.1.3) are an important part of SD training.
I am on a quest for a graph that not only allows quantitative comparison of levels and rates, but also makes the causality immediately obvious, even to an untrained viewer. Traditional approaches simply superpose the three time series:
http://www.ventanasystemsinc.com/private/dan/tradit.jpg
While this does show the size of the rates (over an interval) compared to the level, it falls short because the viewer must still mentally ""carry"" the net inflow up and add it to the total.
I've made three attempts to combine inflow, outflow, and stock time series in a single design that shows what is happening. My first attempt:
http://www.ventanasystemsinc.com/privat ... tempt1.jpg
highlights turnover or churn, but with the drawback of spatially separating inflow and outflow so they cannot be easily compared to one another. The second attempt:
http://www.ventanasystemsinc.com/privat ... tempt2.jpg
uses sloped whiskers to indicate what the total would have been at each point, if there were only an inflow (top whiskers), or only an outflow (bottom whiskers). While this version properly indicates all time rates of change by slopes, and makes them comparable to each other as well as to the level, it is creepily insect-like. It is also redundant: once two of the inflow, outflow, and level are graphed, the third is implied. The last attempt:
http://www.ventanasystemsinc.com/privat ... tempt3.jpg
removes that redundancy by plotting just the level and the outflow. The inflow is indicated by the vertical distance from the end of the whisker to the point above it, and 'remaining' is separated from 'new' by the ends of the whiskers. This is my favorite so far, but it is quite different from familiar graphs, and I am stumped for the clever legend, caption or other annotation required to explain it to the untutored viewer.
The above can be improved, or perhaps there is a better, completely different approach. I posted this question to Edward Tufte's forum on information design (http://www.edwardtufte.com/bboard/q-and ... _id=0000yO
-- scroll down to November 14, 2005)
but received few responses. It did make me the number 1 hit, if only for now, of Google searches for ""sloped whiskers"".
Can some wise reader of this list advance the cause?
Many thanks,
Dan Goldner
Posted by Dan Goldner <dan@ventanasystems.com>
posting date Mon, 21 Nov 2005 17:30:35 -0800
It is well established that most people cannot easily perceive the relationships among an inflow, an outflow, and a stock by looking at their time series. Thus, exercises in graphical integration (e.g. Sterman, Business Dynamics, chapter 7.1.3) are an important part of SD training.
I am on a quest for a graph that not only allows quantitative comparison of levels and rates, but also makes the causality immediately obvious, even to an untrained viewer. Traditional approaches simply superpose the three time series:
http://www.ventanasystemsinc.com/private/dan/tradit.jpg
While this does show the size of the rates (over an interval) compared to the level, it falls short because the viewer must still mentally ""carry"" the net inflow up and add it to the total.
I've made three attempts to combine inflow, outflow, and stock time series in a single design that shows what is happening. My first attempt:
http://www.ventanasystemsinc.com/privat ... tempt1.jpg
highlights turnover or churn, but with the drawback of spatially separating inflow and outflow so they cannot be easily compared to one another. The second attempt:
http://www.ventanasystemsinc.com/privat ... tempt2.jpg
uses sloped whiskers to indicate what the total would have been at each point, if there were only an inflow (top whiskers), or only an outflow (bottom whiskers). While this version properly indicates all time rates of change by slopes, and makes them comparable to each other as well as to the level, it is creepily insect-like. It is also redundant: once two of the inflow, outflow, and level are graphed, the third is implied. The last attempt:
http://www.ventanasystemsinc.com/privat ... tempt3.jpg
removes that redundancy by plotting just the level and the outflow. The inflow is indicated by the vertical distance from the end of the whisker to the point above it, and 'remaining' is separated from 'new' by the ends of the whiskers. This is my favorite so far, but it is quite different from familiar graphs, and I am stumped for the clever legend, caption or other annotation required to explain it to the untutored viewer.
The above can be improved, or perhaps there is a better, completely different approach. I posted this question to Edward Tufte's forum on information design (http://www.edwardtufte.com/bboard/q-and ... _id=0000yO
-- scroll down to November 14, 2005)
but received few responses. It did make me the number 1 hit, if only for now, of Google searches for ""sloped whiskers"".
Can some wise reader of this list advance the cause?
Many thanks,
Dan Goldner
Posted by Dan Goldner <dan@ventanasystems.com>
posting date Mon, 21 Nov 2005 17:30:35 -0800