How a graph tells a story -- power restoration after Hurricane Ike
One of my hobbies is the study of information graphics. It's a surprisingly interesting and powerful discipline with applications in many fields. Given how often we use (and abuse) visual communication in everyday life, I think it's important enough for it to be taught alongside good writing and speaking practices.
Unfortunately, information graphics is not taught. In fact, it's almost completely ignored. The caretakers of data visualization are commercial graphic designers, most of whom are (understandably) more interested in decorating than communicating. It's worth looking at an example to show how this does us all a disservice.
On September 13th, 2008, Hurricane Ike devastated the city of Houston's electrical system, knocking out power to over two million households and businesses. Centerpoint Energy, Houston's main electrical provider, rushed into action to repair damaged infrastructure with the goal of restoring power to "the largest number of people in the least amount of time". Centerpoint's plan involved three main stages:
1. Restore power to critical facilities such as hospitals and water treatment plants.
2. Repair central infrastructure that provided power to whole neighborhoods.
3. Repair power lines going to individual homes and blocks.
Centerpoint also sought to meet the demand for information from residents eager to hear about planning and progress. It provided outage estimates at least once per day and sometimes as often as three or four times per day. It also provided maps and a graph. While the maps were low-quality, shifting from format to format and leaving off critical information, I'd like to talk about the graph because it's a perfect example of data graphics gone bad. Here it as, as shown on September 27th:
This graph does everything wrong. Only one data point is shown -- the current number of outages. The history is represented by a straight line, which does not match the actual history. This line is steeply sloped, which gives the (false) impression that recovery was swift and constant. Finally, there is a cutesy picture of houses in the background, which provides no information and makes it harder to read the graph. Overall, it is dishonest and uninformative.
Reconstructing the data from news archives, I created an improved version, which shows the actual history:
This version is better because it tells a story rather than decorating sparse and redundant information. It also helps us to critically examine the results of Centerpoint's restoration plan. Since Centerpoint is a monopoly utility provider it is especially important that customers keep an eye on them.
Let's look at the key features of the graph. Restoration went most quickly during the first few days and slowed down after that. There was a stall at day 7 (September 20th) when Centerpoint switched over to repairing large fuses serving hundreds of customers, which required a lot of tree clearing. After this, there was a jump followed by a mostly constant rate of restoration until Centerpoint declared completion on October 2nd.
From this, we can see how Centerpoint's plan worked out. Compare the red line (a second-order trendline fit to the data) to the blue lines, which show what a linear rate of restoration would have looked like starting at days 1, 3, and 8. Power was restored faster early in the process than it was later. Another way of showing this is to look at the rate of restoration:
Further analysis might look at the effect of bringing in more line crews from around the country or compare Centerpoint to other electricity providers in the area.
Good graphs give a broad and insightful view of the data, and should be considered an indispensable tool of communication. If you'd like to learn more about data graphics, a good place to start is Edward Tufte's books. I like The Visual Display of Quantitative Information, but if you're less interested in numerical data then check out Envisioning Information.
Unfortunately, information graphics is not taught. In fact, it's almost completely ignored. The caretakers of data visualization are commercial graphic designers, most of whom are (understandably) more interested in decorating than communicating. It's worth looking at an example to show how this does us all a disservice.
On September 13th, 2008, Hurricane Ike devastated the city of Houston's electrical system, knocking out power to over two million households and businesses. Centerpoint Energy, Houston's main electrical provider, rushed into action to repair damaged infrastructure with the goal of restoring power to "the largest number of people in the least amount of time". Centerpoint's plan involved three main stages:
1. Restore power to critical facilities such as hospitals and water treatment plants.
2. Repair central infrastructure that provided power to whole neighborhoods.
3. Repair power lines going to individual homes and blocks.
Centerpoint also sought to meet the demand for information from residents eager to hear about planning and progress. It provided outage estimates at least once per day and sometimes as often as three or four times per day. It also provided maps and a graph. While the maps were low-quality, shifting from format to format and leaving off critical information, I'd like to talk about the graph because it's a perfect example of data graphics gone bad. Here it as, as shown on September 27th:
This graph does everything wrong. Only one data point is shown -- the current number of outages. The history is represented by a straight line, which does not match the actual history. This line is steeply sloped, which gives the (false) impression that recovery was swift and constant. Finally, there is a cutesy picture of houses in the background, which provides no information and makes it harder to read the graph. Overall, it is dishonest and uninformative.
Reconstructing the data from news archives, I created an improved version, which shows the actual history:
This version is better because it tells a story rather than decorating sparse and redundant information. It also helps us to critically examine the results of Centerpoint's restoration plan. Since Centerpoint is a monopoly utility provider it is especially important that customers keep an eye on them.
Let's look at the key features of the graph. Restoration went most quickly during the first few days and slowed down after that. There was a stall at day 7 (September 20th) when Centerpoint switched over to repairing large fuses serving hundreds of customers, which required a lot of tree clearing. After this, there was a jump followed by a mostly constant rate of restoration until Centerpoint declared completion on October 2nd.
From this, we can see how Centerpoint's plan worked out. Compare the red line (a second-order trendline fit to the data) to the blue lines, which show what a linear rate of restoration would have looked like starting at days 1, 3, and 8. Power was restored faster early in the process than it was later. Another way of showing this is to look at the rate of restoration:
Further analysis might look at the effect of bringing in more line crews from around the country or compare Centerpoint to other electricity providers in the area.
Good graphs give a broad and insightful view of the data, and should be considered an indispensable tool of communication. If you'd like to learn more about data graphics, a good place to start is Edward Tufte's books. I like The Visual Display of Quantitative Information, but if you're less interested in numerical data then check out Envisioning Information.