While I am all for data visualization and an avid advocate of scientific data communication among the mass population, I would like to give a gentle reminder about how data visualization can be weaponized to manipulate human perception. This is about misleading graphs.
A very talked-about graph created by Reuters which took the readers on a convoluted path, leaving them bewildered and struggling to comprehend the intended message is the one shown below.
At first glance, it feels as if the graph is trying to say that the number of gun deaths dropped hugely from 2005 after Florida enacted its ‘stand your ground’ law. This is because our eyes are used to see the y-axis starting from zero at the bottom of the graph. But, the actual information is exactly the opposite, gun deaths increased from approximately 550 to 850 from 2005 to 2007! If we look carefully, the y-axis of the graph is labeled upside down with a red area fill to create a dramatic bloody effect. The creator probably aimed for an artistic impact rather than trickery, it did lead many folks to believe that Florida became safer after the law came into effect. Get the full scoop on the graph and the stir it caused in this article- https://visualisingdata.com/2014/04/the-fine-line-between-confusion-and-deception/
There are several ways in which misleading graphs can be generated. Let me show the most popular misleading graphs and how anyone can easily identify those!
Scaling:
Look at the graph at the top created by Fox News showcasing changes in the unemployment rate during Obama’s presidency. The y-axis does not start at zero, distorting the perception of the rate drop. Additionally, the value for November is inaccurately represented. The corrected graph reveals a clear decline from 9.0% to 8.6%.
Non-zero baseline:
Starting the axis at a value greater than zero can distort the perception of changes or differences. Look at this graph used to demonstrate this issue. The graph starts at 44.5% instead of ‘0’. Visually, it gives the idea that the popularity of PS3 is 3 times higher than the Xbox 360, while the reality is the difference between the % popularity is only 49–46 = 3%.
Truncation:
Cutting off a portion of the axis creates an illusion of a larger or smaller effect. The line plot below depicts how the distances of the women’s long jump in the Olympics increased with a steep slope from 1984 to 1988, and then had quite a drop in 1992 and another drop in 2000. But, plotting the same graph without the break in the y-axis shows that the distances have not changed much from 1984 to 2000.
Misleading labels:
Using misleading labels or units that distort the interpretation of the data. The first interval of the y-axis of the bar plot below is 20 followed by 5. The graph seemingly conveys that the price of a basketball match is 4 times higher than a hockey or a baseball game. While it is true that the ticket for the basketball match is higher than the rest of the two, the unequal intervals of the y-axis are exaggerating the information.
Missing information
The following graph is based on the tuition fees and earnings of a college graduate in the U.S. The graph talks about whether it is worth taking loans for a college degree. At first glance, it seems clear to readers that the answer is a big no, that the earnings of a 4-year degree are no match to the costs of a 4-year degree. But, the graph left out the information on ‘earning without a college degree’, or ‘earning with only a high school degree’. From the U.S. census data, the earnings of the latter in 2010 were way lesser than the $45,000/year which a college graduate earned. Another obvious fact is that the earnings shown here are per-year earnings. The 4-year expenses of tuition and boarding is a one-shot expenditure, whereas, a college graduate will most likely earn for about 43 years if we assume the graduating age is 22 and the retirement age is 65. So, a graduate will make way more than the expenditure, which can not be seen from this graph, giving us very misleading information.
Pieces of a Pie Chart are not the correct sizes
The pie chart below requires no explanation. It’s a pie chart generated by Fox News to show the percent share of the supporters of the three candidates Palin, Romney, and Huckabee during the 2012 presidential run. Values of the three slices do not add up to 100%!
Another example of pie chart distortion is the use of 3D pie charts, where it is difficult to compare the sizes of the slices of the pie chart. The same analogy to the ‘exploded’ pie chart. Pie charts work best in the simplest form without any kind of visual distortion effect.
Size of icons:
The sizes of images used in Pictographs differ for the categories being graphed. For example, in the following graph for the ticket prices for different sports, the sizes of the icons of the basketball, the baseball, and the hockey puck are shown. Simply looking at the graph, the super zoomed icon of the hockey puck is making us believe that the ticket to the hockey game is twice as expensive compared to the baseball match. But, if we replace the icons with a bar plot, it appears that the ticket price of a hockey game is almost the same as the basketball match.
Another example of size distortion is if the size of the icon is expanded in both x and y-axes to display the differences, it magnifies the effects way more than the vertical expansion only.
An additional example of such pictographs is shown below. This shows how the expanded size of an icon gives away a wrong impression. The bigger-sized icon of the ice cream cones will make us believe that ice cream cones are the biggest sales of this store when the reality is, it is the fries that are the biggest sales!
Check carefully when you are looking at visuals if you are not already! Here are more examples and resources on misleading graphs-
1. http://passyworldofmathematics.com/misleading-graphs/
2. https://www.statisticshowto.com/probability-and-statistics/descriptive-statistics/misleading-graphs/
3. https://www.youtube.com/watch?v=ETbc8GIhfHo&t=11s
4. A short tour of bad graphs, C. J. SchwarzDepartment of Statistics and Actuarial Science, Simon Fraser University
https://alg.manifoldapp.org/api/proxy/ingestion_sources/339ee58b-7d40-4bc2-aa1d-0061aa89bf09
5. https://teachersinstitute.yale.edu/curriculum/units/2008/6/08.06.06/4