A bar chase plot is an animated plot of a ranked bar plot over time, usually with horizontal bars. As the ranking changes over time, the bars change positions thus, the animation shows a 'race to the top'.
I first encountered such a graph in a BBC graphic about COVID-19 cases (at bottom of page). I've reproduced it with country-assigned colors using the daily data compiled by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University.
The animation is seemingly appealing as a way to convey the changes in, say the top 10 categories over time where each of them seems to be racing for the top.
Because it is a visualization, it forces the reader to use visual clues to extract information, e.g. to answer the basic question "What is this gif showing?".
Here is an example:
Each image in the gif dislays the bar values of the top 10 of some categories over time, here for the COVID-19 data, the categories are countries or US states. For all the daily timeseries in the data, these values are positive. Over time some categories will disappear from the top 10 ranking, while others will stay in the top 10 for many days. When this occurs, the data is increasing. Yet, some bars will shrink. This leads to a complete disconnect viz the numbers (bar lengths) that are displayed: the visual clue contradicts the data.
As a producer (coder) of such a report, this 'strangeness' is quite understandable: it is the ranking itself that is visualized. Yet as a consumer (reader), it is completely psychotic as there is no resolution possible between the visual clues and the numerical information that is displayed. Therefore, this visualization SHOULD NOT be used as is.
Since the point of this gif is to show rankings over time, that is all it should display: adding values will lead to the wierdness described above. Thus, a solution is to only show the categories (names) switching position in the top 10 slots.
The caveat about the appropriateness of such a 'race to the top' would remain, though: it's probably fine in the conext of games or sports data, but not for pandemic tolls!
See this notebook.
- Requirements
Python=3.6.7
IPython=7.13.0
pillow=7.0.0
matplotlib=3.1.1
Numpy=1.18.1
Pandas=0.24.2
- Using a combination of these updated versions produces a "degraded gif" where the text and ticklabels of the axes are displayed in an apparent lower resolution:
Python=3.7.6
IPython=7.16.1
pillow=7.2.0
matplotlib=3.3.1
I have not yet determined which library is the 'culprit', my guess is either pillow or matplotlib.
- This is an example of the degradation: degraded gif
