Nice analysis of racial composition of police forces

The Washington Post has a good idea. Using Census data, they computed the proportion of police force who are white and the corresponding proportion of citizens who are white, in different cities.

In the following scatter plot, they singled out North Charleston, SC where the police force is 85% white but the citizens are only 40% white: (Link to the interactive chart.)

This plot itself is well done, with helpful coloring and labels.

One must be careful about "story time": it's easy to infer from the graph that blue dots mean worse racial tension but that interpretation requires an assumption not proven in the data. (What is missing is the correlation between this data and some other data measuring tension.)

The secret to reading this chart is to look at the slopes of lines from the origin to each point. Above the 45-degree diagonal separating the blue dots from the gray are the cities where the police is more white than the people. The steeper the line to the origin, the more unrepresentative. Once you pass the 45-degree line, do the reverse.

The slope is really the metric of X police per Y residents. So the two dimensions can be collapsed into one. With the one dimension, I'd try a histogram view. If you find the data, let me know. Or just post it to the comments.

Comments

[Michael Schettler]

Hi Kaiser;
The chart buries the lede.
I think the untold story in this data is that the majority of the US population falls in the dark blue section - people of colour under-represented on their local police forces. The "cities" in the light blue zone mostly consist of small towns and cities that one has never heard of.
Maybe a scaled bubble chart with some arbitrary (higher) absolute population cut-off, or if we want to know we could compare three charts, Small, Medium and Large populations.
Thoughts?
Michael

[Jack Hughes]

I think Michael raises a good point that I read as "are the percentages significant for all data points?". However, the Vis uses the term cities, rather than towns, which would imply a higher minimum population to be included.

[junkcharts]

Michael: I do think the histogram view might highlight their lede more.

As to whether to use cities or populations as the unit of analysis, it's an analyst's choice. If you choose populations, then you might as well forget about the majority of smaller cities and towns because they won't matter in the scheme of things. This is exactly the same issue as using a standard map or a cartogram. The Washington Post is doing the same thing that the New York Times did in the tile map.

Bubble charts are difficult things. I suspect that you will have a lot of trouble because the large bubbles will overwhelm the little ones. You'd have to sketch it to see though.

[Chris Pudney]

Here are the data:
http://www.washingtonpost.com/wp-srv/special/national/police-staffing/data/staffing-data.js

[qning]

I created a new viz with a few changes:

https://public.tableau.com/shared/FZ6K75ZW6?:display_count=yes

Size of dots correspond to total city population.
Slider along top to filter city size.
Clickable list of cities to highlight selected city.

(thanks for the link to the data, Chris - I was getting it from census.gov!)

[Michael Schettler]

Hi qning;
Beautiful interactive chart. I had to slide down to about 380,000 population to find the first blue point, Atlanta Georgia (I assume that this is Metro Atlanta). At this population level there are more than 40 orange points.
Said differently, there are NO cities above 385,000 where the police are less white than the population. Pretty powerful conclusion.
Michael