1
$\begingroup$

I'm learning to use Google Charts API in order to play around with data from UNODOC statistics. I found this excel file providing information about homicide count and rates from countries in the world in the years 2000-2013. It shows a count of homicide and rate for every country, and I'd like to produce a clear and meaningful visualization comparing the count and the rate of one random country (USA for example), in an insightful way (may be it won't be possible to provide insight, but at least I'd like to compare the percentage with the count without generating confusion).

And one more thing...

One doubt that struck me when looking at the excel table is: how is the percentage calculated? In the U.S.A., the count homicide in 2000 was 15586 per 100000 population, that would make a ~15% percentage, but instead I find a 5.5%. How was that calculated?

I hope that this is the right place to ask such questions!

$\endgroup$
  • 1
    $\begingroup$ In the excel spreadsheet it lists rates per 100,000 population, not percentages. There are more than 100,000 people in the US! A good visual solution is simply a scatterplot. Look up Hans Rosling and Gapminder for a motivating example. $\endgroup$ – Andy W Oct 14 '15 at 17:16
  • $\begingroup$ aha you're right! It was 5.5 every 100000. I didn't read that well. Thank you $\endgroup$ – Cris Cole Oct 14 '15 at 17:29
  • $\begingroup$ About the visualization... a scatterplot would display the homicide count, what about the rate? $\endgroup$ – Cris Cole Oct 14 '15 at 17:32
  • 2
    $\begingroup$ You can display the homicide count (or the population estimate) on the X axis and the homicide rate on the Y axis. This will produce a funnel like visualization, see here for an example of homicide rates in US counties. The places with smaller populations will show more variability. $\endgroup$ – Andy W Oct 14 '15 at 17:37
1
$\begingroup$

A good way to visualize count data and something like a trait is to use graduated circles (like Hans Rosling's Gapminder) to represent the count data and a color coding to represent the proportion of the trait. For your example, you could do something like take the quintiles of the pop counts (this gives you the 5 circle sizes) and you can divide the homicide rate into a discrete number of categories. You don't want a lot of different categories of colors, it can get difficult to interpret. In a tool like Tableau, you can hover over the circles to see the actual count and percent data, the circle size and color are more to give a gut feel of the impact. The circles are good because then counties that have a large land mass, but small population are not distorting your visualization. In the US, the counties in the West will dwarf your analysis if you color code the whole land mass. A tip that I think is useful is instead of using the typical quintile function that divides the pop count range in 5 equal pieces, you should use a method called Jenks Natural Breaks. If you cannot do that, then I recommend taking the log of the counts, then doing the quintiles. The pop count data has a long, long, long tail. If you don't adjust for that, you'll end up with almost everything being the same sized circle. Good luck! oh, also, when rendering the circles, make the size increase from each of the circle sizes about 4 times bigger.

$\endgroup$
  • 2
    $\begingroup$ I think this answer would really benefit from including an example image. $\endgroup$ – Silverfish Dec 2 '15 at 22:01
  • $\begingroup$ agreed, I'm looking now, will post when I find a good example. thanks! $\endgroup$ – RLB Dec 2 '15 at 22:44
  • $\begingroup$ census.gov/dataviz/visualizations/039 be sure to hit the "run" arrow so you can see this over time. Imagine if these circles had transparent colors (dk green, light green, yellow, orange, red) indicating the homicide rate and the size of the circle indicating the county population. $\endgroup$ – RLB Dec 2 '15 at 22:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.