One of my daily read feeds is a list of the day’s government statistical releases. Today, I spotted a release on the Operation of police powers under the Terrorism Act 2000, quarterly update to September 2015, which included an annes on Arrests and outcomes, year ending September 2015:
I tweeted a link to doc, and Michael/@fantasticlife replied with a comment it might look interesting as a Sankey diagram…
Hmm…
So here’s a quick sketch generated using SankeyMATIC:
I took the liberty of adding an extra “InSystem’ step into the chart to account for the feedback look of the bailed arrests.
Here’s the data I used:
Arrested [192] InSystem Arrested [115] Released without charge Arrested [8] Alternative action InSystem [124] Charged InSystem [68] Released on bail Charged [111] Terrorism Related Charged [13] Non-terrorism related Terrorism Related [36] Prosecuted.t Terrorism Related [1] Not proceeded against Terrorism Related [74] Awaiting prosecution Non-terrorism related [6] Prosecuted.n Non-terrorism related [2] Not proceeded against Non-terrorism related [5] Awaiting prosecution Prosecuted.t [33] Convicted (terrorism related) Prosecuted.t [2] Convicted (non-terrorism related) Prosecuted.t [1] Acquitted Prosecuted.n [5] Convicted (non-terrorism related) Prosecuted.n [1] Acquitted
Looking at the diagram, I find the placement of the labels quite confusing and I’m not really sure what relate to what. (The numbers, for example…) It would also be neater if we could capture flows still “in-the system”, for example by stopping the Released on bail element at the same depth as the Charged elements, and also keeping the Awaiting prosecution element short of the right hand side. (Perhaps bail and awaiting elements could be added into a “limbo” field?)
So – nice idea, but as soon as you look at it you see that a quick look at trivial sketch immediately identifies all sorts of other issues that you need to take into account to make the diagram informatively glanceable…
Hmmm…
Thinks.. SankeyMATIC is a d3.js app… it would be nice if I could drag the elements in the generator to may the diagram a bit clearer… maybe I can?
Only that’s wrong too… because the InSystem label applies to the boundary to the left, and the Bail label to the right… So we need to tweak it a bit more…
In fact, you may notice that the labels seem to be applied left and right justified according to different rules? Hmmm… Not so simple again…
How about if I take out the insterstitial value I added?
That’s perhaps a bit clearer? And all goes some way to showing how constructing a graphic is generally an iterative process, scaffolding the evolution of the diagram as you go, as you learn to see it/read it from different perspectives and tweak it to try to clarify particular communicative messages? (Which in this case, for me, was to try to tease out how far through the process various flows had got, as well as clearly identify final outcomes…)
Other things we could do to try to improve the graphic are experiment a bit more with the colour schemes. But that’s left as an exercise for the reader…;-)
OUseful.Info, the blog... 
I find it misguiding – the fact that almost half of the people are still in the system, not charged with anything yet (released on bail / awaiting prosecution) is important and not really visible. Otherwise it would look like only 1% of the people arrested actually had anything to do with terrorism. Is there a possibility to go back to the raw or historical data? I understand that it takes time to prepare court trial for the prosecution (but at the same time – there is a limit to how long this is allowed to take, right?) – so this is kind of occluding what is really going on?
@Fridolin – not sure I understand? In the last diagram the flow through the system is identified and shows folk are still in the system (i.e. they haven’t made it to the right hand margin). Re: original data – I was working purely with the data from the flow chart – I didn’t dig around to see what data tables were published.
The right hand margin is a little misleading also in that the height of the axis has been stretched cf. the left hand axis because the the flow lines have been separated a little. With a bit more control over the diagram, the vertical ordering of flows could also be tweaked, grouping flows that have passed through the system and flows still in it.