I work for a school district and when we wanted to show individual student entrance and exit enrollment events during a school year we settled on the Sankey Diagram! Power BI made it pretty easy!
Brilliant! In terms of labeling, I’d type the %s (what percent of the school is contained in that bar?) written at the _end points_ of each flow-bar. That would show the conditional probabilities directly. Also total student body sizes written next to the schools at the outer edges of the diagrams. Thank you!
If you add a third dimension you could also represent time. And see total students grow and shrink, and see the transitions grow and shrink. It might look a bit cluttered about.
Interesting idea! Perhaps a 'Sankey-Chain' would be better way to view the time-dependent version of the plot. Each instance of the chain will be a version of the plot at each 'timestep'... just saying.
As much as I like a good Sankey, when I try to use it for bonus points it usually ends up being super complex to build and disappointingly close to looking like a bowl of spaghetti 😅
Ah interesting new diagram... but, I am finding these Sankey Diagrams (at least so far) very difficult to follow. I find them incredibly confusing and, since I can't possibly "hover" interactively, I would be at a huge loss as a "viewer". Since you are the "presenter", it is easy for you to hover over each spot and show the tool tip. But as a viewer, I am sort of forced to constantly focus on your current position in order to follow ... that means I can't take "notes"... I have to have eyes on every second -- whereas, a more traditional diagram is more easily digested visually, and typically contains visible labels along the axis traditional XY axis. Maybe you can add persistent labels and numbers? I would be hard pressed to believe that you can provide more, better organized data using these vs some more standard method with some imaginative improvements. Nevertheless, I do appreciate new ways to visualize data. Perhaps you can demonstrate a comparison of different graphs in a future video, and maybe address these perceived shortcomings.
Hey thanks a ton for your thoughts! I think the Sankey diagram could really benefit from some static labels and we should note that, even with these, it’s a compromise between clear information presentation and elegant data visualization. Perhaps for applications where clarity is desired above all else a set of tables or bar charts would be preferred but when a more engaging presentation is desired, the Sankey diagram offers extra elegance with a small but tangible loss in clarity.
I would love to see more about data visualization. Thanks for this, hope to see more!
More to come!
I work for a school district and when we wanted to show individual student entrance and exit enrollment events during a school year we settled on the Sankey Diagram! Power BI made it pretty easy!
That is awesome!
Two thumbs up regarding the interpretation of Sankey diagram in terms of conditional probability
Thanks! I like that interpretation too
another day another thing i learnt. Thank you.
Glad to hear it!
That's something nice to use when you have a contingency table.
Interesting application!
Brilliant!
In terms of labeling, I’d type the %s (what percent of the school is contained in that bar?) written at the _end points_ of each flow-bar. That would show the conditional probabilities directly. Also total student body sizes written next to the schools at the outer edges of the diagrams.
Thank you!
Great suggestions!
I do not like them, but your explanation was excellent.
If you add a third dimension you could also represent time. And see total students grow and shrink, and see the transitions grow and shrink. It might look a bit cluttered about.
I absolutely love that idea but am also scared of it looking like a scary web 😂 would love to see a basic 3d Sankey in real life tho!
Interesting idea!
Perhaps a 'Sankey-Chain' would be better way to view the time-dependent version of the plot. Each instance of the chain will be a version of the plot at each 'timestep'... just saying.
As much as I like a good Sankey, when I try to use it for bonus points it usually ends up being super complex to build and disappointingly close to looking like a bowl of spaghetti 😅
You’re not wrong 😂 I think it’s best used with fewer than 5 sources and targets otherwise it’s the Spaghetti Diagram
Ah interesting new diagram... but, I am finding these Sankey Diagrams (at least so far) very difficult to follow. I find them incredibly confusing and, since I can't possibly "hover" interactively, I would be at a huge loss as a "viewer".
Since you are the "presenter", it is easy for you to hover over each spot and show the tool tip. But as a viewer, I am sort of forced to constantly focus on your current position in order to follow ... that means I can't take "notes"... I have to have eyes on every second -- whereas, a more traditional diagram is more easily digested visually, and typically contains visible labels along the axis traditional XY axis. Maybe you can add persistent labels and numbers?
I would be hard pressed to believe that you can provide more, better organized data using these vs some more standard method with some imaginative improvements.
Nevertheless, I do appreciate new ways to visualize data. Perhaps you can demonstrate a comparison of different graphs in a future video, and maybe address these perceived shortcomings.
Hey thanks a ton for your thoughts! I think the Sankey diagram could really benefit from some static labels and we should note that, even with these, it’s a compromise between clear information presentation and elegant data visualization. Perhaps for applications where clarity is desired above all else a set of tables or bar charts would be preferred but when a more engaging presentation is desired, the Sankey diagram offers extra elegance with a small but tangible loss in clarity.
Generate this with a LLM?
Which part?