Mikael Eriksson, miker@kth.se, 2019-03-29 Can using Sankey diagrams to visualize when and where students fail courses during their education be a relevant tool for teachers? Mikael Eriksson, miker@kth.se, 2019-03-29
Background Background in computer science ~ 5 years background in teaching, ~ 90 hp pedagogy studies Been involved in working with student data on and off from 2012 Current work, since 2018, a minor part of my employment (~15%) conducted at math department at SCI (School of Engineering Sciences)
Origin of idea June 2018 ”Jag tänkte att man borde kunna kolla hur det ser ut efter misslyckandet för studenten, och få en bild av i hur stor utsträckning misslyckanden på de olika kurserna leder till att studenten hoppar av”
Tracking where students have trouble with courses Operational definition of the ”trouble” := The student enrolled on course X but either Never completed it Or had a considerable delay (more than one semester) before completing it compared to the nominal time *
Algorithm Builds a graph where: The nodes represents courses The value/thickness of the edges represents the accumulated amount of ”trouble” between courses
Add all ”trouble” instances to a graph For each student in the data set, if the student had trouble with course 1 and course 2, and they occur after each other, add 1 to the edge between those courses +1, +1, +1, … Semester 2 Semester 1 Course 2 Course 1
A Sankey diagram with a twist, because… The same student can add to multiple inputs / outputs, so that sum of inputs != sum of outputs +1 Course 3a +1 Course 1 Course 2 +1 Course 3b
Full graph is visually intractable
After pruning the graph *
Far from an exact result Pruning edges in graphs for visualization purposes is a complex issue and is still an active research area I do 2 things in pruning step: Remove edges that are too small to be interesting * Remove cycles
Interactive diagrams In this presentation the examples are static images, but platforms for rendering diagrams come with interactive abilities: When you hover the mouse over a path it is high lighted and the edge information displayed on top SankeyMatic (sankeymatic.com) – comes with limitations * D3.js ”is a JavaScript library for producing dynamic, interactive data visualizations in web browsers”
More examples
Another examples
Causes are uncertain Why do students fail courses? Student councellors say students point to General life troubles (home, relations, health) They blame bad teachers and courses Low motivation to study Bad studying technique, has a hard time with the transition to university studies What about other causes? I remember I had times when I had to struggle with ”restkurser”
Could it be used to improving education? May be difficult to generate key indicators or automatic inferenses from graphs because causes are uncertain, but Flows could be investigated further by other means Teachers do not want top-down demands to investigate But plain curiousity may inspire some to take a closer look and pondering about possible actions Just the knowledge that such diagrams are possible may motivate further thinking
Possible next steps Work to make the graph algorithm more rigorous and resulting diagram more reliable Publish / distribute diagrams to those who are curious to to know more and maybe in taking some action based on the information they provide If you are curious to have diagrams for some program you can contact me at miker@kth.se