SSTiC 2014: Mark Guzdial on CS education

How better to break my blogging drought than live blogging the always awesome Mark Guzdial’ lectures at SSTiC (he is, btw, worth the trip and price of admissions).

Far more people programming as part of their job than are pro programmers.

30%-50% failure/withdrawal rate in first CS course.

I asked how do we know this is bad rather than good. Mark says there’s evidence we can did better. At next slide showed AP results for calc and CS and CS sucks and is flat in comparison. That seems like strong evidence contrary.

As expected, we’re going to use media!

And there he fires up Squeak! Yay!

A whistling demo! Fun!

Mark: it needs to be bigger and you need to slow down 😉 Also, having the comparison between sounds readily available would be cool.

(Sean Bechhofer should see this!)

Calculus for digitizing sound. Someone’s theorem was critical. Fun fact about Sgt Pepper’s: they included a dog only audible sound at the end; we can hear it on the CD due to downsampling!

(Argh…connectivity killed me. Lots of cool stuff:

Peer instruction has dramatic results…in physics.

They we did a bunch of image stuff with peer instruction.)

LiveCoding: two types — demos and live algorithmic music.

Live coding is part of best practice for engaging women. Reduces fear of programming and imposter syndrome.

Spatial intelligence and learning center.

High spatial intelligence do well in CS. Low spatial intelligence goes into business and education!

Train specially low spatial intel engineering students early has great benefits with high transfer.

Kinaesthetic learning activities.

Gestures and sketches are better than premade diagrams because (from bio) incongruity between diagram and mental model inhibits learning. (Uli will love this.)

Computer science unplugged.


Using music to get the idea of linked lists.

Multimodality (channels get overloaded so use multiple channels)

Important of worked examples. 7-10 examples before doing (or slightly more complicated patterns). worked examples take a third of the time.

Apprenticeship model isn’t efficient.

Labelled subgoals in worked examples.

Amazing improvement.

Parson’s problems. Give all lines of code and drag into place.