This is an edited version of a talk I gave last week in Brisbane, at ICOTS – the International Conference on Teaching Statistics.
I’d like to acknowledge that this talk was written on the unceded lands of the Bunurong people, and is being delivered on those of the Turrbal and Jagera people. I recognise the First Nations people of this land as our first scientists, environmentalists, teachers, and storytellers, who we could learn a lot from, if we choose to listen.
I was out at a tech meetup last week when I mentioned I was going to a conference on statistics education this week. “Sounds riveting” said someone, with the deepest sarcasm. The thing is, statistics are riveting. Statistics are powerful. As Brett Sutton recently said on my podcast, Make Me Data Literate, in answer to the question “What excites you about data?”
“It’s the ability to reveal new things and through that telling of a story, bring us to new insights, that hopefully mean something to us in the world, that is a revelation about us as human beings, is a revelation about society, about the fascinating world we live in. There is so much out there that is yet to be discovered, but there is also a whole bunch of stuff that is yet to be fully understood. And it’s in data that gets us to new understandings that can be fascinating and exciting and revelatory. And I think that’s the magic of it all.”
I run the Australian Data Science Education Institute. We are a national charity dedicated to building STEM, critical thinking, and data literacy skills for all Australian students. We build resources and train teachers, because it scales better than working directly with kids.
But I made a mistake, putting data science into the name of our organisation. Because data science is scary. Possibly even scarier than statistics!
If I advertise a data science workshop I get no signups. If I call the same workshop STEM, or Problem Based Learning, it goes bananas.
But you don’t need scary level programming or stats skills to start people off in data science. You need data literacy. You need authenticity. You need relevance.
There is a popular idea that if you teach core skills with toy datasets – neat, clean, simple datasets with only one possible analysis, only one possible question, with only one possible answer – you build core skills that can later be applied to real datasets. But teaching stats using toy datasets is like teaching people how to engage their core muscles while having them lift nothing heavier than a teddy bear. A small teddy bear, at that. It’s nonsensical. They get out into the real world having never lifted anything more substantial than a teddy bear, and the first time they lift anything heavy they pop something. Now that core strength matters, they haven’t got it. They don’t even know what it is, how to get it, or how to use it effectively once they have it. Teaching data science with toy datasets is just like that.
So I have a template for making statistics relevant to school kids, building in STEM skills, and, more importantly than ever, Critical Thinking.
- Find a problem in your own community.
- Measure it.
- Analyse and communicate those measurements.
- Devise a solution. Implement the solution.
- Measure it again to see how well it worked.
Not if it worked. How well it worked. Who it helped. Who it harmed. What could be improved.
At every step of the way, critical thinking is essential. There’s no such thing as a perfect measurement, so what’s wrong with your measurements? If you’re measuring traffic, is this a rostered day off for local building sites? Is the nearby uni on semester break? Is the nearby trainline out of action so it has buses replacing trains? Is the high school off campus for their sports carnival? Often they’ll have to come up with a way of measuring things. How do you measure traffic? Number of cars? Time spent at the traffic light? How big the queues are? Time taken to travel a particular route? Is this a “normal” day? What is a normal day anyway? How do you measure litter? Count pieces? Count types? Measure weight? By location? By volume?
Even a straightforward measurement like “how many people attended this talk” is complex. Define “attended this talk”? Are we counting people who came into the room at all, even if they took one look at me and left? Are we counting people who stayed at least half way? What if they went to the loo and came back? What if they stay for all but the last five minutes? We make definitions and assumptions, and the data is messy and complicated, even when you think it’s not. And that’s exactly the kind of problem kids need for meaningful, relevant learning.
Ok, so assume we’ve measured the problem. Now we have to analyse it. With a toy dataset like the neat clean sports or sales tables you see as examples in textbooks, analysis is simple. There’s only one question the dataset can answer, and really only one way to find it. But with a real dataset, you have to define your categories (for litter, you might sort it by type – plastic, aluminium, paper, compostable – or by weight, or even volume), think about what questions the dataset can answer (where do we find the most litter? What is most litter made of? What days are worst? Who drops the most litter – do we have that information? How could we get it? )
What problem do we want to solve here? That will help us figure out what questions we’re asking, what data we need, and how to find that answer. Now we have to state our definitions and our assumptions, and identify potential weaknesses in our analysis.
We can’t just take a cookie cutter process, apply the formula, and look the answer up in the back of the text book. We can’t even use AI, because AI can’t analyse, it can only pattern match, and there are no patterns to match for these unique, local problems.
Once we’ve analysed the problem, we communicate those results to the class, the school, maybe the whole community, depending on the problem we’re looking at. That involves graphs, but graphs as communication tools, not just technical objects. So the choice of graph is dependent on what you are trying to communicate with the data, not just a technically correct graph for that dataset.
Now we’re in the solutions phase – interesting, creative, and authentic. What might make a difference to this problem? The class comes up with a solution, or groups come up with many solutions and then vote on the one they want to implement, it’s a flexible process. You’ll need to factor in any relevant constraints – budget, time, realistic approaches.
You may have seen the “plan a trip to Mars” or “solve a water crisis in an African country” type problem solving exercises that some schools do, but they can’t be measured, and the solutions can’t be implemented, so they’re pretty hollow. Even worse than teaching core strength with teddy bear lifting. With real projects, the students have to implement their solution. Or maybe even implement several solutions and compare them!
And this is the really important part: Now they measure it again to see how well it works! Imagine if we routinely did that for any problem we solved, any policy we implemented, any program a government announced with great fanfare. Imagine what would change!
Because it’s real world measurement, part of the process has to be considering what’s different between the first set of measurements and the second. Was it a windy day? Were the grade 5’s out on camp? Was it a different day of the week, or time of day? If you’re counting birdlife or biodiversity, could the different weather have made a difference? If you’re looking at energy efficiency, was it hotter, colder, or windier for one set of measurements than another? How can you be sure any changes were due to your intervention, rather than to other things that might have changed independently of you?
Once again, not something AI can answer, and not something you can look up in the back of the book.
You can use any level of technology and tools in this process. You can build whole sensor systems, or simply count by hand. You can graph using matplotlib in Python or you can stack blocks. The technology isn’t important, though it can be taught using these projects. What’s important is the thinking. What’s more, it’s resilient to AI, precisely because the thinking is what we’re teaching. AI can’t do that, it can only pattern match, and there are no patterns for these unique, local problems.
Like Brett said, Data gives us the ability to reveal new things, to gain new insights into ourselves and the world. And that’s the magic of it all. If we’re not showing kids the magic of it all, how can we expect them to love it like we do?