SDS 382: Manage Cognitive Load in Data Science

Podcast Guest: Kirill Eremenko

July 10, 2020

Welcome back to the FiveMinuteFriday episode of the SuperDataScience Podcast! 

Today we’re talking about cognitive load.  
You may have heard cognitive load as being bad for your audience when unstructured. We want to make sure our presentation is structured in a way that minimizes cognitive load. That’s partially untrue. I’m working on my new Tableau course so I’ve been doing research on this. Where we discuss short term memory in the course, I discovered there are three types of cognitive load: intrinsic, extraneous, and germane.
So, cognitive load is the amount of short term memory resources required to process a certain task. Intrinsic cognitive load is associated with the complexity of the topic at hand, such as simple addition vs calculus. We can’t change a topic’s complexity, but we can break it down and compartmentalize it into steps or a story. Extraneous cognitive load is associated with irrelevant parts of a topic, or how information is presented. In germane cognitive load, it is all about the organization of information in the context of existing knowledge. This is how new information becomes long term memory. And for that, you want to maximize this germane cognitive. 
To help boost this germane cognitive load you can employ various techniques. You want your audience to be using that form of cognitive load. Something good to keep in mind for data visualization practices.
ITEMS MENTIONED IN THIS PODCAST:
DID YOU ENJOY THE PODCAST?
  • Are you employing the right balance of cognitive loads in your data visualizations?
  • Download The Transcript
  • Music Credit: Light by Krakn [NCS Release]

Podcast Transcript

(00:04):
This is FiveMinuteFriday, Manage Cognitive Load in Data Science. 

(00:15):
Welcome back to the SuperDataScience Podcast. Everybody’s super excited to have you back here on board. Today, we’ve got an exciting topic. We’re talking about cognitive load. Now, you may have heard about cognitive load and that it’s bad for your audience, and specifically, we’re talking about here, once you have insights, once you’ve done all the analytics, once you’ve done the modeling, you’ve got the insights, and now you want to present them to your audience, how do you structure your presentation? So that’s what today is about. So you may have heard that cognitive load is bad and that we want to make sure that our presentations are structured in a way that we minimize cognitive load. Now, that’s true, but only partially. 

(00:54):
I was doing some research for the course that I’m recording right now, which is the Tableau Certified Professional Course, the most advanced certification on Tableau, and a part of it is all about visualization best practices, because in order to be an effective communicator in data science it’s not just enough to build great visuals and get insights, it’s also important to know how to communicate. So I was doing research and in the section where we talk about short term memory, one of the topics, or one of the most important topics, is cognitive load. So it turns out that there are actually three types of cognitive load. There’s intrinsic cognitive load, extraneous cognitive load, and germane cognitive load. And today I wanted to share some of the insights I found over the past few days about cognitive load because I think it’s very important. 
(01:44):
Whether or not you’re doing Tableau, whichever tool you’re using, however, you’re presenting your insights, it’s very important for data scientists to present and thereby it’s important to understand how audiences process what we show them. So with that, here we go. First of all, let’s define cognitive load. Cognitive load is the amount of short term memory resources that are required to process a certain task. But we are specifically talking about short term memory here, how once the attention ear is on our audience’s mind, once information is in their mind, once we’ve grabbed their attention, they’ve taken the information in, now what happens in short term memory? Well, now let’s look at the three types of cognitive load. The first one is intrinsic cognitive load. This cognitive load is associated with the complexity of the topic. For instance, if you compare teaching somebody how to add two plus two together versus teaching somebody a differential equation. 
(02:44):
There’s absolutely no doubt that these two topics have vastly differing complexities and it’s going to require much more cognitive load to understand how to do a differential equation. And it actually used to be thought in field of cognitive psychology that this intrinsic cognitive load is actually immutable. But the reality is that, while we cannot change the complexity of a topic, we can actually break a topic into parts and then explain those parts separately and then put it all together later on. And that will help break down that cognitive load into steps. And that’s what we can also do in data science when we have insights and it’s a complex series of insights, or it’s a complex phenomenon that we’re describing, or it just requires a lot of moving parts to be put together in an audience’s mind, rather than putting it all into one chart or one dashboard, we can use something that’s called a story. 
(03:44):
In Tableau, actually, you have a whole feature called Create a Story or Storyline. But even if you’re not using Tablo, you could do this in PowerPoint. Basically, break down your insights into a story where you give your audience everything step by step and then you put it all together at the end. That way, you’re not bombarding them with this intrinsic cognitive load all in one go. It’s an important technique to break your insights into a story. The second type of cognitive load is the extraneous cognitive load, and extraneous cognitive load is all the irrelevant stuff. It’s basically how we structure our visualizations, how we present the data. So here’s a great example. For instance, let’s say you want to explain to somebody what a square is. You could go about telling them all of all the lines and the corners in a square and spend maybe five minutes explaining verbally what a square is. Or you could just go up to a whiteboard and draw a square, and it’ll take you one second to explain what a square is. 
(04:50):
Here, the medium of the explanation plays a big role and that is up to the person explaining to select the medium. So are other things, even if you’re staying in one medium, which is in the case of data science, most likely visual presentations, in this medium, you have lots of choices. You can use a bar chart, you can use a map, you can use a scatterplot, you can put two scatterplots together, you can put 15 maps on one visualization, or you can put one map and add a filter so people can click through different options of the same map and adjust it, give them some control over it. So there’s lots of ways you can actually control these visualizations that you’re creating and the goal should be to minimize this extraneous cognitive load. Basically, cognitive load that not necessarily has to be there given that you find a much better way or just a better, simpler way to convey the same information. So there’s lots of way of conveying this information. Some will have high extraneous cognitive load, some will have low extraneous cognitive load, and we want to find the low cognitive load approaches. 
(06:04):
And finally, there’s the germane cognitive load. This cognitive load is actually the good type of cognitive load, something that not many people talk about. Everybody thinks cognitive load is bad. Actually, there’s this good cognitive load, germane cognitive load, and it is all about the organization of information by integrating and connecting it with existing knowledge. This is effectively how our audience takes what we’re presenting to them and turns that into longterm memory. So this is basically where our audience takes what we are presenting to them, and either turns it into long term memory, understands what to do with it, takes business decisions, and so on. So, effectively, we want our audience to maximize this germane cognitive load wherever possible. And in order to do that, on the one hand, we have to minimize the other types of cognitive load, because there’s a limited capacity for cognitive load and if it’s going to be used up for intrinsic or extraneous cognitive load, not much is going to be left for germane cognitive load. 
(07:08):
So we want to minimize intrinsic or extraneous to leave room for germane cognitive load. And at the same time, we also want to help with this germane cognitive load, help guide our users, and help them retain this information. For instance, an example of how you could help with that is by using mnemonics to help people memorize things. For instance, also, repetition is another good example of that. If you help people practice certain information, they’ll retain it better, they’ll understand it better, they’ll understand what to do with it better. So there’s quite a few techniques there, but the point is that germane cognitive load is actually good and we want people to be using that type of cognitive load. 
(07:53):
So there we go. That’s cognitive load in a nutshell, something to keep in mind when you’re doing or creating your visualizations, running your presentations to your audience. Always think about intrinsic cognitive load and how you can space it out through storytelling. Think about extraneous cognitive load. Are you actually presenting in the best possible way using the best possible medium and the best possible charts and graphs and so on? That’s the creative part of data science about the art of presenting insights. That’s what we want to do. Minimize extraneous cognitive load. And germane cognitive load, that’s the one where our audience understands those insights and memorizes them and decides what to do with them in a business sense or whatever other applied sense. So we want to promote that type of cognitive load and what we can do here is things like mnemonics, repetition, practice, and so on. 
(08:48):
There we go. Hope you enjoyed today’s episode. I look forward to seeing you back here next time. Until then, happy analyzing. 
Show All

Share on

Related Podcasts