SDS 384: 10 Tips to Become a Master Presenter

(00:15): Welcome back to the SuperDataScience podcast, everybody. Super pumped to have you back here on the show. Today, we’re going to be talking about presentation. This is a very important skill in data science. By now, you probably have heard me talk about it many times if you’ve been listening to this podcast or taking our courses. It’s one of those skills that separates an outstanding data scientist from a great data scientist. So you can be very good at coding, creating models, deriving insights, even asking the right questions. But unless you can present the insights to the stakeholders who are going to be making those decisions based on the insights, then you will always have to either rely on others to do that for you or you won’t really stand out. You’ll just be another data scientist who does a great, fantastic job, but isn’t that shining superstar that can explain those complexities to a non-technical audience. An extremely important skill, regardless of what area of data science you want to specialize in, whether it’s machine learning, or AI, visualizations, BI, data mining, data preparation, whatever it is, always presentation is super important. 

(01:34):
All right, now that hopefully you’re convinced, I wanted to mention that in the previous FiveMinuteFriday, we talked about managing cognitive load in data science, and there we identified three types of cognitive load. So just to recap, there’s intrinsic cognitive load relating to the complexity of the task that you’re explaining. There’s extraneous cognitive load, which is the unnecessary type of cognitive load that you can and want to eliminate from your visualizations, from your presentations. And finally, there is the germane cognitive load, which you want people to have because that’s the one that allows them to process information, understand what to do with it, and retain it in long-term memory. 

(02:16):
So today, we’re going to build on top of that, so if you haven’t listened to our podcast, I recommend checking it out. But we’re going to build on top of that. And we’re going to look at 10 tips. Actually, these tips come from the course I’m working on right now. It’s a course to help people prepare for the most advanced Tableau examination, the Tableau Certified Professional exam. And also, this part of the course is about visual best practices. And so there, we’ve got 10 tips and today I want to share them on this podcast because while researching them, I learned a lot myself and I think it would be really cool to share them here so that everybody can use them. 

(02:55):
So these are 10 tips about how to reduce the unnecessary cognitive load. So we remember that germane cognitive load is good. We want people to have that. But intrinsic cognitive load, which we cannot remove, we just attach the complexity of task, we can break it into parts. And extraneous cognitive load, we definitely can remove it. So these tips will be focusing mostly on extraneous cognitive load and a bit on the intrinsic. 

(03:23):
So here we go. Tip number one is to use chunking. So there was a cognitive psychologist, George Miller, who actually gave rise to this area of cognitive load. So Miller proposed the number seven, that actually through his studies, he discovered that people can retain about seven plus-minus two items in their short-term memory at any given point in time. So therefore, try to chunk things up as much as you can. For instance, try remembering this number right now, 177631482. So it will you take a while. It will be really hard to remember it and really hard to store in your memory. You probably already might have lost it from your short-term memory. Short-term memory lasts about 15 to 30 seconds. So the number was 177631482. 

(04:23):
But how can we use chunking to help our short term memory or our listeners’ short term memory, to process that number? Well, what if we chunk it up? Instead of dealing with, what was that? Four plus five… Nine separate elements. So 177631482 is nine separate elements that you need to remember, so nine bits of information. If we chunk it up, we can make it less. So for instance, 1776 is the year when America gained its independence. 314 is the first three digits of pi, and 82 is the age at which Frank Sinatra died. So now if you have these three things in mind, year of independence, pi, and Frank Sinatra’s image, then now you only have to remember three things. That’s an example of chunking, a very powerful method to help your audience process information in their short-term memory. But let’s move on. So that was just the first tip. 

(05:25):
Second tip, give control. For instance, often I see visualizations with a grid of four by four maps at the same time, portraying different things about, for example, the United States, how a certain behavior or certain geodemographic aspect is present across different age groups, for example, in the different States of the United States. Rather than having 16 maps on the visualization at the same time, why not put one map and have a filter or have a way for people to change around what they’re seeing? So give control to your audience of the visualization if you can. On the other hand, if you can’t give control, or if that’s not your intention, break it down into a story. This is tip number three. Rather than showing a lot of insights at the same time, break them down into a story, and that way you can break down the intrinsic cognitive load so people only have to deal with a certain part of your insights at a time. 

(06:30):
And really, storytelling in data science is one of the most powerful skills you can possibly possess because it breaks down a complex insight into chunks. Or not even chunks. We already used the word chunks in tip number one. It breaks it into small bits, more small steps that the audience has to process one at a time. And then the whole story builds in their minds. So it’s all about building that full story in their mind, but doing it skillfully. So break it down into a story if you can’t give them control, or if you need to tell a story rather than interacting with visualization. Break it into a story rather than throwing all the insights at them at the same time. 

(07:04):
Tip number four is use color sparingly. Don’t overuse colors. Again, it’s another item for short-term memory to process. Anything above five colors or seven colors at most in a visualization is probably overdoing it. So try to use few colors, but also don’t use too few colors because you’ll end up using the same colors for different purposes. You want to avoid that as well. You want to avoid misleading your audience. So be careful of how many colors you use and also keep color-blind people in mind. That’s very important. 8% of males on the planet are color-blind and 0.5% of females on the planet are color-blind. So it’s quite a lot of people who are color-blind. Definitely keep them in mind. For instance, Christmas colors, red and green, are not a good idea together. Avoid that at all costs. 

(07:54):
Tip number five, avoid redundant coding. Only use it when necessary redundant coding is like if you have a line chart and you decide to, in addition to it showing through the axis, where something is… For instance how GDP has grown or dropped off. By looking at the Y axis, you can tell that already. In addition to that, you would use the width of the line would show the same thing. So the line would get wider as it goes up and get narrow as it goes down. And then in addition to that you could also use color, it becomes darker as it goes up because more saturated and so that means higher GDP, and becomes lighter as it goes down. So that’s a triple redundant coding, absolutely unnecessary in that case. 

(08:37):
The only one time that comes to mind that redundant coding is very useful and necessary is with tables like a highlight table. If you have just a table of numbers, really hard to navigate. If you have a table that’s highlighted and the background of every cell is highlighted based on the magnitude of the number, much easier to navigate. That’s an example of redundant coding that actually helps reduce the extraneous cognitive load. Otherwise, if it’s not helping reduce it, it’s probably helping increase it, and we want to avoid that. 

(09:09):
Tip number six, integrate the legends into your visualization. So wherever you can… For instance, you’re comparing diesel versus petrol. You can put a label, a legend, saying, “Okay, on this chart, diesel is black and a petrol is green.” But instead of that, you could say in the title of your chart, you could say “Diesel versus petrol” and make diesel black and petrol green. And then in your chart, you can really just use those colors. So you don’t have to have an extra legend when it’s given in the title. So you can combine things like that. See when you can combine legends or integrate them into your chart. For instance, you could just put the word diesel at the very end of the line for diesel and the word petrol at the very end of the line for petrol. So you don’t have to have a separate box for the color legend. And again, that helps reduce extraneous cognitive load. 

(10:00):
Number seven, tip number seven. Maximize the data-ink ratio. One of the most powerful principles from Edward Tufte’s list of principles, and that is about maximizing data to ink ratio. So the amount of data represented by ink on your chart should be as much as high as possible because basically you want to avoid, as much as you can, any ink or any visuals on your chart that are unrelated to the data that you’re showing. The less of that, the better it is. So it’s always going to be less than a hundred percent if we can express in percentage terms, but the closer you get to that level, the better.

(10:45):
Also there’s another type of ratio. So that was data-ink ratio, but there’s also signal-noise ratio. So sometimes you might have something on a chart that is actually representing data. So it’s satisfies the data-ink ratio principle, but it’s actually not the signal that you’re looking for. They might be a lot of noise on the chart. So let’s say you’re visualizing a certain parameter about one or two or three of the states of the United States, but then you’re showing all… 

(11:17):
So let’s say you’re visualizing how the sales of a company are going in the United States. And you’re visualizing three states because that’s where most of the sales are going well. But then you’re showing all the other states that have very insignificant, marginal sales, tiny sales that don’t really need to be shown. That’s not part of your story, but you’re still showing them. So that satisfies the data-ink show. It’s still data, it’s ink representing data, but it doesn’t satisfy the signal-noise ratio. So it’s noise that is distracting your audience from seeing what they’re supposed to be seeing. It really depends on the situation, the problem that you’re solving. Maybe you need to show that there are not many sales in those other states. Maybe that’s required. But if it’s not then really a question, what is the signal in your case? What is the noise in your case? 

(12:06):
Tip number eight, master tool tips and annotations. Very powerful techniques. Basically when and how do you add text or accompanying little elements like pop-ups and things like that to your visualizations so that they don’t distract, but they actually add to the experience of the user? 

(12:32):
A classic example that I really like is if you’re visualizing survey results, for instance, you have 10 survey questions and you want to visualize them in a matrix. For instance, you have 10 questions that were asked in different universities. So on the Y axis you would have different university names. On the X axis, you would have the different survey questions. Well, those survey questions are probably very long. They’re not going to fit into your headers. So instead of trying to fit them in, squash them in there, you can put question one, question two, question three and so on, and then have a popup so when a person hovers over the insights, the popup says what that question actually means, the text of the question. Or have an annotation on the side or a comment explaining each one of those questions. So that’s a powerful technique also to reduce cognitive load. 

(13:23):
Tip number nine, simpler charts are often better. And this comes from the Cleveland and McGill study ranking, which is called the ranking of elementary perceptual tasks and the specific of a quantitative information. We spoke about this, I believe it was on episode 329. Yes. Episode 329 with Isaac Reyes, who is a consultant in data, storytelling, data visualization. If you want to check it out, check out that episode, episode 329. In a nutshell, there is a ranking of what people are very good at and what people are not so good at in terms of analyzing visuals. 

(14:00):
And the top three things that people are best at are looking at scatter-plots and line charts, basically comparing individual values against an Y axis. Number two is looking at scatter-plots and line charts with shifted Y axes, so basically you have two scatter-plots side by side, but one is a bit above the other. The axes are identical, but one is just shifted up or down. People are still really good at that. And number three, third place is for bar charts. People are really good at seeing things from bar charts. So those are the top three, and then after that you get all the things like angles, angles are pie charts; area, area is bubble charts and tree maps and all the other things. And even stacked bar charts are, I think, only on fourth place, right? So they’re not in the top three. And different color shading, that’s at the very bottom. 3D objects, that’s also at the very bottom. Volumes is really hard for humans to get insights or quantitative comparisons out of volumes. 

(15:02):
So that’s the top three. The top three are scatter-plots, line charts, shifted scatter-plots and line charts across the Y axis, and bar charts. So those are the simplest things that are easiest for humans to process in their working memory, which is also synonymous to short-term memory. Try and use those when you can. Yes, sometimes we want to create amazing works of art. Fantastic. There’s a time for that, a time and a place, but at the same time, sometimes we just need to get the insights across. Don’t over-complicate it. Simpler is often better. Use the simpler charts when you can. 

(15:38):
And finally tip number 10. Very powerful tip. Simple tip, but powerful tip. Use the chart title to your advantage. Don’t just put a title like “petrol versus diesel prices”. Everybody understands by looking at the chart it’s petrol versus diesel prices. Ask a question. So the tip is ask questions in your titles. So for instance, ask a question like… What insight do you want to get across? Is the goal of this analysis, or have you discovered that petrol prices have been growing faster than diesel prices? So ask the question. Have petrol prices been growing faster than diesel prices? And let your audience use your visualization to guide themselves to the answer. That will get them to interact more. 

(16:25):
It’ll still tell them what the chart’s about. So you’ll get that part checked off, but also guide their attention into that specific question, rather than allowing it to scatter across a whole visualization and maybe all kinds of comments and other things that might be present on the chart or on the day when you’re presenting. The question will help them narrow their focus into a specific part of your visualization, zoom in to that, and they’ll be able to answer it better. 

(16:52):
So another example that is, have you ever seen a subway map of the New York subway system, or of the Berlin subway system or Tokyo subway system? It’s complex. It’s super complex. You get lost in it if you just try to look at it and read. But if you have a question in mind, or if somebody asks you a question, “How do I get from station A to station B?” and they’re on different lines, in different parts of the map, instantly your attention is laser-focused. Your brain throws away anything that’s irrelevant. If it’s a matter of combining the red, green, and blue lines to get from where you were to where you need to go, it’s going to disregard any orange, yellow lines, brown lines, all those paths the trains can go, and parts of these networks. They’ll be disregarded. Instantly it becomes much easier for you to answer that, or navigate this complex visualization or these complex insights, because you have a question in mind to answer. 

(17:59):
So help your audience be in the same position. Ask them a question and use the title to do that. That’s a very powerful technique. Especially if you’re attending a visualization or presentation and somebody on their chart asks the question in the title, you can bet that that person has done some studies in the space of presentation, of soft skills, of visualization, that it’s not their first rodeo, that they’ve done this before, and they actually know how to do it. So you’re dealing with a pro. So be a pro yourself and use those titles to guide your audience. It’s a very subtle, but very powerful technique. 

(18:38):
So quickly to recap, the 10 tips to make the most out of your visualizations and presentations and master this skill, this art, are… Here you go. The 10 tips. Use chunking, give control, break into a story. Tip number four, use color sparingly. Avoid redundant encoding. Integrate the legends. Tip number seven, maximize data-ink ratio and also signal-noise ratio. Master tool tips and annotations. Simpler charts are often better. And finally, tip number 10, use titles to ask questions. 

(19:16):
So there you go. Those were the 10 tips to help guide your audience in the best way possible. I hope you enjoyed this podcast and I look forward to seeing you back here next time. Until then, happy analyzing.