Another important concept to introduce when we talk about channel effectiveness is the concept of discriminability. What does it mean? Well, it means that when you look at how you encode information with one channel, you also have to consider how many distinct values within this channel the viewer is going to be able to perceive. That's a very important characteristic. Let me give you a few examples. For instance, if you're using the area of a bubble to represent a quantity, you want to make sure that you know how many distinct values a viewer is able to perceive. The same is true with say, line width. If you have too many different values, the viewers won't be able to perceive many of them, really. As we will see in a moment, this also depends on other contextual factors. Very similarly with color. If you have too many colors, you won't be able to distinguish between all of these colors. So that's the concept of discriminability. And it's a very, very important one. In general, discriminability depends on a number of parameters. So it depends on intrinsic properties of the channel typically following the same ranking as the accuracy ranking that we have seen before. But they also depend on many other things. They depend on the spatial arrangement of the visual marks. It depends on size, and it also depends on cardinality. Let me show you these three aspects through a few examples. So let's start with the effect of spatial arrangement. So here I have, once again, a series of bubbles of different sizes, and when they are presented this way, aligned, you can very easily distinguish their size. But once they are arranged in a somewhat more random layout which is typically what you have in this organization, in real data, it's much, much harder to distinguish all of these differences. So you can see that there is an effect of layout or spacial arrangement, similarly with color. So here we have lots of different colors. And when they're arranged in a well organized grid, it's somewhat easy to distinguish between them. It's actually pretty easy. But when I use colors, the same set of colors, in a layout where it's much more spread out and random, it's much harder to identify which colors are the same. So that's a problem, and it's very important to keep it in mind, when you are designing or using visualizations. The second aspect is the effect of size. In this affects specially color. Here is a scatterplot showing data coming from food products data set. So every single dot is a food product and as you can see, the dots are colored, and they are colored according to a number of categories of food products. Now in this case, I, on purpose, designed these scatter plots in a way that the dots are very small. And because of that, comparing the colors is really hard. Try to do it for a moment, and look also at the legend that you see on the right. There are many cases where it's confusing and it's not clear which color is which. But once I make these dots bigger, much bigger, it becomes much, much easier to compare the colors. So here is an example that shows you the effect of size with exactly the same channel and exactly the same colors, if the marks are very small, comparing the colors is going to be much, much harder than when you have marks that are bigger. Lastly, there is also an effect of cardinality. What is cardinality? Cardinality means how many different values you want to be able to represent. And I have, once again, an example with color. So the color palette and items that you have on the left side is much, much easier to perceive than the one that you have on the right-hand side. Why? Well, because on the right-hand side you have too many distinct values and there are higher chances that you mix one with the other. So cardinality also plays a major role. So in summary, the screening ability is about the property of a channel to create visual representations in which a certain number of values can be discriminated. And it depends on the accuracy of the channels, but also on a number of other contextual parameters. And this is very important to keep in mind. So discriminability depends on the spatial arrangement, size and cardinality.
