This is Lecture 2 for Lesson 5. Let's talk about colors on maps.

So, colors on thematic maps should really correspond to the types of data you're

trying to show. So, do you have numerical data or do you

have categorical data? Did you take a census and count up a

bunch of things and come up with a rate? Or do you just have the overall

preference for let's say pizza toppings or somehting like that per county.

And color schemes are designed then to include three major types that correspond

to whether or not you have numerical or categorical data.

You have sequential color schemes where you show less stuff to more stuff.

You have diverging color schemes where you're showing stuff above and below an

average. They have qualitative color schemes which

use a different color to signify different category.

So, let's say apples and oranges, and pizza, curry [UNKNOWN] and tonkatsu.

And I'll show you some examples now. So, let's start with a bad example, just

so that I could pound it into your head, make sure you don't do this ever when you

make a map. This is a sequential set of data here

showing the internet users per 100 people per country.

So, it's less of that to more of that per country.

But I've applied a qualitative color scheme just to kind of throw you off

here. A lot of people do this sort of stuff

though as a general rule because they don't know any better.

But why would you have purple being more than green, which is more than orange,

which is more than yellow, and then more than red.

You could kind of learn how to read this map.

But in general, it's very hard to pick out right away which places are high,

right, and which places are low. Or what's the average?

you can't do it very easily. You have to really study this to figure

it out. Contrast that to the appropriate use of

sequential colors here. This is sequential data, so let's look at

sequential colors. This uses five different variations in

hue, lightness, and saturation, to show less internet users per 100 people, and

then more at the high end. And you could see right away where the

low places are versus where the high places are.

It's pretty simple. You can do the same kind of thing, using

a diverging color scheme, which I think are quite useful.

So, what you do with diverging colors is kind of similar but you pick a middle

range. And you say this is the average category.

And then, you make that maybe white or gray or something rather neutral because

the average is not that interesting to you.

But what you want to show here is two ends of the spectrum.

So, we've got green scheme kind of going upward from the average so that you can

see the high areas. And then a purple sort of, I don't know

what you call it, mauve color, below average, you can look at low places.

And I think divergent schemes like this are great for showing really, you know,

what's above and below normal. And then, this is the appropriate use of

a qualitative color scheme. Let's say I had every country in the

world vote on one preferred pizza topping.

So, you had to have like have a contest and pick one for each country.

In this case, you know, Australia and United States and Canada love them with

some sausage. And Brazil's really excited about

pineapple, along with Saudi Arabia and Russia, which totally makes sense, right?

So, qualitative colors make sense here. Because a named hue that's you know, red,

or purple, or green, or whatever it is. You know, corresponds to the fact that we

have very different answers here that really aren't comparable or it's not less

or more, right? Pineapple is not more or less than

pepperoni. They're just different things.

So, in that case when you have different things you're trying to show that are

categorically different, qualitative colors make a lot of sense.

Speaking of color, I want to talk about rainbows and why rainbows kill people.

So, rainbow color schemes, which are also called spectral color schemes, that's the

right way to call them, are commonly misused on maps.

Weather maps are some of the worst offenders, where you show, show

precipitation amounts from less to more by showing green going to orange, going

to red, going to yellow to purple to white sometimes at the very top.

It doesn't make any sense. It should just show less to more color,

because it's less to more rain, right? Almost every heat map you see floating

around these days also uses these spectral color schemes.

And they're bad for the reasons I've already mentioned.

They use qualitative colors to show sequential or divergent data.

So, is blue more than yellow? Is purple more than blue?

You might have a legend that tells you the answer to that question.

But it's very hard for people to learn those things.

And it's certainly difficult to pick out a visual pattern right away on a map, if

you've got that kind of thing going on. Furthermore, rainbows can actually kill

people. And that's only a little bit hyperbolic

on my part. A recent study showed that doctors

interpreting medical images of the heart that used spectral color schemes, made

worse decisions about the patients' outcomes.

Than they did using the simple sequential color schemes.

So, I think they actually can kill people.

Here's an example of a poorly designed rainbow color schemed map just so you

never make them. hopefully you'll never make them anyway.

This shows tweets about Mitt Romney and President Obama during the presidential

election in 2012, and this is just a density map.

these are geo-located tweets, so it should just show west to more of these

tweets, right? Tweets about the same topic.

And it looks really cool with this rainbow color scheme, right?

It looks very lively and, and sort of engaging and dynamic, right?

It looks like you want to understand what's going on here, but take a closer

look at it and try to make sense out of it for a second.

Look closely at the color scale. There's orange, sort of, down towards the

lowest end. And red is at the top end.

And orange and red look similar to each other.

But they're on opposite ends of the actual spectrum.

And then, let's see. Blue is more than green.

Which is, green is more than yellow, though.

So, how would you know for sure, right away, what was the low spots on the map

and what were the sort of hot spots on the map?

I, I mean, it, it looks very compelling, at first, but it's actually almost

impossible to interpret if you really try to go beyond, you know, the presence or

absence of data. Contrast that to this map, which just

shows less to more purple. This is the appropriate way to deal with

this kind of data. Here, I'm showing density, so I just

want to show less to more color, of one color.

And now, I can see right away where I don't have as many of the tweets, and

where I have lots of tweets. So, there you go.

Rainbows kill people. Stop using them.

Data classification is another aspect of thematic map design that's important to

cover in class. So, assigning observations to categories

is called data classification. And there are three major types that I

want you to know about. There's equal interval classification,

quantile classification, and natural break classification.

And there's plenty of other methods beyond that, but it's sort of outside of

the scope of the class to go into that in this, this sort of explanation.

So, the first method that is shown here at the top right is equal interval

classification. What you care about here is establishing

an interval in terms of the value range. So, I, if I have 50 observations along a

set of values like I've shown here, I just decide that I want five classes.

So, I split it up into equal categories, zero to 20, 20 to 40, and so on.

That's the value range, sort of driving the classification there.

In contrast, quantiles, I want to make categories the same size in terms of the

number of observation, not the value range.

So, I don't care what the observations sort of said about, you know, did a

particular place get 22 on a certain score or whatever.

I don't care about that, I just care about having, in this case, five

categories of equal size in terms of number of observations.

So, each category in this case has ten observations.

I count up to ten, and then I make a break, and then I make the next class.

And so on. Finally, natural breaks, in the bottom

right corner, this method is driven by some fancy mathematics.

So, you let an algorithm decide where the so-called natural breaks in the data

distribution are, and it actually works pretty well in practice.

you don't need to understand the algorithm, because it's just not germane

to this class. Let's look at some map examples of each

of these methods though, so you can see how they work.

In this case, I've got some fake data because I love making fake data.

showing the proportion of people that are admirers of beautiful Audi station wagons

like I am. And this map, in particular, shows an

equal interval classification. So, what I care about here, if you look

at the legend is neatness and, and sort of the value range for each category.

So, the first category is zero to 20 people per 100 and so on.

21 to 40, etcetera. And if you look next to that little set

of categories there in the legends, you can see I've actually printed here the

number of states that are in each class. And if you pay attention to this in the

subsequent maps you'll see that it varies quite a bit depending on the kind of

classification I use. So, in this case, in the lowest category

has only six states in it, whereas some of the higher categories have over ten.

Let's look at quantiles using the same exact data.

So here, remember with quantiles, all I want to do is put the same number of

observations in each class. So, I count up the first ten states, and

then I put them in a class. And then, I go for the next ten

observations and put them in a class. And then, the value range can vary.

And there's not an equal interval in between them.

So, the first category here is zero to 30.

The second one is 31 to 45. So, it's much smaller than the first

category. And so on.

And you could see though, that the number of states per class, which is what

quantiles do, stays the same, right? So, you end up with a nice, visually

balanced map when you do quantile classification.

Finally, you've got natural breaks. And all these differences are a little

bit subtle, but you can tell different stories using the different

classification methods. Natural breaks is using some math, behind

the scenes, to try to figure out where the so-called natural, sort of, down

points are between clumps in the data. And you could see that the class breaks

vary quite a bit, so you have zero to 14, 15 to 32 is the next one.

33 to 49, etcetera. And in the lowest class, there's only

five members in that class. There's only five states in that low

category. But in some of the higher categories,

there's 12 or 13. So, you can end up with very different

visual balances on your maps, depending on the kind of classification you use.

Finally, I feel like I'd be remiss without talking a little bit about

textile maps. You're not going to do a lot of text

design on the maps that you make in this class because of the tools we're using.

But designing textile maps is a huge aspect of cartography and frankly there

could be a whole class just on that. And just to give you a flavor for it,

think about choosing fonts and positioning labels to make a map work for

your entire city that you live in at multiple scales, that would work on a

mobile device regular computer, and in a big printed sheet.

It's really hard work. And there's still a lot of manual effort

that goes into placing labels and making sure those things are readable on a lot

of the maps that you use everyday. And one very simple thing that's kind of

a fun little thing to learn about text design of maps is this graphic shows

here, is positioning labels. And this is something that hopefully you

can carry forward and use in other things you might do.

If you have a point here, like if I have this purple point, and you have to assign

a label to it, and you want to position that label somewhere, the best spot, spot

to put it first is where position one is. And if that position doesn't work because

there's a road crossing it, or a city there, or something else that blocks it,

you can go to position two. And if that doesn't work, then you can

try position three, and four, and five, and six, and so on.

So, this just gives you a way to prioritize label placement.

That's kind of quick and dirty. And hopefully, something you can take

away from this class. So at this point, you've learned about

the geospatial revolution. With spatial thinking, and thinking like

a geographer is all about. Where spatial data comes from how spatial

analysis works. And now, a fair bit about how to design a

good map. Hopefully, this has been an engaging

enterprise for you to take this MOOC. And hopefully what I've done with these

lectures and the written content and the labs and stuff made sense.

And you're able to take some of this forward with you.

And make great maps to share with everybody else.

Appreciate your time and I had a really fun time making this class.

Lesson 5 - Lecture 2

Maps and the Geospatial Revolution

与讲师见面