1
00:00:04,710 --> 00:00:08,720
In this video, we'll see how
to color our points by region

2
00:00:08,720 --> 00:00:12,830
and how to add a linear
regression line to our plot.

3
00:00:12,830 --> 00:00:15,360
Here we have our plot
from the last video,

4
00:00:15,360 --> 00:00:18,300
where we've colored
the points dark red.

5
00:00:18,300 --> 00:00:22,020
Now, let's color the
points by region instead.

6
00:00:22,020 --> 00:00:25,910
This time, we want to add a
color option to our aesthetic,

7
00:00:25,910 --> 00:00:30,360
since we're assigning a variable
in our data set to the colors.

8
00:00:30,360 --> 00:00:34,540
To do this, we can type
ggplot, and then first give

9
00:00:34,540 --> 00:00:36,750
the name of our
data, like before,

10
00:00:36,750 --> 00:00:43,870
WHO, and then in our aesthetic,
we again specify that x = GNI

11
00:00:43,870 --> 00:00:47,720
and y = FertilityRate.

12
00:00:47,720 --> 00:00:52,620
But then we want to add
the option color = Region,

13
00:00:52,620 --> 00:00:56,130
which will color the points
by the Region variable.

14
00:00:56,130 --> 00:01:00,840
And then we just want to add
the geom_point function and hit

15
00:01:00,840 --> 00:01:02,760
Enter.

16
00:01:02,760 --> 00:01:06,630
Now, in our plot, we should
see that each point is colored

17
00:01:06,630 --> 00:01:10,320
corresponding to the region
that country belongs in.

18
00:01:10,320 --> 00:01:13,580
So the countries in
Africa are colored red,

19
00:01:13,580 --> 00:01:17,070
the countries in the
Americas are colored gold,

20
00:01:17,070 --> 00:01:19,350
the countries in the
Eastern Mediterranean

21
00:01:19,350 --> 00:01:22,320
are colored green, etc.

22
00:01:22,320 --> 00:01:23,960
This really helps
us see something

23
00:01:23,960 --> 00:01:25,940
that we didn't see before.

24
00:01:25,940 --> 00:01:28,210
The points from the
different regions

25
00:01:28,210 --> 00:01:32,789
are really located in
different areas on the plot.

26
00:01:32,789 --> 00:01:34,970
Let's now instead
color the points

27
00:01:34,970 --> 00:01:37,970
according to the
country's life expectancy.

28
00:01:37,970 --> 00:01:40,509
To do this, we just
have to hit the up arrow

29
00:01:40,509 --> 00:01:43,520
to get back to our
ggplot line, and then

30
00:01:43,520 --> 00:01:47,410
delete Region and
type LifeExpectancy.

31
00:01:51,080 --> 00:01:54,250
Now, we should see that each
point is colored according

32
00:01:54,250 --> 00:01:57,630
to the life expectancy
in that country.

33
00:01:57,630 --> 00:01:59,570
Notice that before,
we were coloring

34
00:01:59,570 --> 00:02:01,930
by a factor variable, Region.

35
00:02:01,930 --> 00:02:04,800
So we had exactly
seven different colors

36
00:02:04,800 --> 00:02:08,070
corresponding to the
seven different regions.

37
00:02:08,070 --> 00:02:10,930
Here, we're coloring by
LifeExpectancy instead,

38
00:02:10,930 --> 00:02:14,720
which is a numerical variable,
so we get a gradient of colors,

39
00:02:14,720 --> 00:02:16,090
like this.

40
00:02:16,090 --> 00:02:20,020
Lighter blue corresponds to
a higher life expectancy,

41
00:02:20,020 --> 00:02:25,350
and darker blue corresponds
to a lower life expectancy.

42
00:02:25,350 --> 00:02:27,740
Let's take a look at
a different plot now.

43
00:02:27,740 --> 00:02:30,870
Suppose we were interested in
seeing whether the fertility

44
00:02:30,870 --> 00:02:34,130
rate of a country was a good
predictor of the percentage

45
00:02:34,130 --> 00:02:36,850
of the population under 15.

46
00:02:36,850 --> 00:02:39,450
Intuitively, we would
expect these variables

47
00:02:39,450 --> 00:02:41,260
to be highly correlated.

48
00:02:41,260 --> 00:02:43,980
But before trying any
statistical models,

49
00:02:43,980 --> 00:02:46,800
let's explore our
data with a plot.

50
00:02:46,800 --> 00:02:51,640
So now, let's use the ggplot
function on the WHO data again,

51
00:02:51,640 --> 00:02:53,730
but we're going to
specify in our aesthetic

52
00:02:53,730 --> 00:02:57,750
that the x variable
should be FertilityRate,

53
00:02:57,750 --> 00:03:02,390
and the y variable should
be the variable, Under15.

54
00:03:02,390 --> 00:03:04,720
Again, we want to
add geom_point,

55
00:03:04,720 --> 00:03:05,890
since we want a scatterplot.

56
00:03:09,250 --> 00:03:10,620
This is really interesting.

57
00:03:10,620 --> 00:03:13,710
It looks like the variables
are certainly correlated,

58
00:03:13,710 --> 00:03:17,340
but as the fertility rate
increases, the variable,

59
00:03:17,340 --> 00:03:21,770
Under15 starts increasing less.

60
00:03:21,770 --> 00:03:24,720
So this doesn't really look
like a linear relationship.

61
00:03:24,720 --> 00:03:28,390
But we suspect that a log
transformation of FertilityRate

62
00:03:28,390 --> 00:03:29,740
will be better.

63
00:03:29,740 --> 00:03:31,390
Let's give it a shot.

64
00:03:31,390 --> 00:03:33,840
So go ahead and scroll
up in your R console

65
00:03:33,840 --> 00:03:38,260
to the previous line, and
instead of x = FertilityRate,

66
00:03:38,260 --> 00:03:39,680
we want x = log(FertilityRate).

67
00:03:42,990 --> 00:03:45,180
And hit Enter.

68
00:03:45,180 --> 00:03:48,160
Now this looks like a
linear relationship.

69
00:03:48,160 --> 00:03:51,000
Let's try building in a
simple linear regression model

70
00:03:51,000 --> 00:03:54,320
to predict the percentage
of the population under 15,

71
00:03:54,320 --> 00:03:56,960
using the log of
the fertility rate.

72
00:03:56,960 --> 00:04:01,700
So let's call our model,
model, and use the lm function

73
00:04:01,700 --> 00:04:06,400
to predict Under15 using
as an independent variable

74
00:04:06,400 --> 00:04:07,190
log(FertilityRate).

75
00:04:09,810 --> 00:04:13,870
And our data set will be WHO.

76
00:04:13,870 --> 00:04:15,520
Let's look at the
summary of our model.

77
00:04:18,190 --> 00:04:20,190
It looks like the
log of FertilityRate

78
00:04:20,190 --> 00:04:22,890
is indeed a great
predictor of Under15.

79
00:04:22,890 --> 00:04:25,230
The variable is
highly significant,

80
00:04:25,230 --> 00:04:29,320
and our R-squared is 0.9391.

81
00:04:29,320 --> 00:04:31,370
Visualization was
a great way for us

82
00:04:31,370 --> 00:04:34,830
to realize that the log
transformation would be better.

83
00:04:34,830 --> 00:04:37,600
If we instead had just
used the FertilityRate,

84
00:04:37,600 --> 00:04:40,650
the R-squared would
have been 0.87.

85
00:04:40,650 --> 00:04:44,610
That's a pretty significant
decrease in R-squared.

86
00:04:44,610 --> 00:04:47,800
So now, let's add this
regression line to our plot.

87
00:04:47,800 --> 00:04:49,730
This is pretty easy in ggplot.

88
00:04:49,730 --> 00:04:52,420
We just have to
add another layer.

89
00:04:52,420 --> 00:04:55,030
So use the up arrow in
your R console to get back

90
00:04:55,030 --> 00:05:06,640
to the plotting line, and then
add stat_smooth(method = "lm"),

91
00:05:06,640 --> 00:05:08,620
and hit Enter.

92
00:05:08,620 --> 00:05:12,120
Now, you should see a blue
line going through the data.

93
00:05:12,120 --> 00:05:14,320
This is our regression line.

94
00:05:14,320 --> 00:05:18,200
By default, ggplot will
draw a 95% confidence

95
00:05:18,200 --> 00:05:20,760
interval shaded around the line.

96
00:05:20,760 --> 00:05:23,600
We can change this
by specifying options

97
00:05:23,600 --> 00:05:25,730
within the statistics layer.

98
00:05:25,730 --> 00:05:28,920
So go ahead and scroll
up in the R console,

99
00:05:28,920 --> 00:05:36,330
and after method = "lm", type
level = 0.99, and hit Enter.

100
00:05:36,330 --> 00:05:40,060
This will give a 99%
confidence interval.

101
00:05:40,060 --> 00:05:44,430
We could instead take away the
confidence interval altogether

102
00:05:44,430 --> 00:05:52,670
by deleting level = 0.99
and typing se = FALSE.

103
00:05:52,670 --> 00:05:56,270
Now, we just have the
regression line in blue.

104
00:05:56,270 --> 00:05:59,920
We could also change the
color of the regression line

105
00:05:59,920 --> 00:06:07,130
by typing as an option,
color = "orange".

106
00:06:07,130 --> 00:06:11,250
Now, we have an orange
linear regression line.

107
00:06:11,250 --> 00:06:13,980
As we've seen in this
lecture, scatterplots

108
00:06:13,980 --> 00:06:16,340
are great for exploring data.

109
00:06:16,340 --> 00:06:18,300
However, there are
many other ways

110
00:06:18,300 --> 00:06:22,650
to represent data visually,
such as box plots, line charts,

111
00:06:22,650 --> 00:06:26,520
histograms, heat maps,
and geographic maps.

112
00:06:26,520 --> 00:06:28,920
In some cases, it may
be better to choose

113
00:06:28,920 --> 00:06:31,980
one of these other ways
of visualizing your data.

114
00:06:31,980 --> 00:06:34,550
Luckily, ggplot
makes it easy to go

115
00:06:34,550 --> 00:06:37,900
from one type of visualization
to another, simply

116
00:06:37,900 --> 00:06:40,860
by adding the appropriate
layer to the plot.

117
00:06:40,860 --> 00:06:43,440
We'll learn more about other
types of visualizations

118
00:06:43,440 --> 00:06:47,050
and how to create them
in the next lecture.

119
00:06:47,050 --> 00:06:50,420
So what is the edge
of visualizations?

120
00:06:50,420 --> 00:06:52,720
The WHO data that
we used here is

121
00:06:52,720 --> 00:06:56,600
used by citizens,
policymakers, and organizations

122
00:06:56,600 --> 00:06:58,300
around the world.

123
00:06:58,300 --> 00:07:01,600
Visualizing the data
facilitates the understanding

124
00:07:01,600 --> 00:07:04,730
of global health
trends at a glance.

125
00:07:04,730 --> 00:07:08,820
By using ggplot in R, we're
able to visualize data

126
00:07:08,820 --> 00:07:14,110
for exploration, modeling,
and sharing analytics results.