1
00:00:08,970 --> 00:00:11,145
I consider Alberto Cairo's book,

2
00:00:11,145 --> 00:00:15,070
The Truthful Art to be required
reading for any data scientist.

3
00:00:15,070 --> 00:00:19,322
In one book, he has managed to pack in
a plethora of references to experts in

4
00:00:19,322 --> 00:00:24,400
the field while still distilling
down details into memorable lessons.

5
00:00:24,400 --> 00:00:27,486
While we won't have an opportunity
to cover everything he talks about.

6
00:00:27,486 --> 00:00:31,710
I want to talk in this lecture about the
foundation of the book in particular which

7
00:00:31,710 --> 00:00:36,140
is a five part frame work for considering
the qualities of an information graphic.

8
00:00:37,340 --> 00:00:41,470
Importantly, Cairo says that these
qualities are not independent of one

9
00:00:41,470 --> 00:00:43,690
another and that they're interrelated.

10
00:00:43,690 --> 00:00:47,000
So while I tend to be prone to
reductionism, keep in mind that

11
00:00:47,000 --> 00:00:50,510
the different qualities interact with one
another as we go through the lecture.

12
00:00:52,110 --> 00:00:56,450
The first quality of a good
visualization is that it's truthful.

13
00:00:56,450 --> 00:00:59,686
Now Cairo admits that truth could
be considered subjective and

14
00:00:59,686 --> 00:01:03,914
it's certainly on a continuum with some
things being more truthful than others.

15
00:01:03,914 --> 00:01:06,890
And then it might be impossible
to establish an absolute truth.

16
00:01:08,290 --> 00:01:09,720
But as data scientists,

17
00:01:09,720 --> 00:01:14,670
he argues, we have two obligations
when it comes to protecting the truth.

18
00:01:14,670 --> 00:01:18,774
First, we have to be honest with ourselves
when we clean and summarize data.

19
00:01:18,774 --> 00:01:22,482
Are the activities we're engaging in
likely to obscure a message because

20
00:01:22,482 --> 00:01:25,490
of the limitations that we've applied?

21
00:01:25,490 --> 00:01:29,248
We should consider explicitly each
modification that we're making to

22
00:01:29,248 --> 00:01:29,817
the data.

23
00:01:29,817 --> 00:01:32,600
And make sure we are practicing
what he calls self deception.

24
00:01:33,730 --> 00:01:37,160
Skepticism is an important skill for
data scientist.

25
00:01:37,160 --> 00:01:40,590
And we often get so focus on looking for
patterns that we reduce or

26
00:01:40,590 --> 00:01:44,800
present data in ways which are inauthentic
to the phenomena being described.

27
00:01:46,030 --> 00:01:48,310
The second obligation is to our audience.

28
00:01:49,370 --> 00:01:53,360
That is there are techniques and data
science and information visualization,

29
00:01:53,360 --> 00:01:57,350
which we can use to shine light
on to specific pieces of data.

30
00:01:57,350 --> 00:01:59,480
Indeed this is generally
the point of the fields.

31
00:02:00,580 --> 00:02:05,240
But if we omitted the ability of the
reader to explore the phenomena more fully

32
00:02:05,240 --> 00:02:07,979
this will generate doubt and
cast distrust.

33
00:02:09,930 --> 00:02:11,880
Cairo gives us a great examples of this,

34
00:02:11,880 --> 00:02:15,710
using a chart inspired by one put
out by the lobbying association.

35
00:02:15,710 --> 00:02:19,640
The national cable and
telecommunications association.

36
00:02:19,640 --> 00:02:23,035
This association lobbies on
behalf of US cable companies.

37
00:02:23,035 --> 00:02:27,238
And provided a graphic which suggested
that, after regulations were relaxed,

38
00:02:27,238 --> 00:02:30,470
cable companies invested four
times as much in the industry.

39
00:02:31,470 --> 00:02:33,650
Now the original chart
has been taken down, but

40
00:02:33,650 --> 00:02:36,090
Cairo provides an example
of what was visualized.

41
00:02:37,130 --> 00:02:41,497
And we can see here, the title of
the chart alone, was intended to draw

42
00:02:41,497 --> 00:02:46,621
the reader to the conclusion, less
regulation means more industry investment.

43
00:02:46,621 --> 00:02:49,430
But there's some immediate
problems with this graphic.

44
00:02:49,430 --> 00:02:52,030
First, it's unclear whether
the monies are adjusted for

45
00:02:52,030 --> 00:02:55,420
inflation, which is a big issue when
providing charts of financial data.

46
00:02:57,180 --> 00:03:00,630
Second, the time spans for
the two bars are different.

47
00:03:00,630 --> 00:03:04,510
One covers a span of three years, while
the other covers a span of four years.

48
00:03:05,580 --> 00:03:08,908
Think of this through the lens
of Tufts' Live Factor Analysis.

49
00:03:08,908 --> 00:03:12,310
The bars have the same width, but
don't represent the same period of time.

50
00:03:13,360 --> 00:03:15,900
This is a completely
unforgivable design choice.

51
00:03:17,250 --> 00:03:22,355
And finally, there seems to be a bunch of
missing data from years 1997 and 1998.

52
00:03:22,355 --> 00:03:28,490
And the chart stops in the year 2003,
but the information was tweeted in 2014.

53
00:03:28,490 --> 00:03:31,294
What does a more honest
view of the data look like?

54
00:03:31,294 --> 00:03:34,500
Well Cairo tracked this down,
and provided a line graph.

55
00:03:34,500 --> 00:03:36,080
We can see that after regulation,

56
00:03:36,080 --> 00:03:40,050
there were in fact years of
sustained industry investment.

57
00:03:40,050 --> 00:03:44,742
We see a slight drop after deregulation,
then a massive spike in spending.

58
00:03:44,742 --> 00:03:47,957
And then a significant fall
from the period post 2002,

59
00:03:47,957 --> 00:03:50,729
much of which is excluded
from the original chart.

60
00:03:51,760 --> 00:03:55,030
While it's unclear whether the dollar
values are in fact corrected for

61
00:03:55,030 --> 00:03:59,440
inflation, seems pretty unreasonable
to suggest from this data alone

62
00:03:59,440 --> 00:04:03,170
that regulations stems cable company
investment and infrastructure.

63
00:04:04,680 --> 00:04:08,680
The lesson here is that,
while there's maybe no absolute truth,

64
00:04:08,680 --> 00:04:11,900
there are ways of displaying data
which are more truthful than others.

65
00:04:13,620 --> 00:04:17,050
The second quality of a visualization
to consider is whether it is

66
00:04:17,050 --> 00:04:17,700
functional or not.

67
00:04:18,810 --> 00:04:22,885
This again is continuum when we saw
an example previously with the slides on

68
00:04:22,885 --> 00:04:25,560
data-ink-ratio provided
by Darkhorse Analytics.

69
00:04:26,620 --> 00:04:30,400
If you expect your audience would
want to compare two data series.

70
00:04:30,400 --> 00:04:34,850
Direct labeling of them is one
way to increase comprehension.

71
00:04:34,850 --> 00:04:38,380
Well there are many heuristics for
increasing functionality,

72
00:04:38,380 --> 00:04:41,792
I would like you to remember
the user test we did with.

73
00:04:41,792 --> 00:04:47,040
I think now is an exciting time for
understating situated human behavior.

74
00:04:47,040 --> 00:04:50,652
The web has dramatically increased
trainability to test social and

75
00:04:50,652 --> 00:04:54,146
behavioral hypothesis that scale
across diverse populations.

76
00:04:54,146 --> 00:04:58,222
And crowd power and mechanical
tuck are just two example of this.

77
00:04:58,222 --> 00:05:02,490
Understanding how functional a given
approach to a visualization is doesn't

78
00:05:02,490 --> 00:05:06,260
necessarily require expensive
ethnographic research.

79
00:05:06,260 --> 00:05:11,706
It's now much easier to test
hypotheses directly with crowd sources.

80
00:05:11,706 --> 00:05:15,849
The third quality of visualization, Cairo
shares is whether it is beautiful or not.

81
00:05:17,350 --> 00:05:20,950
I think that this requires knowing
a great deal about your audience.

82
00:05:20,950 --> 00:05:24,774
There are different perceptions based
on people's lived experiences and

83
00:05:24,774 --> 00:05:27,576
aspects about themselves
such as gender and culture.

84
00:05:27,576 --> 00:05:31,848
There's a lot to building a beautiful
information visualization experience,

85
00:05:31,848 --> 00:05:35,000
which depends on situational and
contextual factors too.

86
00:05:36,060 --> 00:05:38,680
For instance,
if you're building a chart for a report,

87
00:05:38,680 --> 00:05:42,910
the colors and styles that you might
choose, even down the font choice,

88
00:05:42,910 --> 00:05:46,990
might be more related to beauty than for
instance, function.

89
00:05:46,990 --> 00:05:50,760
But the interplay of these elements is
something which is difficult to ignore.

90
00:05:51,810 --> 00:05:56,520
I wonder is this broad topic of beauty one
of the conflationary factors of the chart

91
00:05:56,520 --> 00:05:58,680
junk memorability issue?

92
00:05:58,680 --> 00:06:03,280
Where as Tufte argued that the duck
junk is distracting and misleading,

93
00:06:03,280 --> 00:06:07,640
the study by Bateman et al showed that it
can make visualizations more memorable.

94
00:06:07,640 --> 00:06:11,270
Is this disconnect an issue of
beauty in information graphics?

95
00:06:12,790 --> 00:06:15,600
Here's one parting thought
on beauty in information

96
00:06:15,600 --> 00:06:17,400
that perhaps drives home this quality.

97
00:06:18,740 --> 00:06:21,967
One of the most popular tech magazines,
well known for

98
00:06:21,967 --> 00:06:25,836
it's full page information graphic
spreads is Wired Magazine.

99
00:06:25,836 --> 00:06:30,200
On Wired's website,
they tag these articles as info porn.

100
00:06:30,200 --> 00:06:33,920
A reference to the visceral reaction
that they hope readers will view

101
00:06:33,920 --> 00:06:35,140
when viewing the articles.

102
00:06:37,200 --> 00:06:40,150
The fourth quality of
visualization from Cairo

103
00:06:40,150 --> 00:06:42,870
is that information graphics
should be insightful.

104
00:06:43,870 --> 00:06:46,280
They shouldn't just replicate
data in tables, but

105
00:06:46,280 --> 00:06:50,990
they should draw the viewer in such that
the viewer has an aha or eureka moment.

106
00:06:52,040 --> 00:06:54,710
I think about this largely
in the context of academic

107
00:06:54,710 --> 00:06:57,130
papers where I do most of my work.

108
00:06:57,130 --> 00:07:01,379
A common issue that novice writers have in
this area is the inclusion of a figure for

109
00:07:01,379 --> 00:07:02,989
every piece of data they have.

110
00:07:02,989 --> 00:07:07,360
This is really unnecessary and adds
bloat and contributes to reader fatigue.

111
00:07:07,360 --> 00:07:11,200
But the right figure drives home
the results of a study quickly.

112
00:07:11,200 --> 00:07:13,730
Scientific peer review is
common in my field, but

113
00:07:13,730 --> 00:07:18,090
readers don't have the large amounts of
time that they need to do full reviews.

114
00:07:18,090 --> 00:07:20,020
When flipping through a submitted paper,

115
00:07:20,020 --> 00:07:24,990
the reviewer sees an image which jumps off
the page and gives them immediate insight.

116
00:07:24,990 --> 00:07:27,730
This sets them up to understand
the more nuanced story

117
00:07:27,730 --> 00:07:29,430
to be told throughout
the rest of the paper.

118
00:07:30,720 --> 00:07:33,650
The final quality of
visualization which Cairo

119
00:07:33,650 --> 00:07:36,520
proposes is information graphics
should be enlightening.

120
00:07:37,770 --> 00:07:39,678
This is different than insightful.

121
00:07:39,678 --> 00:07:44,262
In fact, Cairo argues that this quality
is composed of the previous four,

122
00:07:44,262 --> 00:07:47,633
truthful, functional,
beautiful and insightful.

123
00:07:47,633 --> 00:07:51,189
But he separates and then adds
a social responsibility dimension.

124
00:07:52,640 --> 00:07:55,890
I don't think there's an obvious
answer to this question.

125
00:07:55,890 --> 00:07:57,070
But without a doubt,

126
00:07:57,070 --> 00:08:02,160
Cairo's Five Qualities of Great
Visualizations being truthful, functional,

127
00:08:02,160 --> 00:08:06,780
beautiful, insightful, and
enlightening are well worth considering.

128
00:08:06,780 --> 00:08:08,873
Cairo has more great work, and

129
00:08:08,873 --> 00:08:13,220
following this lecture I have
assigned a reading of his as well as

130
00:08:13,220 --> 00:08:17,971
a discussion question to encourage
you to weigh in on the amorality of

131
00:08:17,971 --> 00:08:22,017
information visualization as
it relates to data science.

132
00:08:22,017 --> 00:08:24,420
And this ends the week of the course.

133
00:08:24,420 --> 00:08:27,019
We've covered a lot of
material very quickly.

134
00:08:27,019 --> 00:08:30,491
But I really do believe that these
two authors Alberto Ciro and

135
00:08:30,491 --> 00:08:34,164
Edward Tufty are fundamental in
building a strong foundation for

136
00:08:34,164 --> 00:08:38,860
the applied data science skills, we'll
focus on in the second week of the course.