1
00:00:12,510 --> 00:00:18,120
Welcome back to the Oxford Mathematics Public Lectures Home Edition. This will be your fifth event.

2
00:00:18,120 --> 00:00:23,820
My name is Alan Guerrieri, and I'm in charge of external relations for the Mathematical Institute.

3
00:00:23,820 --> 00:00:26,190
Special thanks to sponsors.

4
00:00:26,190 --> 00:00:34,230
Market access markets are leading quantitative driven electronic market maker with offices in London, Singapore and New York.

5
00:00:34,230 --> 00:00:39,070
Their ongoing support is crucial in providing you quality content.

6
00:00:39,070 --> 00:00:49,060
The current pandemic crisis, as underlined, the importance of mathematics in numbers not a day goes by without news about the reproduction number.

7
00:00:49,060 --> 00:00:53,380
The number of case, the number of deaths and sadly, the number of death.

8
00:00:53,380 --> 00:01:01,120
All these numbers collected, evaluated and presented as led to much confusion, manipulation and endless argument.

9
00:01:01,120 --> 00:01:09,580
The problem goes straight to the heart of what statistic is, and it is used and sometimes abused in society more than ever.

10
00:01:09,580 --> 00:01:16,060
There is a need for clear explanation of numbers and how they are used in decision making processes.

11
00:01:16,060 --> 00:01:22,180
I'm particularly happy that Tim Harford has accepted the invitation to give a lecture about his issues tonight.

12
00:01:22,180 --> 00:01:28,420
Many of you know Tim to his work and the Financial Times the BBC, but to his books,

13
00:01:28,420 --> 00:01:35,800
is sure more or less on Radio four has been particularly influential in bringing light into the recent darkness tonight.

14
00:01:35,800 --> 00:01:40,300
Tim will talk about his recent book, How to Make the World Add Up.

15
00:01:40,300 --> 00:01:44,290
Thank you very much, Tim, for doing this. Please start now.

16
00:01:44,290 --> 00:01:50,800
Hi, Tim Harford here, and thanks so much to the Oxford Mathematics Department for hosting this.

17
00:01:50,800 --> 00:01:57,820
Talk about my book before I actually start talking about the book and answer the questions that you've sent in on Twitter.

18
00:01:57,820 --> 00:02:05,260
One moment. And if you're watching this after the 14th of October, you can skip ahead a minute.

19
00:02:05,260 --> 00:02:15,400
I wanted to talk about Blackwell's books The Amazing Bookshop on Broad Street, the heart of Oxford, when my book came out.

20
00:02:15,400 --> 00:02:20,470
It sold out absolutely everywhere except Blackwell's books,

21
00:02:20,470 --> 00:02:25,000
who had actually gone to the trouble of having enough confidence in me to order enough books.

22
00:02:25,000 --> 00:02:29,440
They didn't instantly sell out, so I wanted to say thank you to them.

23
00:02:29,440 --> 00:02:40,690
So I've teamed up with them to supply personalised dedications of the book, which isn't very easy in this time of coronavirus.

24
00:02:40,690 --> 00:02:46,060
So if you're actually if you're anywhere in the UK, you can get black girls to send you the book.

25
00:02:46,060 --> 00:02:59,020
If you're local, you can order the book. The book looks like this how to make the world that up, and it's up kind of them to to agree to do this.

26
00:02:59,020 --> 00:03:06,640
But if you email Oxford at Blackwell's Echo Dot UK before the 14th of October,

27
00:03:06,640 --> 00:03:12,880
if you include your name, your contact details, the person you would like me to sign the book for.

28
00:03:12,880 --> 00:03:18,430
And if you've got a short message like Happy Birthday or Merry Christmas is not too early for Christmas,

29
00:03:18,430 --> 00:03:23,020
presents just include that I will come into Blackwell's.

30
00:03:23,020 --> 00:03:29,710
I will sign a whole bunch of books, I will personalise them for you. And then after I've done that, you can get in contact with Blackwell.

31
00:03:29,710 --> 00:03:34,390
As you can go in, you can pick up your book, you can ask them to send it to you.

32
00:03:34,390 --> 00:03:38,200
So please do that before the 14th of October 2020.

33
00:03:38,200 --> 00:03:45,920
And then the book should be available after the 16th of October 2020 to collect or to arrange to deliver,

34
00:03:45,920 --> 00:03:52,460
OK, personalised Blackwell's book message over.

35
00:03:52,460 --> 00:03:58,100
Let's talk about storks and babies where babies come from.

36
00:03:58,100 --> 00:04:04,790
You may have heard the the old saying that the stork delivers your baby brother or your baby sister.

37
00:04:04,790 --> 00:04:12,440
That's where babies come from. And you may alternatively have heard a number of rather wild,

38
00:04:12,440 --> 00:04:18,500
implausible and somewhat disgusting theories about alternative explanations for where babies come from.

39
00:04:18,500 --> 00:04:23,330
You can put those out of your mind because the truth is babies do come from storks.

40
00:04:23,330 --> 00:04:32,090
Storks do deliver babies. And if you doubt that for a moment, be reassured that I can prove that with statistics.

41
00:04:32,090 --> 00:04:41,030
It's very simple. What you need to do is gather together a list of the number of babies born in each country and say,

42
00:04:41,030 --> 00:04:48,770
Europe per year and the best estimate of the breeding population of storks in each country.

43
00:04:48,770 --> 00:04:58,130
And you can just plop them a simple scatterplot babies and storks, and you will see just eyeballing the graph that there appears to be a relationship.

44
00:04:58,130 --> 00:05:07,880
And when you do the maths properly, you will find a very strong correlation, far too strong to have been produced by random chance.

45
00:05:07,880 --> 00:05:13,790
And in case you think, well, why would I trust that guy, Tim Harford, what's he ever done for the world of statistics?

46
00:05:13,790 --> 00:05:18,170
I can assure you that this has survived peer review.

47
00:05:18,170 --> 00:05:27,620
There is, in fact, a published paper entitled Storks Deliver Babies P equals zero point zero zero eight.

48
00:05:27,620 --> 00:05:36,560
And without getting into all technical details, all the zeros after the decimal point mean this relationship is real?

49
00:05:36,560 --> 00:05:44,810
It's not a coincidence. There is a strong correlation between storks and babies.

50
00:05:44,810 --> 00:05:52,530
Now, by this point, you may be wondering how exactly the trick is done, or perhaps you figured out the trick, it's pretty simple.

51
00:05:52,530 --> 00:06:00,030
Think about a place like Monaco or Luxembourg. There's not a lot of room for babies, and there's not a lot of room for storks.

52
00:06:00,030 --> 00:06:04,620
Alternatively, if you mention a country such as Germany or Poland,

53
00:06:04,620 --> 00:06:12,210
loads of space for storks and plenty of space for a high population, which means plenty of space for babies too.

54
00:06:12,210 --> 00:06:20,910
So what this relationship is measuring is perfectly real. But what this relationship is measuring is basically big countries versus small countries.

55
00:06:20,910 --> 00:06:25,170
It's a land area that's really being tracked now.

56
00:06:25,170 --> 00:06:33,450
Congratulations. If you figured out exactly how the trick was done, but if you if you want to understand this sort of statistical conjuring,

57
00:06:33,450 --> 00:06:38,970
there's no better book to read than how to lie with statistics,

58
00:06:38,970 --> 00:06:47,070
which is an elegant little work was published first in 1954 and written by a journalist named Darrel Huff,

59
00:06:47,070 --> 00:06:53,220
and immediately got rave reviews in the New York Times and elsewhere and went on to be so,

60
00:06:53,220 --> 00:06:59,460
they say, the bestselling book about statistics ever published.

61
00:06:59,460 --> 00:07:05,430
And I can see why the book has been so successful. I read it when I was a teenager, I learnt a lot.

62
00:07:05,430 --> 00:07:09,930
There's great. It's funny, it's got cartoons. The examples are really good.

63
00:07:09,930 --> 00:07:15,690
They're really incisive. They're written with with real wit. They have a dodgy grass, have a dodgy maps,

64
00:07:15,690 --> 00:07:24,720
and the whole thing will really help you avoid being fooled by phoney statistics, by people who are trying to deceive you.

65
00:07:24,720 --> 00:07:28,890
And of course, there are plenty of people out there who try to deceive you.

66
00:07:28,890 --> 00:07:35,790
Daryl Huff lays out his stall very clearly. He says the the crooks already know these tricks.

67
00:07:35,790 --> 00:07:40,950
It's the honest folk who have to learn them in self-defence.

68
00:07:40,950 --> 00:07:42,840
So there's nothing wrong with that.

69
00:07:42,840 --> 00:07:53,010
There's nothing wrong with a witty little book about how to lie with statistics, except that over the last 10 years or so,

70
00:07:53,010 --> 00:08:00,720
I've become increasingly uneasy about that book and about what it represents and what

71
00:08:00,720 --> 00:08:07,080
does it say about statistics and what does it say about our attitude to statistics?

72
00:08:07,080 --> 00:08:16,110
The most successful book ever written about them is from cover to cover a warning about misinformation.

73
00:08:16,110 --> 00:08:27,880
Once you start trading down that path, it becomes very easy to imagine that lying with statistics is all anybody ever does.

74
00:08:27,880 --> 00:08:39,130
And I think that's a problem, because let me contrast how to lie with statistics published in 1954 with something else that happened in 1954,

75
00:08:39,130 --> 00:08:45,640
which was that two British epidemiologists, Richard Dole and Austin Bradford Hill,

76
00:08:45,640 --> 00:08:54,220
published some of the first compelling evidence that smoking cigarettes dramatically increases your risk of lung cancer.

77
00:08:54,220 --> 00:09:00,760
Now, this is a really important result, and there were other scientists working on this result around the same time.

78
00:09:00,760 --> 00:09:13,360
But it matters to discover that this very popular product that people use all over the world to discover that that's killing people in large numbers,

79
00:09:13,360 --> 00:09:17,470
that it's extremely dangerous. Well, that's a little piece of information.

80
00:09:17,470 --> 00:09:26,980
And that piece of information has contributed, of course, to substantial declines in rates of smoking in developed countries and a longer life

81
00:09:26,980 --> 00:09:33,640
expectancy and reduced heart disease and reduced lung cancer and all kinds of good things.

82
00:09:33,640 --> 00:09:35,230
And this is a life saving discovery.

83
00:09:35,230 --> 00:09:47,430
This is tens of millions of people having their lives saved by war, by war, by a discovery that was made using statistics.

84
00:09:47,430 --> 00:09:52,530
Because when you think about it, you realise you couldn't have figured this out in any other way.

85
00:09:52,530 --> 00:09:59,550
That was one doctor in Argentina who treated a vast number of patients with throat cancer and lung cancer,

86
00:09:59,550 --> 00:10:04,140
and he noticed that cancer rates were on the increase.

87
00:10:04,140 --> 00:10:08,430
But actually, even just to notice that it would be an unusual thing,

88
00:10:08,430 --> 00:10:16,050
much easier to notice that lung cancer is dramatically increasing through statistics and by statistics.

89
00:10:16,050 --> 00:10:23,680
I don't mean anything terribly sophisticated to just mean somebody's got to count cases of lung cancer and write them down somewhere.

90
00:10:23,680 --> 00:10:28,750
So even to notice it, even to see that there's a problem requires a statistical perspective,

91
00:10:28,750 --> 00:10:37,990
even even quite a simple one, and to diagnose what is causing this huge increase in lung cancer.

92
00:10:37,990 --> 00:10:45,250
Well, then you really need to ask the questions in a statistical way to ask them very carefully.

93
00:10:45,250 --> 00:10:49,270
And bear in mind that Richard Doll didn't think that cigarettes were to blame.

94
00:10:49,270 --> 00:10:57,520
Bradford Hill and Dole, by the way, both smokers, at least at the beginning of that research, think you know, the research cured them of that habit.

95
00:10:57,520 --> 00:11:03,670
But Richard Dole thought that it was probably something to do with with the tarmac on the roads.

96
00:11:03,670 --> 00:11:06,670
And another theory, quite quite a plausible theory,

97
00:11:06,670 --> 00:11:14,140
is that it was pollution from cars because you think about this huge increase in lung cancer that's taking place through the late 1930s,

98
00:11:14,140 --> 00:11:20,380
through the 1940s, into the 1950s. Well, what else is going on a lot more cars?

99
00:11:20,380 --> 00:11:31,930
So it makes sense. But to really examine and to figure out, is it the cigarettes or is it the cars that requires the statistical lens?

100
00:11:31,930 --> 00:11:36,490
So what Dole and Bradford Hill were doing was not incredibly complicated,

101
00:11:36,490 --> 00:11:45,910
it did require care and it required effort at what stage they sent what they called a question or two about 60000 doctors.

102
00:11:45,910 --> 00:11:55,510
They thought, Oh, if we if we contact all the doctors, that's that's a great group of people to contact for start 80 percent of them smoke.

103
00:11:55,510 --> 00:12:00,610
But also, we don't want to lose track of doctors because they're all on the medical register.

104
00:12:00,610 --> 00:12:05,350
And when a doctor dies, we'll know what he or she died of.

105
00:12:05,350 --> 00:12:08,290
So they're a great group to examine. Well, this, of course,

106
00:12:08,290 --> 00:12:17,110
made them extremely unpopular with doctors because nobody wants to be told that this thing that you're smoking is is killing you.

107
00:12:17,110 --> 00:12:23,460
And at one Dr. Buttonholed Austin, Bradford Hill, I a dinner party and and said, Oh.

108
00:12:23,460 --> 00:12:27,630
So you're the chap who wants us all to quit smoking, are you?

109
00:12:27,630 --> 00:12:33,600
And Bradford Hill responded, No, not at all. I just want to see how you die.

110
00:12:33,600 --> 00:12:40,020
If you quit smoking, I want to see how you die. And if you don't quit smoking, I want to see how you die.

111
00:12:40,020 --> 00:12:49,110
So do what you wish. Quit or keep going. I will chalk up your death anyway, and it will be very useful to me.

112
00:12:49,110 --> 00:12:52,710
I should mention Austin Bradford Hill, originally trained as an economist.

113
00:12:52,710 --> 00:12:57,300
It's where he learnt his charm. So this is important work.

114
00:12:57,300 --> 00:13:07,230
This is vital work. But now let's think about these two visions of statistics that have emerged in the same year 1954.

115
00:13:07,230 --> 00:13:18,600
You've got Darel Health, who says it's a trick. I'm going to show you where the magician hides the rabbit before pulling it out of a hat.

116
00:13:18,600 --> 00:13:27,090
And it's great fun and it's clever and we can appreciate it, but don't ever take it seriously.

117
00:13:27,090 --> 00:13:38,900
Don't ever think of it as more than a game. In the very same year, Richard Doll, an Austin, Bradford Hill, using it not as a trick, but as a tool,

118
00:13:38,900 --> 00:13:49,960
a tool to understand the world, and they realised that this is not a game or if it is a game, the stakes are incredibly high.

119
00:13:49,960 --> 00:13:56,620
And my aim in writing how to make the world add up is above everything else.

120
00:13:56,620 --> 00:14:03,940
To urge you not to be seduced by this vision of Daryl Huff, where statistics are a trick.

121
00:14:03,940 --> 00:14:14,160
And instead to embrace the perspective of Richard Doll and Austin Bradford Hill to use statistics as a tool like

122
00:14:14,160 --> 00:14:24,200
an astronomer uses a telescope or radiologist uses a uses an X-ray machine or air traffic control use radar.

123
00:14:24,200 --> 00:14:31,660
There are certain things about the world. That we can't perceive in any other way.

124
00:14:31,660 --> 00:14:35,830
And we need statistics to show us what's going on.

125
00:14:35,830 --> 00:14:40,870
And I think the virus, the pandemic has underlined that point.

126
00:14:40,870 --> 00:14:45,490
I mean, it's a it's a heck of a way to be proved right about what you were trying to say.

127
00:14:45,490 --> 00:14:50,740
But to understand where the virus is and who's got it and how quickly it's spreading.

128
00:14:50,740 --> 00:14:55,000
And is it? Is it spreading or is it in retreat?

129
00:14:55,000 --> 00:15:00,580
And how dangerous is it? And who is it dangerous to? And how can we fight it?

130
00:15:00,580 --> 00:15:08,560
What are the treatments that are most effective? These are all one way or another statistical questions.

131
00:15:08,560 --> 00:15:17,710
And I think we learnt back in February March, when we were scrambling to make incredibly high stakes decisions with basically very little information.

132
00:15:17,710 --> 00:15:20,560
We learnt that having the data matters,

133
00:15:20,560 --> 00:15:30,180
having the statistics matters and when they're not available or whether they're patchy and incomplete, we suffer as a result.

134
00:15:30,180 --> 00:15:35,020
So it makes me angry to just hear people going, oh yeah, lies damned lies and statistics,

135
00:15:35,020 --> 00:15:43,540
so you can prove anything with statistics because this is a life saving tool, a life saving tool.

136
00:15:43,540 --> 00:15:57,840
And we should not take it for granted. So one of the things that I'm trying to do in the book is to give people a little bit more confidence to

137
00:15:57,840 --> 00:16:06,450
use that tool for themselves to to teach you certain habits of mind that I've picked up in 13 years,

138
00:16:06,450 --> 00:16:10,350
presenting more or less questions to ask things.

139
00:16:10,350 --> 00:16:18,810
To be curious about that will help you evaluate the claims that you see and think more clearly about them.

140
00:16:18,810 --> 00:16:26,460
One friend of mine, Matt Parker, the stand up mathematician, said, Oh so teams, so you've written this book about statistics.

141
00:16:26,460 --> 00:16:34,860
And I said, Well, no, I haven't. I haven't written a book about statistics. I've written a book about how to think about the world.

142
00:16:34,860 --> 00:16:40,800
And for me, it's just statistics are an important tool for thinking about the world,

143
00:16:40,800 --> 00:16:46,380
and I really believe that it's not as complicated as we often make it seem.

144
00:16:46,380 --> 00:16:52,230
The questions we need to ask don't require a great deal of technical expertise.

145
00:16:52,230 --> 00:16:59,150
Now. I spent some time talking about what happened in 1954,

146
00:16:59,150 --> 00:17:03,650
and there is a reason why I'm so obsessed with this sort of period of history in

147
00:17:03,650 --> 00:17:10,340
the in the early 1950s is because that was also the time at which the tobacco

148
00:17:10,340 --> 00:17:16,580
companies started realising that the scientific evidence was coming in showing

149
00:17:16,580 --> 00:17:20,990
that their product was dangerous and they had to figure out their own response.

150
00:17:20,990 --> 00:17:25,880
And the way they responded, I think was quite brilliant in a in a dark way.

151
00:17:25,880 --> 00:17:33,890
I mean, think about the problem you face. Your product is probably killing the people who consume it.

152
00:17:33,890 --> 00:17:38,200
Think about how anxious people get about even even small things.

153
00:17:38,200 --> 00:17:47,510
We worry about things like gluten these days, and this is a product that is killing people in large numbers,

154
00:17:47,510 --> 00:17:53,810
like thousands of people every week are dying. What on earth is your response to that?

155
00:17:53,810 --> 00:18:01,140
And what the tobacco industry realised was that people who smoke?

156
00:18:01,140 --> 00:18:08,730
They would like to believe that the product is safe, they would like to believe that this thing that they enjoy,

157
00:18:08,730 --> 00:18:15,880
that they find quite social, that they would find it difficult to quit. They'd like to believe that it is not killing them.

158
00:18:15,880 --> 00:18:24,730
So they want reasons to believe, and you don't need to prove that cigarettes are safe.

159
00:18:24,730 --> 00:18:32,290
All you need to do is give them reason to doubt that they're dangerous.

160
00:18:32,290 --> 00:18:37,290
Scepticism, doubt is the weapon.

161
00:18:37,290 --> 00:18:46,140
I think this was a brilliantly simple realisation when you look at the evidence, it does seem that doubt has a particular kind of power.

162
00:18:46,140 --> 00:18:52,890
There's lots of evidence in the book on motivated reasoning and people reaching conclusions that they want to reach.

163
00:18:52,890 --> 00:19:03,900
But one of the very first studies that I discuss by two psychologists called Carey Edwards and Edward Smith shows that negative arguments,

164
00:19:03,900 --> 00:19:09,590
doubtful arguments. They flow with a particular fluidity.

165
00:19:09,590 --> 00:19:15,380
That it's sure people are pretty good at coming up with reasons to believe what they want to believe,

166
00:19:15,380 --> 00:19:24,230
but they're really good at coming up with reasons to disbelieve what they want to disbelieve, to doubt what they want to doubt.

167
00:19:24,230 --> 00:19:32,270
And that's why slogans about lies, damn lies and statistics or fake news, they're so powerful.

168
00:19:32,270 --> 00:19:39,060
You get to this point where we've all had enough of experts because the experts are telling us things we don't want to see.

169
00:19:39,060 --> 00:19:45,280
We don't want to believe. Now we've seen these sorts of.

170
00:19:45,280 --> 00:19:54,790
Doubt seeding strategies, we've seen this sort of approach used in all kinds of politics and debate over climate change and so on.

171
00:19:54,790 --> 00:19:57,170
But it started with cigarettes.

172
00:19:57,170 --> 00:20:05,060
It started with the cigarette companies sounding very reasonable, saying, Well, look, the world's a very uncertain place.

173
00:20:05,060 --> 00:20:11,110
It's true. Experts disagree. It's true. More research is needed.

174
00:20:11,110 --> 00:20:15,610
That's true. I mean, it sounds very reasonable. Sounds very scientific.

175
00:20:15,610 --> 00:20:21,970
But in the end, you very quickly get from the Royal Society's motto Nullius in verba.

176
00:20:21,970 --> 00:20:26,830
Take nobody's word for it to. Nobody knows anything.

177
00:20:26,830 --> 00:20:34,460
And if nobody knows anything, you can believe whatever you want.

178
00:20:34,460 --> 00:20:39,440
In the nineteen sixties, nineteen sixty five, how I remember rightly,

179
00:20:39,440 --> 00:20:49,230
the US Congress held a hearing to decide whether cigarette packets should have health warnings on them.

180
00:20:49,230 --> 00:20:54,020
Was a Senate hearing, and they called in all kinds of expert witnesses,

181
00:20:54,020 --> 00:21:02,760
the epidemiologists and others to discuss the risks of smoking and the pros and cons of putting these health warnings on.

182
00:21:02,760 --> 00:21:16,160
And one of the witnesses. Who was called sat down in front of the senators and started to explain that there's a correlation.

183
00:21:16,160 --> 00:21:20,400
Between storks and babies.

184
00:21:20,400 --> 00:21:31,630
And this correlation is pretty robust, but of course, that's not because storks actually deliver babies, it's because larger places.

185
00:21:31,630 --> 00:21:44,080
Have plenty of room for babies and plenty of room for storks. The expert witness, his name was Darrel Huff, author of How to Lie with Statistics,

186
00:21:44,080 --> 00:21:53,380
because his brand of scepticism searing into cynicism was absolutely perfect for the tobacco industry.

187
00:21:53,380 --> 00:21:57,910
It's exactly the kind of message they wanted, so they hired him.

188
00:21:57,910 --> 00:22:02,770
They paid him to work on a sequel called How to Line with Smoking Statistics.

189
00:22:02,770 --> 00:22:16,330
And they persuaded him to testify in front of the Senate. And the senator leading the the committee hearing said, do you honestly mean to tell us,

190
00:22:16,330 --> 00:22:26,780
you think there is this casual, a connexion between cigarettes and lung cancer, as there is between storks and babies?

191
00:22:26,780 --> 00:22:36,300
And Daryl Huff replied, Well, the two seem to me to be about the same.

192
00:22:36,300 --> 00:22:48,690
That's when I really start to worry about how to lie with statistics about what it represents because it's right to maintain a healthy scepticism.

193
00:22:48,690 --> 00:22:54,720
It's right to notice that statistics are often used to deceive us.

194
00:22:54,720 --> 00:23:05,310
But when we descend into that mode of thinking that it's all a trick, it's all a joke that you can't believe anything very quickly.

195
00:23:05,310 --> 00:23:10,350
You end up siding with a guy who sits in front of the Senate commission and says

196
00:23:10,350 --> 00:23:16,920
that cigarettes and lung cancer is basically just the same as storks and babies.

197
00:23:16,920 --> 00:23:23,430
We, we end up in this sort of defensive crouch where we're basically not willing to believe anything on its merits.

198
00:23:23,430 --> 00:23:30,060
We're not willing to weigh up the evidence. We're just scrabbling around for reasons to believe whatever we want to believe.

199
00:23:30,060 --> 00:23:38,760
And I think that's a shame if I don't think it's a tragedy and I don't think it's that hard to push back and to stand up for ourselves and to say,

200
00:23:38,760 --> 00:23:47,430
actually, this stuff isn't so difficult. I can tell a clear argument from an unclear argument.

201
00:23:47,430 --> 00:23:55,440
I can tell the difference between someone who's seeking after truth and someone who's just trying to win some political debate or score a few points.

202
00:23:55,440 --> 00:24:01,800
I can tell the difference between truth and lies. It's not that hard.

203
00:24:01,800 --> 00:24:07,920
There is an old story about. Galileo and his telescope.

204
00:24:07,920 --> 00:24:18,570
The time he was looking at the moons of Jupiter, the rings of Saturn, the Catholic Church was persecuting him and he said to the Cardinals.

205
00:24:18,570 --> 00:24:24,040
No, I've got my telescope here. Just just take a look.

206
00:24:24,040 --> 00:24:31,780
I'll show you I'll show you what I see about the Solar System and that the Cardinals wouldn't look.

207
00:24:31,780 --> 00:24:37,810
They thought the telescope itself was full of trickery. Some kind of magicians to.

208
00:24:37,810 --> 00:24:41,830
Now, that story has been exaggerated over the years,

209
00:24:41,830 --> 00:24:52,570
but we still tell it and we tell it in this sort of smug way that we dismiss these outdated beliefs and this these religious extremists.

210
00:24:52,570 --> 00:25:00,970
I don't think we should be so smug. Because I see people behaving like that around me every day.

211
00:25:00,970 --> 00:25:07,560
Remember, I described statistics as as like a telescope, the way to see facts about the world.

212
00:25:07,560 --> 00:25:12,960
Facts about the human brain. The human body. About the economy, about the environment.

213
00:25:12,960 --> 00:25:19,260
About this incredibly complex world in which we live that way to see things we can't see in any other way.

214
00:25:19,260 --> 00:25:28,830
So that's why I described them as a telescope. And every day I see people who won't pick up the statistical telescope, they won't use it.

215
00:25:28,830 --> 00:25:36,830
They can look through it for exactly the same reason that allegedly the Cardinals wouldn't look to Galileo's.

216
00:25:36,830 --> 00:25:46,360
Because they were afraid of being tricked. So don't be afraid, I urge you to have a bit more confidence in yourself.

217
00:25:46,360 --> 00:25:49,390
I don't think it's as hard as we sometimes make it out.

218
00:25:49,390 --> 00:25:57,680
And the 10 rules of thumb that I've outlined in the book, I hope, are rules that anybody can use.

219
00:25:57,680 --> 00:26:03,290
So have some confidence in your ability to distinguish truth from fiction.

220
00:26:03,290 --> 00:26:08,000
Have some confidence in your ability to think clearly about the world.

221
00:26:08,000 --> 00:26:13,730
A lot of it is simply about wanting to know, wanting to understand.

222
00:26:13,730 --> 00:26:19,520
And I think all of us should be willing to pick up this statistical telescope.

223
00:26:19,520 --> 00:26:28,560
And to look around, you'll be amazed what we see. Thanks for listening.

224
00:26:28,560 --> 00:26:40,950
I have received many, many questions on Twitter because I asked some of the questions were along the lines of Why are you taking questions on Twitter?

225
00:26:40,950 --> 00:26:47,160
Why can't you give the talk live? Well, you know, there are different ways to give talks.

226
00:26:47,160 --> 00:26:52,500
We're all trying to experiment and adapt in this, this COVID era.

227
00:26:52,500 --> 00:26:58,860
I'm really delighted to have been asked to speak at the Maths Institute, and I hope you've enjoyed what I've said.

228
00:26:58,860 --> 00:27:05,730
But I do have these questions and I wanted to leave quite a bit of time for questions because the questions that have come in, I think,

229
00:27:05,730 --> 00:27:13,650
are a really great way to get into some of the ideas in the book and some of the practical tips that I outline that I outline in the book.

230
00:27:13,650 --> 00:27:18,690
I should wave the book around. Shouldn't I go out to make the world that up?

231
00:27:18,690 --> 00:27:24,990
10 Rules for thinking differently about numbers. So, OK, questions.

232
00:27:24,990 --> 00:27:35,460
Question from Sarah Brighton. What are the best questions to ask when we see statistics used in the news to help understand them?

233
00:27:35,460 --> 00:27:40,070
It's a great question, partly because basically the the 10 rules in the book, the 10,

234
00:27:40,070 --> 00:27:45,120
they're not really Ten Commandments, the ten habits of mind, plus the golden rule at the end of the book.

235
00:27:45,120 --> 00:27:49,680
I mean, they are my answer to the question. The book is basically the answer to this question.

236
00:27:49,680 --> 00:27:54,130
What questions should we ask and when we see statistics used in the news?

237
00:27:54,130 --> 00:28:02,580
Let me give you a few examples. Take a little bit of time to to to focus on this question because it is so central to the idea of the book.

238
00:28:02,580 --> 00:28:09,420
So the first question I would ask, I think, might surprise some people.

239
00:28:09,420 --> 00:28:16,110
So I begin the first chapter of the book by telling the story of Abraham produced Abraham.

240
00:28:16,110 --> 00:28:24,450
Radius was a great art critic in the late 19th and early 20th century, and I describe him sitting in his in his villa in Monaco,

241
00:28:24,450 --> 00:28:32,400
enjoying a well-deserved retirement and enjoying the admiration of the art world for his expertise.

242
00:28:32,400 --> 00:28:43,810
When in the nineteen thirties, a charming Dutch lawyer named Gerard Boone visited him and showed him a painting.

243
00:28:43,810 --> 00:28:52,150
And asked his opinion because Boone said this painting has recently been discovered in from a private collector,

244
00:28:52,150 --> 00:28:55,510
and we think it may be by your hands for me.

245
00:28:55,510 --> 00:29:04,470
Yeah, that's for me, the great Dutch interior painter and Mr. Brady, as you are the world's leading expert on Vermeer.

246
00:29:04,470 --> 00:29:12,900
So could you tell us what what you think of the painting and Abraham Radius was spellbound, almost literally.

247
00:29:12,900 --> 00:29:18,570
He he later wrote shortly afterwards wrote in an art magazine,

248
00:29:18,570 --> 00:29:29,100
Burlington He that when he saw this painting, he said I had difficulty controlling my emotion.

249
00:29:29,100 --> 00:29:32,970
He said he felt that he was in the presence, not just of of a Vermeer,

250
00:29:32,970 --> 00:29:40,680
but the masterpiece of Johannes for me, of Delft, quite different from all his other paintings.

251
00:29:40,680 --> 00:29:52,110
And yet, every inch of Amir. He also described it as uncorrupt just the Dutch word for virginal and pure and uncorrupted and untouched,

252
00:29:52,110 --> 00:29:56,340
which was ironic because basically it was a fraud and it was not only a fraud,

253
00:29:56,340 --> 00:30:05,550
it was a really nasty, vicious fraud by a really nasty, vicious man whom I describe in the book.

254
00:30:05,550 --> 00:30:13,620
And it wasn't even a very good painting. And the thing the one thing you can say about Your Highness for me is that paintings are good.

255
00:30:13,620 --> 00:30:18,980
I he's really good at what he does. The paintings are magical.

256
00:30:18,980 --> 00:30:30,470
So how was it that Abraham Brady was this great critic, this great expert was fooled by this crude forgery that wouldn't fool you or me.

257
00:30:30,470 --> 00:30:35,750
And the simple answer is in that line, he wrote for Burlington Magazine.

258
00:30:35,750 --> 00:30:44,510
When I saw the painting, I had difficulty controlling my emotion by exploiting Abraham Brady as his wishful thinking,

259
00:30:44,510 --> 00:30:51,110
by painting the picture that he knew Abraham Radius wanted to see.

260
00:30:51,110 --> 00:30:55,100
The forger managed to bypass Brady as his expertise.

261
00:30:55,100 --> 00:31:01,220
It didn't matter how much he knew about the technical details of painting.

262
00:31:01,220 --> 00:31:08,180
He wanted to be fooled, and in fact, his expertise in this particular case only made things worse.

263
00:31:08,180 --> 00:31:16,370
She gave him new reasons to believe in a painting that he should never have believed in you, or I would never have fallen for this.

264
00:31:16,370 --> 00:31:23,180
But Abraham Brady, as the expert did. Now why am I telling you a story about art forgery?

265
00:31:23,180 --> 00:31:28,880
Why am I beginning the first chapter of my book, which is supposed to be a book about numbers, right?

266
00:31:28,880 --> 00:31:35,600
That has no numbers in it? That is purely about, well, about painting and about perception.

267
00:31:35,600 --> 00:31:39,590
It's because of emotional reactions come first.

268
00:31:39,590 --> 00:31:46,040
If we've learnt anything from the last few years, the Brexit referendum, the election of Donald Trump, we've learnt that.

269
00:31:46,040 --> 00:31:54,320
We believe what we believe because of who we are and how we feel, our emotional reactions and our preconceptions.

270
00:31:54,320 --> 00:32:00,410
So there's no point in me writing a book that will tell you how to solve technical problems in statistics

271
00:32:00,410 --> 00:32:09,890
if it can also give you the emotional tools to get past your wishful thinking and see the world clearly.

272
00:32:09,890 --> 00:32:20,390
I think the one thing that we've all learnt over the last few years is that what we believe is overwhelmingly determined by what we want to believe.

273
00:32:20,390 --> 00:32:26,780
We're all influenced by ah, by our friends, by our cultural identity, by our political preconceptions.

274
00:32:26,780 --> 00:32:38,030
And so, of course, our emotions matter. So the very first question I would ask in answer to Sarah is how is this statistical claim making me feel?

275
00:32:38,030 --> 00:32:44,360
Is it making me feel angry or defensive or vindicated?

276
00:32:44,360 --> 00:32:50,270
Can I do I see this as as some ammunition for some argument that I want to make?

277
00:32:50,270 --> 00:32:56,690
Because if that's how you're feeling, you're probably not thinking clearly. Of course, we should be influenced by our emotions.

278
00:32:56,690 --> 00:33:00,860
Our emotions are important. We're social beings. It's fine to have preconceptions.

279
00:33:00,860 --> 00:33:05,990
It's fine to pay attention to what our friends think, but we have to be aware of that.

280
00:33:05,990 --> 00:33:12,770
So I'm advocating statistical mindfulness. I realise I now sound like yoga with a calculator.

281
00:33:12,770 --> 00:33:18,110
Statistical mindfulness. Notice how the claim makes you feel.

282
00:33:18,110 --> 00:33:28,910
And if you're if you're if you are aware of your emotions, if you're aware of that emotional reaction, you take a few seconds to observe it.

283
00:33:28,910 --> 00:33:30,530
Maybe let it subside.

284
00:33:30,530 --> 00:33:38,550
I think you're going to think more clearly about how to then go about evaluating the claim and whether you should be sharing it or not.

285
00:33:38,550 --> 00:33:45,020
And second question I would ask, it's a serious question. What are the best questions to ask when we see a number in the news?

286
00:33:45,020 --> 00:33:53,730
The second question to ask is what does the name actually mean by which I mean something quite specific, like what is actually being counted?

287
00:33:53,730 --> 00:34:06,940
So for example, we see these of these daily coronavirus cases reported every day in the UK to understand that.

288
00:34:06,940 --> 00:34:13,600
You know this these are official cases, they're being processed, the tests are being done,

289
00:34:13,600 --> 00:34:17,950
and the tests may be being processed with a delay of several days.

290
00:34:17,950 --> 00:34:25,750
They might have a test gathered before the weekend, and the test is that being the test results published after the weekend?

291
00:34:25,750 --> 00:34:33,070
That sort of is just straightforward knowledge about what this number represents versus, for example,

292
00:34:33,070 --> 00:34:39,490
the Office for National Statistics Infection Survey, which is this is not an official case count.

293
00:34:39,490 --> 00:34:48,910
This is an attempt to randomly sample the population and to estimate how many people out there have infections.

294
00:34:48,910 --> 00:34:51,070
It's a different methodology, I would say.

295
00:34:51,070 --> 00:34:58,540
Probably a better methodology for estimating how how much the virus is out there just to know what's being measured,

296
00:34:58,540 --> 00:35:02,540
what it means during the financial crisis.

297
00:35:02,540 --> 00:35:08,650
I got so, so many claims that mixed up the deficit and the debt you could argue.

298
00:35:08,650 --> 00:35:12,040
People would argue about whether the claim was true, whether the number was right.

299
00:35:12,040 --> 00:35:17,730
And so you can actually even understand what it is that you're talking about.

300
00:35:17,730 --> 00:35:27,670
Or another example, it's in the book an article in The Guardian that talked about suicide and self-harm.

301
00:35:27,670 --> 00:35:31,150
And actually suicides are different from self-harm,

302
00:35:31,150 --> 00:35:41,710
and the definition that was being used for self-harm was was actually ambiguous that the article didn't discuss what was meant by self-harm.

303
00:35:41,710 --> 00:35:45,280
And then I had to go and talk to the researchers and the researchers in the end said, Well, actually,

304
00:35:45,280 --> 00:35:53,650
we don't know what's meant by self-harm either and self-harm as as defined by the the people who responded to our survey.

305
00:35:53,650 --> 00:35:57,040
So it's whatever they think now, it's nothing wrong with that.

306
00:35:57,040 --> 00:36:02,710
But just understanding what the definition is, what's being measured, what's being described before,

307
00:36:02,710 --> 00:36:09,370
before you get into the mass, before you get into evaluating the number, that's really important.

308
00:36:09,370 --> 00:36:15,850
And the third question that I would ask a very simple question is how does this fit into the bigger context?

309
00:36:15,850 --> 00:36:23,590
So for example, if you if you see that there have been 6000 daily cases of coronavirus to ask questions like,

310
00:36:23,590 --> 00:36:28,240
well, how does that compare to a week ago?

311
00:36:28,240 --> 00:36:31,780
Is there a rolling average? How does that compare to other countries?

312
00:36:31,780 --> 00:36:36,070
What about France? What about Spain? How does it compare to six months ago?

313
00:36:36,070 --> 00:36:43,330
Can we make the comparison with six months ago? Are there other reasons to believe that the comparison doesn't work?

314
00:36:43,330 --> 00:36:47,440
Now, of course, you know, as an individual.

315
00:36:47,440 --> 00:36:54,190
Some of this stuff is a lot of hard work. You don't want to have to go through all this work every time you see a number,

316
00:36:54,190 --> 00:36:59,020
but you should certainly see some evidence that somebody else is doing that work for you.

317
00:36:59,020 --> 00:37:05,590
So if you're reading a social media post, listening to the radio, watching the TV, reading a newspaper article,

318
00:37:05,590 --> 00:37:14,350
if you're hearing that sort of explanation, that sort of context, what was down, what was measured is the number going up or going down.

319
00:37:14,350 --> 00:37:16,900
What is it per person or per million people?

320
00:37:16,900 --> 00:37:24,520
If you're seeing that, I would be reassured that you're hearing from someone who's trying to help you understand the world.

321
00:37:24,520 --> 00:37:30,550
And if you're not getting that context, you're you're hearing from someone who's trying to get you excited or riled

322
00:37:30,550 --> 00:37:35,870
up or sort of win your vote or defend themselves against a political attack.

323
00:37:35,870 --> 00:37:40,700
And that's not the same thing, and it's not helpful. Right?

324
00:37:40,700 --> 00:37:45,310
Another question. Thanks, Sarah, for that question. Dave Bradshaw.

325
00:37:45,310 --> 00:37:54,910
This leads into the previous question. Dave Bradshaw asks how much of a worry is it when he puts this in square in scare quotes?

326
00:37:54,910 --> 00:37:55,900
How much of a worry is it?

327
00:37:55,900 --> 00:38:06,310
When leading epidemiologists put out misleading information about the infection fatality rate of the virus never bother to admit the mistake,

328
00:38:06,310 --> 00:38:09,610
and he links to a particular video.

329
00:38:09,610 --> 00:38:17,020
But I'm not so interested in that particular video because there's lots of this about there are we focus on the virus fragments.

330
00:38:17,020 --> 00:38:22,750
There are a lot of people around, some of whom are eminently qualified and some of whom.

331
00:38:22,750 --> 00:38:25,720
It's not that amateurs or their politicians,

332
00:38:25,720 --> 00:38:35,290
but a lot of people about who have become very concerned with winning an argument with making a particular case for a particular view of the world.

333
00:38:35,290 --> 00:38:43,060
And whenever I see that I, I start to worry because it becomes harder to really evaluate what's going on,

334
00:38:43,060 --> 00:38:49,180
becomes harder to see clearly becomes very tempting to score cheap points.

335
00:38:49,180 --> 00:38:51,740
It becomes very tempting to play for people's emotions.

336
00:38:51,740 --> 00:38:59,260
And as we already saw with my answer to Sarah, you know, getting people emotional does not help clear thinking.

337
00:38:59,260 --> 00:39:05,890
And you start getting tempted to cherry pick data to strip things of context and to be unfair.

338
00:39:05,890 --> 00:39:16,570
So some of the. Some of this sort of debating this point scoring is done by experts with some expertise, but it still makes me uncomfortable.

339
00:39:16,570 --> 00:39:19,450
And some of it is just complete nonsense.

340
00:39:19,450 --> 00:39:32,770
So for example, on the side of people who who say we need to be very, very worried indeed about about the virus, I see lots of stuff about long COVID.

341
00:39:32,770 --> 00:39:44,060
In other words, people who have lasting and maybe serious symptoms who nevertheless didn't die, but very little about how common this is.

342
00:39:44,060 --> 00:39:46,120
And we need better data, of course.

343
00:39:46,120 --> 00:39:54,340
But if you're just tweeting examples of lupus, this person and this person suffered this long running debilitating disease.

344
00:39:54,340 --> 00:39:58,090
We haven't really learnt anything because millions of people have had the virus.

345
00:39:58,090 --> 00:40:00,700
Tens of thousands of people have died of the virus.

346
00:40:00,700 --> 00:40:07,600
The fact that you can give me a single example of someone who's suffering long term health effects, it doesn't help me understand.

347
00:40:07,600 --> 00:40:11,740
Now I want to understand long home and I want to know how prevalent it is.

348
00:40:11,740 --> 00:40:16,030
I think it could be a very serious problem, and it's certainly something we should investigate.

349
00:40:16,030 --> 00:40:24,070
But I want to explore that with the mindset of someone who's seeking after truth rather than someone who's trying to make a particular point.

350
00:40:24,070 --> 00:40:30,850
And you see on the other side of the debate, the lockdown sceptics, I think they often call themselves again.

351
00:40:30,850 --> 00:40:38,080
I'm not very loud and we've skipped from one point to another or the infection fatality rate might be super low because

352
00:40:38,080 --> 00:40:46,930
maybe everyone's kind of already had it or we just on the brink of herd immunity or died with COVID vs. died of COVID.

353
00:40:46,930 --> 00:40:55,390
And now all this nonsense about false positives and in each case, there's somewhere underneath those arguments.

354
00:40:55,390 --> 00:40:59,710
There's a really important point to make. False positives are a thing.

355
00:40:59,710 --> 00:41:03,070
Asymptomatic, undetected cases are a thing.

356
00:41:03,070 --> 00:41:11,500
Herd immunity might be more prevalent than we think, but when it's used again to make a particular argument, we're getting stupider.

357
00:41:11,500 --> 00:41:19,540
We're not getting smarter. So it disturbs me when I see anybody arguing for a particular case rather than just trying to explore the truth.

358
00:41:19,540 --> 00:41:29,230
And you can see it's not hard to see. What you know, who's who's doing, what goes on on this sort of thing.

359
00:41:29,230 --> 00:41:40,390
A question from Sally, Steve and Sally Stephens says should it be mandatory for MPs to have some statistics training?

360
00:41:40,390 --> 00:41:47,980
Probably not because, you know, in the end, it's a democracy and we voters need to take responsibility for who we elect.

361
00:41:47,980 --> 00:41:55,540
Right. But I think it would certainly be helpful for MPs to have some statistics training two separate points.

362
00:41:55,540 --> 00:42:02,380
One is that we just need a more diverse group of MPs. And by diversity, I refer to all kinds of things.

363
00:42:02,380 --> 00:42:04,990
You know the things that you're probably thinking of.

364
00:42:04,990 --> 00:42:11,020
So and people from different ethnic backgrounds and from different classes, different parts of the country,

365
00:42:11,020 --> 00:42:20,380
different educational backgrounds, men and women, different sexual orientation and peace with disabilities?

366
00:42:20,380 --> 00:42:27,130
We need all of that. We also need peace with different educational backgrounds.

367
00:42:27,130 --> 00:42:34,600
So a lot of lawyers in parliament and quite a lot of people who studied history and classics and even a few who studied people like me,

368
00:42:34,600 --> 00:42:37,180
there's nothing wrong with any of that.

369
00:42:37,180 --> 00:42:47,560
But it would be good also to see MPs with mathematical training, statistical backgrounds, chemistry, physics, biology, medicine.

370
00:42:47,560 --> 00:42:53,530
We need all of these things. The world is a complicated place. The decisions that MVP's have to make on our behalf,

371
00:42:53,530 --> 00:43:02,360
they're complex and you're only going to get the best decision making when you have that real mix of different expertise in the room.

372
00:43:02,360 --> 00:43:14,740
So it's that second point to make is the research done by Phil Tetlock, Barbara Melas and Don Moore became famous as the super forecasting research.

373
00:43:14,740 --> 00:43:22,150
There's a great book called Super Forecasting by BI Tetlock, and Dan Gardner is his co-author.

374
00:43:22,150 --> 00:43:29,950
One of the things that they find is that even quite brief statistical training, like an hour of statistical training,

375
00:43:29,950 --> 00:43:37,840
really helps forecasters to make better predictions about the world more robust predictions and really understand what's going on.

376
00:43:37,840 --> 00:43:42,310
So there probably is a case for some statistical training.

377
00:43:42,310 --> 00:43:46,540
I don't think it's that hard. I mean, another thing we could do is maybe we could.

378
00:43:46,540 --> 00:43:52,210
Just everyone could mail a copy of how to make the world add up to their local MP.

379
00:43:52,210 --> 00:44:01,240
Great idea. Now you suggest it. But this stuff doesn't always require a Ph.D. in statistics or advanced mathematics.

380
00:44:01,240 --> 00:44:09,850
A lot of it is basic critical thinking skills, plus the motivation to understand what's happening rather than try to win an argument.

381
00:44:09,850 --> 00:44:17,320
And of course, MPPs are highly motivated to try to win arguments, which is a problem and not a problem I can entirely blame them for.

382
00:44:17,320 --> 00:44:25,480
That's just the way things are, right? Oh, question from Will Moy, who runs full fact.

383
00:44:25,480 --> 00:44:30,070
What stat do you wish you knew and why?

384
00:44:30,070 --> 00:44:35,170
She will know if Will is aware of this, but he's in the book,

385
00:44:35,170 --> 00:44:51,040
quoted as saying that we know more about golf than we do about victims of serious crimes such as rape and sexual assault and assault and murder.

386
00:44:51,040 --> 00:44:54,910
Why do we know more about golf than we know about the victims of these crimes?

387
00:44:54,910 --> 00:45:04,000
And because the the survey that measures physical activity, sporting activity is much bigger than the crime survey,

388
00:45:04,000 --> 00:45:12,910
so it's higher resolution enables you to make statements about local areas, enables you to make statements about rare activities.

389
00:45:12,910 --> 00:45:19,120
But this isn't because some civil servant or some politician at some stage said, You know what?

390
00:45:19,120 --> 00:45:24,520
We've got the statistics budget, and we really think sports are more important than crime.

391
00:45:24,520 --> 00:45:32,560
It's just because the The Victims of Crime Survey has been running for a long time at a particular size.

392
00:45:32,560 --> 00:45:36,820
And then when the London Olympics came along, there's this big push for sporting participation,

393
00:45:36,820 --> 00:45:43,990
a big budget to to encourage sporting participation and therefore a big survey of sporting participation.

394
00:45:43,990 --> 00:45:49,720
Nothing wrong with that, but just goes to show that we sometimes we make these argument.

395
00:45:49,720 --> 00:45:56,860
We collect these statistics almost by accident. We're not very considered about what we measure and what we don't.

396
00:45:56,860 --> 00:46:05,020
And I should shout out for an iPad. Will Smith's blog Missing numbers, I think, is missing numbers dot org, but you can find it,

397
00:46:05,020 --> 00:46:11,410
which basically just tries to track things that maybe the government should be measuring and and isn't.

398
00:46:11,410 --> 00:46:20,110
And in particular, things the government used to measure and then stopped measuring and often didn't even notify anybody to stop measuring it.

399
00:46:20,110 --> 00:46:27,850
The data there was just disappeared. For power not to collect numbers is is very important.

400
00:46:27,850 --> 00:46:35,620
All that said, OK, so to answer Will's question, what state do you wish you knew and why?

401
00:46:35,620 --> 00:46:43,780
Well, when I was finishing off the book in late April, the answer was very clear it was the infection fatality rate,

402
00:46:43,780 --> 00:46:53,500
and I wanted to know the infection fatality rate of coronavirus because it was still, I think, highly uncertain and hugely consequential.

403
00:46:53,500 --> 00:46:55,450
Some people were emailing me and saying,

404
00:46:55,450 --> 00:47:05,320
you said in your latest column that the infection fatality rate might be one percent, but it's clearly one in two thousand.

405
00:47:05,320 --> 00:47:14,450
I mean it. If it's one in two thousand, everybody in the UK had coronavirus twice in March,

406
00:47:14,450 --> 00:47:20,480
April and May, there is no way to make the deaths come out in the other way.

407
00:47:20,480 --> 00:47:25,100
This is clearly not one in two thousand so that that was the number I really wanted.

408
00:47:25,100 --> 00:47:33,140
I think we've done a lot of work figuring out what it is. I think we now know it was about one percent and it has been coming down.

409
00:47:33,140 --> 00:47:38,360
It's been coming down because older people who are much more vulnerable and people who

410
00:47:38,360 --> 00:47:43,250
are vulnerable for other reasons are doing a better job of hiding from the virus,

411
00:47:43,250 --> 00:47:50,000
and we're doing a better job of shielding them from the virus. And also because treatment is is getting better.

412
00:47:50,000 --> 00:47:56,600
Possibly also because we're doing a better job of shielding people from from high exposure to the virus so that

413
00:47:56,600 --> 00:48:01,970
these various things are helping to bring the infection fatality rate down from one percent to half a percent,

414
00:48:01,970 --> 00:48:12,660
which is obviously important. Still, a really nasty virus and incredibly dangerous for four people over the age of 70.

415
00:48:12,660 --> 00:48:17,870
So that's the number I that's what I would have said when I was writing the book.

416
00:48:17,870 --> 00:48:22,910
If you'd asked me what number now? So there's a lot in the book about algorithm.

417
00:48:22,910 --> 00:48:30,080
There's a long chapter on algorithms, and one of the things I really wish we knew more about was the effectiveness of

418
00:48:30,080 --> 00:48:37,850
algorithms that our politicians are often deploying to make decisions about us.

419
00:48:37,850 --> 00:48:43,040
So we have algorithms that are deciding who gets bail and who doesn't get bail.

420
00:48:43,040 --> 00:48:50,690
So we saw this summer algorithms that replacing A-level and GCSE exams.

421
00:48:50,690 --> 00:48:57,380
And I think the algo shambles of the summer really indicated part of the problem that

422
00:48:57,380 --> 00:49:05,780
somebody had managed to persuade politicians that this very painful decision that made,

423
00:49:05,780 --> 00:49:10,310
which was to cancel everyone's exams didn't have to be painful at all because there

424
00:49:10,310 --> 00:49:15,410
was going to be this magic algorithm that would just give everyone the right grade,

425
00:49:15,410 --> 00:49:17,750
which when you think about it, it's clearly impossible.

426
00:49:17,750 --> 00:49:24,820
How could an algorithm give you the right grade for an exam that you haven't sat and you're not going to say it is impossible?

427
00:49:24,820 --> 00:49:31,250
And I think if politicians had recognised that back in March, they could very quickly have said,

428
00:49:31,250 --> 00:49:38,540
OK, what are we going to do about the fact that there is no way to give people that grades fairly?

429
00:49:38,540 --> 00:49:43,490
What are we going to do to survive a world where we don't have that information anymore could be postponed?

430
00:49:43,490 --> 00:49:48,920
The exams can be start expanding higher education?

431
00:49:48,920 --> 00:49:53,450
What do we do? But instead, people just thought, well, the algorithm will be fine.

432
00:49:53,450 --> 00:49:59,660
So I want politicians to have a better sense of what these algorithms can and can't do.

433
00:49:59,660 --> 00:50:07,550
But that then leads into this missing statistic that is so often missing that will boy asks about,

434
00:50:07,550 --> 00:50:13,340
which is very often we don't have any proof of effectiveness of an algorithm.

435
00:50:13,340 --> 00:50:19,580
We don't see algorithms evaluated with, for example, with rigorous randomised trial,

436
00:50:19,580 --> 00:50:24,020
which you would often want to do if you've got a medical decision making algorithm or a

437
00:50:24,020 --> 00:50:30,260
policing algorithm to compare what the algorithm is doing vs. what human judgement is doing,

438
00:50:30,260 --> 00:50:36,980
and to say, yeah, the algorithm is or is not making the decision more quickly, more accurately,

439
00:50:36,980 --> 00:50:46,010
etc. So you made these grand claims for these algorithms that are commercially secret that can't be evaluated by independent experts.

440
00:50:46,010 --> 00:50:52,910
We want to see what's going on inside the algorithm. We want independent scrutiny and as much as anything else,

441
00:50:52,910 --> 00:50:59,810
we want proof that it actually works and we don't take vaccines or drugs without proof that they work.

442
00:50:59,810 --> 00:51:07,130
Why on earth should we let an algorithm make decisions without proof that it works?

443
00:51:07,130 --> 00:51:11,030
I'm looking at my clock. I've talked for too long. I've talked for far too long.

444
00:51:11,030 --> 00:51:16,040
I'm going to shut up now and thank you for everyone who. To everyone who sent in questions on Twitter.

445
00:51:16,040 --> 00:51:21,290
I'm sorry, I didn't get to all of them. There were loads of good ones. The book answers some of them.

446
00:51:21,290 --> 00:51:26,300
You can always drop me an email if there are others that you want answered, and I will do my best.

447
00:51:26,300 --> 00:51:34,040
Thank you so much to the Oxford Maths Department, the Mathematical Institute, for inviting me to speak.

448
00:51:34,040 --> 00:51:39,080
Thanks again to Blackwell's books for supporting how to make the world that up.

449
00:51:39,080 --> 00:51:44,480
And final word to all of you. Be curious about the world.

450
00:51:44,480 --> 00:51:54,320
Try to think clearly. Try to ask the questions that you want answered whenever you see one of these statistics.

451
00:51:54,320 --> 00:51:59,810
Don't use statistics as weapons don't stand for other people using them as weapons.

452
00:51:59,810 --> 00:52:05,930
Use them instead. As a kind of telescope to see the world more clearly, you can do it.

453
00:52:05,930 --> 00:52:13,370
I believe in you. I think we can all do it, and we're going to make it easier for each other if we all band together on this point.

454
00:52:13,370 --> 00:52:35,219
So thanks very much.