Speaker: 00:00:02

Folks, you may know it by now, I am a huge

sports fan.

2

00:00:09,082 --> 00:00:12,712

So needless to say that this episode was

like being in a candy store for me.

3

00:00:12,712 --> 00:00:16,362

Well, more appropriately, in a chocolate

store.

4

00:00:16,362 --> 00:00:21,622

Paul Sabin is so knowledgeable that this

conversation was an absolute blast for me.

5

00:00:21,622 --> 00:00:26,182

In it, Paul discusses his experience with

non -stats stakeholders in sports

6

00:00:26,182 --> 00:00:31,054

analytics and the challenges of convincing

them to adopt evidence -based decisions.

7

00:00:31,054 --> 00:00:35,354

He also explains his soccer power ratings

and projections model, which uses a

8

00:00:35,354 --> 00:00:39,374

Bayesian approach and expected goals, as

well as the importance of understanding

9

00:00:39,374 --> 00:00:43,254

player value in difficult to measure

positions and the need for more accessible

10

00:00:43,254 --> 00:00:46,174

and digestible sports analytics for fans.

11

00:00:46,174 --> 00:00:50,494

We also touch on the impact of budget on

team performance in American sports and

12

00:00:50,494 --> 00:00:54,974

the use of plus -minus models in

basketball and American football.

13

00:00:54,974 --> 00:00:59,494

Paul is a senior fellow at the Wharton

Sports Analytics and Business Initiative

14

00:00:59,494 --> 00:01:00,654

and I like truer

15

00:01:00,654 --> 00:01:04,694

in the Department of Statistics and Data

Science at the Wharton School of the

16

00:01:04,694 --> 00:01:06,594

University of Pennsylvania.

17

00:01:06,754 --> 00:01:11,214

He has spent his entire career as a sports

analytics professional, teaching and

18

00:01:11,214 --> 00:01:13,914

leading sports analytics research

projects.

19

00:01:13,914 --> 00:01:20,614

This is Learning Visions Statistics,

episode 108, recorded April 11, 2024.

20

00:01:28,598 --> 00:01:42,018

Welcome to Learning Bayesian Statistics, a

podcast about Bayesian inference, the

21

00:01:42,018 --> 00:01:45,558

methods, the projects, and the people who

make it possible.

22

00:01:45,558 --> 00:01:47,818

I'm your host, Alex Andorra.

23

00:01:47,818 --> 00:01:52,418

You can follow me on Twitter at Alex

underscore Andorra, like the country, for

24

00:01:52,418 --> 00:01:53,968

any info about the show.

25

00:01:53,968 --> 00:01:56,418

LearnBayStats .com is Laplace to me.

26

00:01:56,418 --> 00:01:57,326

Show notes.

27

00:01:57,326 --> 00:02:01,546

becoming a corporate sponsor, unlocking

Bayesian Merge, supporting the show on

28

00:02:01,546 --> 00:02:03,946

Patreon, everything is in there.

29

00:02:03,946 --> 00:02:05,806

That's LearnBasedStats .com.

30

00:02:05,806 --> 00:02:10,186

If you're interested in one -on -one

mentorship, online courses, or statistical

31

00:02:10,186 --> 00:02:15,366

consulting, feel free to reach out and

book a call at topmate .io slash alex

32

00:02:15,366 --> 00:02:17,266

underscore and dora.

33

00:02:17,266 --> 00:02:21,126

See you around, folks, and best Bayesian

wishes to you all.

34

00:02:24,694 --> 00:02:28,974

Welcome to Learning Vagin Statistics.

35

00:02:54,542 --> 00:02:58,082

a full conversation in French as we just

had before recording.

36

00:02:58,462 --> 00:02:59,182

Well done.

37

00:02:59,182 --> 00:03:00,702

It used to be though.

38

00:03:00,762 --> 00:03:02,942

Go back two to three hundred years.

39

00:03:02,942 --> 00:03:06,062

Maybe you just don't go to Africa enough.

40

00:03:06,062 --> 00:03:09,182

That's where French is spoken a lot now

too.

41

00:03:09,222 --> 00:03:10,342

Exactly.

42

00:03:10,342 --> 00:03:14,801

But other than that, you can see French

used to be a very international language

43

00:03:14,801 --> 00:03:20,642

because in my travels, almost all the time

people tell me, yeah, I studied French in

44

00:03:20,642 --> 00:03:21,762

high school.

45

00:03:21,762 --> 00:03:24,302

And the only thing they can say is just a

few words.

46

00:03:24,302 --> 00:03:27,462

Which is normal, like if you don't use it,

right?

47

00:03:27,462 --> 00:03:31,902

But yeah, you can see that because French

is still, or was still taught in high

48

00:03:31,902 --> 00:03:34,162

school and now less and less.

49

00:03:34,282 --> 00:03:37,262

So yeah, so well done Paul for that.

50

00:03:37,262 --> 00:03:41,462

I know, I don't think French is an easy

language to learn.

51

00:03:41,462 --> 00:03:43,152

What has been your experience?

52

00:03:43,152 --> 00:03:44,882

I'm actually very curious.

53

00:03:45,902 --> 00:03:50,892

You know, it's hard to say, so this is a

statistics pod or data science podcast.

54

00:03:50,892 --> 00:03:53,492

So I guess I can't really, I can't really

compare it to anything else.

55

00:03:53,492 --> 00:03:57,102

That's the only other language I've

learned besides my native English.

56

00:03:57,102 --> 00:04:03,062

So, you know, I guess, you know, one

sample size for me, I took it in high

57

00:04:03,062 --> 00:04:03,742

school as well.

58

00:04:03,742 --> 00:04:04,742

I hated it.

59

00:04:04,742 --> 00:04:11,082

I had, so, you know, coming from America,

you know, so the reason I chose, you know,

60

00:04:11,082 --> 00:04:15,076

seventh grade is when I had to choose

whether I was taking French or Spanish.

61

00:04:15,086 --> 00:04:19,466

And I'm the youngest of four kids in my

family growing up.

62

00:04:19,466 --> 00:04:23,266

And my older siblings told me that the

Spanish teacher was really mean.

63

00:04:23,266 --> 00:04:26,166

And that's originally why I took took

French.

64

00:04:26,866 --> 00:04:29,126

and then I took it for the required two to

three years.

65

00:04:29,126 --> 00:04:30,266

And then I was done.

66

00:04:30,266 --> 00:04:34,566

I had in high school, I had this teacher

from Belgium and I still remember her

67

00:04:34,566 --> 00:04:39,666

name, Madame Vendon Plus, and I couldn't

stand her, but come, come to find out

68

00:04:39,666 --> 00:04:42,986

looking back in life that she was actually

a really nice person.

69

00:04:42,986 --> 00:04:45,030

She was just Belgian.

70

00:04:45,102 --> 00:04:51,982

And the cultural, you know, like Americans

think they're the best and the French

71

00:04:51,982 --> 00:04:56,642

language in Europe people also think

they're the best because they ruled the

72

00:04:56,642 --> 00:05:01,222

world in the 17 and 1800s and America felt

like they've ruled the world for the last

73

00:05:01,222 --> 00:05:02,642

100 years.

74

00:05:02,642 --> 00:05:06,602

And so when you get into a room together

and you think both of your cultures are

75

00:05:06,602 --> 00:05:09,262

superior, you know, that doesn't go well

together.

76

00:05:09,282 --> 00:05:10,798

But actually, so after that, I didn't...

77

00:05:10,798 --> 00:05:11,698

speak French at all.

78

00:05:11,698 --> 00:05:15,738

And then I did church service for my

church for two years and I lived in

79

00:05:15,738 --> 00:05:19,918

Montreal, I lived in Quebec, not actually

in the city, I lived in a lot of rural

80

00:05:19,918 --> 00:05:21,398

small town.

81

00:05:21,398 --> 00:05:23,528

And so I studied French really hard.

82

00:05:23,528 --> 00:05:27,098

I had to learn the very strong Quebecois

accent.

83

00:05:27,158 --> 00:05:32,788

And then when I went back to school, it's

when I like really honed in my French.

84

00:05:32,788 --> 00:05:37,638

I was very conversational, could speak

very fluently in Quebec, but then, you

85

00:05:37,638 --> 00:05:39,182

know, I had to learn the grammar a little

bit more.

86

00:05:39,182 --> 00:05:39,722

in depth.

87

00:05:39,722 --> 00:05:43,182

So then I studied French as well at

university as well.

88

00:05:43,182 --> 00:05:47,622

So, you know, immersing yourself and the

actually like learning languages because

89

00:05:47,622 --> 00:05:50,922

when I learned it in school, it didn't

never made sense to me.

90

00:05:50,922 --> 00:05:55,161

But when I studied it on my own and I

studied conjugation and all these things,

91

00:05:55,161 --> 00:05:56,902

it became kind of like a math problem.

92

00:05:56,902 --> 00:06:00,942

And so when I would speak a sentence in my

head, I'd always be like, I need a

93

00:06:00,942 --> 00:06:01,602

subject.

94

00:06:01,602 --> 00:06:03,182

I need to conjugate the verb.

95

00:06:03,182 --> 00:06:06,766

And then I need to say like what I'm, you

know, just

96

00:06:06,766 --> 00:06:08,646

do an adverb or an adjective after it.

97

00:06:08,646 --> 00:06:12,206

And like it made sense in my head, but

that's not how I was taught in school.

98

00:06:12,206 --> 00:06:15,736

I was taught, I had to memorize all these

words, like everything in the kitchen.

99

00:06:15,736 --> 00:06:16,936

How do you say dishwasher?

100

00:06:16,936 --> 00:06:18,066

How do you say refrigerator?

101

00:06:18,066 --> 00:06:19,656

How do you say fork?

102

00:06:19,656 --> 00:06:20,786

How do you say spoon?

103

00:06:20,786 --> 00:06:25,026

I couldn't learn like that, but at like

living and like thinking about French as a

104

00:06:25,026 --> 00:06:28,736

math equation, it made sense in my head

and I was able to pick it up.

105

00:06:28,736 --> 00:06:29,926

You know, sure.

106

00:06:29,926 --> 00:06:33,276

I made tons of mistakes and embarrassed

myself, but it wasn't too bad.

107

00:06:33,276 --> 00:06:34,346

And that's how you learn.

108

00:06:34,346 --> 00:06:34,736

Yeah.

109

00:06:34,736 --> 00:06:36,398

So I'm guessing.

110

00:06:36,398 --> 00:06:40,618

Like from that answer, I'm guessing people

already know why I invited you on the

111

00:06:40,618 --> 00:06:41,318

podcast.

112

00:06:41,318 --> 00:06:44,098

Very nerdy answer, your put languages,

that's perfect.

113

00:06:44,098 --> 00:06:45,458

Thanks a lot.

114

00:06:45,458 --> 00:06:47,468

And yeah, I completely relate actually.

115

00:06:47,468 --> 00:06:54,698

I learned English and German in high

school and yeah, kind of the same.

116

00:06:54,698 --> 00:06:58,698

I always hated formal language learning.

117

00:06:58,878 --> 00:07:02,798

And like in the end I learned these

languages and Spanish that was the same

118

00:07:02,798 --> 00:07:06,542

and Italian that was the same, just going

to the country basically.

119

00:07:06,542 --> 00:07:12,562

And yeah, as you were saying, I think also

what it adds is you've got skin in the

120

00:07:12,562 --> 00:07:13,122

game.

121

00:07:13,122 --> 00:07:17,522

You're in the country, you're having a

conversation with someone.

122

00:07:17,522 --> 00:07:21,862

If you're not able to talk, you look

extremely stupid.

123

00:07:21,862 --> 00:07:27,182

So it's a very good incentive for the

brain to step up and learn.

124

00:07:27,182 --> 00:07:28,602

And that's really awesome.

125

00:07:28,602 --> 00:07:31,214

And then when you are in the situation

that you...

126

00:07:31,214 --> 00:07:33,434

don't know what to say, you remember that.

127

00:07:33,434 --> 00:07:36,814

And then when you learn, this is what I

should have said, it sticks with you

128

00:07:36,814 --> 00:07:40,554

because it has an emotional attachment to

it.

129

00:07:40,554 --> 00:07:40,814

Yeah.

130

00:07:40,814 --> 00:07:41,174

Yeah.

131

00:07:41,174 --> 00:07:42,254

No, exactly.

132

00:07:42,254 --> 00:07:47,394

And I mean, and that's going to be a good

segue to my first question to you, but I

133

00:07:47,394 --> 00:07:55,454

think it's also one of the situations in

life, where you can really, feel and see

134

00:07:55,454 --> 00:07:56,634

your brain learning.

135

00:07:56,634 --> 00:07:59,994

So that's why I also really love learning

new languages and going to countries to do

136

00:07:59,994 --> 00:08:01,134

that because.

137

00:08:01,134 --> 00:08:03,834

Like you arrive in the country, you don't

know how to say anything.

138

00:08:03,834 --> 00:08:08,954

And in just a few weeks, your brain starts

picking up stuff and you can really,

139

00:08:08,954 --> 00:08:14,054

really feel your brain doing its amazing

work that it's been like conditioned to do

140

00:08:14,054 --> 00:08:16,074

from years of evolution.

141

00:08:16,074 --> 00:08:19,174

And to me, that's just absolutely

incredible that the brain is able to do

142

00:08:19,174 --> 00:08:19,734

that.

143

00:08:19,734 --> 00:08:24,714

Even when you're like in your thirties and

beyond, you can do that.

144

00:08:24,714 --> 00:08:27,914

And it's just, I found that absolutely

incredible.

145

00:08:27,914 --> 00:08:30,798

And that's kind of like a Bayesian.

146

00:08:30,798 --> 00:08:35,938

neural network, you know, so I mean, see

that segue, I should definitely have a

147

00:08:35,938 --> 00:08:36,798

podcast.

148

00:08:37,958 --> 00:08:40,678

So actually talking about base.

149

00:08:41,118 --> 00:08:45,038

Yeah, I invited you on the podcast because

you do absolutely awesome work on sports

150

00:08:45,038 --> 00:08:46,798

modeling.

151

00:08:46,818 --> 00:08:52,238

And people know that I'm a big fan of a

lot of sports.

152

00:08:52,238 --> 00:08:54,058

I love modeling sports and so on.

153

00:08:54,058 --> 00:08:55,958

So I'm super happy to have you here.

154

00:08:55,958 --> 00:08:59,790

And I have a list of questions that is

embarrassingly long.

155

00:08:59,790 --> 00:09:05,470

But maybe can you tell us if you are

actually yourself using some basic

156

00:09:05,470 --> 00:09:09,330

methods, if you're familiar with those or

not?

157

00:09:09,510 --> 00:09:13,670

And yeah, in general, what does that look

like in your work?

158

00:09:13,850 --> 00:09:14,390

Yeah.

159

00:09:14,390 --> 00:09:16,830

So yeah, I mean, just a quick background

about myself, right?

160

00:09:16,830 --> 00:09:23,410

I've worked in sports, what we call sports

analytics for almost 10 years now.

161

00:09:24,650 --> 00:09:27,566

Out of actually, I was getting my PhD.

162

00:09:27,566 --> 00:09:34,736

And statistics, and I, you got, there was

this job opportunity at ESPN, you know,

163

00:09:34,736 --> 00:09:38,506

which is a sports broadcasting television

channel in the U S and a few other

164

00:09:38,506 --> 00:09:38,906

countries.

165

00:09:38,906 --> 00:09:45,066

And, you know, I got the job offer to work

on their sports analytics team where

166

00:09:45,066 --> 00:09:50,146

essentially what the team there does is

make forecasts so that, you know, they can

167

00:09:50,146 --> 00:09:54,426

show on TV, you know, on the bottom line,

like who's expected to win, or they can,

168

00:09:54,426 --> 00:09:56,782

we will run simulations on.

169

00:09:56,782 --> 00:09:59,502

you know, who's likely to win the

championship, you know, all throughout the

170

00:09:59,502 --> 00:09:59,982

season.

171

00:09:59,982 --> 00:10:03,722

And so, you know, you can tell stories

with that saying, you know, the team was

172

00:10:03,722 --> 00:10:04,931

just like the beginning of the season.

173

00:10:04,931 --> 00:10:08,282

No one thought they were going to be any

good, but just look how it, you know, they

174

00:10:08,282 --> 00:10:09,682

got better or the opposite.

175

00:10:09,682 --> 00:10:13,562

Like they were supposed to be really good

and everything just went wrong.

176

00:10:13,562 --> 00:10:19,922

And so in my field in sports modeling, I

would think actually you can't, you can't

177

00:10:19,922 --> 00:10:21,152

do it without being Bayesian.

178

00:10:21,152 --> 00:10:24,814

And so when I would interview people, I'd

always focus on, on those.

179

00:10:24,814 --> 00:10:28,354

So as people coming out of school,

sometimes they don't always learn Bayesian

180

00:10:28,354 --> 00:10:30,654

methods very well.

181

00:10:30,654 --> 00:10:34,814

And the reason is in sports, sample sizes

are very small and you have to make

182

00:10:34,814 --> 00:10:37,874

forecasts with very limited data.

183

00:10:38,094 --> 00:10:42,734

And the great thing about Bayesian is

statistics is that you actually have more

184

00:10:42,734 --> 00:10:43,164

data.

185

00:10:43,164 --> 00:10:44,334

You just haven't observed it.

186

00:10:44,334 --> 00:10:48,110

You have expertise or you have opinions,

but those opinions actually matter.

187

00:10:48,110 --> 00:10:51,930

And so maybe we'll get into this, but I'm

actually a very strong advocate because of

188

00:10:51,930 --> 00:10:55,450

my field of being a subjective Bayesian

analysis.

189

00:10:55,450 --> 00:10:59,610

It's okay to insert some information into

your models and it usually makes them

190

00:10:59,610 --> 00:11:00,910

better.

191

00:11:01,910 --> 00:11:02,260

Yeah.

192

00:11:02,260 --> 00:11:03,690

Well, awesome.

193

00:11:03,990 --> 00:11:07,110

couldn't have dreamt better and I have to

fully structure.

194

00:11:07,110 --> 00:11:10,630

I didn't know Paul was going to answer

that because that's not really, I haven't

195

00:11:10,630 --> 00:11:14,910

seen that in your, you know, on your

website or else,

196

00:11:15,086 --> 00:11:18,786

So before, while preparing the episode, I

didn't know if you were already using

197

00:11:18,786 --> 00:11:20,346

Bayesian methods or else.

198

00:11:20,346 --> 00:11:23,546

But definitely, definitely happy to hear

that.

199

00:11:23,546 --> 00:11:26,146

And so that people know that was not a

conspiracy.

200

00:11:26,146 --> 00:11:29,286

I didn't know anything that Paul was going

to say.

201

00:11:30,486 --> 00:11:32,686

OK, so that's awesome.

202

00:11:33,086 --> 00:11:37,446

So I'm an open source developer, so I'm

always very curious about the stack you're

203

00:11:37,446 --> 00:11:38,446

using.

204

00:11:39,046 --> 00:11:45,312

What are you using actually when you're

doing Bayesian analysis of a spot model?

205

00:11:46,094 --> 00:11:50,194

So in my career, I almost always use R and

Stan.

206

00:11:50,194 --> 00:11:53,714

So if I'm doing Bayes analysis, I write a

lot of Stan code.

207

00:11:55,154 --> 00:11:58,254

It's gotten easier with the Chat GPT.

208

00:11:58,474 --> 00:12:00,714

It doesn't do it all the way, right?

209

00:12:00,714 --> 00:12:04,934

But if it's like, hey, I want to build

this kind of model, it'll at least give me

210

00:12:04,934 --> 00:12:05,724

a good framework.

211

00:12:05,724 --> 00:12:10,914

And then I can adjust it and edit it as I

want from there.

212

00:12:10,994 --> 00:12:11,694

Yeah.

213

00:12:11,714 --> 00:12:12,434

Yeah.

214

00:12:12,614 --> 00:12:15,662

And I mean, for sure, you cannot go wrong

with the.

215

00:12:15,662 --> 00:12:17,522

with R and Stan.

216

00:12:17,522 --> 00:12:19,022

So yeah, definitely.

217

00:12:19,022 --> 00:12:26,902

And we've had the, one of the creators of

Stan, Andrew Gellman, was back on the

218

00:12:26,902 --> 00:12:29,662

podcast a few weeks ago.

219

00:12:29,742 --> 00:12:35,382

It was not released yet, but through time

travel, it's gonna have been released when

220

00:12:35,382 --> 00:12:36,702

your episode is out.

221

00:12:36,702 --> 00:12:42,042

So folks, you can go back to - Right,

because I am definitely a lesser draw than

222

00:12:42,042 --> 00:12:44,602

Andrew Gellman is, but that's great.

223

00:12:44,846 --> 00:12:52,646

No, yeah, so if people are curious about

what Andrew has been up to, lastly, it's

224

00:12:52,646 --> 00:12:56,066

the third time he's been on the show and

he just released a new book, Active

225

00:12:56,066 --> 00:12:58,256

Statistics, that I definitely recommend.

226

00:12:58,256 --> 00:13:00,646

It's really fun to read.

227

00:13:00,746 --> 00:13:05,246

It's like, it's how to teach statistics

with stories, which actually relates to

228

00:13:05,246 --> 00:13:11,146

something you just said, Paul, about the,

like, cool and fun way to relate

229

00:13:11,146 --> 00:13:12,934

statistics to...

230

00:13:12,974 --> 00:13:20,234

non -stats people was to be able to tell

stories about a team's probability of

231

00:13:20,234 --> 00:13:23,034

winning or any forecast like that.

232

00:13:23,034 --> 00:13:27,774

So that's definitely interesting to hear

you talk about that.

233

00:13:27,894 --> 00:13:35,134

And actually I'm curious because I've been

following that field of spots analytics

234

00:13:35,134 --> 00:13:40,176

for a few years and I've seen it

personally mature.

235

00:13:40,206 --> 00:13:44,886

quite a lot and evolved quite a lot when

it comes to the technology and the data

236

00:13:44,886 --> 00:13:46,066

availability.

237

00:13:46,086 --> 00:13:52,906

So I'm curious what an expert like you

think about that evolution of technology

238

00:13:52,906 --> 00:13:58,046

and data availability and how that changed

the landscape of Spots Analytics.

239

00:13:59,246 --> 00:14:02,576

Yeah, I mean, it's exploded in the last 10

to 15 years.

240

00:14:02,576 --> 00:14:07,822

So I mean, if people are familiar with the

book slash movie Moneyball, which is

241

00:14:07,822 --> 00:14:10,092

20, about 20 years, the book is about 20

years old now.

242

00:14:10,092 --> 00:14:13,002

The movie is about 12, 13 years old now.

243

00:14:13,782 --> 00:14:19,642

you know, back then in baseball, baseball

was the sport that sort of took off in

244

00:14:19,642 --> 00:14:20,252

sports analytics.

245

00:14:20,252 --> 00:14:22,502

I mean, for a couple of reasons.

246

00:14:22,502 --> 00:14:25,142

One, the game is very discreet.

247

00:14:25,142 --> 00:14:27,322

So their start and their stopping points.

248

00:14:27,322 --> 00:14:28,522

So you can measure.

249

00:14:28,522 --> 00:14:29,052

Right.

250

00:14:29,052 --> 00:14:32,742

Discrete events very well in baseball, but

two, like they're the only sport that

251

00:14:32,742 --> 00:14:36,500

actually had a really long running data

set.

252

00:14:36,558 --> 00:14:39,938

And that went back and they've been

keeping statistics in baseball and you can

253

00:14:39,938 --> 00:14:46,178

actually go back to the 1800s and find out

how people were playing baseball in 1895.

254

00:14:46,178 --> 00:14:47,698

No other sport has that.

255

00:14:47,698 --> 00:14:50,288

So that's, that's probably the reason why

baseball took off.

256

00:14:50,568 --> 00:14:54,438

but since then, you know, every sport for

a while after that, every sport had what

257

00:14:54,438 --> 00:14:57,638

we call play by play data, which is like,

this is what happens.

258

00:14:57,638 --> 00:15:01,478

Soccer had a, a version that was called

event data.

259

00:15:01,478 --> 00:15:02,670

So would people would.

260

00:15:02,670 --> 00:15:06,350

watch a game and every time someone

touched the ball or made a pass, they

261

00:15:06,350 --> 00:15:10,990

would mark, the ball was touched here on

the field and it was passed to there or

262

00:15:10,990 --> 00:15:12,670

they dribbled from here to there.

263

00:15:12,670 --> 00:15:18,070

So it was, they kind of were discretizing

soccer in a way to make it a similar

264

00:15:18,070 --> 00:15:18,809

format.

265

00:15:18,809 --> 00:15:22,150

But then about 10 years ago, we started

getting this player tracking data, which

266

00:15:22,150 --> 00:15:27,650

is the location of everybody and the ball

or the puck on the field, you know,

267

00:15:27,650 --> 00:15:30,090

depending on the sport, 10 to 25 times per

second.

268

00:15:30,090 --> 00:15:32,334

And that's drastically changed.

269

00:15:32,334 --> 00:15:34,934

the methodologies and things that are

used.

270

00:15:34,934 --> 00:15:39,174

So, I mean, Bayesian analysis was great

for this play by play data or even, you

271

00:15:39,174 --> 00:15:44,034

know, game by game data and measuring how,

how players or teams performed.

272

00:15:44,034 --> 00:15:48,174

And then now we've started getting such

huge data sets that, you know, more of the

273

00:15:48,174 --> 00:15:51,654

computer science world, neural networks,

things like that started becoming much

274

00:15:51,654 --> 00:15:56,194

more prevalent in sports analysis just

because the data sets were so massive.

275

00:15:56,194 --> 00:15:58,134

Not that statistics doesn't play a role.

276

00:15:58,134 --> 00:15:58,974

It still does.

277

00:15:58,974 --> 00:15:59,854

And I think.

278

00:15:59,854 --> 00:16:02,654

People sometimes overly rely on these

black box methods.

279

00:16:02,654 --> 00:16:06,594

They don't think about the implications or

the biases in the data, which are still

280

00:16:06,594 --> 00:16:07,214

important.

281

00:16:07,214 --> 00:16:11,504

But we have these huge amounts of data now

and it's just exploded to like, you know,

282

00:16:11,504 --> 00:16:17,474

if you want all the data in a season in

the NFL, it's like over one terabyte of

283

00:16:17,474 --> 00:16:22,194

locations of everybody on every field, 20,

every play of 25 times a second.

284

00:16:22,194 --> 00:16:22,994

It's just massive.

285

00:16:22,994 --> 00:16:23,214

Right.

286

00:16:23,214 --> 00:16:27,874

So it's, it's really changed the way

people have done things.

287

00:16:27,874 --> 00:16:28,270

Right.

288

00:16:28,270 --> 00:16:31,660

And we started going from really simple

questions to huge big questions.

289

00:16:31,660 --> 00:16:36,270

And the funny thing is now, I actually

think with the data being so large, people

290

00:16:36,270 --> 00:16:39,270

are now actually going back to answering

more simple questions.

291

00:16:39,270 --> 00:16:41,730

Like we're not trying to measure

everything all at once.

292

00:16:41,730 --> 00:16:46,330

Let's try to measure very specific things

that we weren't able to measure before.

293

00:16:46,330 --> 00:16:47,430

Hmm.

294

00:16:47,430 --> 00:16:50,560

Yeah, that is definitely interesting.

295

00:16:51,250 --> 00:16:55,810

and is that so first.

296

00:16:56,270 --> 00:17:03,750

Is that availability of data, massive

availability of data, the case in all the

297

00:17:03,750 --> 00:17:04,990

sports industry?

298

00:17:04,990 --> 00:17:09,330

Or is it more, well, the most historical

ones, as you were saying, maybe more

299

00:17:09,330 --> 00:17:10,130

baseball.

300

00:17:10,130 --> 00:17:15,610

I know the data set are more massive there

and maybe other sports like soccer are

301

00:17:15,610 --> 00:17:21,250

less prevalent, the data set are less

prevalent, less massive, or is that a

302

00:17:21,250 --> 00:17:21,920

uniform trend?

303

00:17:21,920 --> 00:17:22,650

First question.

304

00:17:22,650 --> 00:17:25,165

And then second question is,

305

00:17:25,165 --> 00:17:26,525

Where does that data leave?

306

00:17:26,525 --> 00:17:32,665

Is that mostly open source or is that

still quite close source data?

307

00:17:33,025 --> 00:17:33,685

Yeah.

308

00:17:33,685 --> 00:17:38,125

So I mean, baseball is usually like the

cutting edge of everything because they

309

00:17:38,125 --> 00:17:39,905

had a head start.

310

00:17:40,065 --> 00:17:44,685

And basketball and then like kind of

American football, international soccer

311

00:17:44,685 --> 00:17:49,505

football and hockey kind of trail behind.

312

00:17:49,785 --> 00:17:53,565

But the data sets now in all those sports

are very massive.

313

00:17:53,565 --> 00:17:55,054

Hockey just got

314

00:17:55,054 --> 00:18:00,794

The NHL just got their player puck

tracking data just a couple of years ago.

315

00:18:01,634 --> 00:18:05,634

Now baseball and basketball have moved on

beyond just knowing where players are on

316

00:18:05,634 --> 00:18:06,454

the field.

317

00:18:06,454 --> 00:18:09,224

They actually have data of what's called

pose data.

318

00:18:09,224 --> 00:18:14,054

So they know where different joints and

their arms and the legs are of every

319

00:18:14,054 --> 00:18:16,594

player on the field or on the court.

320

00:18:16,594 --> 00:18:18,174

So that data is massive.

321

00:18:18,174 --> 00:18:19,734

It's massive everywhere.

322

00:18:19,734 --> 00:18:24,902

There's companies that are trying to

collect new data based on

323

00:18:26,064 --> 00:18:30,934

video, so they're using computer vision

algorithms to do that, but largely to

324

00:18:30,934 --> 00:18:32,954

answer your second question.

325

00:18:33,374 --> 00:18:35,344

This is not open source data.

326

00:18:35,344 --> 00:18:38,914

So the old school data, the play by play

data is open source.

327

00:18:38,914 --> 00:18:43,934

You can find that on every sport pretty

much via an open source mechanism now.

328

00:18:43,934 --> 00:18:49,934

But this huge, these huge data sets of the

tracking of the players, you know, 10 to

329

00:18:49,934 --> 00:18:51,834

25 times per second.

330

00:18:51,834 --> 00:18:53,454

It's usually all closed source.

331

00:18:53,454 --> 00:18:54,510

There are a few.

332

00:18:54,510 --> 00:18:59,490

releases of that here and there, you know,

the NFL does a competition where they

333

00:18:59,490 --> 00:19:04,230

release some of that data each year, like

a very small set.

334

00:19:04,870 --> 00:19:07,750

and a few other leagues have done

something similar as well.

335

00:19:07,750 --> 00:19:10,570

If they know that's, that's kind of gives

you a taste.

336

00:19:10,970 --> 00:19:14,570

if you have money, there are companies

that try to create that data themselves

337

00:19:14,570 --> 00:19:16,310

and they'll sell it to you.

338

00:19:16,310 --> 00:19:20,870

But you know, that's usually pretty

expensive for an individual person to buy.

339

00:19:21,190 --> 00:19:23,290

So again, just that.

340

00:19:24,044 --> 00:19:24,814

I see.

341

00:19:24,814 --> 00:19:25,314

Okay.

342

00:19:25,314 --> 00:19:26,214

Yeah, interesting.

343

00:19:26,214 --> 00:19:27,354

Definitely.

344

00:19:27,854 --> 00:19:31,624

Because like data is kind of oil in our

industry, right?

345

00:19:31,624 --> 00:19:38,514

So it's definitely interesting to know

what's the state of the supply of oil in a

346

00:19:38,514 --> 00:19:39,594

way.

347

00:19:39,814 --> 00:19:48,094

Maybe for people who are less versed in in

sports modeling, can you give us an

348

00:19:48,094 --> 00:19:53,038

example of how analytical insights have

349

00:19:53,038 --> 00:19:59,498

directly influenced team strategy or

player selection in one of your consulting

350

00:19:59,498 --> 00:20:00,390

roles.

351

00:20:01,934 --> 00:20:02,214

Yeah.

352

00:20:02,214 --> 00:20:05,034

So I mean, I'll just kind of talk broadly

at first.

353

00:20:05,074 --> 00:20:08,104

I mean, so sometimes it's just the most

basic things, right?

354

00:20:08,104 --> 00:20:13,074

So like in basketball, people shoot three

pointers more because all they did is

355

00:20:13,074 --> 00:20:18,153

figured out the expected value was larger

for three point shot than it was for most

356

00:20:18,153 --> 00:20:19,214

two point shots.

357

00:20:19,214 --> 00:20:22,174

Not, not those layups and the dunks,

right?

358

00:20:22,174 --> 00:20:23,584

Those are very high percentages.

359

00:20:23,584 --> 00:20:30,074

So the expected value of a, of a high

percentage times two is, you know, is, is

360

00:20:30,074 --> 00:20:30,584

pretty good.

361

00:20:30,584 --> 00:20:32,230

But then even if.

362

00:20:32,494 --> 00:20:36,254

The percentage drops off a lot when you

multiply it by three to get the expected

363

00:20:36,254 --> 00:20:37,294

value of a three point shot.

364

00:20:37,294 --> 00:20:39,034

You know, it's also pretty good.

365

00:20:39,034 --> 00:20:42,134

So that means basketball has changed

drastically because of that.

366

00:20:42,134 --> 00:20:48,154

and in my roles, I guess, you know, I

think in a lot of sports, there's just

367

00:20:48,154 --> 00:20:49,434

been a lot of open questions.

368

00:20:49,434 --> 00:20:50,864

People kind of move one way.

369

00:20:50,864 --> 00:20:56,414

And then I think actually, I think the

sports analysis does really good job of

370

00:20:56,414 --> 00:20:58,990

tackling very easy problems first.

371

00:20:58,990 --> 00:21:02,430

But then I think there's actually a

tendency for the analysts themselves to be

372

00:21:02,430 --> 00:21:07,690

overconfident in their analysis and

they're not factoring in all of the

373

00:21:07,690 --> 00:21:10,130

sources of variation that might be there.

374

00:21:10,130 --> 00:21:16,790

And something I'm also very curious about

it is what's your experience with non

375

00:21:16,790 --> 00:21:18,670

-stats stakeholders?

376

00:21:18,670 --> 00:21:25,410

So coaches, scouts, players, how do they

typically respond to the analytics and the

377

00:21:25,410 --> 00:21:28,334

insights you provide and other...

378

00:21:28,334 --> 00:21:33,934

differences in reception across sports,

maybe across roles.

379

00:21:34,734 --> 00:21:35,114

Yeah.

380

00:21:35,114 --> 00:21:38,734

So, I mean, it really does vary as in all

things, there's variance.

381

00:21:38,734 --> 00:21:45,894

There are some typically younger, you

know, coaches or scouts that are a little

382

00:21:45,894 --> 00:21:49,794

bit more receptive than people who have

been doing something for a long time.

383

00:21:49,794 --> 00:21:51,034

And I think that's just human nature.

384

00:21:51,034 --> 00:21:52,394

You're used to doing things a certain way.

385

00:21:52,394 --> 00:21:53,374

You don't like.

386

00:21:54,222 --> 00:21:57,982

You know, to stereotype, you don't like

some young person coming and telling you

387

00:21:57,982 --> 00:21:58,952

how to do your job.

388

00:21:58,952 --> 00:21:59,122

Right.

389

00:21:59,122 --> 00:22:01,402

So you have to be really careful about

that.

390

00:22:02,022 --> 00:22:06,942

and the, and the funny thing is, you know,

everything that I have learned or, you

391

00:22:06,942 --> 00:22:12,682

know, I believe in, in terms of making

data driven decisions and don't

392

00:22:12,682 --> 00:22:17,562

overestimate based on small sample sizes

goes out the window when I'm trying to

393

00:22:17,562 --> 00:22:20,782

convince a stakeholder of something.

394

00:22:20,782 --> 00:22:21,934

So for example,

395

00:22:21,934 --> 00:22:26,874

If I have a model and I want them to use

it, and I think it's going to help them.

396

00:22:26,874 --> 00:22:33,174

Of course, I've done the analysis to say,

you know, what over the long run, how it

397

00:22:33,174 --> 00:22:36,214

would improve our efficiency, or if we

make a decision in this way, it'd be

398

00:22:36,214 --> 00:22:38,174

better process, et cetera.

399

00:22:38,294 --> 00:22:42,074

I've done that analysis and I've done it

over a larger sample size.

400

00:22:42,074 --> 00:22:47,334

But when I, when I tell them what they

want to know is they want confirmation

401

00:22:47,334 --> 00:22:48,844

bias, right?

402

00:22:48,844 --> 00:22:51,726

They love confirming their beliefs.

403

00:22:51,726 --> 00:22:57,706

So in order to get them to, agree with

what you're saying, it, this works so much

404

00:22:57,706 --> 00:23:02,946

more better than saying, you know, out of

the thousand players that I did this in,

405

00:23:02,946 --> 00:23:07,586

you know, you only were correct 60 % of

the time, but my model would have been

406

00:23:07,586 --> 00:23:08,626

correct 70%.

407

00:23:08,626 --> 00:23:10,186

Like they don't want to hear that.

408

00:23:10,186 --> 00:23:13,146

They essentially say, well, my model, you

know, you love this player.

409

00:23:13,146 --> 00:23:14,186

So does my model.

410

00:23:14,186 --> 00:23:18,546

I find the one guy, even if it's literally

only one person, they're like, yeah.

411

00:23:18,546 --> 00:23:20,014

Like, if your model can.

412

00:23:20,014 --> 00:23:23,514

If your model can see that, then it must

be doing something right.

413

00:23:23,514 --> 00:23:26,454

And then it's like, then they start to

trust you a little bit.

414

00:23:26,454 --> 00:23:31,834

And over time you give them little pieces,

little crumbs of a cookie that they can

415

00:23:31,834 --> 00:23:34,074

help, you know, get confidence in.

416

00:23:34,074 --> 00:23:39,074

And then, you know, then is when you share

with them, okay, well, but it's also

417

00:23:39,074 --> 00:23:42,864

suggesting this, which is different than

what you've been doing in the past.

418

00:23:42,864 --> 00:23:43,084

Right?

419

00:23:43,084 --> 00:23:48,046

So you don't ever start with, you know,

trust me.

420

00:23:48,046 --> 00:23:50,506

because you might be wrong, because you're

a human.

421

00:23:50,506 --> 00:23:53,866

I mean, like, you know, humans always make

mistakes, but we usually don't think we

422

00:23:53,866 --> 00:23:56,026

make as many mistakes as we do.

423

00:23:56,586 --> 00:24:01,686

And so I found just over time is if you

get people to trust you by confirming

424

00:24:01,686 --> 00:24:03,466

their prior held beliefs, right?

425

00:24:03,466 --> 00:24:04,626

It's another Bayesian concepts.

426

00:24:04,626 --> 00:24:09,846

If you can confirm their prior beliefs,

they're going to accept your future

427

00:24:09,846 --> 00:24:16,166

recommendations or future things that the

model might suggest more than if you start

428

00:24:16,166 --> 00:24:16,718

with.

429

00:24:16,718 --> 00:24:18,518

the differences upfront.

430

00:24:18,518 --> 00:24:20,648

And so that's like a little bit of human

bias, right?

431

00:24:20,648 --> 00:24:22,398

That you have just learned over time.

432

00:24:22,398 --> 00:24:26,638

And some things are just really hard for

people to accept, but over time, if you

433

00:24:26,638 --> 00:24:29,618

get people to trust you and you build that

relationship, there's a lot of human

434

00:24:29,618 --> 00:24:33,678

elements here and then they trust your

work by confirming their prior held

435

00:24:33,678 --> 00:24:38,258

beliefs, then they'll trust you and open

up a little bit more to being a little bit

436

00:24:38,258 --> 00:24:40,228

more open -minded about other things as

well.

437

00:24:40,228 --> 00:24:42,698

Because then like, okay, well, I know

you're not an idiot.

438

00:24:42,698 --> 00:24:45,486

Like you could speak my language some.

439

00:24:45,486 --> 00:24:49,026

now I might be more open to learning a

little bit of your language.

440

00:24:49,026 --> 00:24:53,806

And that's just sort of a human

relationship thing that you have to always

441

00:24:53,806 --> 00:24:54,790

work on.

442

00:24:56,714 --> 00:25:00,174

Yeah, that is very interesting.

443

00:25:00,234 --> 00:25:04,434

And I'm very, yeah, I'm always very

interested to hear about that because I

444

00:25:04,434 --> 00:25:10,794

also face clients daily and have to

explain models to them.

445

00:25:10,794 --> 00:25:17,814

And so as you were saying, that definitely

varies a lot in interactions to the model.

446

00:25:18,154 --> 00:25:25,902

But that negative wisdom of maybe

indulging the...

447

00:25:25,902 --> 00:25:33,142

the confirmation bias at the beginning and

then slowly go towards a bit more of

448

00:25:33,142 --> 00:25:34,282

speaking the truth.

449

00:25:34,282 --> 00:25:35,052

It's very interesting.

450

00:25:35,052 --> 00:25:39,682

I had not thought of that, but that's

yeah, definitely I can see that being a

451

00:25:39,682 --> 00:25:49,242

valid strategy when you also are in front

of someone who doesn't really understand

452

00:25:49,242 --> 00:25:52,522

the value of the modeling, I would say.

453

00:25:53,002 --> 00:25:54,606

Whereas when I

454

00:25:54,606 --> 00:25:59,706

encounter clients who are already

convinced of what the models can do for

455

00:25:59,706 --> 00:25:59,996

them.

456

00:25:59,996 --> 00:26:04,666

They are usually looking for contradicting

what they already think.

457

00:26:04,666 --> 00:26:07,866

And that's when they find the model

interesting.

458

00:26:07,866 --> 00:26:10,206

So I find that really, really cool to see.

459

00:26:10,206 --> 00:26:13,666

The contradictions are really where

there's value, right?

460

00:26:13,666 --> 00:26:17,026

But there's no value in a model if no one

uses it, right?

461

00:26:17,026 --> 00:26:20,760

Even if the model is really good, if no

one uses it, it has zero value.

462

00:26:20,974 --> 00:26:25,254

If they use it, the contradictions are

valuable if they're right, correct?

463

00:26:25,254 --> 00:26:30,994

So in soccer analysis, you know, I've

spent my career doing lots of different

464

00:26:30,994 --> 00:26:33,824

sports, but there's this sort of, this

applies to every sport.

465

00:26:33,824 --> 00:26:37,154

In basketball, we can call it the LeBron

test and soccer, we'll call it the messy

466

00:26:37,154 --> 00:26:40,394

test, where it's essentially, if you build

a model and it's trying to evaluate

467

00:26:40,394 --> 00:26:46,746

players and messy is not like one of the

top players in your model, then.

468

00:26:46,798 --> 00:26:49,778

You're not going to share it with anybody

because no one's going to believe you.

469

00:26:49,778 --> 00:26:49,928

Right.

470

00:26:49,928 --> 00:26:54,018

That's like the first thing everyone does

is like, okay, well is messy up top.

471

00:26:54,098 --> 00:26:57,498

And if like, if messy is near the top,

then like people, at least they'll listen

472

00:26:57,498 --> 00:26:58,678

to you a little bit longer.

473

00:26:58,678 --> 00:26:58,818

Right.

474

00:26:58,818 --> 00:27:00,418

But they're not going to listen to you at

all.

475

00:27:00,418 --> 00:27:02,898

If you're like, yeah, messy is an okay

player.

476

00:27:03,238 --> 00:27:03,478

Right.

477

00:27:03,478 --> 00:27:04,928

Like I don't care what your model says.

478

00:27:04,928 --> 00:27:05,058

Right.

479

00:27:05,058 --> 00:27:05,588

That's wrong.

480

00:27:05,588 --> 00:27:05,878

Right.

481

00:27:05,878 --> 00:27:07,168

That's that, that's what people believe.

482

00:27:07,168 --> 00:27:09,358

So it's like a little bit of like, I need

to feed you like, no, no, no.

483

00:27:09,358 --> 00:27:12,818

Like I'm taking a different approach than

what you do, but you know, my approach

484

00:27:12,818 --> 00:27:14,488

also thinks that messy is the best.

485

00:27:14,488 --> 00:27:14,978

Right.

486

00:27:14,978 --> 00:27:16,278

And then I'm like, it's okay.

487

00:27:16,278 --> 00:27:16,846

You know,

488

00:27:16,846 --> 00:27:17,856

Okay, yeah, we agree.

489

00:27:17,856 --> 00:27:19,546

He is really good.

490

00:27:20,986 --> 00:27:23,386

Yeah, it's like a sniff test, right?

491

00:27:23,386 --> 00:27:27,326

And it's like, in a way, it's like, well,

that's a strong prior.

492

00:27:27,406 --> 00:27:30,046

And it's like, it's saying, well, I have a

very strong prior.

493

00:27:30,046 --> 00:27:31,446

That message is really good.

494

00:27:31,446 --> 00:27:36,086

To convince me, otherwise you're going to

need really, really good data.

495

00:27:36,086 --> 00:27:41,386

It's like, well, the earth is very

probably somewhat round.

496

00:27:41,506 --> 00:27:44,430

It's going to be very hard for you to...

497

00:27:44,430 --> 00:27:48,250

move that prior from me and telling me

it's not, in a way.

498

00:27:48,250 --> 00:27:48,690

Yeah.

499

00:27:48,690 --> 00:27:51,310

And in sports, people have really strong

priors, right?

500

00:27:51,310 --> 00:27:54,770

So, you know, those sniff tests do really

matter.

501

00:27:54,770 --> 00:27:57,950

And as a modeler, even for myself, like,

I'm a human.

502

00:27:57,950 --> 00:27:59,170

So like, I do the same thing.

503

00:27:59,170 --> 00:28:02,350

If I'm building a model, I always want to

see the results.

504

00:28:02,350 --> 00:28:05,810

And it's like, I don't look at the median,

like I do, but I don't look at who the

505

00:28:05,810 --> 00:28:08,650

median result is in my model half the

time.

506

00:28:08,650 --> 00:28:11,630

I usually look at the best and I look at

the worst.

507

00:28:11,630 --> 00:28:15,350

And if I don't understand it, then I'm

like, maybe my model is doing something

508

00:28:15,350 --> 00:28:15,990

wrong.

509

00:28:15,990 --> 00:28:19,630

And I'm all like, gonna, I'm going to dive

in a little bit more.

510

00:28:19,630 --> 00:28:23,250

If it like confirms my prior held beliefs,

I'm like, it's probably correct.

511

00:28:23,250 --> 00:28:23,520

Right.

512

00:28:23,520 --> 00:28:26,370

And even as a modeler, right, you have to

be careful of that.

513

00:28:26,370 --> 00:28:31,550

But at the same time in sports, you know,

it's like I said, subjective analysis can

514

00:28:31,550 --> 00:28:32,290

be helpful.

515

00:28:32,290 --> 00:28:35,770

It's because people's subjective and I'm

like, there's wisdom.

516

00:28:35,770 --> 00:28:38,822

People coaches have been playing a game

for.

517

00:28:39,118 --> 00:28:43,358

20, 30, or coaching a game for 20 or 30

years to think that they don't have

518

00:28:43,358 --> 00:28:46,658

something to offer a model is kind of

crazy in my opinion.

519

00:28:46,658 --> 00:28:51,838

They might have biases and of course they

do, but their information that they can

520

00:28:51,838 --> 00:28:53,618

provide is useful.

521

00:28:54,378 --> 00:28:55,938

Yeah, definitely.

522

00:28:55,938 --> 00:28:59,958

And that's where we go back to what we

were talking about at the beginning in the

523

00:28:59,958 --> 00:29:02,908

value of Bayesian inference in that

context.

524

00:29:02,908 --> 00:29:08,462

Because if you can leverage that deep and

hard -hearned knowledge,

525

00:29:08,462 --> 00:29:15,722

from the coaches, from the scouts, and add

that to your model, it's like getting the

526

00:29:15,722 --> 00:29:16,782

best of both worlds.

527

00:29:16,782 --> 00:29:21,802

And that can make your analysis extremely

powerful and useful, as you were saying.

528

00:29:21,842 --> 00:29:22,252

Yeah.

529

00:29:22,252 --> 00:29:24,782

And people have done studies like this,

I've done studies like this.

530

00:29:24,782 --> 00:29:30,902

If you build a model just on the data and

ignore the human element, right?

531

00:29:30,902 --> 00:29:34,662

Or if you build a model just on human and

scouting analysis and ignore the other

532

00:29:34,662 --> 00:29:35,540

data.

533

00:29:35,694 --> 00:29:35,954

Right.

534

00:29:35,954 --> 00:29:39,364

Neither one of those is going to do as

well as when you combine both.

535

00:29:39,364 --> 00:29:42,074

And that's really, that's what, you know,

that's Bayesian analysis is you're

536

00:29:42,074 --> 00:29:47,734

combining subjective belief with objective

data and then making forecasts based on

537

00:29:47,734 --> 00:29:48,024

them.

538

00:29:48,024 --> 00:29:54,974

And we know that if you have priors that

are not really, really bad, a subjective

539

00:29:54,974 --> 00:30:01,194

Bayesian forecast is going to have smaller

error than a data, you know, what we call

540

00:30:01,194 --> 00:30:02,684

maximum likelihood forecast, right.

541

00:30:02,684 --> 00:30:04,434

And stats terms, right.

542

00:30:04,434 --> 00:30:05,294

Or.

543

00:30:05,294 --> 00:30:11,174

You know, just the human one, just the no

data, but, you know, feelings forecast as

544

00:30:11,174 --> 00:30:11,554

well, right?

545

00:30:11,554 --> 00:30:14,514

So there's the combination of the two,

always does better.

546

00:30:15,074 --> 00:30:15,454

Yeah.

547

00:30:15,454 --> 00:30:15,934

Yeah.

548

00:30:15,934 --> 00:30:16,474

Yeah.

549

00:30:16,474 --> 00:30:20,654

Preaching, preaching to the choir here for

sure.

550

00:30:21,194 --> 00:30:26,954

And actually, I think that's a good time

now in the episode to get a bit more

551

00:30:26,954 --> 00:30:34,174

nerdy, if we can, because I've seen you,

so you've obviously worked extensively

552

00:30:34,174 --> 00:30:34,766

with.

553

00:30:34,766 --> 00:30:41,426

soccer analytics and you have an

interesting soccer power ratings and

554

00:30:41,426 --> 00:30:48,606

projections on your website that I'm gonna

link to in the show notes but can you tell

555

00:30:48,606 --> 00:30:54,346

us about it and what makes these

projections unique in your perspective in

556

00:30:54,346 --> 00:31:01,466

evaluating team and player performance and

don't be afraid to dig into the nerdy

557

00:31:01,466 --> 00:31:03,182

details because...

558

00:31:03,182 --> 00:31:05,362

My audience definitely liked that.

559

00:31:05,362 --> 00:31:05,832

Yes.

560

00:31:05,832 --> 00:31:06,142

Sure.

561

00:31:06,142 --> 00:31:06,992

I'll dig in.

562

00:31:06,992 --> 00:31:09,202

So what's on my website is...

563

00:31:09,202 --> 00:31:10,422

Sorry if you can hear my dog there.

564

00:31:10,422 --> 00:31:16,402

What's on my website is perhaps the most

simple power ratings forecast that I've

565

00:31:16,402 --> 00:31:17,482

ever done.

566

00:31:17,482 --> 00:31:20,902

So I say that, not that it's like stupid

or anything.

567

00:31:20,902 --> 00:31:26,162

So when I was at ESPN, I build power

ratings in American football, both

568

00:31:26,162 --> 00:31:30,094

professional and collegiate, and

basketball, professional and collegiate.

569

00:31:30,094 --> 00:31:33,944

and hockey, I mean, like almost every

sport, right?

570

00:31:33,944 --> 00:31:38,714

So what's on my website, I'll explain the

model very simply is it's a Bayesian model

571

00:31:38,714 --> 00:31:42,114

where you have an effect for each team,

right?

572

00:31:42,114 --> 00:31:48,974

And the response variable is the expected

goals for each team.

573

00:31:48,974 --> 00:31:53,854

So usually when we do a power ratings and

we're trying to estimate for a team, you

574

00:31:53,854 --> 00:31:55,398

know, there's two sort of.

575

00:31:55,438 --> 00:31:59,058

things that we're trying to estimate their

offensive ability and their defensive

576

00:31:59,058 --> 00:32:04,198

ability and then you assume essentially

that their overall team ability, you know,

577

00:32:04,198 --> 00:32:08,178

if it's a linear model, right is the

combination of their offense and their

578

00:32:08,178 --> 00:32:09,778

defensive abilities.

579

00:32:09,778 --> 00:32:12,498

Okay, so you so essentially in each match,

right?

580

00:32:12,498 --> 00:32:18,098

You have essentially two rows of data

where you have the expected goals for the

581

00:32:18,098 --> 00:32:21,058

one team and then the expected goals for

the other and the reason we use expected

582

00:32:21,058 --> 00:32:23,182

goals, although I actually have

583

00:32:23,182 --> 00:32:24,732

lot of issues with the expected goals.

584

00:32:24,732 --> 00:32:31,022

They are a better indicator of how, how

good the team performed on offense than

585

00:32:31,022 --> 00:32:32,592

just the raw number of goals.

586

00:32:32,592 --> 00:32:32,812

And right.

587

00:32:32,812 --> 00:32:34,262

I don't need to go into details, right?

588

00:32:34,262 --> 00:32:39,062

It's essentially a, it's an expected value

as opposed to an observation from a

589

00:32:39,062 --> 00:32:44,082

Poisson distribution, which soccer scores

roughly, roughly reflect a Poisson or

590

00:32:44,082 --> 00:32:46,162

pretty close to a Poisson distribution,

right?

591

00:32:46,162 --> 00:32:48,582

The expected goals is that expectation.

592

00:32:48,642 --> 00:32:52,790

And so essentially I have a hierarchical

Bayesian model where I actually.

593

00:32:52,878 --> 00:32:54,458

I actually do a few things.

594

00:32:54,458 --> 00:32:58,478

So I actually assume the expected goals is

the mean of a Poisson distribution.

595

00:32:58,478 --> 00:33:04,518

The observed goals is the actual outcome

of the Poisson distribution.

596

00:33:04,518 --> 00:33:09,018

And then I fit a linear model essentially

where I look, okay, I have team A was on

597

00:33:09,018 --> 00:33:10,938

offense, team B was the opponent.

598

00:33:10,938 --> 00:33:13,958

And this was team A's expected goals.

599

00:33:13,958 --> 00:33:17,578

And I'm essentially fitting a regression

model, right?

600

00:33:17,578 --> 00:33:20,288

A Bayesian regression model where I have

individual team effects.

601

00:33:20,288 --> 00:33:22,382

I have a prior on each team.

602

00:33:22,930 --> 00:33:25,710

each team's offense and each team's

defense.

603

00:33:25,710 --> 00:33:29,110

And that prior, you know, rough, I don't

have to get too crazy.

604

00:33:29,110 --> 00:33:34,630

You know, I just use a normal distribution

and, and, you know, sometimes I actually,

605

00:33:34,630 --> 00:33:38,550

when I code in Stan, I actually like

using, distribution was a little, a little

606

00:33:38,550 --> 00:33:39,370

bit thicker tails.

607

00:33:39,370 --> 00:33:42,770

But I think for this model, I was just

trying to go simple, normal distribution

608

00:33:42,770 --> 00:33:47,650

prior with a mean, you know, for my

expected, essentially each team's expected

609

00:33:47,650 --> 00:33:50,382

goals per game, on offense versus.

610

00:33:50,382 --> 00:33:54,982

Defense right and the defensive value I

usually use I usually do the subtraction

611

00:33:54,982 --> 00:33:59,862

So it's team the offensive team minus the

defensive team and that way The the

612

00:33:59,862 --> 00:34:05,582

defensive team's value is is is higher if

they're a good defense So essentially if

613

00:34:05,582 --> 00:34:09,802

team a's, you know expect the goals and

they in a game against an average opponent

614

00:34:09,802 --> 00:34:17,562

is like 1 .5 and the defense was Average

expected goals in the game was you know

615

00:34:17,562 --> 00:34:19,822

that they allowed was 1 .4

616

00:34:19,822 --> 00:34:22,922

then you would say, the difference is like

0 .1, okay.

617

00:34:22,922 --> 00:34:28,522

I also include effects for being at home

in this model.

618

00:34:28,522 --> 00:34:29,942

I think, actually, I think that's all I

do.

619

00:34:29,942 --> 00:34:33,822

But in other models I've done, you can

look at things such as how much rest

620

00:34:33,822 --> 00:34:36,122

they've had since their last match.

621

00:34:36,182 --> 00:34:39,162

You can look at the difference between

each team's rest.

622

00:34:39,162 --> 00:34:40,942

And those are not linear effects, right?

623

00:34:40,942 --> 00:34:44,102

You have to do some sort of nonlinear

effects for that, right?

624

00:34:44,102 --> 00:34:49,454

Because like one day of rest is, two days

of rest is not,

625

00:34:49,454 --> 00:34:52,094

Like the difference between two days of

rest and one day of rest is very different

626

00:34:52,094 --> 00:34:54,254

than seven days versus eight days of rest,

right?

627

00:34:54,254 --> 00:34:58,234

Seven and eight days of rest are pretty

much the same thing, but two and one is

628

00:34:58,234 --> 00:34:59,344

very different, right?

629

00:34:59,344 --> 00:35:04,194

Like much bigger effect for having two

days of rest than just one day of rest.

630

00:35:04,194 --> 00:35:09,454

And so you can do things like that, or how

far away they had to travel, those sorts

631

00:35:09,454 --> 00:35:09,934

of things.

632

00:35:09,934 --> 00:35:13,434

Now in European soccer, that's not a huge

deal, because especially in the

633

00:35:13,434 --> 00:35:16,686

competitions within each country, no team

is traveling that far.

634

00:35:16,686 --> 00:35:18,866

But in American sports, it is a pretty big

deal.

635

00:35:18,866 --> 00:35:22,366

Like, you know, you, you have to fly five,

six hours across the country on short

636

00:35:22,366 --> 00:35:22,786

notice.

637

00:35:22,786 --> 00:35:25,726

Like that can, that can really affect

performance.

638

00:35:26,346 --> 00:35:29,646

and, and other things, like I said, I

don't have this in the soccer model, but

639

00:35:29,646 --> 00:35:34,306

I, if anyone's interested in modeling

sports outcomes, that people typically

640

00:35:34,306 --> 00:35:39,766

tend to overlook is the, I liked always a

big proponent of elevation, meaning that

641

00:35:39,766 --> 00:35:43,886

if there are certain sports where there

are certain teams that play at higher

642

00:35:43,886 --> 00:35:45,070

altitudes,

643

00:35:45,070 --> 00:35:48,650

And if you're not used to playing at

higher altitudes, it's actually a very

644

00:35:48,650 --> 00:35:52,070

noticeable effect in a model that you're

going to have a lower offensive output and

645

00:35:52,070 --> 00:35:55,470

you'll actually allow more points on the

other end due to fatigue.

646

00:35:55,470 --> 00:36:00,170

And so the United States, it's the teams

that are playing in Colorado and in Utah.

647

00:36:00,170 --> 00:36:03,810

But in Europe, it could be the teams that

have to go to Switzerland or the teams

648

00:36:03,810 --> 00:36:09,990

that have to go to some of these alpine

regions that are higher up in altitude.

649

00:36:10,062 --> 00:36:13,312

In Mexico, if you have to go to Mexico

City, it's extremely high.

650

00:36:13,312 --> 00:36:14,292

Or Colombia, right?

651

00:36:14,292 --> 00:36:17,502

I mean, depending on what you're doing,

these are very high altitude places that

652

00:36:17,502 --> 00:36:21,322

have shown to have a measurable impact on

an opponent's performance.

653

00:36:21,922 --> 00:36:23,922

Yeah, that's very fun.

654

00:36:23,922 --> 00:36:25,992

My God, I love those kind of models.

655

00:36:25,992 --> 00:36:27,522

That's so much fun.

656

00:36:27,742 --> 00:36:35,242

And I would also guess that, I mean, at

least my per would be that there is a

657

00:36:35,242 --> 00:36:39,854

reverse mechanism also for teams who are

used to playing altitude.

658

00:36:39,854 --> 00:36:45,254

Do they get a boost of performance when

they play closer to the C level?

659

00:36:45,254 --> 00:36:51,314

Because they could have had adaptation

that make them better when they go to the

660

00:36:51,314 --> 00:36:52,434

C level.

661

00:36:52,994 --> 00:36:53,194

Yeah.

662

00:36:53,194 --> 00:36:55,524

I mean, I think there's certainly science

behind that.

663

00:36:55,524 --> 00:37:00,714

I found that is a lot harder to show in a

model than the reverse.

664

00:37:00,714 --> 00:37:03,934

Not that it might not be there, but I

think the effect size, if it is there, is

665

00:37:03,934 --> 00:37:08,034

definitely smaller than the reverse.

666

00:37:08,274 --> 00:37:08,584

Yeah.

667

00:37:08,584 --> 00:37:09,198

That's what...

668

00:37:09,198 --> 00:37:10,618

That's what I would expect to like.

669

00:37:10,618 --> 00:37:17,578

I think the effect is here mainly because,

well, I've seen it.

670

00:37:17,598 --> 00:37:25,218

Like it seems to be pretty well seated in

the science literature, but that doesn't

671

00:37:25,218 --> 00:37:26,938

mean the effect is big.

672

00:37:27,478 --> 00:37:28,718

So yeah.

673

00:37:28,918 --> 00:37:29,138

Yeah.

674

00:37:29,138 --> 00:37:33,118

I mean, I'm a runner and I know that all

of the distance runners that are training

675

00:37:33,118 --> 00:37:37,198

for marathons that are elites and

professionals, they all train at higher

676

00:37:37,198 --> 00:37:38,228

altitudes, right?

677

00:37:38,228 --> 00:37:38,894

For the...

678

00:37:38,894 --> 00:37:44,074

six weeks leading up to a competition and

then they travel to the competition at a

679

00:37:44,074 --> 00:37:44,974

lower altitude.

680

00:37:44,974 --> 00:37:48,354

And, you know, they think they have an

oxygen performance boost due to that.

681

00:37:48,354 --> 00:37:49,274

Yeah.

682

00:37:49,274 --> 00:37:49,614

Yeah.

683

00:37:49,614 --> 00:37:53,594

Kind of like legal oxygen doping, legal

blood doping.

684

00:37:53,594 --> 00:37:53,694

Yeah.

685

00:37:53,694 --> 00:37:54,374

Yeah, exactly.

686

00:37:54,374 --> 00:37:54,694

Yeah.

687

00:37:54,694 --> 00:37:54,774

Yeah.

688

00:37:54,774 --> 00:37:59,034

I mean, I think it seems to be pretty much

proven.

689

00:37:59,034 --> 00:38:04,614

I would say maybe it has more of an impact

on individual spots like marathon running

690

00:38:04,614 --> 00:38:07,982

or else, because it's more like, you know,

it's just like,

691

00:38:07,982 --> 00:38:13,682

Even if you're winning just a few tenths

of a second, well, it can help you have a

692

00:38:13,682 --> 00:38:19,962

better time in the end because, well, at

this level, just having the smallest

693

00:38:19,962 --> 00:38:26,462

increase in performance could be the

difference between first and second place.

694

00:38:26,822 --> 00:38:36,082

But maybe that's harder to see such a

small effect on a collective spot, a

695

00:38:36,082 --> 00:38:37,582

collective game because, well,

696

00:38:37,582 --> 00:38:39,462

Maybe there are some...

697

00:38:39,462 --> 00:38:41,152

Maybe it's just not an addition.

698

00:38:41,152 --> 00:38:43,882

Maybe it's actually the effect cancel out.

699

00:38:43,882 --> 00:38:47,582

So in the end, you don't really see a big

effect.

700

00:38:48,162 --> 00:38:49,302

But that would be...

701

00:38:49,302 --> 00:38:49,942

Yeah.

702

00:38:49,942 --> 00:38:52,922

I'd love to do an experiment on that.

703

00:38:52,922 --> 00:38:54,142

Like an RCT.

704

00:38:54,142 --> 00:38:56,102

That would be so much fun.

705

00:38:57,282 --> 00:38:57,632

Yeah.

706

00:38:57,632 --> 00:38:59,782

Well, good luck trying to do experiments

in sports.

707

00:38:59,782 --> 00:39:00,682

It's hard.

708

00:39:00,682 --> 00:39:01,742

Yeah, I know.

709

00:39:01,742 --> 00:39:02,402

I know.

710

00:39:02,402 --> 00:39:03,242

But that...

711

00:39:03,242 --> 00:39:05,742

I mean, if the multiverse exists...

712

00:39:05,742 --> 00:39:08,922

Then there is a universe where we can do

that kind of experiments.

713

00:39:08,922 --> 00:39:12,702

And my god, these scientists must have so

much fun.

714

00:39:14,942 --> 00:39:20,842

And yeah, so thanks a lot, first, for

detailing the model that clearly and in so

715

00:39:20,842 --> 00:39:21,602

much details.

716

00:39:21,602 --> 00:39:23,042

That's super cool.

717

00:39:23,342 --> 00:39:27,082

So the results of the model are in a cool

dashboard on your website.

718

00:39:27,082 --> 00:39:32,522

Do you have the model and data available

freely, maybe on your GitHub, that we can

719

00:39:32,522 --> 00:39:34,292

put in the show notes?

720

00:39:35,534 --> 00:39:36,454

Yeah, I'm not sure.

721

00:39:36,454 --> 00:39:40,964

I think my GitHub, I don't know if my

GitHub model is in the model.

722

00:39:40,964 --> 00:39:41,524

It's on GitHub.

723

00:39:41,524 --> 00:39:44,614

I don't know if it's private or not, but I

can let you know.

724

00:39:45,414 --> 00:39:49,194

You know, I use actually open source data

for that.

725

00:39:49,194 --> 00:39:51,264

So I, I, let me double check.

726

00:39:51,264 --> 00:39:55,734

I can actually double check and get back

to you after the show on if, yeah, if I

727

00:39:55,754 --> 00:39:58,494

could have it in my public GitHub or not.

728

00:39:58,494 --> 00:39:59,564

So, yeah.

729

00:39:59,564 --> 00:39:59,844

Yeah.

730

00:39:59,844 --> 00:40:05,326

But essentially it uses the, there's a

package called world football R and.

731

00:40:05,326 --> 00:40:08,626

It uses data from there to build the

model.

732

00:40:08,626 --> 00:40:13,346

So some of that data is just from, it's

scraped from like transfer market.

733

00:40:13,666 --> 00:40:17,666

so I use, I use, I didn't really talk

about how I set priors means for each of

734

00:40:17,666 --> 00:40:24,366

the teams, but very, a very simple, very

simple, hierarchical model is essentially

735

00:40:24,366 --> 00:40:32,286

just to use the expenditures of the club

and use that as a prior mean for how good

736

00:40:32,286 --> 00:40:34,206

the club will be going into the season.

737

00:40:34,206 --> 00:40:34,542

And, and.

738

00:40:34,542 --> 00:40:40,442

Unlike some other sports in soccer, world

football, how much a club spends is very

739

00:40:40,442 --> 00:40:45,062

highly correlated with how successful they

are, which makes sense, but it's not true

740

00:40:45,062 --> 00:40:47,082

necessarily in like baseball.

741

00:40:47,562 --> 00:40:52,342

So, do you see these effects of budget?

742

00:40:52,342 --> 00:40:57,292

So, yeah, first, before I go on a follow

up question, yeah, for sure.

743

00:40:57,292 --> 00:40:58,962

Get back to me after the show.

744

00:40:58,962 --> 00:41:03,758

And if that's possible, we'll put that in

the show notes because I'm sure.

745

00:41:03,758 --> 00:41:06,718

A lot of listeners will be interested in

checking that out.

746

00:41:06,718 --> 00:41:10,578

I personally will be very interested in

checking that out, definitely.

747

00:41:10,658 --> 00:41:12,598

So that'd be awesome.

748

00:41:12,758 --> 00:41:21,558

And second, that effect of budget that you

see on the performance of a team.

749

00:41:21,918 --> 00:41:28,258

And so I guess in football performance

mean number of expect expectation of games

750

00:41:28,258 --> 00:41:29,538

won.

751

00:41:30,638 --> 00:41:32,418

Do you see that on Curse?

752

00:41:32,418 --> 00:41:33,382

Do you see?

753

00:41:33,454 --> 00:41:39,894

that much of an effect also in a closed

league system like the MLS?

754

00:41:40,714 --> 00:41:45,214

Or is that so because my prior would be

the effect of budget would be even

755

00:41:45,234 --> 00:41:51,814

stronger in open leagues like we have in

Europe because it's like there is no

756

00:41:51,814 --> 00:41:53,694

compensation mechanism, right?

757

00:41:53,694 --> 00:41:58,994

Clubs can go down and usually in Europe

the strongest clubs are the historical

758

00:41:58,994 --> 00:41:59,854

clubs.

759

00:41:59,854 --> 00:42:04,214

or the new clubs are just the ones that

were lucky to be bought by very, very

760

00:42:04,214 --> 00:42:06,494

healthy shareholders.

761

00:42:06,734 --> 00:42:11,234

And like, there is not a lot of switching

of the hierarchy and changing of the

762

00:42:11,234 --> 00:42:14,754

hierarchy, mainly because of budget, as

you were saying.

763

00:42:14,754 --> 00:42:20,134

But I would think that maybe the effect of

budget is less strong in a closed league

764

00:42:20,134 --> 00:42:21,474

like the MLS.

765

00:42:21,474 --> 00:42:22,224

Is that true?

766

00:42:22,224 --> 00:42:24,934

Is that something you see or is it

something that's still in the air?

767

00:42:25,806 --> 00:42:26,346

Yes.

768

00:42:26,346 --> 00:42:30,326

So I haven't looked specifically at the

MLS, but in general in American sports,

769

00:42:30,326 --> 00:42:35,206

which all have closed leagues, the budget,

well, for various reasons, the budget

770

00:42:35,206 --> 00:42:37,326

effects are not super strong.

771

00:42:37,326 --> 00:42:42,226

So, you know, in American baseball, there

is no spending limit.

772

00:42:42,226 --> 00:42:46,066

So in some American sports, like the NFL

and football, like there's a salary cap,

773

00:42:46,066 --> 00:42:47,976

meaning you can't spend more than a

certain amount.

774

00:42:47,976 --> 00:42:52,086

So there is no relationship between

overall spending and winning because

775

00:42:52,086 --> 00:42:55,346

everyone has to spend a minimum and

there's a maximum.

776

00:42:55,406 --> 00:42:58,596

In baseball, there is no limit.

777

00:42:58,596 --> 00:42:59,406

There's a tax.

778

00:42:59,406 --> 00:43:02,526

If you spend too much money, they do tax

you.

779

00:43:02,666 --> 00:43:04,506

But there's still not a huge correlation.

780

00:43:04,506 --> 00:43:06,926

And then in MLS, like I said, I'm not

entirely sure.

781

00:43:06,926 --> 00:43:10,846

Most of the clubs, they are constrained

about how much they can spend.

782

00:43:11,286 --> 00:43:14,506

And so there isn't as much variance also

in spending.

783

00:43:14,506 --> 00:43:19,066

So like, you know, Messi going to Inter

Miami, it wasn't that Inter Miami could

784

00:43:19,066 --> 00:43:20,430

pay him a lot of money.

785

00:43:20,430 --> 00:43:24,510

They actually, you know, there's a couple

of exemptions that an MLS club could use

786

00:43:24,510 --> 00:43:26,050

to pay an international player.

787

00:43:26,050 --> 00:43:29,010

They have, they're called, you know, a

couple of exemption players they have.

788

00:43:29,010 --> 00:43:34,250

And that's originally started when David

Beckham went to Los Angeles and they kind

789

00:43:34,250 --> 00:43:38,570

of made that rule essentially just so he

could, they could afford paying him what

790

00:43:38,570 --> 00:43:43,310

he was used to or close to what he was

used to being paid in Europe.

791

00:43:43,970 --> 00:43:45,490

and, and the MLS is still kind of the

case.

792

00:43:45,490 --> 00:43:49,646

You have one or two players you're allowed

to have on these exemptions and.

793

00:43:49,646 --> 00:43:55,786

The way Messi was able to make it work is

he's getting paid from Apple for his

794

00:43:55,786 --> 00:43:58,866

Apple's broadcasting the MLS games.

795

00:43:58,866 --> 00:44:04,346

So they're paying him essentially to play

in the MLS because they're hoping, more

796

00:44:04,346 --> 00:44:07,596

people are going to watch our broadcasts

are going to pay us.

797

00:44:07,596 --> 00:44:09,716

And so we're going to give you a

percentage of that.

798

00:44:09,716 --> 00:44:12,946

And that's where actually a lot of his

salary or like his earnings are coming

799

00:44:12,946 --> 00:44:18,366

from is from a, a deal with Apple versus

the actual MLS club in Miami, which can

800

00:44:18,366 --> 00:44:19,566

only pay him so much.

801

00:44:19,566 --> 00:44:23,966

So my guess is, my prior is, I haven't

looked specifically at the MLS with this,

802

00:44:23,966 --> 00:44:27,846

but my prior is yes, that there isn't a

huge relationship in the MLS between

803

00:44:27,846 --> 00:44:30,726

winning and spending just because there's

not much of a variance.

804

00:44:30,726 --> 00:44:34,926

In order to see those correlations, you

have to have a large enough variance in

805

00:44:34,926 --> 00:44:37,716

the spending to notice the relationship,

right?

806

00:44:37,716 --> 00:44:38,406

So.

807

00:44:38,926 --> 00:44:41,306

Yeah, definitely interesting.

808

00:44:42,046 --> 00:44:46,638

I mean, I love also looking at these, you

know, the...

809

00:44:46,638 --> 00:44:55,498

how the structure of a league impacts the

show and the wins is extremely

810

00:44:55,498 --> 00:44:56,028

interesting.

811

00:44:56,028 --> 00:45:00,558

That can seem very nerdy and I think

that's my political science training that

812

00:45:00,558 --> 00:45:06,678

kicks back here, but really how you

structure the game also makes the game

813

00:45:06,678 --> 00:45:09,698

what it is and the results and the show

you're going to get.

814

00:45:10,298 --> 00:45:15,378

I find that extremely interesting to see

how the American games, the US games are

815

00:45:15,378 --> 00:45:16,370

structured.

816

00:45:16,398 --> 00:45:24,138

Because ironically, it's a system where

there is much more social transfers, if

817

00:45:24,138 --> 00:45:28,558

you want, like we have in Europe for

social security and health and education.

818

00:45:28,558 --> 00:45:33,338

American sports are socialist, and

European sports are capitalist.

819

00:45:33,338 --> 00:45:37,118

But typically, we consider Americans to be

more capitalist and the Europeans to be

820

00:45:37,118 --> 00:45:38,058

more socialist.

821

00:45:38,058 --> 00:45:41,078

So it's an interesting inversion.

822

00:45:41,078 --> 00:45:41,478

Yeah.

823

00:45:41,478 --> 00:45:42,718

No, definitely.

824

00:45:42,838 --> 00:45:45,006

And I mean, I think...

825

00:45:45,006 --> 00:45:48,366

Honestly, that's going to be interesting

in the coming years to see what's

826

00:45:48,366 --> 00:45:51,986

happening on the European side because

there are more and more debates about

827

00:45:51,986 --> 00:45:59,266

whether we should have a closed European

wide league, which would basically be an

828

00:45:59,266 --> 00:46:02,266

extension of the current Champions League.

829

00:46:02,346 --> 00:46:07,406

And honestly, I think it's going to take

that road because more and more

830

00:46:07,926 --> 00:46:12,946

championship, at least all the

championship, I would say, for the

831

00:46:12,946 --> 00:46:14,574

exception of the Premier League.

832

00:46:14,574 --> 00:46:17,094

get more and more concentrated on just a

few clubs.

833

00:46:17,094 --> 00:46:22,314

And just from time to time, you have one

club that bumps onto the top, like

834

00:46:22,314 --> 00:46:27,634

Leverkusen this year in Germany, Monaco in

France a few years ago, Montpellier.

835

00:46:27,634 --> 00:46:29,834

But that's like really exceptions.

836

00:46:29,834 --> 00:46:34,084

And in the end, you almost always get the

same clubs that win all the time.

837

00:46:34,084 --> 00:46:39,564

And so the idea of open leagues is not

really true for the top of the leagues.

838

00:46:39,564 --> 00:46:43,694

It's definitely true for the bottom, but

the big clubs never go down.

839

00:46:43,694 --> 00:46:44,110

And...

840

00:46:44,110 --> 00:46:48,190

And so I think at some point, this

illusion of the open leagues is going to

841

00:46:48,190 --> 00:46:53,770

disappear and probably we'll get a

European wide championship where like

842

00:46:53,770 --> 00:46:58,970

basically the leagues are going to get a

bit more even because I think it's better

843

00:46:58,970 --> 00:47:00,970

for the show and that's going to make more

money.

844

00:47:00,970 --> 00:47:03,490

And in the end, I think that's what the

question is also.

845

00:47:03,490 --> 00:47:05,750

Yeah, you might be right, but I hope, I

hope not.

846

00:47:05,750 --> 00:47:11,350

I really, as an American, always have

dreamed of Americans doing relegation and

847

00:47:11,350 --> 00:47:12,750

promotion just because...

848

00:47:12,750 --> 00:47:16,900

You know, in America, we have this problem

where we call it tanking, right?

849

00:47:16,900 --> 00:47:20,810

Because we have the socialist draft system

where if the worst teams are incentivized

850

00:47:20,810 --> 00:47:24,150

to lose because they know they're not

going to win.

851

00:47:24,150 --> 00:47:29,570

So they want to get the best possible

players in the draft the next season.

852

00:47:29,570 --> 00:47:33,050

And so they're incentivized, you know, to,

to lose a little bit more.

853

00:47:33,050 --> 00:47:36,370

And so that really does kind of, you know,

the promotion relegation is nice because

854

00:47:36,370 --> 00:47:40,710

it solves that, you know, if you keep

losing, you lose a lot of money because

855

00:47:40,710 --> 00:47:42,052

you get sent down.

856

00:47:42,480 --> 00:47:47,410

so everyone's motivated even at the bottom

of each league to keep winning games,

857

00:47:47,410 --> 00:47:47,730

right?

858

00:47:47,730 --> 00:47:48,680

As much as possible.

859

00:47:48,680 --> 00:47:50,920

Otherwise they lose a lot of money.

860

00:47:50,920 --> 00:47:54,890

And in American leagues with the closed

system, it's like, well, Hey, you know,

861

00:47:54,890 --> 00:47:56,530

it's actually, we talk about sick sickle.

862

00:47:56,530 --> 00:47:59,950

He, and one thing that sports analytics

analytics have done is essentially say,

863

00:47:59,950 --> 00:48:06,810

it's really hard to go from an American

sport being an average team to a really

864

00:48:06,810 --> 00:48:07,580

good team.

865

00:48:07,580 --> 00:48:09,318

And the reason is.

866

00:48:09,518 --> 00:48:10,728

is the draft system.

867

00:48:10,728 --> 00:48:15,358

So in the draft system, people are always

overconfident in how good the players are,

868

00:48:15,358 --> 00:48:20,258

but there's really thick right tails of

how good a player can be.

869

00:48:20,258 --> 00:48:23,958

So when you get a new player who's young

and you can draft them at the top of the

870

00:48:23,958 --> 00:48:27,908

draft, they might not pan out, but they

also have a really thick right tail,

871

00:48:27,908 --> 00:48:31,958

meaning that if they do pan out, you could

go from being one of the worst teams to

872

00:48:31,958 --> 00:48:34,018

one of the best teams really quickly.

873

00:48:34,018 --> 00:48:35,206

And so,

874

00:48:35,662 --> 00:48:38,982

You know, it's this other analysis of

like, well, if you don't ever have an

875

00:48:38,982 --> 00:48:43,182

option opportunity to draft someone in a

position where there's that right tail,

876

00:48:43,182 --> 00:48:47,642

where, you know, once out of every five

years, you get a player who's transcends

877

00:48:47,642 --> 00:48:52,562

everyone else that comes in, then you

can't move up from average to really good,

878

00:48:52,562 --> 00:48:55,482

but you can go from being bad to really

good.

879

00:48:55,482 --> 00:48:59,742

So often teams and the smarter teams, if

they're really good, they say really good.

880

00:48:59,742 --> 00:49:03,082

But once they start noticing the players

are getting older, they just trade

881

00:49:03,082 --> 00:49:04,174

everybody away.

882

00:49:04,174 --> 00:49:08,394

They get rid of all their best players and

they just stink for a year or two and

883

00:49:08,394 --> 00:49:09,744

hopefully they can get some good draft.

884

00:49:09,744 --> 00:49:11,474

They get a lot of draft picks.

885

00:49:11,474 --> 00:49:11,594

Essentially.

886

00:49:11,594 --> 00:49:14,974

They try to trade their players away, get

more draft picks, and then it becomes a

887

00:49:14,974 --> 00:49:15,694

sample size problem.

888

00:49:15,694 --> 00:49:21,074

And it says, well, if we have more draft

picks, our probability of getting someone

889

00:49:21,074 --> 00:49:22,834

on the right tail goes up.

890

00:49:22,834 --> 00:49:25,954

And so that's all we're going to do is

we're just going to increase our odds of

891

00:49:25,954 --> 00:49:27,114

getting that right tail player.

892

00:49:27,114 --> 00:49:30,714

And if we get that player, then we'll be

good again.

893

00:49:31,134 --> 00:49:31,274

Yeah.

894

00:49:31,274 --> 00:49:31,734

Yeah.

895

00:49:31,734 --> 00:49:32,494

It's like.

896

00:49:32,494 --> 00:49:34,494

buying a lot of lottery tickets.

897

00:49:34,494 --> 00:49:36,294

Yeah, that's what they're doing.

898

00:49:36,574 --> 00:49:38,754

Yeah, now that's fascinating.

899

00:49:38,894 --> 00:49:41,794

Yeah, I wasn't aware of these effects.

900

00:49:41,794 --> 00:49:43,694

That's super interesting.

901

00:49:43,934 --> 00:49:48,734

Because basically, what you're saying is

there is an incentive to be extreme,

902

00:49:48,734 --> 00:49:49,194

basically.

903

00:49:49,194 --> 00:49:53,254

Either you want to be among the top ones

or you want to be among the worst ones.

904

00:49:53,254 --> 00:49:55,794

But being in the middle is the worst,

actually.

905

00:49:55,794 --> 00:49:56,794

It is the worst.

906

00:49:56,794 --> 00:49:57,134

Yeah.

907

00:49:57,134 --> 00:49:57,454

Yeah.

908

00:49:57,454 --> 00:49:59,614

That is extremely interesting.

909

00:49:59,614 --> 00:50:01,254

And that's...

910

00:50:01,710 --> 00:50:06,490

Yeah, I mean, I actually don't know which

system I prefer.

911

00:50:06,670 --> 00:50:12,310

Honestly, I'm just saying I think Europe

is getting, is going there because we have

912

00:50:12,310 --> 00:50:16,330

more and more basically concentration of

the wealth at the very top of the leagues

913

00:50:16,330 --> 00:50:19,070

and that's going to make the national

leagues less and less interesting

914

00:50:19,070 --> 00:50:20,310

basically.

915

00:50:21,110 --> 00:50:26,950

But I don't know either if I prefer the

European wide championship.

916

00:50:27,050 --> 00:50:30,382

Well, I think I would prefer European wide

championship.

917

00:50:30,382 --> 00:50:33,342

for sure, but I think it would be great to

have it still open.

918

00:50:33,342 --> 00:50:36,542

So where you could have, you know, like

basically countries would become regions

919

00:50:36,542 --> 00:50:42,222

and then you get from like, if you, if

you're in the best in France, basically in

920

00:50:42,222 --> 00:50:45,342

one year, then you get to the highest

level, which is the European one.

921

00:50:45,342 --> 00:50:49,622

And then if you're among the worst, you

get down to your country the next year.

922

00:50:49,622 --> 00:50:55,262

I think that would be very fun because

the, like, especially now that players can

923

00:50:55,262 --> 00:50:58,478

be traded very easily between the, the...

924

00:50:58,478 --> 00:51:01,758

continental Europe because it's basically

the same country legally.

925

00:51:01,758 --> 00:51:07,178

That also makes sense that the teams, you

know, basically meeting PSG versus

926

00:51:07,178 --> 00:51:14,518

Barcelona is much more tied than PSG

versus literally any team in France.

927

00:51:14,578 --> 00:51:17,298

So yeah, that's going to be very

interesting.

928

00:51:17,558 --> 00:51:24,494

But at the same time, I'm very, yeah, I

love hearing about the wrong incentives.

929

00:51:24,494 --> 00:51:27,074

at the same time of the closed system.

930

00:51:27,074 --> 00:51:28,594

So thanks a lot for that.

931

00:51:28,594 --> 00:51:30,354

That's food for thought.

932

00:51:30,774 --> 00:51:36,234

And that's again, like that's very close

to two elections, actually, like how you

933

00:51:36,234 --> 00:51:39,394

count the votes impacts the winner.

934

00:51:39,394 --> 00:51:46,274

And so here, like really in sports to how

you structure your game has an impact on

935

00:51:46,274 --> 00:51:47,014

the winners.

936

00:51:47,014 --> 00:51:52,914

And I think it's extremely important to

keep in mind because in the end, like how

937

00:51:52,914 --> 00:51:53,838

the

938

00:51:53,838 --> 00:52:01,678

the organization, so the MLS in the US or

the UEFA in Europe have actually huge

939

00:52:01,678 --> 00:52:04,678

power over the game.

940

00:52:06,258 --> 00:52:09,498

Well, thanks for that political science

parenthesis.

941

00:52:09,498 --> 00:52:12,638

I wasn't expecting that, but that's

definitely super interesting.

942

00:52:13,018 --> 00:52:17,838

To get back to the modeling because time

is running by and I definitely want to ask

943

00:52:17,838 --> 00:52:22,286

you about the plus minus models because

you're using that also to...

944

00:52:22,286 --> 00:52:25,006

estimate player value in American

football.

945

00:52:25,006 --> 00:52:26,385

So I'm curious about that.

946

00:52:26,385 --> 00:52:28,506

What is that kind of model?

947

00:52:28,506 --> 00:52:33,406

Is that mainly for American football that

you're using that also for other sports?

948

00:52:33,406 --> 00:52:37,986

Or if it's only for American football, why

is that particularly tailored to that

949

00:52:37,986 --> 00:52:38,906

sport?

950

00:52:39,486 --> 00:52:39,766

Yeah.

951

00:52:39,766 --> 00:52:45,906

So plus minus models actually are

originated in basketball and they're, they

952

00:52:45,906 --> 00:52:46,936

work the best in basketball.

953

00:52:46,936 --> 00:52:48,166

They're not perfect.

954

00:52:48,166 --> 00:52:51,746

And that sort of the concept in basketball

is you have 10 players on the court at

955

00:52:51,746 --> 00:52:52,526

each.

956

00:52:52,526 --> 00:52:54,886

at each moment and they substitute in and

out.

957

00:52:54,886 --> 00:52:58,706

But while those 10 players are on the

court, you know how many points are scored

958

00:52:58,706 --> 00:53:00,286

for each team, right?

959

00:53:00,286 --> 00:53:04,706

So, you know, five players on the offense

side and five players on defensive side.

960

00:53:04,706 --> 00:53:09,086

There's essentially just a big linear

model and you look at and you want to

961

00:53:09,086 --> 00:53:12,746

adjust for how long they're on the court

or how many possessions they were on the

962

00:53:12,746 --> 00:53:13,136

court for.

963

00:53:13,136 --> 00:53:16,846

So you can say, okay, these 10 players are

on the court for two and a half minutes.

964

00:53:16,846 --> 00:53:21,186

And in those two and a half minutes, this

team scored six points and their team

965

00:53:21,186 --> 00:53:22,318

scored four points.

966

00:53:22,318 --> 00:53:26,678

And essentially what you're doing then is

a plus minus model, essentially.

967

00:53:26,678 --> 00:53:31,618

So sometimes you might see in a, in a

statistic after the game, like the total

968

00:53:31,618 --> 00:53:36,518

difference in the net points for the team

when a player was on the court versus when

969

00:53:36,518 --> 00:53:37,438

they're not.

970

00:53:37,438 --> 00:53:40,238

Well, that's not too useful because

there's a lot of correlations, right?

971

00:53:40,238 --> 00:53:42,738

You're playing with someone else a lot.

972

00:53:42,738 --> 00:53:46,538

So what we call an adjusted plus minus

model, right, is a linear model that then

973

00:53:46,538 --> 00:53:51,618

tries to fit those player effects of, you

know, you get a one when you're on the

974

00:53:51,618 --> 00:53:52,014

court.

975

00:53:52,014 --> 00:53:53,984

on offense and negative one year on

defense.

976

00:53:53,984 --> 00:53:56,354

And we look at your team's efficiency,

right?

977

00:53:56,354 --> 00:54:01,634

Your points divided by some denominator,

whether it's minutes or possessions.

978

00:54:01,634 --> 00:54:01,834

Okay.

979

00:54:01,834 --> 00:54:04,124

And that's sort of the basketball thing

over time.

980

00:54:04,124 --> 00:54:08,574

They realized, okay, well, there's so much

correlation between who is playing

981

00:54:08,574 --> 00:54:09,174

together.

982

00:54:09,174 --> 00:54:10,664

We need to adjust for that.

983

00:54:10,664 --> 00:54:12,254

So they used ridge regression.

984

00:54:12,254 --> 00:54:15,554

And so that would divvy up the credit a

little bit better.

985

00:54:15,554 --> 00:54:18,542

And you know, ridge regression is very

good at when there's

986

00:54:18,542 --> 00:54:22,102

A lot of multicollinearity or correlation

between two effects, right?

987

00:54:22,102 --> 00:54:25,822

And on the basketball team or all

basketball players, you have teammates

988

00:54:25,822 --> 00:54:29,262

that play a lot together and they don't

play with other people a lot.

989

00:54:29,282 --> 00:54:33,642

But Ridge Regression has done a decently

good job in basketball over a big sample

990

00:54:33,642 --> 00:54:37,022

of estimating how effective players are.

991

00:54:37,022 --> 00:54:40,902

And if you look at these things, you'll

see, we talked about the sniff test.

992

00:54:40,902 --> 00:54:43,694

In 2012, LeBron is the number one player.

993

00:54:43,694 --> 00:54:46,914

And he's the number one player for a lot

of the years, not so much anymore because

994

00:54:46,914 --> 00:54:47,964

he's older, et cetera.

995

00:54:47,964 --> 00:54:48,144

Right.

996

00:54:48,144 --> 00:54:50,794

But that's sort of those sniff tests that

we get.

997

00:54:50,834 --> 00:54:55,644

Well, some people in, in basketball and

I'm proponent of this, like, you know,

998

00:54:55,644 --> 00:55:00,894

this is a Bayesian podcast is that ridge

regression, you know, for those unfamiliar

999

00:55:00,894 --> 00:55:04,864

is, is a frequentist way to write a

Bayesian model.

Speaker: 1000

00:55:04,864 --> 00:55:09,474

That's very specific where you have a

normal prior on each player with a mean

Speaker: 1001

00:55:09,474 --> 00:55:10,474

zero.

Speaker: 1002

00:55:10,474 --> 00:55:10,774

Okay.

Speaker: 1003

00:55:10,774 --> 00:55:12,270

And that's ridge regression.

Speaker: 1004

00:55:12,270 --> 00:55:16,310

So we think about it from that perspective

with adjusted plus minus models.

Speaker: 1005

00:55:16,350 --> 00:55:20,030

What happens when you have a normal prior

with mean zero is that when you have

Speaker: 1006

00:55:20,030 --> 00:55:24,710

players that play less, we shrink more

towards the prior mean.

Speaker: 1007

00:55:24,770 --> 00:55:28,349

And it's only when we have more data for

players that we can deviate from that

Speaker: 1008

00:55:28,349 --> 00:55:29,150

prior mean.

Speaker: 1009

00:55:29,150 --> 00:55:33,490

Well, one thing we know about sports is if

you're not playing as much, that actually

Speaker: 1010

00:55:33,490 --> 00:55:34,630

is pretty useful information.

Speaker: 1011

00:55:34,630 --> 00:55:35,670

And what does that tell us?

Speaker: 1012

00:55:35,670 --> 00:55:36,990

You're not very good.

Speaker: 1013

00:55:36,990 --> 00:55:39,570

Because if you're good, you're going to

play more.

Speaker: 1014

00:55:39,570 --> 00:55:41,710

And if you're bad, you play less.

Speaker: 1015

00:55:41,710 --> 00:55:45,900

So other people have come around and, you

know, in the last 10, 15 years and said,

Speaker: 1016

00:55:45,900 --> 00:55:50,090

okay, well, instead of a ridge regression

model for basketball, we should do a

Speaker: 1017

00:55:50,090 --> 00:55:51,070

Bayesian regression model.

Speaker: 1018

00:55:51,070 --> 00:55:55,650

And instead of having a mean zero for a

player, we should have a mean of something

Speaker: 1019

00:55:55,650 --> 00:55:56,120

else.

Speaker: 1020

00:55:56,120 --> 00:55:58,730

So there's a few different versions that

people have done.

Speaker: 1021

00:55:58,730 --> 00:56:04,270

One thing, a very simple version is say

just everybody has a mean prior mean of,

Speaker: 1022

00:56:04,270 --> 00:56:06,630

you know, what we call a replacement

player.

Speaker: 1023

00:56:06,630 --> 00:56:07,110

Okay.

Speaker: 1024

00:56:07,110 --> 00:56:08,750

Someone that doesn't play very much.

Speaker: 1025

00:56:08,750 --> 00:56:10,574

If you're really good and you play a lot.

Speaker: 1026

00:56:10,574 --> 00:56:14,633

It doesn't matter what the prior mean is

too much because the data is going to

Speaker: 1027

00:56:14,633 --> 00:56:15,794

overwhelm the prior.

Speaker: 1028

00:56:15,794 --> 00:56:20,294

But if you don't play very much, we're

going to stick with that sort of negative

Speaker: 1029

00:56:20,294 --> 00:56:23,354

prior mean because it means you're below

average.

Speaker: 1030

00:56:23,354 --> 00:56:25,144

And so that's one thing you can do.

Speaker: 1031

00:56:25,144 --> 00:56:28,174

A more sophisticated thing sometimes

people will do is they'll build a

Speaker: 1032

00:56:28,174 --> 00:56:33,874

hierarchical model where you have

essentially a, a prior mean that is based

Speaker: 1033

00:56:33,874 --> 00:56:37,454

on other statistics that we observe.

Speaker: 1034

00:56:37,474 --> 00:56:40,270

So how many points you score or how many

assists you have.

Speaker: 1035

00:56:40,270 --> 00:56:46,610

And those that's called a box, a box score

prior mean or a box score plus minus.

Speaker: 1036

00:56:46,610 --> 00:56:47,790

So that's sort of the basketball.

Speaker: 1037

00:56:47,790 --> 00:56:49,190

So we gave you the what plus minus models.

Speaker: 1038

00:56:49,190 --> 00:56:51,210

So that's sort of the basketball approach.

Speaker: 1039

00:56:51,229 --> 00:56:51,850

Now.

Speaker: 1040

00:56:51,920 --> 00:56:56,190

Basketball is really nice because you have

lots of games in the NBA.

Speaker: 1041

00:56:56,190 --> 00:57:00,550

You play every team at least twice and you

substitute a lot and there's lots of

Speaker: 1042

00:57:00,550 --> 00:57:01,450

scoring.

Speaker: 1043

00:57:01,450 --> 00:57:06,690

Now my work in American football tried to

address a lot of these issues in American

Speaker: 1044

00:57:06,690 --> 00:57:07,080

football.

Speaker: 1045

00:57:07,080 --> 00:57:08,886

You don't play every team.

Speaker: 1046

00:57:09,262 --> 00:57:11,182

you don't substitute very much.

Speaker: 1047

00:57:11,182 --> 00:57:16,322

And if you do play, you only play with

certain people like all the time.

Speaker: 1048

00:57:16,322 --> 00:57:18,812

And then there's not a lot of scoring

compared to basketball.

Speaker: 1049

00:57:18,812 --> 00:57:22,742

There's some scoring, but you know,

there's, you know, American football point

Speaker: 1050

00:57:22,742 --> 00:57:24,462

scoring is unique, right?

Speaker: 1051

00:57:24,462 --> 00:57:27,542

You get six or seven points for a

touchdown, you get three points for a

Speaker: 1052

00:57:27,542 --> 00:57:30,902

field goal, you know, and then on more

rare occasions, you get these two point

Speaker: 1053

00:57:30,902 --> 00:57:32,182

safeties.

Speaker: 1054

00:57:32,202 --> 00:57:32,878

Yeah.

Speaker: 1055

00:57:32,878 --> 00:57:38,258

So there's roughly maybe 10 scoring events

in an American football game versus in

Speaker: 1056

00:57:38,258 --> 00:57:42,318

basketball where you have, you know, a

hundred to a hundred.

Speaker: 1057

00:57:42,318 --> 00:57:45,968

So there's, you know, about each two to

three points, each one there's, you know,

Speaker: 1058

00:57:45,968 --> 00:57:48,428

80 to 120 scoring events in a basketball

game.

Speaker: 1059

00:57:48,428 --> 00:57:48,598

Right.

Speaker: 1060

00:57:48,598 --> 00:57:49,988

So these models work a lot better.

Speaker: 1061

00:57:49,988 --> 00:57:53,398

My work in American football has been to

sort of, how do we take the basketball

Speaker: 1062

00:57:53,398 --> 00:57:56,588

model and make some modifications so we

can do a football model?

Speaker: 1063

00:57:56,588 --> 00:57:59,718

And so one of the things that is tricky in

football is.

Speaker: 1064

00:57:59,886 --> 00:58:03,606

that certain positions never get

substituted out.

Speaker: 1065

00:58:03,606 --> 00:58:07,925

So on offense, the quarterback plays every

single play unless they're hurt or they

Speaker: 1066

00:58:07,925 --> 00:58:08,706

stink.

Speaker: 1067

00:58:08,706 --> 00:58:10,296

So they get benched.

Speaker: 1068

00:58:10,296 --> 00:58:14,966

Well, the quarterback also always plays

with the same offensive line as long as

Speaker: 1069

00:58:14,966 --> 00:58:17,286

they're healthy and they don't get

substituted out.

Speaker: 1070

00:58:17,286 --> 00:58:22,146

So how does a model separate credit when

the same players are on the field all the

Speaker: 1071

00:58:22,146 --> 00:58:22,336

time?

Speaker: 1072

00:58:22,336 --> 00:58:27,470

And so my work in that was sort of to use

Bayesian statistics and take the...

Speaker: 1073

00:58:27,470 --> 00:58:31,590

the Bayesian regression model where we had

a prior mean, I used some information to

Speaker: 1074

00:58:31,590 --> 00:58:36,690

inform the prior mean for each player, but

I also did this unique thing where I

Speaker: 1075

00:58:36,690 --> 00:58:37,510

shrink.

Speaker: 1076

00:58:37,510 --> 00:58:43,810

So the prior variance is a function and is

actually, there's one prior variance for

Speaker: 1077

00:58:43,810 --> 00:58:48,410

all players and then it's multiplied by

another parameter, which is unique for the

Speaker: 1078

00:58:48,410 --> 00:58:50,150

position that they play.

Speaker: 1079

00:58:50,150 --> 00:58:54,610

And so quarterbacks have a different

shrinkage parameter, essentially, or prior

Speaker: 1080

00:58:54,610 --> 00:58:56,238

variance than.

Speaker: 1081

00:58:56,238 --> 00:58:58,278

a different position.

Speaker: 1082

00:58:58,478 --> 00:59:02,218

And then instead of just looking at

scoring plays in football, we have what we

Speaker: 1083

00:59:02,218 --> 00:59:03,868

call is expected points added.

Speaker: 1084

00:59:03,868 --> 00:59:09,098

So at each play, we look at on average,

how many points are you going to score if

Speaker: 1085

00:59:09,098 --> 00:59:11,238

you have the ball in this position?

Speaker: 1086

00:59:11,238 --> 00:59:13,578

And I look at the difference between two

plays, right?

Speaker: 1087

00:59:13,578 --> 00:59:18,378

And that tells you essentially how much

value you got in the result of the play.

Speaker: 1088

00:59:18,438 --> 00:59:22,238

So instead of using every scoring play, I

just use every single play in football.

Speaker: 1089

00:59:22,238 --> 00:59:24,782

And I do this unique shrinkage.

Speaker: 1090

00:59:24,942 --> 00:59:28,901

dependent on position and doing that, and

it's a huge model.

Speaker: 1091

00:59:28,901 --> 00:59:32,582

So I did this in college football, which

has way too many parameters because

Speaker: 1092

00:59:32,582 --> 00:59:34,342

there's like 16 ,000 kids.

Speaker: 1093

00:59:34,342 --> 00:59:39,652

But even in the NFL, I've done this and

you get interesting results.

Speaker: 1094

00:59:39,652 --> 00:59:42,632

Sometimes they match up with what you

think, sometimes they don't.

Speaker: 1095

00:59:42,632 --> 00:59:45,642

But the interesting thing is you can

actually estimate how much you should

Speaker: 1096

00:59:45,642 --> 00:59:47,182

shrink each position.

Speaker: 1097

00:59:47,182 --> 00:59:52,222

And so actually the model is nice because

it essentially tells you how much of the

Speaker: 1098

00:59:52,222 --> 00:59:53,902

variance in the outcome of the play.

Speaker: 1099

00:59:53,902 --> 00:59:58,622

is dependent on how good players are

across different positions.

Speaker: 1100

00:59:58,762 --> 01:00:02,942

So in football, we all know that

quarterbacks are the most impactful

Speaker: 1101

01:00:02,942 --> 01:00:05,062

position in the game.

Speaker: 1102

01:00:05,062 --> 01:00:10,542

And I did give somewhat subjective priors,

but not with, I still left a lot of

Speaker: 1103

01:00:10,542 --> 01:00:14,862

uncertainty around and the model very well

could see and estimate that quarterbacks

Speaker: 1104

01:00:14,862 --> 01:00:20,182

are in fact the most important position

because you shrink them the less they have

Speaker: 1105

01:00:20,182 --> 01:00:21,742

the largest variance.

Speaker: 1106

01:00:21,742 --> 01:00:23,556

So.

Speaker: 1107

01:00:23,886 --> 01:00:24,816

You could look at that.

Speaker: 1108

01:00:24,816 --> 01:00:29,206

If you look at the most impactful players

in football, it should be a quarterback.

Speaker: 1109

01:00:29,206 --> 01:00:33,866

But in the same measure, the worst players

in football are also quarterbacks because

Speaker: 1110

01:00:33,866 --> 01:00:37,906

in order to negatively hurt your team, you

can only hurt your team really a lot.

Speaker: 1111

01:00:37,906 --> 01:00:40,966

If you're a quarterback compared to other

positions, I mean, every position you can

Speaker: 1112

01:00:40,966 --> 01:00:43,846

hurt your team, but no one can hurt a team

as much as a bad quarterback hurts their

Speaker: 1113

01:00:43,846 --> 01:00:44,566

team.

Speaker: 1114

01:00:44,566 --> 01:00:47,276

Just like a good quarterback can help

their team better.

Speaker: 1115

01:00:47,276 --> 01:00:52,866

So that's sort of like a kind of rough

overview of, of my plus minus modeling in

Speaker: 1116

01:00:52,866 --> 01:00:53,614

football.

Speaker: 1117

01:00:53,614 --> 01:00:59,784

I think I do have, when I wrote the paper,

I have a version of that written in Stan.

Speaker: 1118

01:00:59,784 --> 01:01:04,354

The data set itself was not public, but I

did have a version of the Stan model

Speaker: 1119

01:01:04,354 --> 01:01:08,194

written and uploaded on my GitHub that you

can look at.

Speaker: 1120

01:01:08,194 --> 01:01:10,194

It's pretty massive.

Speaker: 1121

01:01:10,474 --> 01:01:15,694

In recent years, I've tried to expand it

and to do a state space model type

Speaker: 1122

01:01:15,694 --> 01:01:16,074

version.

Speaker: 1123

01:01:16,074 --> 01:01:19,022

So I have effects for each player for each

season over time.

Speaker: 1124

01:01:19,022 --> 01:01:20,932

Yeah, that was exactly what I meant.

Speaker: 1125

01:01:20,932 --> 01:01:23,822

Computationally, that gets a little bit

trickier.

Speaker: 1126

01:01:23,822 --> 01:01:28,342

And my dataset, actually, I was able to

scrape some data for that.

Speaker: 1127

01:01:28,342 --> 01:01:29,742

And then actually, I can't anymore.

Speaker: 1128

01:01:29,742 --> 01:01:32,622

The NFL just stopped releasing that.

Speaker: 1129

01:01:32,622 --> 01:01:34,492

So that work is on hold for now.

Speaker: 1130

01:01:34,492 --> 01:01:38,842

But I probably need to find a graduate

student that can help me finish it.

Speaker: 1131

01:01:40,162 --> 01:01:43,102

Yeah, definitely we should put that in the

show notes.

Speaker: 1132

01:01:43,462 --> 01:01:44,642

That's super interesting.

Speaker: 1133

01:01:44,642 --> 01:01:46,414

Your paper in the...

Speaker: 1134

01:01:46,414 --> 01:01:49,354

and the link to the GitHub repo.

Speaker: 1135

01:01:49,674 --> 01:01:51,554

That's for sure.

Speaker: 1136

01:01:52,014 --> 01:02:01,094

And that makes me think a recent episode I

did, and also a recent interest of mine, I

Speaker: 1137

01:02:01,094 --> 01:02:09,514

started contributing to that package

called Baseflow, where that's precisely

Speaker: 1138

01:02:09,794 --> 01:02:15,074

that could be useful in your case here,

because your model structure doesn't

Speaker: 1139

01:02:15,074 --> 01:02:15,950

change.

Speaker: 1140

01:02:15,950 --> 01:02:19,430

If I understand correctly, because well,

once you have the model structure, it's

Speaker: 1141

01:02:19,430 --> 01:02:20,770

kind of like a physics model.

Speaker: 1142

01:02:20,770 --> 01:02:25,040

It's not going to change when you have new

data, but the data sets do change.

Speaker: 1143

01:02:25,040 --> 01:02:27,230

So you have new data sets coming in.

Speaker: 1144

01:02:27,230 --> 01:02:33,210

And so that's where probably using these

kind of inference that's called amortized

Speaker: 1145

01:02:33,210 --> 01:02:38,550

Bayesian inference could be extremely

useful because you would basically, if the

Speaker: 1146

01:02:38,550 --> 01:02:42,320

bottle, the computational bottleneck would

just happen once.

Speaker: 1147

01:02:42,729 --> 01:02:46,192

That would be when you train the deep

neural network.

Speaker: 1148

01:02:46,222 --> 01:02:50,522

to learn the posterior structure and

parameters.

Speaker: 1149

01:02:50,582 --> 01:02:57,342

So instead of MCMC, you're using the deep

neural network to learn the posterior.

Speaker: 1150

01:02:57,342 --> 01:03:03,442

But then once you have trained the deep

neural network, then it's like doing

Speaker: 1151

01:03:03,442 --> 01:03:05,502

posterior inference is trivial.

Speaker: 1152

01:03:06,182 --> 01:03:12,782

And so for that kind of models where you

have a lot of data, but the model is the

Speaker: 1153

01:03:12,782 --> 01:03:13,882

same.

Speaker: 1154

01:03:14,318 --> 01:03:18,378

That's a very good use case for amortized

Bayesian inference.

Speaker: 1155

01:03:18,598 --> 01:03:21,538

So that could be something very

interesting here.

Speaker: 1156

01:03:21,958 --> 01:03:22,058

Yeah.

Speaker: 1157

01:03:22,058 --> 01:03:22,798

Yeah.

Speaker: 1158

01:03:22,798 --> 01:03:23,078

Yeah.

Speaker: 1159

01:03:23,078 --> 01:03:23,178

Yeah.

Speaker: 1160

01:03:23,178 --> 01:03:28,518

Happy to tell you more about that

afterwards if you're interested.

Speaker: 1161

01:03:28,518 --> 01:03:32,498

But yeah, I've started digging into that,

and that's super fun for sure.

Speaker: 1162

01:03:32,498 --> 01:03:36,118

So yeah, and I think this is a cool use

case.

Speaker: 1163

01:03:36,958 --> 01:03:37,308

Awesome.

Speaker: 1164

01:03:37,308 --> 01:03:41,646

Well, I still have a few questions, but

can I?

Speaker: 1165

01:03:41,646 --> 01:03:45,986

We are getting short on time, so can I

keep you a bit longer?

Speaker: 1166

01:03:46,146 --> 01:03:47,426

Yeah, just a few more minutes.

Speaker: 1167

01:03:47,426 --> 01:03:47,566

Sure.

Speaker: 1168

01:03:47,566 --> 01:03:48,406

Yeah.

Speaker: 1169

01:03:48,406 --> 01:03:48,786

Okay.

Speaker: 1170

01:03:48,786 --> 01:03:49,366

Awesome.

Speaker: 1171

01:03:49,366 --> 01:03:49,986

Yeah.

Speaker: 1172

01:03:49,986 --> 01:03:56,626

So actually, I'd like to pick your brain

about now talking a bit more about the

Speaker: 1173

01:03:56,626 --> 01:03:57,666

future.

Speaker: 1174

01:03:57,666 --> 01:03:59,066

I'm curious.

Speaker: 1175

01:03:59,066 --> 01:04:01,846

So let me fuse two questions.

Speaker: 1176

01:04:01,846 --> 01:04:09,606

So first, I'm curious what, where do you

see the field of spots analytics heading

Speaker: 1177

01:04:09,606 --> 01:04:11,438

in the next years?

Speaker: 1178

01:04:11,438 --> 01:04:13,038

five to 10 years.

Speaker: 1179

01:04:13,278 --> 01:04:19,118

And also sub question is other spots,

specific spots where you see significant

Speaker: 1180

01:04:19,118 --> 01:04:21,958

potential for growth in analytics.

Speaker: 1181

01:04:23,918 --> 01:04:25,378

Yeah, those are, those are good questions.

Speaker: 1182

01:04:25,378 --> 01:04:27,478

I think they go kind of hand in hand.

Speaker: 1183

01:04:27,478 --> 01:04:33,718

You know, I think it's hard to, it's hard

if I could predict the future, right?

Speaker: 1184

01:04:33,718 --> 01:04:36,857

I would probably have a different job.

Speaker: 1185

01:04:37,258 --> 01:04:39,258

I'd probably be retired.

Speaker: 1186

01:04:39,258 --> 01:04:41,216

But.

Speaker: 1187

01:04:41,486 --> 01:04:47,366

You know, I think a lot of the future is

going to be catching up to, you know,

Speaker: 1188

01:04:47,366 --> 01:04:52,776

sports like soccer, American football,

hockey going to be catching up.

Speaker: 1189

01:04:52,776 --> 01:04:56,846

And I think a lot of the growth is

actually going to be making sports

Speaker: 1190

01:04:56,846 --> 01:05:00,866

analytics more digestible for just

everyday people.

Speaker: 1191

01:05:00,866 --> 01:05:02,306

So the fans, right.

Speaker: 1192

01:05:02,306 --> 01:05:03,686

And that's happened over time, right?

Speaker: 1193

01:05:03,686 --> 01:05:07,266

You watched a broadcast of a, of a soccer

game,

Speaker: 1194

01:05:07,854 --> 01:05:10,144

20 years ago, no one talked about expected

goals.

Speaker: 1195

01:05:10,144 --> 01:05:12,314

Now, most broadcasts will show it.

Speaker: 1196

01:05:12,314 --> 01:05:13,634

They might not always talk about it.

Speaker: 1197

01:05:13,634 --> 01:05:14,534

They'll show it.

Speaker: 1198

01:05:14,534 --> 01:05:19,234

Like I said, expected goals, it's better

than just showing the score, but there's a

Speaker: 1199

01:05:19,234 --> 01:05:22,094

lot to be left undone.

Speaker: 1200

01:05:22,094 --> 01:05:26,714

I think in the future, there's going to be

a lot of sports analytics that's really

Speaker: 1201

01:05:26,714 --> 01:05:29,974

much focused on expected values to date.

Speaker: 1202

01:05:29,974 --> 01:05:35,354

And not enough has been focused on

distributions and variance around

Speaker: 1203

01:05:35,354 --> 01:05:36,654

estimates.

Speaker: 1204

01:05:36,654 --> 01:05:41,334

And so I think once one place it's going

to have to end up going.

Speaker: 1205

01:05:41,344 --> 01:05:44,104

and part of the reason is, right, we, we

talk about neural networks.

Speaker: 1206

01:05:44,104 --> 01:05:49,014

Neural networks are very good at expected

values, with really large data sets.

Speaker: 1207

01:05:49,014 --> 01:05:50,284

It's a lot harder, right?

Speaker: 1208

01:05:50,284 --> 01:05:54,754

Modeling variance is a lot harder in

anything than modeling and expectations.

Speaker: 1209

01:05:54,754 --> 01:05:57,544

So I think catching up on some of those

things.

Speaker: 1210

01:05:57,544 --> 01:06:02,094

And I think also, like I said, taking a

step back and I think, you know, there's

Speaker: 1211

01:06:02,094 --> 01:06:04,694

been a lot of good work that has been

done, but I think we're going to find a

Speaker: 1212

01:06:04,694 --> 01:06:05,550

few things that.

Speaker: 1213

01:06:05,550 --> 01:06:08,850

Hey, maybe we were a little bit

overconfident, right?

Speaker: 1214

01:06:08,850 --> 01:06:12,450

And with everything in sports, it's always

about game theory.

Speaker: 1215

01:06:12,450 --> 01:06:18,310

So even if something is optimal today,

that strategy is not always going to be

Speaker: 1216

01:06:18,310 --> 01:06:20,030

optimal in the future.

Speaker: 1217

01:06:20,030 --> 01:06:25,570

And so if you, if, you know, in basketball

for a sec, we talked about three pointers.

Speaker: 1218

01:06:25,570 --> 01:06:28,350

Of course, three pointers are really good

right now because they have higher

Speaker: 1219

01:06:28,350 --> 01:06:32,830

expected value, but you know, defensively

players are learning to play against three

Speaker: 1220

01:06:32,830 --> 01:06:34,446

pointers better than they used to.

Speaker: 1221

01:06:34,446 --> 01:06:38,606

or in American football, the numbers have

said you should pass the ball more.

Speaker: 1222

01:06:38,606 --> 01:06:42,385

Well, now the defenses are learning how to

defend it better.

Speaker: 1223

01:06:42,385 --> 01:06:45,216

And so running is going to be more

important than it used to be.

Speaker: 1224

01:06:45,216 --> 01:06:45,366

Right.

Speaker: 1225

01:06:45,366 --> 01:06:47,446

And so these things are always going to

change.

Speaker: 1226

01:06:47,446 --> 01:06:50,626

And so in five to 10 years, I don't know

exactly what it's going to be, but I think

Speaker: 1227

01:06:50,626 --> 01:06:55,446

in some ways, you know, you might find

some analytics person in 10 years giving

Speaker: 1228

01:06:55,446 --> 01:07:00,166

exact opposite advice of what we're seeing

now, just because the game has evolved.

Speaker: 1229

01:07:00,166 --> 01:07:01,606

The game has changed.

Speaker: 1230

01:07:02,030 --> 01:07:04,250

And so now you should do something else,

right?

Speaker: 1231

01:07:04,250 --> 01:07:05,570

To get an edge.

Speaker: 1232

01:07:06,130 --> 01:07:09,450

and so I think the growth is in twofold.

Speaker: 1233

01:07:09,450 --> 01:07:12,310

We're always staying on the cutting edge

of like, what's next.

Speaker: 1234

01:07:12,310 --> 01:07:15,870

Sometimes that's going back to where you

were.

Speaker: 1235

01:07:16,000 --> 01:07:21,490

and like I said, making the numbers more

digestible for the everyday consumer.

Speaker: 1236

01:07:21,850 --> 01:07:24,440

you know, it's, it's one thing you and I,

we can talk about models.

Speaker: 1237

01:07:24,440 --> 01:07:26,280

I had to do this at ESPN all the time.

Speaker: 1238

01:07:26,280 --> 01:07:29,230

I can't talk about prior distributions on

TV.

Speaker: 1239

01:07:29,230 --> 01:07:29,560

Right?

Speaker: 1240

01:07:29,560 --> 01:07:31,840

So how do we explain these things?

Speaker: 1241

01:07:31,840 --> 01:07:31,990

Right?

Speaker: 1242

01:07:31,990 --> 01:07:35,150

And I think what's really going to be key

is over time, this has happened already,

Speaker: 1243

01:07:35,150 --> 01:07:38,810

but it's going to keep on happening that

the analysts themselves are going to be

Speaker: 1244

01:07:38,810 --> 01:07:41,990

much more data literate than they have

been in the past.

Speaker: 1245

01:07:41,990 --> 01:07:45,990

Not just because they have more people

working with them or they're younger.

Speaker: 1246

01:07:45,990 --> 01:07:51,170

Also the analysts in the future is going

to be able to use AI to do their own

Speaker: 1247

01:07:51,170 --> 01:07:51,870

analysis.

Speaker: 1248

01:07:51,870 --> 01:07:57,084

And that could be scary because they might

make some bad assumptions.

Speaker: 1249

01:07:57,230 --> 01:08:01,130

but they're also going to be more data

savvy and they could load up a data set

Speaker: 1250

01:08:01,130 --> 01:08:02,440

and use an AI tool.

Speaker: 1251

01:08:02,440 --> 01:08:06,830

And even if they can't code to get

insights that, you know, I used to have to

Speaker: 1252

01:08:06,830 --> 01:08:10,510

write some code to get them and now they

can just do it themselves.

Speaker: 1253

01:08:10,510 --> 01:08:10,740

Right.

Speaker: 1254

01:08:10,740 --> 01:08:14,150

And so that's, I think somewhere else that

teams and coaches are going to be able to

Speaker: 1255

01:08:14,150 --> 01:08:16,490

do more analysis on their own.

Speaker: 1256

01:08:16,490 --> 01:08:19,180

And it's not that the data people aren't,

aren't needed.

Speaker: 1257

01:08:19,180 --> 01:08:22,530

In fact, they're going to be needed even

more to make sure that the coach isn't

Speaker: 1258

01:08:22,530 --> 01:08:24,280

missing an assumption, right.

Speaker: 1259

01:08:24,280 --> 01:08:26,894

That he needs to be thinking about of the

structure of the data.

Speaker: 1260

01:08:26,894 --> 01:08:28,094

Cause he might just be, great.

Speaker: 1261

01:08:28,094 --> 01:08:29,184

Now I can run a regression.

Speaker: 1262

01:08:29,184 --> 01:08:29,794

I don't even know.

Speaker: 1263

01:08:29,794 --> 01:08:32,034

I don't even need to know how to code it.

Speaker: 1264

01:08:32,034 --> 01:08:32,414

Right.

Speaker: 1265

01:08:32,414 --> 01:08:33,054

that's great.

Speaker: 1266

01:08:33,054 --> 01:08:35,534

But are you thinking about this?

Speaker: 1267

01:08:35,534 --> 01:08:35,654

Right.

Speaker: 1268

01:08:35,654 --> 01:08:38,934

And so there's going to be a lot of

education about using some of these tools

Speaker: 1269

01:08:38,934 --> 01:08:42,784

better and every, but everyone's going to

have their access to it.

Speaker: 1270

01:08:42,784 --> 01:08:42,894

Right.

Speaker: 1271

01:08:42,894 --> 01:08:45,854

It's going to be so much more accessible

in the future than it has been in the

Speaker: 1272

01:08:45,854 --> 01:08:46,914

past.

Speaker: 1273

01:08:46,914 --> 01:08:47,494

Yeah.

Speaker: 1274

01:08:47,494 --> 01:08:48,174

Yeah.

Speaker: 1275

01:08:48,174 --> 01:08:49,194

Yeah.

Speaker: 1276

01:08:49,694 --> 01:08:50,194

yeah, for sure.

Speaker: 1277

01:08:50,194 --> 01:08:53,144

Completely, completely agree with that.

Speaker: 1278

01:08:53,144 --> 01:08:54,954

and that's also something I'm very

passionate about.

Speaker: 1279

01:08:54,954 --> 01:08:56,750

That's also what these show.

Speaker: 1280

01:08:56,750 --> 01:08:57,750

is here, right?

Speaker: 1281

01:08:57,750 --> 01:09:05,490

It's to have the bridge between the

modelers and the known stats people be

Speaker: 1282

01:09:05,490 --> 01:09:06,600

easier, in a way.

Speaker: 1283

01:09:06,600 --> 01:09:09,370

And that's something I really love doing

also in my job, basically being that

Speaker: 1284

01:09:09,370 --> 01:09:14,150

bridge between the really nitty gritty

details of the model.

Speaker: 1285

01:09:14,150 --> 01:09:19,410

And then, OK, now that we have the model,

how do we explain to the people who are

Speaker: 1286

01:09:19,410 --> 01:09:23,890

actually going to consume the model

results what the model can do, what it

Speaker: 1287

01:09:23,890 --> 01:09:26,573

cannot do, and how we can?

Speaker: 1288

01:09:26,573 --> 01:09:30,293

make decisions based on that, that

hopefully are going to be better decisions

Speaker: 1289

01:09:30,293 --> 01:09:31,793

than we used to make.

Speaker: 1290

01:09:31,793 --> 01:09:33,583

And also, how do we update our decisions?

Speaker: 1291

01:09:33,583 --> 01:09:37,453

Because, well, the game changes, as you

said so well.

Speaker: 1292

01:09:37,453 --> 01:09:42,273

So yeah, for sure, all that stuff is

absolutely crucial.

Speaker: 1293

01:09:42,273 --> 01:09:45,433

And I like using the metaphor of the

engine and the car, right?

Speaker: 1294

01:09:45,433 --> 01:09:48,573

It's like building the model is the engine

of the car.

Speaker: 1295

01:09:48,573 --> 01:09:52,133

So surely, you want the best engine

possible, but you also need a very cool

Speaker: 1296

01:09:52,133 --> 01:09:55,393

car, because otherwise, nobody's going to

want your engine.

Speaker: 1297

01:09:55,393 --> 01:09:55,950

And so...

Speaker: 1298

01:09:55,950 --> 01:10:00,670

like then building all the communication

around the model, the visualizations,

Speaker: 1299

01:10:00,670 --> 01:10:05,050

things like that, extremely important

because then in the end, as you were

Speaker: 1300

01:10:05,050 --> 01:10:10,130

saying at the beginning of the show, if

the model isn't used, well, that's not a

Speaker: 1301

01:10:10,130 --> 01:10:11,830

very good investment.

Speaker: 1302

01:10:13,950 --> 01:10:14,550

Yeah.

Speaker: 1303

01:10:14,550 --> 01:10:20,030

So I would have literally, I would have a

lot more questions if they are on my list,

Speaker: 1304

01:10:20,030 --> 01:10:24,142

but we are going to call it a show poll

because I don't want to keep you...

Speaker: 1305

01:10:24,142 --> 01:10:28,062

three hours, you've already been very

generous with your time.

Speaker: 1306

01:10:28,142 --> 01:10:32,122

You can come back to the show anytime if

you want to, if you have a cool new

Speaker: 1307

01:10:32,122 --> 01:10:34,302

project you want to talk about for sure.

Speaker: 1308

01:10:34,622 --> 01:10:38,932

Yeah, maybe we can record the French

version of the podcast sometime, you know.

Speaker: 1309

01:10:38,932 --> 01:10:39,842

yeah, yeah.

Speaker: 1310

01:10:39,842 --> 01:10:41,092

I'll definitely be down for that.

Speaker: 1311

01:10:41,092 --> 01:10:44,922

You know, someone who will be very happy

is my mother.

Speaker: 1312

01:10:44,922 --> 01:10:48,222

She's always asking me, so when are you

going to do the French version of your

Speaker: 1313

01:10:48,222 --> 01:10:49,962

courses in your podcasting zone?

Speaker: 1314

01:10:49,962 --> 01:10:52,722

I'm like, that's not going to happen, mom.

Speaker: 1315

01:10:54,078 --> 01:10:58,978

Maybe that's what moms are for though.

Speaker: 1316

01:10:59,138 --> 01:11:00,398

Exactly.

Speaker: 1317

01:11:03,178 --> 01:11:07,438

Before letting you go, Paul, I'm going to

ask you the last two questions.

Speaker: 1318

01:11:07,438 --> 01:11:11,818

I ask every guest at the end of the show

because it's a Beijing show, so what

Speaker: 1319

01:11:11,818 --> 01:11:15,878

counts is not the individual point

estimate, but the distribution of the

Speaker: 1320

01:11:15,878 --> 01:11:17,018

responses.

Speaker: 1321

01:11:17,498 --> 01:11:21,838

First question, if you had unlimited time

and resources, which problem?

Speaker: 1322

01:11:21,838 --> 01:11:23,332

would you try to solve?

Speaker: 1323

01:11:29,838 --> 01:11:30,058

Good.

Speaker: 1324

01:11:30,058 --> 01:11:30,708

That's a good question.

Speaker: 1325

01:11:30,708 --> 01:11:34,478

You sent me this ahead of time and I spent

a couple seconds and I was like, man, I

Speaker: 1326

01:11:34,478 --> 01:11:35,438

don't know.

Speaker: 1327

01:11:35,438 --> 01:11:37,848

But I, it's tough.

Speaker: 1328

01:11:37,848 --> 01:11:39,458

There's so many questions in sports.

Speaker: 1329

01:11:39,458 --> 01:11:40,218

Yeah.

Speaker: 1330

01:11:40,218 --> 01:11:41,038

I know.

Speaker: 1331

01:11:41,038 --> 01:11:48,658

I, I mean, my, one of my passions is

American football and I just keep going

Speaker: 1332

01:11:48,658 --> 01:11:49,358

back.

Speaker: 1333

01:11:49,358 --> 01:11:53,338

So I could tell, I love American football

and I love soccer, international football.

Speaker: 1334

01:11:53,338 --> 01:11:53,738

Right.

Speaker: 1335

01:11:53,738 --> 01:11:57,582

And both of those games, understanding.

Speaker: 1336

01:11:57,582 --> 01:12:02,082

There's certain positions that are just

really hard to understand how valuable

Speaker: 1337

01:12:02,082 --> 01:12:03,162

they are.

Speaker: 1338

01:12:03,162 --> 01:12:05,482

And so in soccer, it's like the midfield.

Speaker: 1339

01:12:05,482 --> 01:12:09,102

It's we know you need a good midfielder,

but how do you measure that?

Speaker: 1340

01:12:09,102 --> 01:12:10,582

That's a really hard problem.

Speaker: 1341

01:12:10,582 --> 01:12:13,522

And in football, there's a lot of

positions in American football.

Speaker: 1342

01:12:13,522 --> 01:12:14,852

There's a lot of positions like that as

well.

Speaker: 1343

01:12:14,852 --> 01:12:16,352

So I probably go somewhere along those.

Speaker: 1344

01:12:16,352 --> 01:12:20,942

Like I want to, I want to discover and

measure the value in these really hard to

Speaker: 1345

01:12:20,942 --> 01:12:25,202

measure, traits and values and these two

sports.

Speaker: 1346

01:12:25,202 --> 01:12:26,902

Yeah.

Speaker: 1347

01:12:27,182 --> 01:12:28,862

Yeah, I definitely understand.

Speaker: 1348

01:12:29,582 --> 01:12:35,901

The battle for the middle is extremely

important always in soccer.

Speaker: 1349

01:12:35,962 --> 01:12:42,102

And if you look at all the teams which win

the Champions League, so the Holy Grail,

Speaker: 1350

01:12:42,102 --> 01:12:48,262

like the Super Bowl of the soccer world,

almost all the time they have an amazing

Speaker: 1351

01:12:48,262 --> 01:12:54,302

and impressive pair or three players as

midfielders.

Speaker: 1352

01:12:54,302 --> 01:12:56,222

And that's like a sine qua non.

Speaker: 1353

01:12:56,222 --> 01:12:57,262

But...

Speaker: 1354

01:12:57,262 --> 01:13:03,062

As you were saying, it's extremely hard to

come up with a metric that's going to not

Speaker: 1355

01:13:03,062 --> 01:13:08,982

only explain why the midfielders are good,

but also help you constantly choose

Speaker: 1356

01:13:08,982 --> 01:13:12,562

midfielders that will increase your

probability of winning the Champions

Speaker: 1357

01:13:12,562 --> 01:13:13,062

League.

Speaker: 1358

01:13:13,062 --> 01:13:17,702

And I'm seeing that as a very frustrated

Paris fan because that's been years since

Speaker: 1359

01:13:17,702 --> 01:13:21,622

Thiago Mota basically retired that we're

looking for a number six.

Speaker: 1360

01:13:21,622 --> 01:13:25,562

So the play, the midfielder just before

the defense and we're still looking for

Speaker: 1361

01:13:25,562 --> 01:13:26,722

him.

Speaker: 1362

01:13:26,842 --> 01:13:27,576

Yeah.

Speaker: 1363

01:13:28,270 --> 01:13:31,970

So please, Paul, let me know when you're

done with that.

Speaker: 1364

01:13:31,990 --> 01:13:32,510

Yeah.

Speaker: 1365

01:13:32,510 --> 01:13:36,980

Well, unfortunately, there's several

really good French midfielders.

Speaker: 1366

01:13:36,980 --> 01:13:38,930

They just don't play for PSG.

Speaker: 1367

01:13:38,930 --> 01:13:40,790

I know.

Speaker: 1368

01:13:40,790 --> 01:13:41,250

I know.

Speaker: 1369

01:13:41,250 --> 01:13:43,330

Not a lot of French players stay in

France.

Speaker: 1370

01:13:43,330 --> 01:13:46,430

That's why I'm telling you, we need a

European wide league.

Speaker: 1371

01:13:46,430 --> 01:13:51,670

Many more players would stay in France and

play for PSG, I guess.

Speaker: 1372

01:13:52,090 --> 01:13:56,750

And second question, if you could have

dinner with any great scientific mind.

Speaker: 1373

01:13:56,750 --> 01:14:00,110

dead, alive or fictional, who would it be?

Speaker: 1374

01:14:00,110 --> 01:14:01,030

Fictional?

Speaker: 1375

01:14:01,030 --> 01:14:03,880

I haven't really thought about fictional

scientific minds.

Speaker: 1376

01:14:03,880 --> 01:14:08,470

That is a good question.

Speaker: 1377

01:14:10,650 --> 01:14:11,716

Geez.

Speaker: 1378

01:14:17,038 --> 01:14:18,298

Man.

Speaker: 1379

01:14:19,238 --> 01:14:23,418

Well, I mean, I thought you were going to

answer very fast.

Speaker: 1380

01:14:23,418 --> 01:14:27,018

Actually, that one, I thought you were

going to answer Bill James like super

Speaker: 1381

01:14:27,018 --> 01:14:27,598

fast.

Speaker: 1382

01:14:27,598 --> 01:14:28,518

Bill James.

Speaker: 1383

01:14:28,518 --> 01:14:28,838

Yeah.

Speaker: 1384

01:14:28,838 --> 01:14:30,538

Well, I've met Bill James.

Speaker: 1385

01:14:30,538 --> 01:14:31,038

So, okay.

Speaker: 1386

01:14:31,038 --> 01:14:34,478

So I have dinner with him, but I have met

him.

Speaker: 1387

01:14:34,478 --> 01:14:40,558

I'll go a little, how liberal are you with

the word scientific mind here?

Speaker: 1388

01:14:40,558 --> 01:14:41,118

Yeah.

Speaker: 1389

01:14:41,118 --> 01:14:46,278

So I think scientific mind, I think

Galileo, I think Newton, I think Einstein,

Speaker: 1390

01:14:46,278 --> 01:14:46,518

right?

Speaker: 1391

01:14:46,518 --> 01:14:47,278

Like,

Speaker: 1392

01:14:47,278 --> 01:14:54,368

You know, those are all, but I'm sure from

the sports world, from the sports world,

Speaker: 1393

01:14:54,368 --> 01:15:00,458

there is a former football player that

very few people have ever heard of and his

Speaker: 1394

01:15:00,458 --> 01:15:01,838

name is Virgil Carter.

Speaker: 1395

01:15:01,838 --> 01:15:07,518

And the reason why I love him, he played

in the seventies is that he wrote a paper

Speaker: 1396

01:15:07,518 --> 01:15:11,878

about expected points in football while he

was playing in the NFL.

Speaker: 1397

01:15:11,958 --> 01:15:15,566

And it was sort of the first sports

analytics.

Speaker: 1398

01:15:15,566 --> 01:15:20,586

ever done in American football and he was

a player in American football at the same

Speaker: 1399

01:15:20,586 --> 01:15:21,326

time.

Speaker: 1400

01:15:21,326 --> 01:15:22,956

So very, not very well known.

Speaker: 1401

01:15:22,956 --> 01:15:23,826

He's still alive.

Speaker: 1402

01:15:23,826 --> 01:15:26,846

I don't know him at all, but he would be a

really cool person.

Speaker: 1403

01:15:26,846 --> 01:15:31,846

If I go like classical, scientific,

scientific minds, I would, I would

Speaker: 1404

01:15:31,846 --> 01:15:38,086

probably, maybe Gauss like, Hey, this

distribution that has your name is like

Speaker: 1405

01:15:38,086 --> 01:15:41,006

used everywhere and it's very useful.

Speaker: 1406

01:15:41,006 --> 01:15:43,246

So I probably, I would stick with him.

Speaker: 1407

01:15:43,466 --> 01:15:45,650

Normal distributions.

Speaker: 1408

01:15:45,742 --> 01:15:48,682

counseling distributions, like the rule of

world nowadays.

Speaker: 1409

01:15:48,682 --> 01:15:52,402

So I'd probably stick with that if I were

to go traditional scientific mind.

Speaker: 1410

01:15:52,402 --> 01:15:52,582

Yeah.

Speaker: 1411

01:15:52,582 --> 01:15:53,242

Yeah.

Speaker: 1412

01:15:53,242 --> 01:15:54,212

Now good choices.

Speaker: 1413

01:15:54,212 --> 01:15:55,322

Good choices.

Speaker: 1414

01:15:55,322 --> 01:15:58,072

I am amazed about that Virgil Carter

story.

Speaker: 1415

01:15:58,072 --> 01:15:59,642

That's so amazing.

Speaker: 1416

01:15:59,642 --> 01:16:00,042

Yeah.

Speaker: 1417

01:16:00,042 --> 01:16:05,682

So if anybody knows Virgil Carter, please

contact us and we'll try to get that

Speaker: 1418

01:16:05,682 --> 01:16:06,642

dinner for Paul.

Speaker: 1419

01:16:06,642 --> 01:16:11,802

If you do that, I'll definitely be here to

grab the dinner and have a conversation

Speaker: 1420

01:16:11,802 --> 01:16:15,062

with Virgil because like having someone

like that on the show would be absolutely

Speaker: 1421

01:16:15,062 --> 01:16:15,694

amazing.

Speaker: 1422

01:16:15,694 --> 01:16:16,774

I love that story.

Speaker: 1423

01:16:16,774 --> 01:16:17,874

That's so amazing.

Speaker: 1424

01:16:17,874 --> 01:16:21,234

It's like, you know, the myth of the

philosopher king.

Speaker: 1425

01:16:21,234 --> 01:16:24,034

Well, here is like the myth of the

scientist player.

Speaker: 1426

01:16:24,034 --> 01:16:26,253

It's just like, I love that.

Speaker: 1427

01:16:26,474 --> 01:16:27,394

Yeah.

Speaker: 1428

01:16:27,394 --> 01:16:28,234

that's fantastic.

Speaker: 1429

01:16:28,234 --> 01:16:28,974

Damn.

Speaker: 1430

01:16:28,974 --> 01:16:30,514

Thanks a lot, Paul.

Speaker: 1431

01:16:30,614 --> 01:16:32,014

Let's call it a show.

Speaker: 1432

01:16:32,014 --> 01:16:33,294

Thanks for having me.

Speaker: 1433

01:16:33,294 --> 01:16:35,694

Yeah, that was amazing.

Speaker: 1434

01:16:35,874 --> 01:16:41,034

As usual, we'll put resources and a link

to your website in the show notes for

Speaker: 1435

01:16:41,034 --> 01:16:42,494

those who want to dig deeper.

Speaker: 1436

01:16:42,494 --> 01:16:45,678

Thanks again, Paul, for taking the time

and being on this show.

Speaker: 1437

01:16:46,126 --> 01:16:48,614

Thanks once again, I really enjoyed it.

Speaker: 1438

01:16:52,558 --> 01:16:56,258

This has been another episode of Learning

Bayesian Statistics.

Speaker: 1439

01:16:56,258 --> 01:17:01,218

Be sure to rate, review, and follow the

show on your favorite podcatcher, and

Speaker: 1440

01:17:01,218 --> 01:17:06,138

visit learnbaystats .com for more

resources about today's topics, as well as

Speaker: 1441

01:17:06,138 --> 01:17:10,858

access to more episodes to help you reach

true Bayesian state of mind.

Speaker: 1442

01:17:10,858 --> 01:17:12,798

That's learnbaystats .com.

Speaker: 1443

01:17:12,798 --> 01:17:17,618

Our theme music is Good Bayesian by Baba

Brinkman, fit MC Lass and Megharam.

Speaker: 1444

01:17:17,618 --> 01:17:20,798

Check out his awesome work at bababrinkman

.com.

Speaker: 1445

01:17:20,798 --> 01:17:21,966

I'm your host.

Speaker: 1446

01:17:21,966 --> 01:17:22,966

Alex and Dora.

Speaker: 1447

01:17:22,966 --> 01:17:27,226

You can follow me on Twitter at Alex

underscore and Dora like the country.

Speaker: 1448

01:17:27,226 --> 01:17:32,286

You can support the show and unlock

exclusive benefits by visiting patreon

Speaker: 1449

01:17:32,286 --> 01:17:34,466

.com slash LearnBasedDance.

Speaker: 1450

01:17:34,466 --> 01:17:36,886

Thank you so much for listening and for

your support.

Speaker: 1451

01:17:36,886 --> 01:17:42,806

You're truly a good Bayesian change your

predictions after taking information and

Speaker: 1452

01:17:42,806 --> 01:17:45,946

if you think and I'll be less than

amazing.

Speaker: 1453

01:17:45,946 --> 01:17:49,262

Let's adjust those expectations.

Speaker: 1454

01:17:49,262 --> 01:17:54,662

Let me show you how to be a good Bayesian

Change calculations after taking fresh

Speaker: 1455

01:17:54,662 --> 01:18:00,702

data in Those predictions that your brain

is making Let's get them on a solid

Speaker: 1456

01:18:00,702 --> 01:18:02,502

foundation