Speaker: 00:00:05

Today's episode takes us into the dynamic

intersection of Bayesian statistics and

2

00:00:09,966 --> 00:00:12,109

sports analytics with

3

00:00:13,326 --> 00:00:19,169

Patrick Ward, the Director of Research and

Analysis for the Seattle Seahawks.

4

00:00:19,169 --> 00:00:24,481

With a rich background that spans from the

Nike Sports Research Lab to teaching

5

00:00:24,481 --> 00:00:28,953

statistics, Patrick brings a wealth of

knowledge to the table.

6

00:00:28,953 --> 00:00:34,626

In our discussion, Patrick delves into how

these methods are revolutionizing the way

7

00:00:34,626 --> 00:00:39,118

we understand player performance and

manage injury risks in professional

8

00:00:39,118 --> 00:00:39,818

sports.

9

00:00:39,818 --> 00:00:43,300

He sheds light on the particular

challenges of translating

10

00:00:43,300 --> 00:00:48,853

complex Beijing concepts for coaches and

team managers who may not be versed in

11

00:00:48,853 --> 00:00:54,846

statistical methods but need to leverage

these insights for strategic decisions.

12

00:00:54,846 --> 00:00:58,967

Patrick also walks us through the

practical aspects of applying Beijing

13

00:00:58,967 --> 00:01:02,099

stats in the high -stakes world of the

NFL.

14

00:01:02,099 --> 00:01:06,071

From selecting the right players to

optimizing training loads, he illustrates

15

00:01:06,071 --> 00:01:11,133

the profound impact that thoughtful

statistical analysis can have on a team's

16

00:01:11,133 --> 00:01:13,114

success and players'

17

00:01:13,302 --> 00:01:17,946

For those of you who appreciate the blend

of science and strategy, this conversation

18

00:01:17,946 --> 00:01:22,609

offers a behind -the -scenes look at the

sophisticated analytics powering team

19

00:01:22,609 --> 00:01:23,689

decisions.

20

00:01:23,710 --> 00:01:29,032

And when he's not dissecting data or

strategizing for the Seahawks, Patrick

21

00:01:29,032 --> 00:01:35,778

enjoys the simple pleasures of reading,

savoring coffee, and playing jazz guitar.

22

00:01:35,854 --> 00:01:43,194

This is Learning Bayesian Statistics,

episode 111, recorded June 19, 2024.

23

00:01:58,934 --> 00:02:04,866

Welcome to Learning Bayesian Statistics, a

podcast about Bayesian inference.

24

00:02:04,866 --> 00:02:08,648

the methods, the projects, and the people

who make it possible.

25

00:02:08,648 --> 00:02:10,769

I'm your host, Alex Andorra.

26

00:02:10,769 --> 00:02:15,200

You can follow me on Twitter at alex

.andorra, like the country.

27

00:02:15,200 --> 00:02:19,432

For any info about the show, learnbasedats

.com is Laplace to be.

28

00:02:19,432 --> 00:02:23,944

Show notes, becoming a corporate sponsor,

unlocking Bayesian Merge, supporting the

29

00:02:23,944 --> 00:02:26,595

show on Patreon, everything is in there.

30

00:02:26,595 --> 00:02:28,526

That's learnbasedats .com.

31

00:02:28,526 --> 00:02:32,988

If you're interested in one -on -one

mentorship, online courses, or statistical

32

00:02:32,988 --> 00:02:33,976

consulting,

33

00:02:33,976 --> 00:02:39,070

Feel free to reach out and book a call at

topmate .io slash alex underscore and

34

00:02:39,070 --> 00:02:39,551

dora.

35

00:02:39,551 --> 00:02:42,964

See you around folks and best patient

wishes to you all.

36

00:02:42,964 --> 00:02:47,818

And if today's discussion sparked ideas

for your business, well, our team at Pimc

37

00:02:47,818 --> 00:02:50,079

Labs can help bring them to life.

38

00:02:50,079 --> 00:02:52,321

Check us out at pimc -labs

39

00:02:56,674 --> 00:03:02,575

Hello, my dear patients, I have some

exciting personal news to share with you.

40

00:03:02,596 --> 00:03:08,157

I am thrilled to announce that I have

recently taken on a new role as a senior

41

00:03:08,157 --> 00:03:11,118

applied scientist with the Miami Marlins.

42

00:03:11,118 --> 00:03:15,699

In this position, I'll be diving even

deeper into the world of sports analytics,

43

00:03:15,699 --> 00:03:20,701

leveraging Bayesian modeling, of course,

to enhance team performance and player

44

00:03:20,701 --> 00:03:21,741

development.

45

00:03:22,087 --> 00:03:28,001

And honestly, this move is so exciting to

me and solidifies my commitment to

46

00:03:28,001 --> 00:03:31,723

advancing the application of Beijing stats

and sports.

47

00:03:31,964 --> 00:03:36,727

if you find yourself in Miami or if you're

curious about the intersection of Beijing

48

00:03:36,727 --> 00:03:42,492

methods and baseball or team sports in

general, don't hesitate to reach out.

49

00:03:42,492 --> 00:03:44,773

OK, back to the show now.

50

00:03:48,193 --> 00:03:51,875

Patrick Ward, welcome to Learning Bayesian

Statistics.

51

00:03:52,736 --> 00:03:53,597

Thanks for having me.

52

00:03:53,597 --> 00:03:56,739

I listen to every episode.

53

00:03:56,739 --> 00:04:02,303

I think every year at the end of the year,

Spotify tells me that it's one of my

54

00:04:02,303 --> 00:04:04,464

highly listened to podcasts.

55

00:04:04,464 --> 00:04:06,926

So it's pleasure to be here.

56

00:04:06,926 --> 00:04:07,466

Hopefully.

57

00:04:07,466 --> 00:04:10,288

I don't know if I can live up to your

prior.

58

00:04:10,348 --> 00:04:14,031

You've had some pretty big timers, but

yeah.

59

00:04:14,031 --> 00:04:14,801

No, yeah.

60

00:04:14,801 --> 00:04:17,833

So first, thanks a lot for being such

61

00:04:17,909 --> 00:04:19,690

faithful listener.

62

00:04:20,191 --> 00:04:22,113

I definitely appreciate that.

63

00:04:22,113 --> 00:04:28,018

And I'm always amazed at the diversity of

people who listen to the show.

64

00:04:28,018 --> 00:04:32,841

That's really awesome.

65

00:04:33,142 --> 00:04:38,066

And also I want to thank Scott Morrison,

put us in contact.

66

00:04:38,847 --> 00:04:41,840

Scott is working at the Miami Marlins.

67

00:04:41,840 --> 00:04:46,133

He's a fellow colleague now.

68

00:04:46,285 --> 00:04:49,956

That's change for me, that's great change.

69

00:04:49,956 --> 00:04:56,048

I'm extremely excited about that new step

in my life.

70

00:04:56,048 --> 00:04:59,959

But today we're not going to talk a lot

about baseball.

71

00:04:59,959 --> 00:05:02,910

We're going to talk a lot about US

football.

72

00:05:02,910 --> 00:05:08,011

So today, European listeners, when you

hear football, we're going to talk about

73

00:05:08,011 --> 00:05:14,413

American football, the one with a ball

that looks like a rugby ball.

74

00:05:15,933 --> 00:05:17,894

And so Patrick, we're going to talk about

that.

75

00:05:17,894 --> 00:05:22,715

But first, as usual, I want to talk a bit

more about you.

76

00:05:22,755 --> 00:05:26,036

Can you tell the listeners what you're

doing nowadays?

77

00:05:26,036 --> 00:05:33,598

So I gave your title, your bio in the

intro, but maybe like tell us a bit more

78

00:05:33,818 --> 00:05:38,379

in the flesh what you're doing and also

how you ended up doing what you're doing.

79

00:05:38,519 --> 00:05:38,969

Yeah.

80

00:05:38,969 --> 00:05:43,391

Well, currently I'm at the Seattle

Seahawks, which is one of the American

81

00:05:43,391 --> 00:05:45,471

football teams in the NFL.

82

00:05:45,829 --> 00:05:50,190

And I'm the director of research and

analysis there.

83

00:05:50,370 --> 00:05:54,972

So we kind of work across all of football

operations.

84

00:05:54,972 --> 00:06:02,494

So everything from player acquisition,

front office type of stuff to team based

85

00:06:02,494 --> 00:06:05,494

analysis and opponent analysis.

86

00:06:05,694 --> 00:06:11,916

And just kind of coordinating a research

strategy around how we attack questions

87

00:06:11,916 --> 00:06:15,437

for the key decision makers or the key

stakeholders across

88

00:06:15,461 --> 00:06:21,885

coaching, acquisition, even into player

health and performance and development and

89

00:06:21,885 --> 00:06:23,226

things like that.

90

00:06:23,587 --> 00:06:29,730

And I got here, this is my 10th year, I

got here from Nike.

91

00:06:29,730 --> 00:06:35,254

So I was at Nike in the sports research

lab actually working for nearly two years

92

00:06:35,254 --> 00:06:36,695

as a researcher.

93

00:06:37,176 --> 00:06:39,557

And the way that I got

94

00:06:39,775 --> 00:06:46,710

was I was doing some projects for Nike

around applied sports research and they

95

00:06:46,710 --> 00:06:50,503

had just at the time, I think they had

just become like the biggest sponsor of

96

00:06:50,503 --> 00:06:54,476

the newly minted National Women's Soccer

League.

97

00:06:54,476 --> 00:06:57,848

And they said, we want to do something

around this.

98

00:06:57,848 --> 00:07:00,200

And so, you we were kind of kicking around

ideas.

99

00:07:00,200 --> 00:07:05,393

And one of the ideas we had was what if we

went out and we tested all of the women in

100

00:07:05,393 --> 00:07:07,865

the league, like tested them

101

00:07:07,979 --> 00:07:10,660

sprinting and jumping and power output and

things like that.

102

00:07:10,660 --> 00:07:15,031

And then we could basically build like

archetypes and that would be useful for,

103

00:07:16,211 --> 00:07:20,733

you know, like apps in your watch and on

your phone and girls could in the field

104

00:07:20,733 --> 00:07:23,513

could compare themselves to their favorite

athletes and stuff.

105

00:07:23,513 --> 00:07:24,724

So they let us do it.

106

00:07:24,724 --> 00:07:30,305

And they sent me on the road for an entire

off season, the entire off season training

107

00:07:30,305 --> 00:07:32,176

of the national women's soccer league.

108

00:07:32,176 --> 00:07:36,427

went around the country to every single

team, myself and four colleagues, and we

109

00:07:36,427 --> 00:07:37,407

tested.

110

00:07:37,417 --> 00:07:38,918

every woman in the league.

111

00:07:38,918 --> 00:07:45,271

And so we had the largest data set on

women's soccer players that anyone could

112

00:07:45,271 --> 00:07:45,571

have.

113

00:07:45,571 --> 00:07:50,024

So we did some conference presentations

and things like that with that data.

114

00:07:50,024 --> 00:07:56,728

And lo and behold, Nike was there and the

Seattle Seahawks called down to Nike and

115

00:07:56,728 --> 00:07:59,849

said, hey, we hear there's this test

battery and we'd love to see what our

116

00:07:59,849 --> 00:08:00,770

players do on it.

117

00:08:00,770 --> 00:08:05,693

And so I went up and I did a project for

them around that.

118

00:08:05,693 --> 00:08:06,513

And

119

00:08:07,085 --> 00:08:10,385

then they kind of just said like, what if

you just did this kind of stuff all the

120

00:08:10,385 --> 00:08:10,725

time?

121

00:08:10,725 --> 00:08:14,425

And so that's how I started out 10 years

ago.

122

00:08:14,425 --> 00:08:18,165

And I basically started out just in

applied physiology, which was my

123

00:08:18,165 --> 00:08:18,585

background.

124

00:08:18,585 --> 00:08:24,445

And I was doing like wearables, wearable

tech for the team, like GPS and

125

00:08:24,445 --> 00:08:26,225

accelerometry and things like that.

126

00:08:26,225 --> 00:08:33,325

And then that kind of progressed into

draft analysis and player evaluation and

127

00:08:33,325 --> 00:08:33,985

things like that.

128

00:08:33,985 --> 00:08:37,069

And it just kind of growing until

129

00:08:37,069 --> 00:08:39,009

Yeah, 10 years later, here we are.

130

00:08:39,009 --> 00:08:43,949

Yeah, that's a great, yeah, it's it's a

great, uh, background.

131

00:08:43,949 --> 00:08:49,389

love it because I mean, definitely it

seems like you've been into sports since

132

00:08:49,389 --> 00:08:58,069

you are, uh, at least a college graduate,

but also there is a, uh, a bit of

133

00:08:58,069 --> 00:09:00,109

randomness in this.

134

00:09:00,129 --> 00:09:01,429

So sorry, love that.

135

00:09:01,429 --> 00:09:06,931

Uh, of course, as a, as a fellow Bayesian,

always, always interested in.

136

00:09:06,931 --> 00:09:11,545

in the random parts of anybody's journey.

137

00:09:11,545 --> 00:09:20,583

Actually, how much of the Bayesian stats

do you have in that journey and also in

138

00:09:20,583 --> 00:09:22,134

your current work?

139

00:09:22,995 --> 00:09:31,683

How Bayesian in a way is your work right

now, but also how were you introduced to

140

00:09:31,683 --> 00:09:33,104

Bayesian stats?

141

00:09:34,094 --> 00:09:38,774

Well, mean, anyone who has watched

American football knows it's a game of

142

00:09:38,774 --> 00:09:40,794

very, very small sample sizes.

143

00:09:40,794 --> 00:09:45,674

So we only play up until two years ago.

144

00:09:45,694 --> 00:09:47,154

We play 17 games now.

145

00:09:47,154 --> 00:09:49,274

We used to only play 16 games.

146

00:09:49,714 --> 00:09:55,054

So unlike most of the other sports,

baseball has got 162, several hundred at

147

00:09:55,054 --> 00:09:59,240

bats, basketball, hockey, 82 games.

148

00:09:59,240 --> 00:10:00,160

many attempts.

149

00:10:00,160 --> 00:10:04,722

Also the players in a lot of these sports

are all doing the same things in baseball.

150

00:10:04,722 --> 00:10:08,334

Aside from the pitchers, everybody's going

to go to the plate and hit in basketball.

151

00:10:08,334 --> 00:10:13,876

Everybody has a chance on the court to

dribble the ball, shoot, score, pass, get

152

00:10:13,876 --> 00:10:15,837

assists, get blocks, et cetera.

153

00:10:15,837 --> 00:10:22,399

Football is really unique because it's a

very tactical game.

154

00:10:23,240 --> 00:10:27,231

There's discrete events in terms of plays,

stop and start.

155

00:10:27,425 --> 00:10:31,258

But because of the tactical nature of it

and one ball, there's only certain

156

00:10:31,258 --> 00:10:32,619

positions that touch the ball.

157

00:10:32,619 --> 00:10:35,590

There's only certain opportunities that

players are going to have.

158

00:10:35,711 --> 00:10:38,133

So that was always an issue.

159

00:10:38,133 --> 00:10:44,116

And when I did my PhD, a big part of my

PhD was using mixed models to look at

160

00:10:45,138 --> 00:10:51,662

physiological differences between players

on the field with GPS and accelerometry.

161

00:10:51,662 --> 00:10:54,854

And I always thought of mixed models, even

though I didn't know it at the time,

162

00:10:54,854 --> 00:10:55,979

because I hadn't really...

163

00:10:55,979 --> 00:10:57,630

learned anything Bayesian yet.

164

00:10:57,630 --> 00:10:58,860

I always thought of mixed models.

165

00:10:58,860 --> 00:11:03,532

I think of them as like this bridge to

Bayesian analysis because you have these

166

00:11:03,532 --> 00:11:09,053

fixed effects which behave like our

population averages, our population base

167

00:11:09,053 --> 00:11:10,234

rates, I guess you could say.

168

00:11:10,234 --> 00:11:15,486

And the random effects are sort of like,

hey, we know something about you or your

169

00:11:15,486 --> 00:11:19,957

group and therefore we know how you

deviate from the population.

170

00:11:19,957 --> 00:11:23,118

And then with those two bits of

information, we're also like, hey, here's

171

00:11:23,118 --> 00:11:27,489

someone new in the population, or maybe

someone that we've only seen or observed

172

00:11:27,489 --> 00:11:29,379

do the thing one time.

173

00:11:29,900 --> 00:11:34,921

best guess therefore is the fixed effects

portion of this until proven otherwise.

174

00:11:35,221 --> 00:11:40,683

So I always had that in the back of my

head going through this, but you know, my

175

00:11:40,683 --> 00:11:44,224

first two or three years in the NFL, we

always just used to kind of throw our

176

00:11:44,224 --> 00:11:49,065

hands up when we see small samples, we'd

be like, yeah, it's this, it's

177

00:11:49,645 --> 00:11:52,605

50%, but it's such a small sample, we

can't really know.

178

00:11:52,605 --> 00:11:57,625

And we didn't really have a good way of

like sorting out what to do with that

179

00:11:57,625 --> 00:11:58,285

information.

180

00:11:58,285 --> 00:12:03,045

Because as you know, know, something like

one out of 10 and 10 out of a hundred and

181

00:12:03,045 --> 00:12:06,885

you know, a hundred out of a thousand,

those are the same proportions, but

182

00:12:06,885 --> 00:12:11,345

different levels of information are

contained within those proportions.

183

00:12:11,665 --> 00:12:16,269

And I stumbled upon a paper, it was like a

19, I think it was like 19.

184

00:12:16,269 --> 00:12:21,069

77 or something by Efron and Morris and it

was called the Stein's paradox and I

185

00:12:21,069 --> 00:12:24,869

probably stumbled on it because I was like

You know, there's so much in Saber Metrics

186

00:12:24,869 --> 00:12:28,949

someone in baseball has probably figured

this out before and so I I was probably

187

00:12:28,949 --> 00:12:34,049

googling something like Small samples

baseball statistics Saber Metrics blah

188

00:12:34,049 --> 00:12:40,749

blah blah and I stumbled upon this paper

about Stein's paradox and The crux of the

189

00:12:40,749 --> 00:12:44,237

paper was if we observe these I think it

was

190

00:12:44,237 --> 00:12:48,817

12 or 18 baseball players through the

first half of the season up to the All

191

00:12:48,817 --> 00:12:49,757

-Star break.

192

00:12:49,757 --> 00:12:53,477

And we see the number of times they went

to the plate and what they're, you know,

193

00:12:53,477 --> 00:12:56,237

and the number of times they hit, we have

a batting average.

194

00:12:56,377 --> 00:13:02,617

If we take the observed batting average

through the first half of the season, how

195

00:13:02,617 --> 00:13:07,457

well does that predict the batting average

at the end of the season?

196

00:13:07,457 --> 00:13:09,957

Meaning now they've gone through the

second half.

197

00:13:10,397 --> 00:13:11,337

And

198

00:13:11,807 --> 00:13:14,959

You look at that and you're like, okay,

let's, you know, what's this all about?

199

00:13:14,959 --> 00:13:18,381

And so the first thing they do is they set

up this argument that like, well, that

200

00:13:18,381 --> 00:13:22,723

doesn't do a very good job because some of

these players batted, you know, five times

201

00:13:22,723 --> 00:13:27,156

or three times, certainly a player who

went three for three has a hundred percent

202

00:13:27,156 --> 00:13:27,835

batting average.

203

00:13:27,835 --> 00:13:31,369

We don't think this is the greatest

baseball player of all time yet, because

204

00:13:31,369 --> 00:13:33,910

we've only seen them do this thing three

times.

205

00:13:34,270 --> 00:13:39,213

So, the basic naive prediction of using

the half

206

00:13:39,213 --> 00:13:43,033

first half of the season to predict the

second half wasn't very good.

207

00:13:43,533 --> 00:13:50,073

And so in that paper, they introduced this

kind of simple Bayesian model of saying,

208

00:13:50,073 --> 00:13:52,853

well, we know something about average

baseball players.

209

00:13:52,853 --> 00:13:55,533

What if we weighted everybody to that?

210

00:13:55,953 --> 00:14:00,953

And lo and behold, that did a bit better

of a job constraining the small sampled

211

00:14:00,953 --> 00:14:04,693

players who had these, you know, a guy

that goes 0 for 10, which is totally

212

00:14:04,693 --> 00:14:07,873

possible in baseball when you have

hundreds of it bats.

213

00:14:08,365 --> 00:14:11,385

We don't think that's the worst hitter in

baseball.

214

00:14:11,405 --> 00:14:16,225

so, you know, constraining those players

told them something about what they

215

00:14:16,225 --> 00:14:18,685

expected to then see at the end of the

season.

216

00:14:18,685 --> 00:14:24,925

And so through that paper, then I found

this blog by David Robinson, who's an R

217

00:14:24,925 --> 00:14:25,545

programmer.

218

00:14:25,545 --> 00:14:30,405

And it was all about like using empirical

Bayesian analysis for baseball.

219

00:14:30,405 --> 00:14:33,645

And then he made it into a nice little

book that you could buy on Amazon for

220

00:14:33,645 --> 00:14:35,785

like, I don't know, $20 or something.

221

00:14:36,629 --> 00:14:39,470

You know, and I read those two things and

I was like, this is incredible.

222

00:14:39,470 --> 00:14:42,591

This is exactly what I've always wanted to

know.

223

00:14:42,591 --> 00:14:46,512

And so like I went in the next day to our

other analysts at the time, there was only

224

00:14:46,512 --> 00:14:46,892

two of us.

225

00:14:46,892 --> 00:14:51,013

And I said, I think I figured out a way we

could solve small sample problems.

226

00:14:51,013 --> 00:14:52,453

And, and that was it.

227

00:14:52,453 --> 00:14:59,055

Like then after that, you really couldn't

convince me otherwise that this wasn't a

228

00:14:59,055 --> 00:15:00,095

great way of thinking.

229

00:15:00,095 --> 00:15:02,736

That doesn't mean that everything we do

has to be Bayesian.

230

00:15:02,736 --> 00:15:06,317

Certainly like there's other things that

we do that are used.

231

00:15:06,317 --> 00:15:10,497

you know, different tools like machine

learning models and neural networks and

232

00:15:10,497 --> 00:15:11,597

things like that.

233

00:15:11,597 --> 00:15:17,057

But certainly when we start thinking about

like decision -making, how do I

234

00:15:17,057 --> 00:15:21,037

incorporate priors, domain expertise?

235

00:15:21,037 --> 00:15:22,737

How do I fit the right prior?

236

00:15:22,737 --> 00:15:28,717

You know, like if you went 0 for 5 and

you're first at bats, let's say in

237

00:15:28,717 --> 00:15:34,357

baseball, but you were a college standout

and you were an amazing

238

00:15:34,829 --> 00:15:40,149

player in the AAA, I probably have a

stronger prior that you're maybe a

239

00:15:40,149 --> 00:15:44,589

slightly better than average baseball

player than if you went 0 for 5 and you

240

00:15:44,589 --> 00:15:48,049

were a horrific college player and you

weren't very good in AAA and you were

241

00:15:48,049 --> 00:15:51,149

really the last person on the bench that

we needed to call.

242

00:15:51,208 --> 00:15:53,329

And maybe that prior is much lower.

243

00:15:53,429 --> 00:16:00,989

so utilizing that information in order to

help us make decisions going forward,

244

00:16:01,569 --> 00:16:02,669

that's really

245

00:16:02,861 --> 00:16:04,481

That was kind of the money for me.

246

00:16:04,481 --> 00:16:06,001

And so how much do we use it?

247

00:16:06,001 --> 00:16:10,461

mean, if we have a new analyst start, one,

you know, one of our new analyst starts,

248

00:16:10,461 --> 00:16:11,981

started two years ago.

249

00:16:11,981 --> 00:16:15,681

I think the first thing was like, how much

do you know about BASE?

250

00:16:15,681 --> 00:16:18,841

And it was like, well, I never really

learned that in school and blah, blah.

251

00:16:18,841 --> 00:16:21,681

And it was like, okay, here's two books.

252

00:16:21,681 --> 00:16:23,161

Here's a 12 week curriculum.

253

00:16:23,161 --> 00:16:27,401

We're going to meet every week and you're

going to do projects and homework and

254

00:16:27,401 --> 00:16:28,021

reading.

255

00:16:28,021 --> 00:16:28,841

And that was it.

256

00:16:28,841 --> 00:16:31,161

Like, it was like, you have to learn this

because this is how we're going to think.

257

00:16:31,161 --> 00:16:32,661

And this is how we're going to,

258

00:16:33,709 --> 00:16:36,951

process information and communicate

information.

259

00:16:38,733 --> 00:16:40,733

Well, what about that?

260

00:16:41,334 --> 00:16:48,990

I told the listeners that we were not

going to talk a lot about baseball, but in

261

00:16:48,990 --> 00:16:50,200

the end we are.

262

00:16:50,741 --> 00:16:52,952

It all comes back to baseball, think.

263

00:16:52,952 --> 00:16:58,346

Yeah, in sports analytics, all comes back

to baseball, Certainly, yeah.

264

00:16:58,446 --> 00:17:01,388

Yeah, okay.

265

00:17:02,879 --> 00:17:12,412

If I understand correctly, was motivated a

lot by low sample sizes and being able to

266

00:17:12,412 --> 00:17:16,393

handle all of that in your models.

267

00:17:16,393 --> 00:17:18,793

That makes a ton of sense.

268

00:17:18,793 --> 00:17:26,486

As a lot of people, I've seen a lot of

clients definitely motivated by a very

269

00:17:26,486 --> 00:17:30,317

practical problem that you were having.

270

00:17:31,721 --> 00:17:36,985

I mean, most of people enter the Beijing

field through that.

271

00:17:38,047 --> 00:17:44,572

Something that I'm actually very curious

about, because like I could keep talking

272

00:17:44,572 --> 00:17:51,798

about that for hours, but I really want to

dive into what you're doing at the

273

00:17:51,919 --> 00:17:59,115

Seahawks and also, you know, like how

Beijing stats is helpful.

274

00:17:59,223 --> 00:18:00,344

to what you guys are doing.

275

00:18:00,344 --> 00:18:05,787

I think it's the most interesting for the

listeners who understand basically how

276

00:18:05,787 --> 00:18:11,341

themselves could they apply patient stats

to their own problems, which are not

277

00:18:11,341 --> 00:18:18,116

necessarily in sports, but I think sports

is a really good field to think about that

278

00:18:18,116 --> 00:18:22,889

because you have a lot of diversity and

you have also a lot of somewhat controlled

279

00:18:22,889 --> 00:18:23,750

experiments.

280

00:18:23,750 --> 00:18:28,293

You have a lot of constraints and that's

always extremely interesting to talk about

281

00:18:28,293 --> 00:18:29,033

that.

282

00:18:29,105 --> 00:18:36,607

Maybe you can start by basically

explaining how patient stats are applied

283

00:18:36,607 --> 00:18:43,689

in your current role for analyzing player

performance and injury risk.

284

00:18:43,689 --> 00:18:51,272

Because now that I work directly in

sports, something I'm starting to

285

00:18:51,272 --> 00:18:55,393

understand is that really player

286

00:18:55,959 --> 00:19:01,290

projecting player performance and also

being able to handle injury risk are two

287

00:19:01,290 --> 00:19:03,451

extremely important topics.

288

00:19:03,451 --> 00:19:05,452

So maybe let's start with that.

289

00:19:05,452 --> 00:19:08,813

What can you tell us about that, Okay.

290

00:19:08,813 --> 00:19:10,543

Let's see.

291

00:19:10,543 --> 00:19:11,813

Which one should I start with?

292

00:19:11,813 --> 00:19:16,555

I guess I'll start with injury risk, I

suppose.

293

00:19:16,555 --> 00:19:17,365

Injury is like...

294

00:19:17,365 --> 00:19:23,556

I mean, this is like a super difficult

problem to solve.

295

00:19:25,291 --> 00:19:27,743

You know, I've written a number of papers

on those.

296

00:19:27,743 --> 00:19:30,584

think you can link to my research gate.

297

00:19:30,605 --> 00:19:33,997

And there's a number of methodology papers

that we've written that have looked at

298

00:19:33,997 --> 00:19:35,247

things like this.

299

00:19:35,768 --> 00:19:40,912

And I think it's complicated because one,

there's like a ton of inter -individual

300

00:19:40,912 --> 00:19:43,764

differences as far as why people get hurt.

301

00:19:43,764 --> 00:19:49,448

There's a ton of things that we probably,

you know, don't know they're important yet

302

00:19:49,448 --> 00:19:54,011

because we can't measure them or we at

least can't measure them in the real world

303

00:19:54,011 --> 00:19:54,665

applied.

304

00:19:54,665 --> 00:19:56,946

setting, maybe in a lab you can.

305

00:19:57,006 --> 00:20:00,728

And then there's other things that we just

don't know because we're like, it's a

306

00:20:01,308 --> 00:20:02,118

epistemic problem.

307

00:20:02,118 --> 00:20:03,279

Like we're just stupid about it.

308

00:20:03,279 --> 00:20:06,831

We're naive that there's other things out

there that maybe we're just unaware of

309

00:20:06,831 --> 00:20:07,611

yet.

310

00:20:07,711 --> 00:20:11,293

And so it's a really hard problem to try

and solve.

311

00:20:11,333 --> 00:20:17,446

So when I see papers that basically come

out and say like an injury prediction

312

00:20:17,446 --> 00:20:20,417

model and they're estimating

313

00:20:20,599 --> 00:20:24,641

prediction as like a one or a zero, like a

yes or a no, like a binary response, and

314

00:20:24,641 --> 00:20:28,442

they give a nice little two by two table

and they talk about how well their model

315

00:20:28,442 --> 00:20:29,353

did.

316

00:20:29,873 --> 00:20:35,185

I'm always like, I don't, how is that

useful to the people who actually have to

317

00:20:35,185 --> 00:20:35,936

do the work?

318

00:20:35,936 --> 00:20:42,589

Because in reality, what we're dealing

with is it's probably not unlike a hedge

319

00:20:42,589 --> 00:20:46,280

fund manager managing the risk of their

portfolio.

320

00:20:46,280 --> 00:20:48,321

And if you think of each player,

321

00:20:48,331 --> 00:20:55,086

or each athlete that you deal with as a

portfolio, they each have some level of

322

00:20:55,086 --> 00:20:55,847

base risk.

323

00:20:55,847 --> 00:21:01,211

So if we know nothing about you, you

really have to have a pretty good handle

324

00:21:01,211 --> 00:21:06,956

in your sport of what's the base rates of

risk of injury for position groups and

325

00:21:06,956 --> 00:21:08,947

players of different age and things like

that.

326

00:21:08,947 --> 00:21:12,099

So that might be an initial model, right?

327

00:21:12,640 --> 00:21:14,361

And then from there...

328

00:21:14,945 --> 00:21:18,137

The players go out and they do things and

they play and they perform and they

329

00:21:18,137 --> 00:21:22,769

compete and they get dinged up and they

take hits and they get, you know, hit by

330

00:21:22,769 --> 00:21:26,850

hit by pitches or they get tackled really

hard or things like that.

331

00:21:26,951 --> 00:21:32,973

And we collect that information and we're

basically just shifting the probabilities

332

00:21:32,973 --> 00:21:36,195

up and down based on what we observe over

time.

333

00:21:36,855 --> 00:21:40,729

And when that probability reaches a

certain threshold.

334

00:21:40,729 --> 00:21:44,701

And of course you could use a posterior

distribution.

335

00:21:44,701 --> 00:21:49,504

So you have an integral of like how much

of the probability distribution is above

336

00:21:49,504 --> 00:21:51,315

or below a certain threshold.

337

00:21:51,595 --> 00:21:56,748

Then you have the opportunity to have a

discussion about when to act or what to

338

00:21:56,748 --> 00:21:57,419

do.

339

00:21:57,419 --> 00:22:06,144

And how you act and when to act is going

to be dependent on your tolerance for risk

340

00:22:06,144 --> 00:22:08,104

or your coach's tolerance for risk.

341

00:22:08,937 --> 00:22:15,032

If it's your best player, if it's the MVP

of your team and it's week two of the

342

00:22:15,032 --> 00:22:21,627

season and the risk probability, or let's

say we're using this as a model.

343

00:22:21,928 --> 00:22:25,239

Some of the stuff that you mentioned Scott

earlier that we've worked on is like

344

00:22:25,239 --> 00:22:30,515

return to play type of models where it's

like, okay, the athlete has, you know, saw

345

00:22:30,515 --> 00:22:33,497

an ankle sprain and we're there rehabbing

346

00:22:34,477 --> 00:22:39,537

And we have a, you know, we have a test or

several tests, a test battery that tells

347

00:22:39,537 --> 00:22:43,617

us where that athlete is on their return

to play timeline.

348

00:22:44,657 --> 00:22:50,997

Um, let's say it's week two of the season

and we say, well, there's a, you know, the

349

00:22:50,997 --> 00:22:54,277

probability distribution, the posterior

distribution looks like this.

350

00:22:54,277 --> 00:22:58,497

Here's the threshold that we'd feel

comfortable releasing this athlete back to

351

00:22:58,497 --> 00:23:00,117

full on competition.

352

00:23:00,117 --> 00:23:03,657

And there's a 30 % chance

353

00:23:04,237 --> 00:23:08,796

they're in good shape and there's a 70 %

chance that they're below that threshold.

354

00:23:09,777 --> 00:23:14,237

In week two of the season, we probably

want to say, you know what?

355

00:23:14,557 --> 00:23:16,917

Let's not take that risk this week.

356

00:23:17,737 --> 00:23:23,337

Let's be a little bit more risk averse

here because it is the best player.

357

00:23:23,697 --> 00:23:29,077

And let's wait till we have more

distribution on the right side of the

358

00:23:29,077 --> 00:23:29,997

threshold.

359

00:23:30,697 --> 00:23:32,437

Alternatively, if it's

360

00:23:32,663 --> 00:23:37,346

final game of the season, it's the Super

Bowl or the World Series or the Champions

361

00:23:37,346 --> 00:23:42,949

League final or something like that,

you're going to probably take that risk

362

00:23:42,949 --> 00:23:44,900

because you need the best player out

there.

363

00:23:44,900 --> 00:23:54,115

And so when I think about injury risk

modeling, what I really think about is how

364

00:23:54,115 --> 00:24:00,589

do we evaluate this individual's current

status?

365

00:24:00,915 --> 00:24:04,006

on our sort of risk score or our risk

distribution.

366

00:24:04,006 --> 00:24:09,708

And when do we feel like we need to

intervene and do something?

367

00:24:09,708 --> 00:24:15,669

And when are we going to feel like, this

is fine and continue training as is.

368

00:24:15,669 --> 00:24:18,240

And I think that's the tricky part.

369

00:24:18,240 --> 00:24:21,321

I don't think it's not easy.

370

00:24:21,321 --> 00:24:24,332

I don't think I've solved anything.

371

00:24:25,092 --> 00:24:27,313

I don't think anyone has, but...

372

00:24:28,053 --> 00:24:33,227

Certainly from the perspective of our

staff, we can all sit down with a

373

00:24:33,227 --> 00:24:37,780

performance staff of strength coaches and

dieticians and strength coaches and

374

00:24:37,780 --> 00:24:40,882

medical people and sit down and have these

conversations.

375

00:24:40,882 --> 00:24:47,086

And what makes it nice about using a

Bayesian approach is that we can also take

376

00:24:47,086 --> 00:24:51,950

into account domain expertise that we

might not have in the data.

377

00:24:52,110 --> 00:24:56,683

So if we sit down on a Monday meeting and

then we say, you know, this player,

378

00:24:56,737 --> 00:25:01,150

This is where they're currently at and

this is their risk status, which I don't

379

00:25:01,150 --> 00:25:02,701

know, I don't really feel comfortable with

that.

380

00:25:02,701 --> 00:25:03,902

How do you feel about it?

381

00:25:03,902 --> 00:25:07,815

And then one of the medical people say,

you know, he's been complaining that his

382

00:25:07,815 --> 00:25:11,127

hamstring feels really tight and he's been

getting treatment every morning.

383

00:25:11,127 --> 00:25:14,970

Well, that's not data that we would be

collecting, but that's valuable domain

384

00:25:14,970 --> 00:25:19,873

information that this individual who's

working with the player now adds to this.

385

00:25:20,234 --> 00:25:23,436

And it's just like anything in

probability.

386

00:25:23,436 --> 00:25:24,617

It's like if we

387

00:25:25,045 --> 00:25:31,648

two or three or four independent sources,

all kind of converging on the same

388

00:25:31,848 --> 00:25:36,640

outcome, on the same end point, we

probably need to feel really good about

389

00:25:36,640 --> 00:25:41,101

making that decision and saying like, hey,

let's do something about this, let's act

390

00:25:41,101 --> 00:25:42,052

now, right?

391

00:25:42,052 --> 00:25:46,984

So that's kind of how we, you know, that's

how I think about it in that, you know,

392

00:25:46,984 --> 00:25:48,555

from that side of things.

393

00:25:49,135 --> 00:25:52,576

From the performance side of things, the

development side of things,

394

00:25:53,517 --> 00:25:57,297

It's probably going to be, I mean, it'd be

way different for you guys in baseball

395

00:25:57,297 --> 00:26:00,217

because you draft a player.

396

00:26:00,217 --> 00:26:05,477

You don't expect them to maybe get to the

major leagues and contribute till 23, 24,

397

00:26:05,477 --> 00:26:07,037

25 years old.

398

00:26:07,037 --> 00:26:12,157

You know, for us, you draft a player and

those are going to be the, you know, next

399

00:26:12,157 --> 00:26:15,597

year they're playing, they're ready, you

know, they're in, they're in the mix.

400

00:26:16,589 --> 00:26:22,049

So in that regard, you'd be thinking of

models that would probably be, in my head,

401

00:26:22,049 --> 00:26:28,609

I would be thinking of it as like models

that are mapping the growth potential of

402

00:26:28,609 --> 00:26:29,309

an individual.

403

00:26:29,309 --> 00:26:33,429

How are they progressing through the minor

leagues, which attributes matter?

404

00:26:33,429 --> 00:26:38,209

And then maybe from there answering

questions like what's the probability that

405

00:26:38,209 --> 00:26:46,089

this player makes 20 starts in the major

leagues or starts for three seasons

406

00:26:46,123 --> 00:26:49,874

whatever end point makes sense to the

decision makers, obviously.

407

00:26:49,994 --> 00:26:52,995

You know, for us, it's more about like

player identification.

408

00:26:52,995 --> 00:26:57,016

And again, football is a, is a sport of

small samples.

409

00:26:57,016 --> 00:27:03,918

And so in their college years, some of

these kids might really only be a starter

410

00:27:03,918 --> 00:27:09,129

or a full -time player in their junior and

senior year, or maybe just their senior

411

00:27:09,129 --> 00:27:10,519

year of college.

412

00:27:10,580 --> 00:27:15,917

Additionally, you know, unlike, unlike the

NFL where

413

00:27:15,917 --> 00:27:20,597

you know, at that highest level, the

talent is much more homogenous.

414

00:27:20,597 --> 00:27:25,977

You get to the college football ranks and

you have just this diversity of talent

415

00:27:25,977 --> 00:27:30,717

where you might have a big time team

playing a really lower level opponent.

416

00:27:30,917 --> 00:27:36,097

And so, you you have to adjust things,

being able to hand off the ball to your

417

00:27:36,097 --> 00:27:39,237

running back who's playing against a very

low level opponent.

418

00:27:39,237 --> 00:27:44,937

And he goes for 500 yards or something

absurd, 200, 300 yards in a game.

419

00:27:46,189 --> 00:27:50,929

that has to be adjusted and weighted in

some way because it's not the same as

420

00:27:50,929 --> 00:27:54,149

going two or 300 yards against a big time

opponent.

421

00:27:54,569 --> 00:27:58,649

And the big time opponents are more

similar to the NFL players that they're

422

00:27:58,649 --> 00:27:59,709

going to play against.

423

00:27:59,709 --> 00:28:05,509

And so, you know, all of these types of

things fit into models and hierarchical

424

00:28:05,509 --> 00:28:11,209

models and Bayesian models, which help us

utilize prior information.

425

00:28:11,209 --> 00:28:15,049

And the other way that the Bayesian models

are useful here

426

00:28:15,649 --> 00:28:21,431

You know, sometimes we're dealing with

information that's incomplete because we

427

00:28:21,431 --> 00:28:23,772

can't observe all of the cases.

428

00:28:23,793 --> 00:28:29,815

You know, for example, in college sport,

division one is the top division.

429

00:28:30,155 --> 00:28:35,837

You know, and then you have FBS and then

they division two and division three.

430

00:28:36,178 --> 00:28:42,060

So if you pull all the division two kids

that have ever made it as a pro athlete,

431

00:28:42,060 --> 00:28:44,011

the list is very small.

432

00:28:44,311 --> 00:28:45,822

but they're kids that made it.

433

00:28:45,822 --> 00:28:50,995

And so if you were to just build a normal

model on this, it would say like, well,

434

00:28:50,995 --> 00:28:55,617

the best players clearly come from these

lower level schools because all of the

435

00:28:55,617 --> 00:28:58,749

ones that we have seen have made it, have

been successful.

436

00:28:59,009 --> 00:29:03,092

And in theory, there's hundreds of

thousands of kids from that level that

437

00:29:03,092 --> 00:29:03,942

have never made it.

438

00:29:03,942 --> 00:29:07,394

So we have to adjust that model in some

way.

439

00:29:07,394 --> 00:29:09,585

We have to weight that prior back down.

440

00:29:09,585 --> 00:29:13,087

Yeah, this guy is really, really good at

that level.

441

00:29:13,783 --> 00:29:18,914

but our prior belief on him making it is

very, very low.

442

00:29:18,914 --> 00:29:22,936

And you mean he'd have to be so

exceptional in order to, and this is where

443

00:29:22,936 --> 00:29:29,797

like, oftentimes people rail on like, use

weekly informative priors, let the data

444

00:29:29,797 --> 00:29:31,008

speak a little bit.

445

00:29:31,008 --> 00:29:34,869

But there are times where in these

situations where I feel like you could

446

00:29:34,869 --> 00:29:38,890

probably put a slightly stronger prior on

this and be like, man, this guy's really

447

00:29:38,890 --> 00:29:43,271

gonna have to do something outstanding to

get outside.

448

00:29:43,271 --> 00:29:47,503

of the distribution that we believe is on

this just given what we know.

449

00:29:47,503 --> 00:29:48,934

Okay, yeah, that's very interesting.

450

00:29:48,934 --> 00:29:50,224

That's a very good point.

451

00:29:50,224 --> 00:29:56,107

Since I, yeah, related to survivor bias in

a way.

452

00:29:56,107 --> 00:30:02,189

How concretely, how do you, how do you

handle these kind of cases?

453

00:30:02,189 --> 00:30:10,433

Is it a matter of using a different prior

for these type of players or something

454

00:30:12,797 --> 00:30:14,838

Try to do this in a few different ways.

455

00:30:14,838 --> 00:30:24,750

One is you try and make basically like

equivalency metrics, like saying if you

456

00:30:24,750 --> 00:30:32,923

did X at this low level, it in some way

relates to Y at this other level.

457

00:30:32,923 --> 00:30:39,005

So you try and normalize players based on

players that you've seen that have moved,

458

00:30:39,005 --> 00:30:40,785

say, between levels

459

00:30:41,037 --> 00:30:41,977

of the game.

460

00:30:41,977 --> 00:30:47,317

so like, again, if you think about it from

a baseball perspective, you know, hitting

461

00:30:47,317 --> 00:30:57,737

40 home runs in AA baseball might be

related to, you know, might be in some way

462

00:30:57,737 --> 00:31:05,597

convert to like 33 home runs in AAA and 24

home runs in the MLB or 12 home runs in

463

00:31:05,597 --> 00:31:07,257

the MLB or whatever it might be.

464

00:31:07,257 --> 00:31:07,497

Right.

465

00:31:07,497 --> 00:31:09,157

So trying to,

466

00:31:09,333 --> 00:31:15,858

identify equivalencies between those that

we can then like constrain everybody.

467

00:31:16,079 --> 00:31:20,863

Other ways is just like, like you said,

like putting a prior on it.

468

00:31:20,863 --> 00:31:26,708

knowing the level that the person is

playing at, you would have like a lower

469

00:31:26,708 --> 00:31:28,569

level of prior.

470

00:31:28,569 --> 00:31:30,811

For example, it's just like playtime.

471

00:31:31,452 --> 00:31:37,816

If I think about playtime and performance

as sort of this,

472

00:31:39,137 --> 00:31:44,241

this kind of like rising curve that goes

to an asymptote of some upper level of

473

00:31:44,241 --> 00:31:45,442

performance.

474

00:31:46,463 --> 00:31:52,338

The players way at the left who have very

small number of observations, it would be

475

00:31:52,338 --> 00:31:57,132

silly to say that my prior for those

players is the league average.

476

00:31:57,193 --> 00:31:59,144

There's a reason why they're not playing

very much.

477

00:31:59,144 --> 00:32:03,198

It's probably because people don't think

they're very good, right?

478

00:32:03,198 --> 00:32:07,063

So somewhere in that curve,

479

00:32:07,063 --> 00:32:11,506

for each of those numbers of observations

across whatever performance metric we're

480

00:32:11,506 --> 00:32:17,169

looking at, there's going to be a specific

prior on that continuous distribution.

481

00:32:17,169 --> 00:32:21,602

And that's where I would, you know, that's

where we would kind of draw a stake in the

482

00:32:21,602 --> 00:32:28,085

ground and say like, we probably think

based on what we know that this player is

483

00:32:28,386 --> 00:32:32,468

closer to these players than he is to

those players.

484

00:32:36,265 --> 00:32:37,726

Okay, yeah, yeah, I see.

485

00:32:37,726 --> 00:32:39,336

Yeah, definitely makes sense.

486

00:32:39,336 --> 00:32:45,289

And yeah, yeah, like that point of play

time already tells you something.

487

00:32:45,890 --> 00:32:53,573

Because if the player plays less, then

very probably already you know you have

488

00:32:53,573 --> 00:32:55,314

information about his level.

489

00:32:55,314 --> 00:33:02,377

And that means he's at least not as good

as the A level players that play much more

490

00:33:03,041 --> 00:33:07,643

The only time you get in trouble with that

is like an endowment effect where if you,

491

00:33:07,643 --> 00:33:11,185

you know, like in major league baseball,

there's been some research on players who

492

00:33:11,185 --> 00:33:18,738

are drafted very high in the first round,

second round get progressed up and through

493

00:33:18,738 --> 00:33:23,800

the minor leagues faster than players who

were drafted lower, even if they don't

494

00:33:23,800 --> 00:33:28,722

outperform those players just because

they're high as a consequence of being a

495

00:33:28,722 --> 00:33:29,992

high draft pick.

496

00:33:32,523 --> 00:33:35,874

That one's a tricky one, but there has to

be, at some point it's like actually, and

497

00:33:35,874 --> 00:33:40,336

this is where like, know, posterior

distributions, you can really, I mean,

498

00:33:40,336 --> 00:33:42,007

it's almost like doing an AB test.

499

00:33:42,007 --> 00:33:45,508

Like we've got two players and what's the

probability that this guy is actually

500

00:33:45,508 --> 00:33:50,340

outperforming the other guy, even though

the other guy might've been, you know, a

501

00:33:50,340 --> 00:33:52,171

higher draft pick or something like that.

502

00:33:52,171 --> 00:33:58,414

And so you try and at least display, you

know, we try and at least display that

503

00:33:58,414 --> 00:34:00,375

visually and have those conversations.

504

00:34:00,375 --> 00:34:01,125

It's,

505

00:34:02,039 --> 00:34:06,673

kind of in my head, at least maybe I'm

wrong, but a nice way of like helping

506

00:34:06,673 --> 00:34:10,897

people understand the uncertainty, you

know, which is really important.

507

00:34:10,897 --> 00:34:15,230

always, maybe it's try, you know, I used

to work with a guy who whenever I would

508

00:34:15,230 --> 00:34:18,013

present some of the stuff at work and he'd

be like, stop doing that.

509

00:34:18,013 --> 00:34:22,717

Like every, every time you present, you

talk about like what the uncertainty and

510

00:34:22,717 --> 00:34:25,759

the assumptions and the limitations are,

like just give them the answers.

511

00:34:25,759 --> 00:34:30,093

And I'm like, well, it's important that

they know what the limitations

512

00:34:30,239 --> 00:34:33,321

and what assumptions are behind this

because we can't, we don't want to talk

513

00:34:33,321 --> 00:34:38,434

past the sale and sell them on something

that, you know, isn't really there.

514

00:34:38,434 --> 00:34:41,916

Like there's been times where I've had to

stop someone and just be like, hold on.

515

00:34:41,916 --> 00:34:44,757

This analysis definitely can't tell us

that.

516

00:34:44,757 --> 00:34:47,699

Like what you're saying right now, it

can't tell us that.

517

00:34:47,699 --> 00:34:51,621

like, let's not, let's not try and make

this more than it is.

518

00:34:52,102 --> 00:34:55,334

And also just, you know, conveying your

uncertainty.

519

00:34:55,334 --> 00:34:58,205

mean, that's just super important because

520

00:34:59,221 --> 00:35:01,362

It's really, really hard.

521

00:35:01,362 --> 00:35:04,233

I mean, we're all going to fail at trying

to identify talent.

522

00:35:04,233 --> 00:35:09,425

It's really hard to identify why one

player is going to succeed over another.

523

00:35:09,485 --> 00:35:14,727

so, you know, in some way it's not binary.

524

00:35:14,727 --> 00:35:16,698

It's not a like, do you like this guy or

not?

525

00:35:16,698 --> 00:35:17,878

Is he good or bad?

526

00:35:17,878 --> 00:35:20,409

Is this guy better or worse than the other

guy?

527

00:35:21,665 --> 00:35:25,467

there's a lot of factors that go into why

someone has success.

528

00:35:25,467 --> 00:35:28,658

And so I think conveying that uncertainty

is really important.

529

00:35:28,658 --> 00:35:33,940

And obviously, the more observations that

we have of you doing the thing, the more

530

00:35:33,940 --> 00:35:37,772

certain we are that this is your true

level of performance.

531

00:35:37,772 --> 00:35:39,202

But it takes a while to get there.

532

00:35:39,202 --> 00:35:41,373

So we have to just be honest about that.

533

00:35:42,674 --> 00:35:44,214

Yeah, yeah.

534

00:35:44,214 --> 00:35:49,186

I think that's actually related to

something I wanted to ask you about also a

535

00:35:49,186 --> 00:35:51,457

bit more generally, you know, but

536

00:35:52,341 --> 00:35:58,443

the most significant challenges that you

face when applying Bayesian stance in, in

537

00:35:58,443 --> 00:36:04,314

sports science and, and how you address

them, because I'm guessing that you, you

538

00:36:04,314 --> 00:36:06,325

already started talking a bit about that.

539

00:36:06,325 --> 00:36:07,675

So, let's go there.

540

00:36:07,675 --> 00:36:11,927

And then, then I have other technical

questions for you, but the kind of, of

541

00:36:11,927 --> 00:36:17,668

models and, and, and usefulness that

Bayesian stance has in your field.

542

00:36:17,668 --> 00:36:21,269

But I think this is a good moment to, to

address these.

543

00:36:21,269 --> 00:36:22,389

questions.

544

00:36:22,570 --> 00:36:29,334

think the biggest or there's a few

challenges.

545

00:36:29,334 --> 00:36:35,948

One challenge is not everybody is excited

about a posterior distribution like you

546

00:36:35,948 --> 00:36:37,518

might be.

547

00:36:37,519 --> 00:36:38,749

Most of the time, they just want an

answer.

548

00:36:38,749 --> 00:36:39,679

Tell me what to

549

00:36:49,578 --> 00:36:52,820

Give me the yes or no, make it binary.

550

00:36:52,820 --> 00:36:55,262

And so that's always tough.

551

00:36:55,262 --> 00:37:00,646

And you're trying to oftentimes convey

this to non -technical audiences or people

552

00:37:00,646 --> 00:37:03,888

who are good at doing other things.

553

00:37:03,888 --> 00:37:06,390

They're not math people or they're not

stats people.

554

00:37:06,390 --> 00:37:07,710

And that's okay.

555

00:37:08,031 --> 00:37:11,954

So that always makes it challenging is why

are you showing me this distribute?

556

00:37:11,954 --> 00:37:15,226

I don't understand what I'm supposed to

take from this.

557

00:37:15,226 --> 00:37:16,297

Just tell me.

558

00:37:16,593 --> 00:37:17,213

What to do?

559

00:37:17,213 --> 00:37:18,374

Tell me which guy's better.

560

00:37:18,374 --> 00:37:19,954

Tell me which guy's worse.

561

00:37:20,214 --> 00:37:21,635

So that's always hard.

562

00:37:21,635 --> 00:37:28,156

And that takes a lot of patience and

communication.

563

00:37:28,437 --> 00:37:33,158

For a while, we used to do just weekly sit

downs with our scouts where we would teach

564

00:37:33,158 --> 00:37:35,419

them about like one stat a week.

565

00:37:35,439 --> 00:37:37,159

And we'd go slow.

566

00:37:37,159 --> 00:37:38,640

And we'd also try and...

567

00:37:43,359 --> 00:37:47,271

as best as possible, relate things back to

the currency that they speak in.

568

00:37:47,271 --> 00:37:52,336

And scouts and coaches, the currency they

speak in is video, not charts and graphs.

569

00:37:52,677 --> 00:37:59,022

So the more that we can connect our

analysis to video cut -ups, because then

570

00:37:59,022 --> 00:38:00,083

they can see it.

571

00:38:00,083 --> 00:38:07,689

And then they understand why a model says

what it says or makes a decision or why it

572

00:38:07,689 --> 00:38:08,580

has assumptions.

573

00:38:08,580 --> 00:38:12,693

And this is also super valuable too,

because they give

574

00:38:12,959 --> 00:38:20,996

And they say, it's, saying that, you know,

the model is saying that, this is, is the

575

00:38:20,996 --> 00:38:24,608

outcome, but I can see why it's because

these four other things happen.

576

00:38:24,608 --> 00:38:25,659

It's like, wow.

577

00:38:25,659 --> 00:38:27,141

Well, we could probably account for that.

578

00:38:27,141 --> 00:38:29,823

And we never, I just didn't know it,

right?

579

00:38:29,823 --> 00:38:33,405

That's why they're domain expert and, and,

and I'm not.

580

00:38:33,596 --> 00:38:34,446

so.

581

00:38:34,947 --> 00:38:39,420

You know, the patience around

communicating stats and numbers is always

582

00:38:39,420 --> 00:38:42,923

difficult and also knowing what people

like.

583

00:38:43,782 --> 00:38:47,805

When I first started, everybody would tell

you, need to have, you know, got to have

584

00:38:47,805 --> 00:38:52,840

an amazing dashboard, got to have like

charts and graphs, you know, and all that

585

00:38:52,840 --> 00:38:53,170

stuff.

586

00:38:53,170 --> 00:38:56,643

And what I found was there was a lot of

people who were like, I don't, what do

587

00:38:56,643 --> 00:38:58,044

you, I don't even know what I'm looking

at.

588

00:38:58,044 --> 00:38:59,035

Like, I hate these things.

589

00:38:59,035 --> 00:39:00,486

Just give me the table of numbers.

590

00:39:00,486 --> 00:39:01,547

It's like, okay.

591

00:39:01,547 --> 00:39:06,641

Well, maybe a table of numbers with just

some conditionally formatted information.

592

00:39:06,641 --> 00:39:11,285

And also, you know,

593

00:39:11,585 --> 00:39:16,867

I have an academic side, I do supervise

PhD students and master students, and I do

594

00:39:16,867 --> 00:39:20,689

teach a master's class in statistics at

college.

595

00:39:22,050 --> 00:39:26,212

So I guess what I'm about to say would,

know, people on the academic side would

596

00:39:26,212 --> 00:39:30,033

hate it, but you have to like recognize

the environment you're in.

597

00:39:30,033 --> 00:39:37,197

And sometimes just like changing the

verbiage helps, like instead of calling

598

00:39:37,197 --> 00:39:38,517

things the...

599

00:39:38,549 --> 00:39:42,271

low credible interval and the high

credible interval, like we just call it

600

00:39:42,271 --> 00:39:43,691

the floor and the ceiling.

601

00:39:43,691 --> 00:39:47,683

And people are like, yeah, this guy's

floor, it's a bit higher than the other

602

00:39:47,683 --> 00:39:48,263

guy's floor.

603

00:39:48,263 --> 00:39:50,574

And that guy's ceiling, this guy's got a

better ceiling.

604

00:39:50,574 --> 00:39:56,277

And like, know, academically you'd get

shot for that, it's like, those kinds of

605

00:39:56,277 --> 00:40:02,019

things go a long way because it brings the

information to the end user.

606

00:40:02,039 --> 00:40:04,740

And if you want them to start to...

607

00:40:06,445 --> 00:40:10,547

take this information into their decision

calculus, you have to get them

608

00:40:10,547 --> 00:40:11,178

comfortable.

609

00:40:11,178 --> 00:40:15,249

And sometimes it's just meeting them with

terminology that helps.

610

00:40:15,430 --> 00:40:20,172

And so I think that's a really, you know,

that's a big one.

611

00:40:20,172 --> 00:40:23,654

Those are big challenges in communicating

this stuff.

612

00:40:24,695 --> 00:40:25,855

Yeah, definitely.

613

00:40:25,855 --> 00:40:27,786

And I resonate with that.

614

00:40:27,786 --> 00:40:32,439

I've had the same issues.

615

00:40:33,700 --> 00:40:36,375

I'll be able to tell.

616

00:40:36,375 --> 00:40:42,030

talk more precisely about sports in a few

months.

617

00:40:42,611 --> 00:40:49,757

But when it comes to a lot of other

fields, whether it's marketing or biostats

618

00:40:49,757 --> 00:40:57,463

or electrical forecasting, yeah, the

issues are related to these.

619

00:40:57,484 --> 00:40:59,205

They're also extremely diverse.

620

00:40:59,205 --> 00:41:00,196

So that's interesting.

621

00:41:00,196 --> 00:41:02,828

You definitely don't have a one size fits

all.

622

00:41:04,599 --> 00:41:08,702

Definitely what's extremely important

basically is to know the model extremely

623

00:41:08,702 --> 00:41:12,514

well from my experience.

624

00:41:12,514 --> 00:41:20,280

And yeah, if you have coded the model

yourself, you usually know it really well

625

00:41:20,280 --> 00:41:27,074

because you spent hours on it to try and

get it to work and understand what it's

626

00:41:27,074 --> 00:41:27,394

doing.

627

00:41:27,394 --> 00:41:30,697

And when it's not able to do as you were

saying, I think it's extremely important

628

00:41:30,697 --> 00:41:33,768

to be able to tell people what the model

cannot tell you.

629

00:41:35,305 --> 00:41:46,704

And yeah, I think these are extremely good

points to try and balance what people are

630

00:41:46,704 --> 00:41:48,646

usually wondering about.

631

00:41:48,646 --> 00:41:53,180

And that's also where I think having the

Bayesian model is extremely interesting,

632

00:41:53,180 --> 00:41:53,420

right?

633

00:41:53,420 --> 00:41:57,873

Because the Bayesian model by definition

is extremely open box and you have to run

634

00:41:57,873 --> 00:41:59,805

it down your assumptions.

635

00:41:59,865 --> 00:42:05,169

And so you know much better what the model

is doing than a black box model.

636

00:42:05,387 --> 00:42:08,250

Yeah, I mean, that's another good point

is.

637

00:42:10,281 --> 00:42:17,705

If you go into a meeting and you have

model outputs and your only reason when

638

00:42:17,705 --> 00:42:21,247

asked, why does it prefer this over that?

639

00:42:21,247 --> 00:42:23,718

Your only reason is because the model said

so.

640

00:42:23,718 --> 00:42:27,110

If people aren't going to be super excited

about that.

641

00:42:27,110 --> 00:42:30,892

knowing why things are happening, know,

this also, you know, I mean, this really

642

00:42:30,892 --> 00:42:37,036

plays into like how you validate and check

your models.

643

00:42:37,036 --> 00:42:39,627

And so buildings, you know, we kind

644

00:42:39,627 --> 00:42:45,200

within that Bayesian sort of world,

building simulations is a big part of it.

645

00:42:45,200 --> 00:42:49,722

And building simulations to see how the

model behaves under different constraints

646

00:42:49,722 --> 00:42:55,506

and different pieces of information,

that's really important because it gives

647

00:42:55,506 --> 00:43:01,769

you useful context to talk about and it

gives you useful information in order to

648

00:43:02,589 --> 00:43:06,982

head things off at the pass when you know

there's gonna be some gotchas and some

649

00:43:06,982 --> 00:43:09,431

trouble if, you

650

00:43:09,431 --> 00:43:11,112

people have certain types of questions.

651

00:43:11,112 --> 00:43:14,904

You can head things off of the past

because you're already aware of them.

652

00:43:15,264 --> 00:43:21,007

Another thing that I do think is really

useful in this and maybe in some of your

653

00:43:21,007 --> 00:43:27,331

prior work in consulting, I'm sure you've

like stumbled on like, or used frameworks

654

00:43:27,331 --> 00:43:29,382

like crisp DM and things like that.

655

00:43:29,382 --> 00:43:34,054

Like in statistics, there's a PPDAC

problem plan, data analysis and

656

00:43:34,054 --> 00:43:35,155

conclusion.

657

00:43:35,155 --> 00:43:39,137

Those types of frameworks help just

because again,

658

00:43:39,831 --> 00:43:45,926

A lot of times we're dealing with non

-technical audiences and they're trying to

659

00:43:45,926 --> 00:43:49,749

give you a question and say like, Hey, can

we look at this?

660

00:43:49,749 --> 00:43:57,115

And oftentimes these things are very vague

and very sort of like, you know, not, not,

661

00:43:57,195 --> 00:43:59,056

not clearly defined.

662

00:43:59,337 --> 00:44:04,812

like, you know, my younger self would take

that and run away and, know, do something

663

00:44:04,812 --> 00:44:07,824

for a week or two and then come back and

be like, Hey, here's this thing, you know,

664

00:44:07,824 --> 00:44:09,245

and you ask about

665

00:44:10,697 --> 00:44:15,479

you know, they're usually like, the reply

is, that's kind of cool, but I was

666

00:44:15,479 --> 00:44:17,909

thinking of it like this and I would do

this with it.

667

00:44:17,909 --> 00:44:22,091

it's like, man, if you, you know, if you

told me that two weeks ago, I would have

668

00:44:22,091 --> 00:44:23,581

done something else.

669

00:44:23,741 --> 00:44:28,233

So using those kinds of frameworks, one,

does a few things.

670

00:44:28,233 --> 00:44:30,313

One, it gives us the opportunity.

671

00:44:30,313 --> 00:44:34,044

Like I always tell our analysts like

question the question, like, you know,

672

00:44:35,365 --> 00:44:36,715

question the question.

673

00:44:36,715 --> 00:44:36,995

Right?

674

00:44:36,995 --> 00:44:40,348

So when they have a question, I'm always

sitting there and I'm like, okay, well,

675

00:44:40,348 --> 00:44:42,330

you know, what would you want to do with

this?

676

00:44:42,330 --> 00:44:45,713

How, do you foresee yourself using it to

make a decision?

677

00:44:45,713 --> 00:44:48,575

What's the cadence that you would need to

access this information?

678

00:44:48,575 --> 00:44:54,119

If I were to get it to you tomorrow, you

know, what would you, what kind of

679

00:44:54,119 --> 00:44:55,160

decision would you want to make?

680

00:44:55,160 --> 00:44:58,443

Like really kind of Socratic questions,

you know, question the question.

681

00:44:58,443 --> 00:45:01,726

And, that does a few things.

682

00:45:01,726 --> 00:45:06,289

One, we get, we get to two, you know,

683

00:45:06,323 --> 00:45:07,974

usually two different results.

684

00:45:07,974 --> 00:45:08,884

Both of them are good.

685

00:45:08,884 --> 00:45:13,996

The first is I get them to then walk

through that five minutes with me and

686

00:45:13,996 --> 00:45:16,597

clearly define what it is they're looking

for.

687

00:45:16,657 --> 00:45:17,818

That's great.

688

00:45:18,198 --> 00:45:22,290

The other result is the opposite, but it's

also a good result, which is we get about

689

00:45:22,290 --> 00:45:25,051

three minutes in and they go, you know

what?

690

00:45:25,051 --> 00:45:26,902

I haven't thought about this well enough.

691

00:45:26,902 --> 00:45:29,423

Let me think through it a bit more and

come back to you.

692

00:45:29,423 --> 00:45:33,804

In which case I didn't waste the time

building things and scraping and cleaning

693

00:45:33,804 --> 00:45:35,655

data and doing all that stuff.

694

00:45:35,927 --> 00:45:41,519

The other thing that those frameworks do

is, and I try and get analysts to think

695

00:45:41,519 --> 00:45:46,572

like this, is utilize each step within

those frameworks as touch points back to

696

00:45:46,572 --> 00:45:48,442

the person who asked you the question.

697

00:45:48,442 --> 00:45:49,763

Hey, this is where we're at.

698

00:45:49,763 --> 00:45:51,123

We've collected this kind of data.

699

00:45:51,123 --> 00:45:52,304

These are the things we're thinking.

700

00:45:52,304 --> 00:45:54,375

These are the features that we're thinking

about using.

701

00:45:54,375 --> 00:45:55,565

What do you think about that?

702

00:45:55,565 --> 00:45:57,426

Anything else you can think of.

703

00:45:57,946 --> 00:46:02,458

By doing that, along each step of the way,

they get to see the model developed.

704

00:46:02,458 --> 00:46:04,669

They get to provide input.

705

00:46:04,781 --> 00:46:08,241

And what that does is it gives them a bit

of ownership over it.

706

00:46:08,241 --> 00:46:14,161

So when you get to the end result, they're

like, geez, this was built exactly in my

707

00:46:14,161 --> 00:46:16,541

vision, and now I'm excited to use it.

708

00:46:16,541 --> 00:46:20,461

And that's a really cool thing too.

709

00:46:21,421 --> 00:46:22,141

Yeah.

710

00:46:22,141 --> 00:46:22,861

Yeah.

711

00:46:22,861 --> 00:46:25,301

Thanks for that detailed answer, Patrick.

712

00:46:25,301 --> 00:46:30,881

I can definitely hear the 10 years of

experience working on that.

713

00:46:30,921 --> 00:46:34,441

That makes me think about a lot of other

things.

714

00:46:35,969 --> 00:46:42,572

Yeah, definitely the same for me, would

say, where my personal evolution has been

715

00:46:42,633 --> 00:46:53,559

trying to really understand the question

the consumer of the model is trying to get

716

00:46:53,559 --> 00:46:54,479

to, right?

717

00:46:54,479 --> 00:46:56,510

Like what actually is your question?

718

00:46:56,510 --> 00:47:01,583

Because you have something in mind, but

maybe the way we're talking about it right

719

00:47:01,583 --> 00:47:04,884

now and the way I have it in mind is not

what you want.

720

00:47:05,909 --> 00:47:12,554

And so, yeah, as you were saying, a good

model is really that's custom made, that's

721

00:47:12,554 --> 00:47:15,716

fine and hard work and that takes time.

722

00:47:15,716 --> 00:47:24,212

so before investing all that time in doing

the model, let's actually make sure we

723

00:47:24,212 --> 00:47:30,166

align and agree on what we're actually

looking at in studying.

724

00:47:30,166 --> 00:47:32,968

That's, think it's extremely important.

725

00:47:33,128 --> 00:47:34,807

Yeah, no doubt.

726

00:47:34,807 --> 00:47:42,193

I think that's often the hardest part,

because it's just getting people to really

727

00:47:42,193 --> 00:47:42,964

define.

728

00:47:42,964 --> 00:47:48,228

that's probably, I mean, that and making

sure that you have good data.

729

00:47:48,529 --> 00:47:49,860

Those are the two biggest things.

730

00:47:49,860 --> 00:47:57,256

The model building part and things like

that sort of happen a little bit easier

731

00:47:57,256 --> 00:47:59,957

once you do the first two things.

732

00:48:00,078 --> 00:48:01,699

That's always the tough part.

733

00:48:01,880 --> 00:48:03,817

Yeah, yeah, yeah.

734

00:48:03,817 --> 00:48:11,139

Actually, continuing on that topic, how do

you communicate these statistical

735

00:48:11,139 --> 00:48:11,649

concepts?

736

00:48:11,649 --> 00:48:15,220

And honestly, a lot of them are really

complex.

737

00:48:15,220 --> 00:48:22,582

So how do you communicate that to non

-stats people in your line of work?

738

00:48:22,582 --> 00:48:27,904

I'm guessing that would be scouts, as you

talked about, coaches, players.

739

00:48:27,904 --> 00:48:30,164

How do you make sure they understand?

740

00:48:31,597 --> 00:48:36,977

what you're doing and in the end are able

to use it because we talked about that in

741

00:48:36,977 --> 00:48:39,737

episode 108 with Paul Sabin.

742

00:48:40,337 --> 00:48:46,317

If your model is awesome but not used,

it's not very interesting.

743

00:48:46,477 --> 00:48:48,476

So yeah, how do you do that?

744

00:48:50,237 --> 00:48:58,257

First, trying to really understand what

kind of cadence this is going to be on.

745

00:48:58,257 --> 00:48:59,617

So some questions.

746

00:48:59,935 --> 00:49:02,166

especially in sport, get asked.

747

00:49:02,347 --> 00:49:10,262

And they're more asked from the knowledge

generation standpoint, meaning that I have

748

00:49:10,262 --> 00:49:11,353

a question.

749

00:49:11,894 --> 00:49:17,908

I think it'll help us with, you know,

updating our priors, our prior beliefs

750

00:49:17,908 --> 00:49:19,178

about the game.

751

00:49:19,379 --> 00:49:20,560

Maybe things have changed.

752

00:49:20,560 --> 00:49:24,182

Maybe rule changes have altered things or

something like that.

753

00:49:24,543 --> 00:49:26,263

Can we study this?

754

00:49:26,664 --> 00:49:28,045

A question like

755

00:49:28,577 --> 00:49:33,559

for knowledge generation requires a

different output than something that's

756

00:49:33,559 --> 00:49:36,020

like weekly or daily consumption.

757

00:49:36,020 --> 00:49:42,983

So if it's for knowledge generation,

that's usually communicated in the form of

758

00:49:42,983 --> 00:49:47,025

like a short written report.

759

00:49:47,425 --> 00:49:52,507

The question at the top, the bottom line

up front, here's the four bullet points,

760

00:49:52,507 --> 00:49:55,018

and then the nitty gritty.

761

00:49:55,018 --> 00:49:57,207

Like this is how we went about studying

it.

762

00:49:57,207 --> 00:50:01,978

charts and graphs and usually it's like a

page or two and a PDF or maybe like an

763

00:50:01,978 --> 00:50:07,880

interactive HTML file that they can see

things and have a table of contents and go

764

00:50:07,880 --> 00:50:09,240

to different sections.

765

00:50:09,921 --> 00:50:17,183

If the question is directed at stuff

that's required to be evaluated weekly or

766

00:50:17,183 --> 00:50:21,664

daily, like I need to see this every week

because we're going to be evaluating a

767

00:50:21,664 --> 00:50:26,093

certain player or an opponent or I need to

see this daily because it's

768

00:50:26,093 --> 00:50:28,374

player health related, something like

that.

769

00:50:28,895 --> 00:50:31,917

We're always thinking in terms of like web

applications.

770

00:50:31,917 --> 00:50:37,921

So how do I get, you now I have to think

through the full stack pipeline of like,

771

00:50:37,921 --> 00:50:39,582

where do we get the data?

772

00:50:39,582 --> 00:50:41,703

Where does it live in the database?

773

00:50:42,624 --> 00:50:44,645

What's the analysis layer?

774

00:50:44,686 --> 00:50:46,347

Kick it out to an output.

775

00:50:46,347 --> 00:50:48,128

Where's that output stored?

776

00:50:48,128 --> 00:50:54,272

And then how does the website ingest that

output and make it consumable?

777

00:50:54,612 --> 00:50:55,973

And for that,

778

00:50:56,073 --> 00:51:01,878

It's usually some form of charts and

graphs and a table.

779

00:51:01,878 --> 00:51:03,749

And usually it's interactive stuff.

780

00:51:03,749 --> 00:51:09,124

So they can sort and filter and hover over

points and access the information.

781

00:51:09,124 --> 00:51:16,630

And again, as best as possible, I'm always

thinking to try and develop that in the

782

00:51:16,630 --> 00:51:18,792

way that they're going to use it.

783

00:51:18,792 --> 00:51:23,386

So like I was sitting down, for example,

today with our director of player health,

784

00:51:23,386 --> 00:51:24,917

and he was like, you

785

00:51:25,131 --> 00:51:33,313

I'd love to have this information daily so

that I can relay it to the new coaching

786

00:51:33,313 --> 00:51:34,164

staff.

787

00:51:34,404 --> 00:51:36,985

And I want to say it, you know, say these

things.

788

00:51:36,985 --> 00:51:38,345

Okay, great.

789

00:51:38,685 --> 00:51:39,926

I have all that information.

790

00:51:39,926 --> 00:51:47,748

I have all of that, those models, but come

over to the whiteboard and draw for me the

791

00:51:47,748 --> 00:51:51,809

path that you want to take to going from

sitting at your desk.

792

00:51:51,849 --> 00:51:55,732

and reading the information from a webpage

to how you want to communicate it.

793

00:51:56,073 --> 00:52:00,396

And as soon as he started drawing it out,

it's like, okay, I know exactly what to do

794

00:52:00,396 --> 00:52:00,667

now.

795

00:52:00,667 --> 00:52:01,478

That's perfect.

796

00:52:01,478 --> 00:52:05,000

Otherwise I would have built something

that in my head I thought would be useful,

797

00:52:05,000 --> 00:52:06,462

but maybe not useful to him.

798

00:52:06,462 --> 00:52:11,806

And then he uses like part of it or maybe

because he's super motivated, he's going

799

00:52:11,806 --> 00:52:13,048

to use it.

800

00:52:13,048 --> 00:52:14,849

And he's also going to,

801

00:52:15,181 --> 00:52:19,441

use like 10 other things to get the other

stuff he wants, but he's a nice guy and he

802

00:52:19,441 --> 00:52:23,001

doesn't want to tell me that it doesn't

have all the things that he needs.

803

00:52:23,121 --> 00:52:26,981

And so then like four weeks later, I walk

in his office, I'm like, what are you

804

00:52:26,981 --> 00:52:27,301

doing?

805

00:52:27,301 --> 00:52:31,240

It's like, oh, I go here and then I get

this information from this webpage, but

806

00:52:31,240 --> 00:52:33,381

then I go to this other three webpages

again.

807

00:52:33,381 --> 00:52:36,561

So, whoa, whoa, whoa, why didn't you just

tell me that?

808

00:52:36,561 --> 00:52:39,201

Like I'll just, I could make this all into

one thing.

809

00:52:39,201 --> 00:52:41,921

Like you don't have to, and so.

810

00:52:42,149 --> 00:52:47,313

That's a really important piece is knowing

how the data is going to be utilized,

811

00:52:47,754 --> 00:52:53,157

making sure that it's exactly in the order

that the decision maker requires it.

812

00:52:53,618 --> 00:52:54,609

Yeah.

813

00:52:54,609 --> 00:52:55,880

Awesome points.

814

00:52:56,040 --> 00:52:56,310

Yeah.

815

00:52:56,310 --> 00:52:57,151

Thanks for that, Patrick.

816

00:52:57,151 --> 00:53:02,395

And I think it's also very valuable to a

lot of listeners because we're talking

817

00:53:02,395 --> 00:53:08,570

about a professional sports team here, but

it is definitely transferable to

818

00:53:08,711 --> 00:53:12,053

basically, I think, any company where

you're working

819

00:53:12,407 --> 00:53:19,100

different people who are using the models

but are not themselves producing the

820

00:53:19,100 --> 00:53:19,660

models.

821

00:53:19,660 --> 00:53:22,291

It's like almost every company out there.

822

00:53:22,291 --> 00:53:27,813

yeah, I think and also from my experience

doing consulting in a lot of different

823

00:53:27,813 --> 00:53:35,206

fields, I can definitely vouch for the

things you've touched on here.

824

00:53:35,206 --> 00:53:37,027

yeah, thanks.

825

00:53:37,027 --> 00:53:40,868

That's definitely, I think, very valuable.

826

00:53:41,899 --> 00:53:47,573

turn back a bit more to the technical

stuff because I see time is running and I

827

00:53:47,573 --> 00:53:56,449

definitely want to touch a bit more on the

spot side of things and how patient stance

828

00:53:56,449 --> 00:54:00,381

is applied in the film.

829

00:54:02,784 --> 00:54:09,688

Obviously a very important part of your

work is, I'm guessing,

830

00:54:11,053 --> 00:54:14,656

drafting players, player selection

processes.

831

00:54:15,977 --> 00:54:21,822

So yeah, how might Bayesian methods be

applied here to improve the drafted

832

00:54:21,822 --> 00:54:25,785

strategies in the player selection

processes?

833

00:54:26,246 --> 00:54:36,555

Yeah, well, again, like I think I said

earlier, everybody's going to miss.

834

00:54:36,555 --> 00:54:39,037

It's impossible to be, you know...

835

00:54:39,319 --> 00:54:43,670

to have a good hit rate and always be

picking, you know, picking players who are

836

00:54:43,670 --> 00:54:47,091

going to reach high level success.

837

00:54:47,091 --> 00:54:51,762

And a lot of that is just because, you

know, performance and talent are extremely

838

00:54:51,762 --> 00:54:52,443

right -tailed.

839

00:54:52,443 --> 00:54:56,714

You know, you have a whole bunch of

players that never make it.

840

00:54:56,714 --> 00:55:01,075

You have a small group that make it and

are good enough to make it.

841

00:55:01,075 --> 00:55:06,116

You have an even smaller group that are

good enough to make it and like really

842

00:55:06,116 --> 00:55:08,007

good to play all the time.

843

00:55:08,007 --> 00:55:09,461

And then you have

844

00:55:09,461 --> 00:55:12,444

a few Hall of Famers sprinkled in, right?

845

00:55:12,444 --> 00:55:13,805

So it's really right -tailed.

846

00:55:13,805 --> 00:55:17,749

it is very hard to do this stuff.

847

00:55:17,749 --> 00:55:27,559

So, you know, understanding or modeling

your uncertainty, that's really important.

848

00:55:27,559 --> 00:55:28,379

And

849

00:55:32,319 --> 00:55:38,523

information from the domain experts, know,

scouts see things on film that we can't

850

00:55:38,523 --> 00:55:41,374

see in numbers and vice versa.

851

00:55:41,594 --> 00:55:48,497

One of the values that we have is we can

process way more players than any one

852

00:55:48,497 --> 00:55:50,639

human can actually watch.

853

00:55:51,400 --> 00:55:58,834

So we have the ability to build models

that can identify players and hopefully

854

00:55:58,834 --> 00:55:59,944

get them,

855

00:56:03,246 --> 00:56:08,131

over to the domain experts who have to

then watch the film and write the reports

856

00:56:08,131 --> 00:56:12,915

and say like, hey, did you know this guy

was really good in these things?

857

00:56:13,516 --> 00:56:16,438

This is his potential ceiling.

858

00:56:16,639 --> 00:56:22,815

And we think that we have, you know, we

think that this would be valuable for our

859

00:56:22,815 --> 00:56:24,005

team, right?

860

00:56:24,246 --> 00:56:27,248

Building models like that, that help us.

861

00:56:29,291 --> 00:56:33,864

Identify talent, give us a range of

plausible outcomes.

862

00:56:34,306 --> 00:56:40,702

One, it helps us get information to the

people who have to watch the film and make

863

00:56:40,702 --> 00:56:41,852

the decisions.

864

00:56:42,233 --> 00:56:50,161

Two, it helps us have discussions about

where the appropriate time to acquire

865

00:56:50,161 --> 00:56:50,941

people

866

00:56:52,831 --> 00:56:56,653

If you're sitting there, obviously, you

know, in the major league draft, major

867

00:56:56,653 --> 00:56:58,514

league baseball draft, it would be the

same thing.

868

00:56:58,514 --> 00:57:02,945

Everybody knows who the first round picks

are and the second round picks.

869

00:57:03,306 --> 00:57:07,368

It's after that, that things become pretty

sparse.

870

00:57:07,368 --> 00:57:16,602

And if you can identify players that have

unique abilities later in the draft, that

871

00:57:16,602 --> 00:57:19,153

opens up a lot of opportunities to,

872

00:57:20,449 --> 00:57:24,012

select players that might be able to

contribute successfully to your team.

873

00:57:24,012 --> 00:57:27,255

And so that's really where those models

help us.

874

00:57:27,255 --> 00:57:34,741

The other area that they help us in is, I

always talk about with our analysts, like,

875

00:57:34,741 --> 00:57:38,143

what is the benchmark that you're trying

to beat?

876

00:57:38,664 --> 00:57:40,916

So every model, like you can't just build

a model.

877

00:57:40,916 --> 00:57:44,549

I mean, I remember one of our analysts,

she had a model and she said, I built a

878

00:57:44,549 --> 00:57:46,140

model and I think it's really good.

879

00:57:46,140 --> 00:57:47,291

And I said, cool.

880

00:57:47,291 --> 00:57:49,713

How well does it do against the benchmark?

881

00:57:50,145 --> 00:57:52,656

She's like, well, what do you mean?

882

00:57:52,656 --> 00:57:56,578

And I was like, well, like how well does

it do against if we just use, let's say

883

00:57:56,578 --> 00:58:01,260

scout grades or if we just use public

perception, how well does it do

884

00:58:01,260 --> 00:58:02,670

historically against that?

885

00:58:02,670 --> 00:58:04,591

She's like, no, no, no, I don't care about

that.

886

00:58:04,591 --> 00:58:09,893

Like this model is just with their stats

and you You know, it's like, no, no, but

887

00:58:09,893 --> 00:58:13,845

you have to care about that because if

it's not better than those things, then

888

00:58:13,845 --> 00:58:14,936

why would we use it?

889

00:58:14,936 --> 00:58:15,476

Right?

890

00:58:15,476 --> 00:58:18,297

You have to be able to beat that

benchmark.

891

00:58:19,315 --> 00:58:26,159

One of the areas where we can really beat

a benchmark is when we combine the domain

892

00:58:26,159 --> 00:58:32,292

experts information with the actual

observed data information.

893

00:58:32,292 --> 00:58:34,634

And a Bayesian model allows us to do that,

right?

894

00:58:34,634 --> 00:58:41,628

It allows us to take down the domain

expert who's maybe scoring the player a

895

00:58:41,628 --> 00:58:45,740

certain way, writing information about the

player.

896

00:58:45,740 --> 00:58:47,881

It allows us to take that information.

897

00:58:48,415 --> 00:58:53,536

mix it with the numbers and get a model

that is, I guess, man and machine, right?

898

00:58:53,536 --> 00:58:59,828

And those models beat our benchmark much

better than any one of these alone, right?

899

00:58:59,828 --> 00:59:06,520

If we just use numbers, never watched any

film, never knew anything about the

900

00:59:06,520 --> 00:59:09,571

player, or if we just use domain expert

information.

901

00:59:09,571 --> 00:59:13,132

When we combine those things, we tend to

do a much better job.

902

00:59:13,132 --> 00:59:17,153

And so that's where Bayesian analysis

really helps us.

903

00:59:17,153 --> 00:59:18,413

And also,

904

00:59:18,717 --> 00:59:23,778

That's where you start to get interesting

discussions about the floor and the

905

00:59:23,778 --> 00:59:24,769

ceiling of a player.

906

00:59:24,769 --> 00:59:29,770

Because now once you run their posterior

distribution and the domain experts

907

00:59:29,770 --> 00:59:36,942

information is in there and you're saying,

yeah, this guy, he's awesome at tackling

908

00:59:36,942 --> 00:59:40,643

and he'll be a great tackler and blah,

blah, blah.

909

00:59:40,883 --> 00:59:42,083

And these are his numbers.

910

00:59:42,083 --> 00:59:47,405

the numeric model says like, yeah, I think

this guy's a pretty good tackler.

911

00:59:48,087 --> 00:59:52,020

Domain experts saying like, no, no, no, I

watched him and he doesn't play against

912

00:59:52,020 --> 00:59:54,713

great competition, but his technique is

really bad.

913

00:59:54,713 --> 00:59:57,605

It's not going to translate against these

bigger players.

914

00:59:58,346 --> 01:00:01,529

It's like, well, that's not information

that maybe our stats would have.

915

01:00:01,529 --> 01:00:07,244

But when we combine those two bits of

information, all of a sudden, our maybe

916

01:00:07,244 --> 01:00:11,517

overly bullish belief in this player gets

brought down a bit.

917

01:00:11,658 --> 01:00:15,841

And utilizing the information like that

918

01:00:15,999 --> 01:00:20,330

is interesting and it also makes it unique

to the people that are in that room, the

919

01:00:20,330 --> 01:00:25,392

domain experts that you have in that room

and things like that.

920

01:00:25,392 --> 01:00:28,152

How you weight those things is really

important.

921

01:00:28,953 --> 01:00:33,144

For our own analytics staff, we'll do

things like we'll build our own separate

922

01:00:33,144 --> 01:00:37,755

models and have our own meetings and we'll

build our own analysis.

923

01:00:37,755 --> 01:00:42,217

So we'll have independent models all

against each other and maybe we'll have

924

01:00:42,217 --> 01:00:44,117

them weighted or we'll use

925

01:00:44,203 --> 01:00:48,487

you know, like triangle prior and build

them together and, you know, mix them

926

01:00:48,487 --> 01:00:51,019

together and get posterior simulations.

927

01:00:51,019 --> 01:00:58,066

And we try and do those things in a way

that allows us to understand all the

928

01:00:58,066 --> 01:01:01,959

plausible outcomes that might be relevant

for this individual.

929

01:01:03,839 --> 01:01:04,690

It's fascinating.

930

01:01:04,690 --> 01:01:05,110

Yeah.

931

01:01:05,110 --> 01:01:13,256

And I really love both that feel the fact

that you have to blend a lot of different

932

01:01:13,256 --> 01:01:14,596

information.

933

01:01:15,557 --> 01:01:22,422

Like the domain knowledge from the scouts,

the benchmark from the markets, the models

934

01:01:22,422 --> 01:01:29,687

that you have in house, also scientific

knowledge of all the scientists that the

935

01:01:29,687 --> 01:01:32,528

team has inside of it.

936

01:01:34,419 --> 01:01:37,530

that makes all that much more complicated,

right?

937

01:01:37,530 --> 01:01:44,273

I'm guessing sometimes as the modeler, you

would probably be like, my God, that'd be

938

01:01:44,273 --> 01:01:50,776

so much easier if we could just run some

very big neural network and that'd be

939

01:01:50,776 --> 01:01:51,636

done.

940

01:01:52,757 --> 01:01:59,479

at the same time, I think it's what makes

the thrill of that field, at least for me,

941

01:01:59,479 --> 01:02:02,250

is that, no, that stuff is really hard.

942

01:02:02,250 --> 01:02:04,087

There is a lot of randomness.

943

01:02:04,087 --> 01:02:09,221

There is a lot of things we don't really

understand either.

944

01:02:09,261 --> 01:02:14,946

And you have to blend all of these

elements together to try and make the best

945

01:02:14,946 --> 01:02:20,931

decisions you can, even though you know

you're not making the optimal decisions,

946

01:02:20,931 --> 01:02:22,331

as you are saying.

947

01:02:22,852 --> 01:02:28,847

And I think it's a fascinating field to

study important decision -making under

948

01:02:28,847 --> 01:02:30,118

uncertainty.

949

01:02:30,919 --> 01:02:31,880

Yeah, for sure.

950

01:02:31,880 --> 01:02:33,881

I think that's the thing that's most

951

01:02:34,931 --> 01:02:36,646

interesting about it to me.

952

01:02:36,646 --> 01:02:44,664

Like, yeah, I think that's the most that

stuff is fascinating just knowing

953

01:02:46,891 --> 01:02:53,636

Yeah, Decision making under uncertainty is

really challenging and I think that's the

954

01:02:53,636 --> 01:02:58,349

thing that makes this the most, you know,

the coolest stuff to work on.

955

01:02:58,369 --> 01:03:00,810

Yeah, yeah, no, definitely.

956

01:03:01,471 --> 01:03:06,615

Actually, maybe a last question on the

technical side.

957

01:03:06,615 --> 01:03:11,928

Now if we look, so we've talked about the

beginning of the career of a player,

958

01:03:11,928 --> 01:03:12,269

right?

959

01:03:12,269 --> 01:03:13,680

Like the draft.

960

01:03:13,680 --> 01:03:15,160

We've talked about...

961

01:03:16,951 --> 01:03:22,805

kind of the whole lifetime of the player,

which is projection, performance

962

01:03:22,805 --> 01:03:26,087

projection over the whole career.

963

01:03:26,467 --> 01:03:29,640

Now I'm wondering about the day -to -day

stuff.

964

01:03:29,640 --> 01:03:37,395

What can Bayesian models tell us here or

how can they help us in predicting the

965

01:03:37,395 --> 01:03:43,118

impact of training loads on the athletes'

wellbeing and performance?

966

01:03:43,919 --> 01:03:46,701

I know, I think it's kind of a frontier

967

01:03:47,221 --> 01:03:53,394

almost all the sports, but I'm curious

what the state of the art here is,

968

01:03:53,394 --> 01:03:55,284

especially in US football.

969

01:03:55,524 --> 01:04:00,867

Yeah, it really is, I think, the sort of

one of the final frontiers, I guess, in

970

01:04:00,867 --> 01:04:01,787

sport.

971

01:04:02,887 --> 01:04:12,341

Team sport is just challenging because you

perform well or you win or you lose due to

972

01:04:12,341 --> 01:04:15,772

a whole bunch of issues that sometimes

973

01:04:16,717 --> 01:04:18,337

have nothing to do with you.

974

01:04:18,337 --> 01:04:22,377

For example, I can train you, you know, we

could train you and you could be very fit

975

01:04:22,377 --> 01:04:23,657

and strong.

976

01:04:23,697 --> 01:04:29,577

And if in the last play of the game, the

quarterback throws the ball to a patch of

977

01:04:29,577 --> 01:04:34,617

grass and you lose, it had nothing to do

with you being fit and strong.

978

01:04:34,617 --> 01:04:38,057

know, counter that to like individual

sport athletes.

979

01:04:38,057 --> 01:04:45,097

If you're a 400 meter runner, a cyclist, a

swimmer, a runner, a marathoner, you know,

980

01:04:45,097 --> 01:04:46,377

physiologically.

981

01:04:46,929 --> 01:04:55,254

If we build you up, we have a much more

direct line between how you develop and

982

01:04:55,254 --> 01:04:58,716

how it directly relates to your

performance.

983

01:04:58,716 --> 01:05:00,777

There's not a lot of other information

there.

984

01:05:00,777 --> 01:05:05,800

No one's trying to tackle you on the bike

or in the pool or something like that.

985

01:05:05,800 --> 01:05:10,782

So that makes, that makes a sport much

more difficult.

986

01:05:10,782 --> 01:05:15,545

Baseball is probably the closest because

even though it is a team

987

01:05:16,447 --> 01:05:20,418

It really is this sort of zero sum duel

between a pitcher and a batter.

988

01:05:20,599 --> 01:05:23,039

And one guy wins and one guy loses.

989

01:05:23,200 --> 01:05:26,461

And the events are very discreet.

990

01:05:27,522 --> 01:05:34,845

The states of the game have been played

out, know, runner on first and second with

991

01:05:34,845 --> 01:05:37,866

two outs, bottom of the third, blah, blah,

blah.

992

01:05:37,866 --> 01:05:42,207

So it's maybe a little bit more clear in

baseball.

993

01:05:42,207 --> 01:05:46,219

I think in the other team sports, in the

kind of invasion sports,

994

01:05:47,734 --> 01:05:50,856

what makes this challenging is

identifying.

995

01:05:50,856 --> 01:05:56,990

I always try and take it back to

identifying the discrete events that we're

996

01:05:56,990 --> 01:06:00,963

trying to, trying to maybe measure

against.

997

01:06:00,963 --> 01:06:04,856

like, for example, I can give you example,

a pretty clear example from basketball.

998

01:06:04,856 --> 01:06:11,750

was talking with a friend in a, in an NBA

team and, he was like, yeah, you know,

999

01:06:11,750 --> 01:06:15,661

our, our, our coach and our scouts and

the, you know,

Speaker: 1000

01:06:15,661 --> 01:06:20,304

coaches, feel like our players don't close

out three pointers fast enough.

Speaker: 1001

01:06:21,146 --> 01:06:26,910

And I was like, well, is that a tactical

problem or is it a physical problem?

Speaker: 1002

01:06:27,731 --> 01:06:30,433

And he's like, well, how would we look at

that?

Speaker: 1003

01:06:30,433 --> 01:06:33,716

And I was like, you have the player

tracking data.

Speaker: 1004

01:06:33,716 --> 01:06:38,580

And if you know every time your team's on

defense, which is easy to know, and you

Speaker: 1005

01:06:38,580 --> 01:06:43,424

know every three pointer that's been shot

against your defense, if you were to take

Speaker: 1006

01:06:43,424 --> 01:06:44,597

that frame,

Speaker: 1007

01:06:44,597 --> 01:06:49,330

out of the player tracking data and maybe

like the frame a second to a second and a

Speaker: 1008

01:06:49,330 --> 01:06:50,290

half before that.

Speaker: 1009

01:06:50,290 --> 01:06:53,762

So all of that information for every one

of those three pointers.

Speaker: 1010

01:06:54,423 --> 01:06:59,256

You have an idea of the relationship

between your player and the player who's

Speaker: 1011

01:06:59,256 --> 01:07:01,207

taking the three point shot.

Speaker: 1012

01:07:01,427 --> 01:07:05,270

You have an idea of the relationship

between your player and the other players

Speaker: 1013

01:07:05,270 --> 01:07:06,710

on his team.

Speaker: 1014

01:07:07,451 --> 01:07:11,265

So you know from a technical, a tactical

standpoint.

Speaker: 1015

01:07:11,265 --> 01:07:15,449

you know what type of like formation or

defense you're trying to run.

Speaker: 1016

01:07:15,449 --> 01:07:20,173

So first things first, are the players in

the right position to close out that three

Speaker: 1017

01:07:20,173 --> 01:07:21,053

pointer?

Speaker: 1018

01:07:21,274 --> 01:07:23,075

Maybe, you know what?

Speaker: 1019

01:07:23,115 --> 01:07:27,219

Our guys consistently mess up the

defensive shape and when they get in

Speaker: 1020

01:07:27,219 --> 01:07:30,661

there, they give too much ground to the

guy shooting a three pointer.

Speaker: 1021

01:07:31,562 --> 01:07:36,877

The other is the physical standpoint of,

well, no, they're in good position, but

Speaker: 1022

01:07:36,877 --> 01:07:39,989

when they go to close it out over that

second and a half,

Speaker: 1023

01:07:40,023 --> 01:07:41,774

They're not fast enough to get there.

Speaker: 1024

01:07:41,774 --> 01:07:43,055

Okay, great.

Speaker: 1025

01:07:43,075 --> 01:07:47,218

Now roll it back to what you can measure

in the gym.

Speaker: 1026

01:07:47,699 --> 01:07:52,292

Is there some measure, let's say on a

force plate of the amount of impulse or

Speaker: 1027

01:07:52,292 --> 01:07:56,825

force under the force time curve that the

player outputs that can tell us something

Speaker: 1028

01:07:56,825 --> 01:08:00,998

about their ability to move rapidly, apply

force into the ground, move rapidly to

Speaker: 1029

01:08:00,998 --> 01:08:02,689

close out that three pointer?

Speaker: 1030

01:08:03,210 --> 01:08:07,293

And maybe if you look at several years

worth of data, you'd find

Speaker: 1031

01:08:08,033 --> 01:08:12,186

The top players on your team all do this

thing really well, and some of the worst

Speaker: 1032

01:08:12,186 --> 01:08:14,798

players at closing out the three do this

thing poorly.

Speaker: 1033

01:08:14,798 --> 01:08:20,072

And so now you have something to say about

like, hey, what if we develop this quality

Speaker: 1034

01:08:20,072 --> 01:08:23,374

in the off season and our players, would

we be able to close out the three pointers

Speaker: 1035

01:08:23,374 --> 01:08:25,526

more effectively, more efficiently?

Speaker: 1036

01:08:25,526 --> 01:08:32,291

And so I think from that standpoint,

linking the development piece to sport,

Speaker: 1037

01:08:33,192 --> 01:08:35,463

team sport, invasion sport.

Speaker: 1038

01:08:35,489 --> 01:08:40,840

You have to really think about the

discrete events of the game and how you

Speaker: 1039

01:08:40,840 --> 01:08:45,291

can kind of tease those out of, let's say

the player tracking data.

Speaker: 1040

01:08:45,292 --> 01:08:50,253

And it's like super hard in something

like, you know, in football, because

Speaker: 1041

01:08:50,253 --> 01:08:52,254

players all do really different things.

Speaker: 1042

01:08:52,254 --> 01:08:56,315

You know, the linebacker does something

totally different than the offensive

Speaker: 1043

01:08:56,315 --> 01:08:56,895

lineman.

Speaker: 1044

01:08:56,895 --> 01:09:02,576

And so you have to really get down to the,

the domain of each of those positions and

Speaker: 1045

01:09:02,576 --> 01:09:05,457

say like, gosh, what are the discrete

events?

Speaker: 1046

01:09:05,569 --> 01:09:11,494

that define what this position does, then

how do we measure success in those?

Speaker: 1047

01:09:11,494 --> 01:09:16,278

And then if we can measure success, how do

we identify the archetype of players who

Speaker: 1048

01:09:16,278 --> 01:09:18,039

are good at those things?

Speaker: 1049

01:09:18,360 --> 01:09:21,983

And then if we can do that, maybe then we

can start to talk about, is this something

Speaker: 1050

01:09:21,983 --> 01:09:23,744

that you can develop in a player?

Speaker: 1051

01:09:23,744 --> 01:09:27,007

Is it something that you have to identify

in a player?

Speaker: 1052

01:09:27,488 --> 01:09:33,252

That's sort of the, in my head, I mean, I

don't know, I could be wrong.

Speaker: 1053

01:09:33,292 --> 01:09:34,483

This is not.

Speaker: 1054

01:09:34,621 --> 01:09:37,952

Nobody, think everybody's trying to figure

this out, but I could be wrong.

Speaker: 1055

01:09:37,952 --> 01:09:41,883

But in my head, that's at least the

process that I would, you know, I try and

Speaker: 1056

01:09:41,883 --> 01:09:44,444

think through when I think about these

things.

Speaker: 1057

01:09:45,764 --> 01:09:46,644

Yeah.

Speaker: 1058

01:09:46,745 --> 01:09:46,975

Yeah.

Speaker: 1059

01:09:46,975 --> 01:09:48,025

It makes a ton of sense.

Speaker: 1060

01:09:48,025 --> 01:09:56,217

mean, and it seems like, yeah, that, and

there are so many areas, open areas of

Speaker: 1061

01:09:56,217 --> 01:09:58,588

research on all of that stuff.

Speaker: 1062

01:09:58,588 --> 01:10:00,839

That's just, just fascinating.

Speaker: 1063

01:10:00,839 --> 01:10:01,505

I'm

Speaker: 1064

01:10:01,505 --> 01:10:06,049

I'm already thinking, that'd be amazing to

have a huge patient model where you have

Speaker: 1065

01:10:06,049 --> 01:10:11,363

all of those topics that we've talked

about.

Speaker: 1066

01:10:11,363 --> 01:10:16,717

Basically, it could be a big patient model

where you have a bunch of likelihoods.

Speaker: 1067

01:10:17,859 --> 01:10:20,371

And yeah, that'd be super fun.

Speaker: 1068

01:10:20,371 --> 01:10:25,465

I'm guessing we're still a bit far from

that, but maybe not too far.

Speaker: 1069

01:10:25,465 --> 01:10:30,029

Hopefully in a few years, that'd be

definitely super fun.

Speaker: 1070

01:10:30,197 --> 01:10:31,878

Yeah, no doubt.

Speaker: 1071

01:10:32,299 --> 01:10:36,782

Yeah, I mean, and that's, definitely

doable.

Speaker: 1072

01:10:36,982 --> 01:10:41,185

But yeah, you need you need really good

data and you need really good structure in

Speaker: 1073

01:10:41,185 --> 01:10:42,226

your model.

Speaker: 1074

01:10:42,887 --> 01:10:48,111

Yeah, that's the part too, is getting

getting good data, know, player tracking

Speaker: 1075

01:10:48,111 --> 01:10:50,172

data is fine.

Speaker: 1076

01:10:50,172 --> 01:10:53,865

I mean, it has errors, you know, people

who think that it's like a panacea, you

Speaker: 1077

01:10:53,865 --> 01:10:55,536

know, it's like, have you really worked

with it?

Speaker: 1078

01:10:55,536 --> 01:10:56,577

I mean, there's

Speaker: 1079

01:10:56,749 --> 01:11:01,009

Sampling at 10 hertz for humans that move

really, really fast.

Speaker: 1080

01:11:01,949 --> 01:11:04,429

Acceleration is a derivative of speed.

Speaker: 1081

01:11:04,429 --> 01:11:09,649

At 10 hertz, people who are moving really

fast, that data gets noisy pretty quick.

Speaker: 1082

01:11:13,349 --> 01:11:19,109

I think one of the things is as we

progress, as the technology keeps

Speaker: 1083

01:11:19,109 --> 01:11:21,469

improving, things get better.

Speaker: 1084

01:11:22,421 --> 01:11:26,404

you get better data and maybe that helps

you also answer some of these questions a

Speaker: 1085

01:11:26,404 --> 01:11:28,225

little bit more specifically.

Speaker: 1086

01:11:29,146 --> 01:11:30,247

yeah.

Speaker: 1087

01:11:30,247 --> 01:11:36,242

And then we'll be able to have our huge

patient model with a lot of different

Speaker: 1088

01:11:36,242 --> 01:11:39,364

likelihoods in there that fit into each

other.

Speaker: 1089

01:11:39,364 --> 01:11:42,987

And then we don't even need to play the

game.

Speaker: 1090

01:11:42,987 --> 01:11:43,937

We don't have to play the game.

Speaker: 1091

01:11:43,937 --> 01:11:46,840

They just let the computers play the game

and it's over.

Speaker: 1092

01:11:46,840 --> 01:11:47,290

We're done.

Speaker: 1093

01:11:47,290 --> 01:11:49,241

Yeah, no.

Speaker: 1094

01:11:49,291 --> 01:11:53,965

No, you still have to play the game

because you still have randomness.

Speaker: 1095

01:11:53,965 --> 01:11:55,346

Then you're like, yeah.

Speaker: 1096

01:11:55,346 --> 01:11:59,880

mean, because otherwise the model is kind

of like if you want kind of a quantum

Speaker: 1097

01:11:59,880 --> 01:12:00,670

state, right?

Speaker: 1098

01:12:00,670 --> 01:12:05,935

Where the model can see the probabilities

of things happening, but then you have to

Speaker: 1099

01:12:05,935 --> 01:12:09,137

open the box and see what is actually

happening.

Speaker: 1100

01:12:09,137 --> 01:12:12,500

So you can have the best model.

Speaker: 1101

01:12:12,900 --> 01:12:16,794

In the end, you still have to play the

game to see what's going to happen because

Speaker: 1102

01:12:16,794 --> 01:12:18,945

it's not deterministic.

Speaker: 1103

01:12:19,181 --> 01:12:21,142

Yeah, thankfully.

Speaker: 1104

01:12:21,443 --> 01:12:22,464

yeah, that's right.

Speaker: 1105

01:12:22,464 --> 01:12:22,884

Yeah.

Speaker: 1106

01:12:22,884 --> 01:12:23,385

Yeah.

Speaker: 1107

01:12:23,385 --> 01:12:28,328

But I mean, it's definitely I always love

doing these these big models.

Speaker: 1108

01:12:28,569 --> 01:12:29,890

And that's definitely doable.

Speaker: 1109

01:12:29,890 --> 01:12:33,993

I've done that for election forecasting,

for instance, where you have several

Speaker: 1110

01:12:34,414 --> 01:12:40,078

likelihoods, one for polls, for instance,

and one for elections.

Speaker: 1111

01:12:40,980 --> 01:12:45,083

So yeah, that's I know that's definitely

doable in the Bayesian framework, because

Speaker: 1112

01:12:45,083 --> 01:12:46,024

I mean, why not?

Speaker: 1113

01:12:46,024 --> 01:12:47,745

It's just part of the big

Speaker: 1114

01:12:48,265 --> 01:12:54,629

of the same big model in a directed S

-secret graph, if you want.

Speaker: 1115

01:12:55,470 --> 01:12:59,693

But yeah, I'm curious to see that done in

spots.

Speaker: 1116

01:12:59,693 --> 01:13:06,398

Maybe we'll get back together for another

episode, Patrick, where we talk about that

Speaker: 1117

01:13:06,398 --> 01:13:08,830

and how we did that.

Speaker: 1118

01:13:08,830 --> 01:13:10,100

That'd be cool.

Speaker: 1119

01:13:10,181 --> 01:13:11,441

Yeah, there you go.

Speaker: 1120

01:13:11,522 --> 01:13:14,783

Yeah, actually, I wanted to ask you to

close us out here.

Speaker: 1121

01:13:17,907 --> 01:13:23,300

about, you you've started talking about

that right now, like some emerging trends

Speaker: 1122

01:13:23,380 --> 01:13:29,043

in sports analytics that you believe will

significantly impact how teams manage

Speaker: 1123

01:13:29,043 --> 01:13:35,606

training, performance, drafting in the

near future.

Speaker: 1124

01:13:36,006 --> 01:13:41,409

And also if there are any spots you see as

more promising than others.

Speaker: 1125

01:13:43,527 --> 01:13:45,659

well, mean, yeah, trends.

Speaker: 1126

01:13:45,659 --> 01:13:46,019

Yeah.

Speaker: 1127

01:13:46,019 --> 01:13:50,782

We talked a lot about that stuff and I

think, you know, better data and better,

Speaker: 1128

01:13:50,782 --> 01:13:52,643

you know, better technology.

Speaker: 1129

01:13:53,624 --> 01:13:56,906

all of those things will, will, I think

will help us.

Speaker: 1130

01:13:56,906 --> 01:14:03,921

I think also it's getting, you know,

getting the decision makers comfortable

Speaker: 1131

01:14:03,921 --> 01:14:08,955

with the utility of some of this stuff,

you know, baseball, has always been a game

Speaker: 1132

01:14:08,955 --> 01:14:10,095

of numbers.

Speaker: 1133

01:14:10,136 --> 01:14:12,717

And, I think early.

Speaker: 1134

01:14:13,101 --> 01:14:22,281

maybe mid 2000s, early 2004, five, six,

seven, you know, releasing data kind of to

Speaker: 1135

01:14:22,281 --> 01:14:28,061

the public, really the first sport to get

player tracking data, things like that.

Speaker: 1136

01:14:28,761 --> 01:14:35,721

I think that opened up a lot of

opportunities for people to do really

Speaker: 1137

01:14:35,721 --> 01:14:42,049

interesting work in the public space,

which then sort of got

Speaker: 1138

01:14:42,049 --> 01:14:49,075

teams interested and then sort of a, you

know, more of a shift in people in the

Speaker: 1139

01:14:49,075 --> 01:14:55,920

front office where, maybe historically it

was ex players who kind of played out

Speaker: 1140

01:14:55,920 --> 01:15:00,564

until they retired and then became scouts

and managers and things like that.

Speaker: 1141

01:15:02,386 --> 01:15:07,490

I think that, you know, that happening in

baseball was a really good thing for that

Speaker: 1142

01:15:07,490 --> 01:15:08,030

sport.

Speaker: 1143

01:15:08,030 --> 01:15:11,813

And I think slowly for the other sports,

that's really

Speaker: 1144

01:15:12,343 --> 01:15:19,856

probably needs to happen because the more

that these things are open and sort of

Speaker: 1145

01:15:19,856 --> 01:15:26,539

curbside, I think the more the decision

makers become comfortable with them and

Speaker: 1146

01:15:26,539 --> 01:15:29,230

can say like, I can see how I would use

this.

Speaker: 1147

01:15:29,230 --> 01:15:32,661

I can see what this might help me with.

Speaker: 1148

01:15:32,721 --> 01:15:38,624

so I think that's never underestimate the

work that you do in the public space

Speaker: 1149

01:15:38,624 --> 01:15:41,805

because I think there's an opportunity to

always.

Speaker: 1150

01:15:42,711 --> 01:15:46,590

you know, help things evolve,

crowdsourcing, guess.

Speaker: 1151

01:15:48,225 --> 01:15:51,547

Yeah, mean, preaching to the choir here.

Speaker: 1152

01:15:52,108 --> 01:15:57,753

Yeah, for me, a lot more of these data

would be open sourced.

Speaker: 1153

01:15:57,753 --> 01:16:06,841

Yeah, I mean, there is also an extremely

interesting trend right now towards open

Speaker: 1154

01:16:06,841 --> 01:16:11,144

sourcing more and more parts of large

language models.

Speaker: 1155

01:16:12,626 --> 01:16:17,015

I think that's going to be extremely

interesting to see that develop because

Speaker: 1156

01:16:17,015 --> 01:16:20,787

At the same time, this is very hard

because these kind of models are just so

Speaker: 1157

01:16:20,787 --> 01:16:21,787

huge.

Speaker: 1158

01:16:22,368 --> 01:16:28,792

You need a lot of computing power to make

them run.

Speaker: 1159

01:16:28,792 --> 01:16:33,434

So I don't know how open source can help

in that, but I know how open source can

Speaker: 1160

01:16:33,434 --> 01:16:40,478

help in the development and sustainability

and trustworthiness and openness of all

Speaker: 1161

01:16:40,478 --> 01:16:40,908

that stuff.

Speaker: 1162

01:16:40,908 --> 01:16:43,159

So that's going to be super interesting.

Speaker: 1163

01:16:43,620 --> 01:16:45,921

And I'm also going to be very interested

in

Speaker: 1164

01:16:46,081 --> 01:16:48,082

the different spots evolve.

Speaker: 1165

01:16:48,402 --> 01:16:54,546

Now that basically the nerds are they are

much more right than before.

Speaker: 1166

01:16:54,546 --> 01:17:01,610

know, so like, probably baseball is going

to be at the forefront of that because

Speaker: 1167

01:17:01,610 --> 01:17:07,293

they just have a lot of, of, know,

advanced in years compared to the other

Speaker: 1168

01:17:07,293 --> 01:17:07,973

sports.

Speaker: 1169

01:17:07,973 --> 01:17:11,535

So it's going to be interesting to see how

things plays out here when it comes to

Speaker: 1170

01:17:11,535 --> 01:17:12,656

data.

Speaker: 1171

01:17:13,216 --> 01:17:14,537

Because at the same

Speaker: 1172

01:17:14,977 --> 01:17:19,810

Not sure it makes a lot of sense for all

the clubs to have their own data

Speaker: 1173

01:17:19,810 --> 01:17:25,082

collection structure if in the end they

just have the same data because you're

Speaker: 1174

01:17:26,503 --> 01:17:34,668

mainly, I think, to gather data, you are

limited, I'm guessing, by the technology

Speaker: 1175

01:17:34,668 --> 01:17:43,833

much more than by the ideas of a coach or

manager or a scientist being like, I

Speaker: 1176

01:17:44,075 --> 01:17:49,488

data, I think in the end, the data

collection is something that can be pretty

Speaker: 1177

01:17:49,488 --> 01:17:58,553

much, you know, collective, but then how

you use the data is more the appropriate

Speaker: 1178

01:17:58,553 --> 01:18:00,033

proprietary stuff.

Speaker: 1179

01:18:00,033 --> 01:18:02,376

It's going to be interesting to see that

out.

Speaker: 1180

01:18:02,376 --> 01:18:05,917

Yeah, no doubt.

Speaker: 1181

01:18:06,138 --> 01:18:06,478

Great.

Speaker: 1182

01:18:06,478 --> 01:18:09,390

Well, Patrick, I've taken a lot of your

time already.

Speaker: 1183

01:18:09,390 --> 01:18:11,761

I need to let you go because...

Speaker: 1184

01:18:11,885 --> 01:18:14,865

You need to drink some coffee.

Speaker: 1185

01:18:14,865 --> 01:18:19,545

definitely need to because that was very

intense.

Speaker: 1186

01:18:19,924 --> 01:18:21,665

But man, so interesting.

Speaker: 1187

01:18:22,205 --> 01:18:26,005

Before letting you go, so I have the last

two questions, of course, as usual.

Speaker: 1188

01:18:26,085 --> 01:18:30,485

You told me before we started the show

that when the season is going to start

Speaker: 1189

01:18:30,485 --> 01:18:36,525

again for you in US football, your days

are going to be extremely busy.

Speaker: 1190

01:18:36,525 --> 01:18:39,545

Like basically working from 5 a .m.

Speaker: 1191

01:18:39,545 --> 01:18:40,545

to 10 p .m.

Speaker: 1192

01:18:40,545 --> 01:18:41,825

or something like that.

Speaker: 1193

01:18:41,997 --> 01:18:43,537

How is that possible?

Speaker: 1194

01:18:43,537 --> 01:18:44,877

when do you sleep?

Speaker: 1195

01:18:45,617 --> 01:18:48,117

We do have some long days.

Speaker: 1196

01:18:48,637 --> 01:18:53,317

It depends on the day of the week and when

the full practice days are.

Speaker: 1197

01:18:53,957 --> 01:19:01,117

Usually, yeah, I'd get in around 4, 45 or

5, have a bit of a workout, and then kind

Speaker: 1198

01:19:01,117 --> 01:19:05,477

of start the day around 6, 30 or 7.

Speaker: 1199

01:19:05,477 --> 01:19:08,257

And it's really long.

Speaker: 1200

01:19:08,257 --> 01:19:10,417

I mean, there's a ton of meetings.

Speaker: 1201

01:19:10,509 --> 01:19:13,511

It's a very tactical sport if you've ever

watched it.

Speaker: 1202

01:19:13,511 --> 01:19:19,816

And so the players are nonstop in and out

of meetings and walk through practices and

Speaker: 1203

01:19:19,816 --> 01:19:23,618

full practices and then more meetings.

Speaker: 1204

01:19:23,618 --> 01:19:30,803

it's all a big, you know, tactical pattern

recognition type of thing.

Speaker: 1205

01:19:30,803 --> 01:19:39,277

And so, you know, we're in, you know,

working on projects and data and getting

Speaker: 1206

01:19:39,277 --> 01:19:47,617

you know, things set up so that model set

up and identifying things in data for the

Speaker: 1207

01:19:47,617 --> 01:19:49,317

staff and things like that.

Speaker: 1208

01:19:49,357 --> 01:19:53,537

it just becomes this really long day.

Speaker: 1209

01:19:53,737 --> 01:19:58,337

And I mean, like, yeah, if we go home at

eight or nine, maybe 930 sometimes, maybe

Speaker: 1210

01:19:58,337 --> 01:20:03,697

10, but I mean, there's people there

that'll stay even later than that, just

Speaker: 1211

01:20:03,697 --> 01:20:06,797

going through film and watching it.

Speaker: 1212

01:20:07,021 --> 01:20:08,561

They are very long days.

Speaker: 1213

01:20:08,561 --> 01:20:12,981

Usually those types of days are about

three days a week and then the other days,

Speaker: 1214

01:20:12,981 --> 01:20:16,961

I might be in there at five and get out at

like five or six.

Speaker: 1215

01:20:16,961 --> 01:20:22,361

So still 12 hour days, but it's a long

week for sure.

Speaker: 1216

01:20:22,901 --> 01:20:24,001

This is brutal.

Speaker: 1217

01:20:24,001 --> 01:20:24,581

Yeah.

Speaker: 1218

01:20:24,741 --> 01:20:31,281

But is it like that during the whole

season or is that mainly the start of the

Speaker: 1219

01:20:31,281 --> 01:20:32,121

season?

Speaker: 1220

01:20:32,401 --> 01:20:34,021

No, that's the season.

Speaker: 1221

01:20:34,021 --> 01:20:35,921

That is

Speaker: 1222

01:20:37,101 --> 01:20:40,421

18 weeks later we have a bye, 17 games

this season.

Speaker: 1223

01:20:41,241 --> 01:20:42,961

Damn, impressive.

Speaker: 1224

01:20:42,961 --> 01:20:48,361

You have to be sharp with your sleep also,

I guess in these weeks.

Speaker: 1225

01:20:48,361 --> 01:20:49,541

You do, yes.

Speaker: 1226

01:20:49,761 --> 01:20:52,501

You try and catch up on the weekends.

Speaker: 1227

01:20:52,961 --> 01:20:54,341

Yeah, damn.

Speaker: 1228

01:20:55,521 --> 01:21:00,781

Awesome, well Patrick, I think it's time

to call it a show.

Speaker: 1229

01:21:01,561 --> 01:21:04,121

Thank you so much, that was amazing.

Speaker: 1230

01:21:04,469 --> 01:21:09,391

Of course, I'm going to ask you the last

two questions, ask every guest at the end

Speaker: 1231

01:21:09,391 --> 01:21:10,531

of the show.

Speaker: 1232

01:21:11,132 --> 01:21:13,413

You knew that was coming, right?

Speaker: 1233

01:21:13,413 --> 01:21:15,694

Yes.

Speaker: 1234

01:21:15,694 --> 01:21:16,734

So what's the first one?

Speaker: 1235

01:21:16,734 --> 01:21:18,094

You know the first one.

Speaker: 1236

01:21:18,455 --> 01:21:23,817

The first one is, if unlimited resources,

what problem would you solve?

Speaker: 1237

01:21:23,817 --> 01:21:25,838

Yeah, unlimited time and resources.

Speaker: 1238

01:21:25,838 --> 01:21:28,259

I'll take one outside of sport, but one

Speaker: 1239

01:21:33,291 --> 01:21:35,062

I witnessed in sport.

Speaker: 1240

01:21:35,062 --> 01:21:42,627

so when I first started, I used to do all

of the GPS stuff, like live on the field.

Speaker: 1241

01:21:42,627 --> 01:21:48,131

Now someone else does it, but coding it or

cutting it up and stuff like that during

Speaker: 1242

01:21:48,131 --> 01:21:48,691

practice.

Speaker: 1243

01:21:48,691 --> 01:21:56,666

And on Friday practices at the time, that

was the day for our make -a -wish, the

Speaker: 1244

01:21:56,666 --> 01:21:58,029

make -a -wish child.

Speaker: 1245

01:21:58,029 --> 01:22:01,632

So they'd have kids that had make a wish

and their wish was to see a practice and

Speaker: 1246

01:22:01,632 --> 01:22:03,123

meet their favorite NFL players.

Speaker: 1247

01:22:03,123 --> 01:22:08,417

And these were usually kids that were, you

know, were small and terminally ill.

Speaker: 1248

01:22:09,179 --> 01:22:16,044

I think the, that's probably the thing

that I would solve because standing there

Speaker: 1249

01:22:16,465 --> 01:22:22,789

and you watch that and you work with all

these guys that are healthy and young.

Speaker: 1250

01:22:22,810 --> 01:22:26,733

And then you see this little kid who never

have a chance to

Speaker: 1251

01:22:27,575 --> 01:22:31,517

healthy and young, but they're just so

happy to meet these guys.

Speaker: 1252

01:22:31,517 --> 01:22:36,219

I think like that's a super unfair thing

for those little kids.

Speaker: 1253

01:22:36,219 --> 01:22:42,281

if I could solve anything, it'd be like

that, you know, kids and cancer and stuff

Speaker: 1254

01:22:42,281 --> 01:22:42,621

like that.

Speaker: 1255

01:22:42,621 --> 01:22:44,372

I think it's just a horrible thing.

Speaker: 1256

01:22:44,372 --> 01:22:50,325

And then your second question is always, I

could have dinner with anyone dead or

Speaker: 1257

01:22:50,325 --> 01:22:51,915

alive, who would it be?

Speaker: 1258

01:22:52,776 --> 01:22:56,457

There's so many good ones, but I think I

would pick...

Speaker: 1259

01:22:56,479 --> 01:23:01,962

a previous guest that you've had, I think

three times, if I'm correct, which is

Speaker: 1260

01:23:01,962 --> 01:23:03,113

Andrew Gellman.

Speaker: 1261

01:23:03,113 --> 01:23:09,556

I think he's fascinatingly interesting and

I think dinner would be pretty amazing.

Speaker: 1262

01:23:10,317 --> 01:23:11,117

Yeah.

Speaker: 1263

01:23:11,117 --> 01:23:12,597

yeah.

Speaker: 1264

01:23:12,597 --> 01:23:15,519

Both good choices, amazing answers.

Speaker: 1265

01:23:15,519 --> 01:23:18,181

Thanks, Patrick.

Speaker: 1266

01:23:18,981 --> 01:23:24,885

I can tell your faithful listeners because

they're like, yeah, you knew the

Speaker: 1267

01:23:24,885 --> 01:23:25,867

questions.

Speaker: 1268

01:23:25,867 --> 01:23:28,939

Like you're taking my job basically, I can

see that.

Speaker: 1269

01:23:30,942 --> 01:23:32,322

No, that's great.

Speaker: 1270

01:23:32,563 --> 01:23:40,250

So Andrew, if you're listening, well, if

you're ever in New York, Patrick will try

Speaker: 1271

01:23:40,250 --> 01:23:41,531

and make that work.

Speaker: 1272

01:23:41,531 --> 01:23:42,552

That'd be fun for sure.

Speaker: 1273

01:23:42,552 --> 01:23:46,805

Yeah, Andrew is always fantastic to talk

to.

Speaker: 1274

01:23:46,805 --> 01:23:49,317

So yeah, that's definitely a great choice.

Speaker: 1275

01:23:49,538 --> 01:23:51,009

Awesome.

Speaker: 1276

01:23:51,009 --> 01:23:53,207

Well, that's it, Patrick.

Speaker: 1277

01:23:53,207 --> 01:23:55,669

Thank you so much for being in the show.

Speaker: 1278

01:23:55,669 --> 01:24:02,784

I really had a blast and learned a lot

about US football because I that's, I

Speaker: 1279

01:24:02,784 --> 01:24:06,677

think that's not the sport I know most

about.

Speaker: 1280

01:24:06,677 --> 01:24:11,310

So definitely thank you so much for taking

the time.

Speaker: 1281

01:24:11,490 --> 01:24:17,384

We'll put resources to your website in the

show notes for those who want to dig

Speaker: 1282

01:24:17,384 --> 01:24:17,725

deeper.

Speaker: 1283

01:24:17,725 --> 01:24:20,916

have a bunch of links over there and

Speaker: 1284

01:24:21,269 --> 01:24:25,122

Thank you again, Patrick, for taking the

time and being on the show.

Speaker: 1285

01:24:25,226 --> 01:24:26,207

Thank you.

Speaker: 1286

01:24:31,565 --> 01:24:35,248

This has been another episode of Learning

Bayesian Statistics.

Speaker: 1287

01:24:35,248 --> 01:24:40,162

Be sure to rate, review, and follow the

show on your favorite podcatcher, and

Speaker: 1288

01:24:40,162 --> 01:24:45,056

visit learnbaystats .com for more

resources about today's topics, as well as

Speaker: 1289

01:24:45,056 --> 01:24:49,840

access to more episodes to help you reach

true Bayesian state of mind.

Speaker: 1290

01:24:49,840 --> 01:24:51,782

That's learnbaystats .com.

Speaker: 1291

01:24:51,782 --> 01:24:56,646

Our theme music is Good Bayesian by Baba

Brinkman, fit MC Lars and Meghiraam.

Speaker: 1292

01:24:56,646 --> 01:24:59,788

Check out his awesome work at bababrinkman

.com.

Speaker: 1293

01:24:59,788 --> 01:25:00,981

I'm your host.

Speaker: 1294

01:25:00,981 --> 01:25:01,821

Alex Andorra.

Speaker: 1295

01:25:01,821 --> 01:25:06,203

You can follow me on Twitter at Alex

underscore Andorra like the country.

Speaker: 1296

01:25:06,203 --> 01:25:11,326

You can support the show and unlock

exclusive benefits by visiting Patreon

Speaker: 1297

01:25:11,326 --> 01:25:13,447

.com slash LearnBasedDance.

Speaker: 1298

01:25:13,447 --> 01:25:15,828

Thank you so much for listening and for

your support.

Speaker: 1299

01:25:15,828 --> 01:25:18,139

You're truly a good Bayesian.

Speaker: 1300

01:25:18,139 --> 01:25:22,831

Change your predictions after taking

information in and if you're thinking of

Speaker: 1301

01:25:22,831 --> 01:25:27,572

me less than amazing, let's adjust those

expectations.

Speaker: 1302

01:25:28,503 --> 01:25:33,536

me show you how to be a good Bayesian

Change calculations after taking fresh

Speaker: 1303

01:25:33,536 --> 01:25:39,640

data in Those predictions that your brain

is making Let's get them on a solid

Speaker: 1304

01:25:39,640 --> 01:25:41,481

foundation