Artwork for podcast Learning Bayesian Statistics
#143 Transforming Nutrition Science with Bayesian Methods, with Christoph Bamberg
Episode 143 • 15th October 2025 • Learning Bayesian Statistics • Alexandre Andorra
00:00:00 01:12:55

Share Episode

Shownotes

Proudly sponsored by PyMC Labs, the Bayesian Consultancy. Book a call, or get in touch!

Our theme music is « Good Bayesian », by Baba Brinkman (feat MC Lars and Mega Ran). Check out his awesome work!

Visit our Patreon page to unlock exclusive Bayesian swag ;)

Takeaways:

  • Bayesian mindset in psychology: Why priors, model checking, and full uncertainty reporting make findings more honest and useful.
  • Intermittent fasting & cognition: A Bayesian meta-analysis suggests effects are context- and age-dependent – and often small but meaningful.
  • Framing matters: The way we frame dietary advice (focus, flexibility, timing) can shape adherence and perceived cognitive benefits.
  • From cravings to choices: Appetite, craving, stress, and mood interact to influence eating and cognitive performance throughout the day.
  • Define before you measure: Clear definitions (and DAGs to encode assumptions) reduce ambiguity and guide better study design.
  • DAGs for causal thinking: Directed acyclic graphs help separate hypotheses from data pipelines and make causal claims auditable.
  • Small effects, big implications: Well-estimated “small” effects can scale to public-health relevance when decisions repeat daily.
  • Teaching by modeling: Helping students write models (not just run them) builds statistical thinking and scientific literacy.
  • Bridging lab and life: Balancing careful experiments with real-world measurement is key to actionable health-psychology insights.
  • Trust through transparency: Openly communicating assumptions, uncertainty, and limitations strengthens scientific credibility.

Chapters:

10:35 The Struggles of Bayesian Statistics in Psychology

22:30 Exploring Appetite and Cognitive Performance

29:45 Research Methodology and Causal Inference

36:36 Understanding Cravings and Definitions

39:02 Intermittent Fasting and Cognitive Performance

42:57 Practical Recommendations for Intermittent Fasting

49:40 Balancing Experimental Psychology and Statistical Modeling

55:00 Pressing Questions in Health Psychology

01:04:50 Future Directions in Research

Thank you to my Patrons for making this episode possible!

Yusuke Saito, Avi Bryant, Ero Carrera, Giuliano Cruz, James Wade, Tradd Salvo, William Benton, James Ahloy, Robin Taylor,, Chad Scherrer, Zwelithini Tunyiswa, Bertrand Wilden, James Thompson, Stephen Oates, Gian Luca Di Tanna, Jack Wells, Matthew Maldonado, Ian Costley, Ally Salim, Larry Gill, Ian Moran, Paul Oreto, Colin Caprani, Colin Carroll, Nathaniel Burbank, Michael Osthege, Rémi Louf, Clive Edelsten, Henri Wallen, Hugo Botha, Vinh Nguyen, Marcin Elantkowski, Adam C. Smith, Will Kurt, Andrew Moskowitz, Hector Munoz, Marco Gorelli, Simon Kessell, Bradley Rode, Patrick Kelley, Rick Anderson, Casper de Bruin, Philippe Labonde, Michael Hankin, Cameron Smith, Tomáš Frýda, Ryan Wesslen, Andreas Netti, Riley King, Yoshiyuki Hamajima, Sven De Maeyer, Michael DeCrescenzo, Fergal M, Mason Yahr, Naoya Kanai, Aubrey Clayton, Jeannine Sue, Omri Har Shemesh, Scott Anthony Robson, Robert Yolken, Or Duek, Pavel Dusek, Paul Cox, Andreas Kröpelin, Raphaël R, Nicolas Rode, Gabriel Stechschulte, Arkady, Kurt TeKolste, Marcus Nölke, Maggi Mackintosh, Grant Pezzolesi, Joshua Meehl, Javier Sabio, Kristian Higgins, Matt Rosinski, Bart Trudeau, Luis Fonseca, Dante Gates, Matt Niccolls, Maksim Kuznecov, Michael Thomas, Luke Gorrie, Cory Kiser, Julio, Edvin Saveljev, Frederick Ayala, Jeffrey Powell, Gal Kampel, Adan Romero, Will Geary, Blake Walters, Jonathan Morgan, Francesco Madrisotti, Ivy Huang, Gary Clarke, Robert Flannery, Rasmus Hindström, Stefan, Corey Abshire, Mike Loncaric, David McCormick, Ronald Legere, Sergio Dolia, Michael Cao, Yiğit Aşık, Suyog Chandramouli and Adam Tilmar Jakobsen.

Links from the show:

Transcript

This is an automatic transcript and may therefore contain errors. Please get in touch if you're willing to correct them.

Transcripts

Speaker:

In this episode, I have the pleasure to sit down with Christophe Bamberg, a former LBS

collaborator and now a brilliant researcher straddling cognitive science, health,

2

:

psychology and data science.

3

:

We dive into his work on appetite regulation and cognition, including a Bayesian

meta-analysis of intermittent fasting and cognitive performance, what the data actually

4

:

say, how effects vary,

5

:

and why small, well-estimated effects can still matter for real people.

6

:

We also talk about research design, prior choices, model checking, and the day-to-day

realities of convincing collaborators to coabation.

7

:

If you care about bringing rigor to behavioral data without losing practical impact, this

conversation is for you.

8

:

This is Learning Visions Statistics, episode 143, recorded September 30, 2025.

9

:

Welcome Bayesian Statistics, a podcast about Bayesian inference, the methods, the

projects, and the people who make it possible.

10

:

I'm your host, Alex Andorra.

11

:

You can follow me on Twitter at alex-underscore-andorra.

12

:

like the country.

13

:

For any info about the show, learnbasedats.com is Laplace to be.

14

:

Show notes, becoming a corporate sponsor, unlocking Beijing Merch, supporting the show on

Patreon, everything is in there.

15

:

That's learnbasedats.com.

16

:

If you're interested in one-on-one mentorship, online courses, or statistical consulting,

feel free to reach out and book a call at topmate.io slash Alex underscore and Dora.

17

:

See you around, folks.

18

:

and best patient wishes to you all.

19

:

And if today's discussion sparked ideas for your business, well, our team at Pimc Labs can

help bring them to life.

20

:

Check us out at pimc-labs.com.

21

:

Hello my dear patients!

22

:

You remember that a few weeks ago, I sent you a survey to tell me what you would like to

hear if I were to ever make live cohorts to learn with you and also teach them?

23

:

Well, this is now happening!

24

:

If you want to learn patient stats live with me, now is the time!

25

:

Because I am excited...

26

:

to do that in partnership with Athletics.

27

:

And I'm going to be one of their early collaborator on their live feature.

28

:

And this is a company with which I share a focus on applied hands-on learning, especially

through Spots case studies, where patient methods, of course, shine.

29

:

And this partnership with Athletics means a few things, but first, it means short live

code first

30

:

workshops with recordings and GitHub repo included.

31

:

Thanks to athletics, that means pre-authenticated Google Cloud virtual machines so you can

model without set-up friction and without environment problems.

32

:

That means sports analytics projects, you can adapt to your team or your work.

33

:

And of course, an ongoing Q &A.

34

:

via my Learning Patients community, you will the Discord with all the patrons if you are

an alumni of the live cohorts.

35

:

We're kicking this off with hierarchical models, what else?

36

:

In PIMC on November 5th and 6th 2025.

37

:

This will be two short sessions, two hours and a half, where you will build an intro prep

38

:

several working multi-level models with posterior checks and stakeholder-ready visuals and

you'll have, as I was saying, all the bells and whistles.

39

:

I will put the links in the show notes for you to learn more and enroll if you want to.

40

:

Of course, if you are a patron of the show, have a 20 % discount that I added in Patreon

and the Discord, so make sure to paste the discount code.

41

:

check out and I'm really really looking forward to doing that first cohort and also to

building more intensives if this pilot helps you deliver impact quickly so please let me

42

:

know in the comments on LinkedIn, on Patreon, on Discord, wherever you can reach me please

let me know what your favorite topics are and what you would like me to teach next time

43

:

I'm really looking forward to this thank you so much for being

44

:

so excited about that, at least as excited as I am from what I've seen.

45

:

So thank you so much for your support and I can't wait to see you in this first cohort.

46

:

And in the meantime, let's now dive into nutrition science with Christophe Bam.

47

:

Christoph Bamberg, Willkommen zur Learningvation Statistics.

48

:

Very nice.

49

:

Thank you for the introduction.

50

:

It's like the only German I have left.

51

:

That's all you Nothing more.

52

:

I was in Berlin the other day and yeah, like, I didn't know my German was coming back, but

still, you it was always like, it's still far in my memory, but it's interesting to see

53

:

how the brain works.

54

:

Yeah.

55

:

guess still somewhere in my brain.

56

:

I can understand fairly decently that the issue is talking is way, way harder as all the

Yeah.

57

:

I mean, I could try my French now, but I don't think that's necessary on the podcast.

58

:

Yeah.

59

:

I will be nice to you and won't make you do the show in French.

60

:

That's better.

61

:

That's actually kinder on the listeners too.

62

:

So.

63

:

Yeah, probably.

64

:

um

65

:

I'm super happy to have you back.

66

:

I mean, not back on the show.

67

:

It feels like you're back on the show because like people may not remember it or know it,

but you were actually part of the LBS staff when you were still a graduate student.

68

:

Uh, I think.

69

:

And so it's been a few years now, but yeah.

70

:

So you were, um, helping out.

71

:

think you were one of the first person helping me out.

72

:

like making the episodes.

73

:

work better, you were in charge of all the marketing stuff that I'm very bad at.

74

:

um You were doing a lot of the stuff under the hood that people don't see that go into

making your podcast happen.

75

:

um So again, thank you so much for all the work you've done.

76

:

um And all the listeners, like if you know Christophe or connect with him on LinkedIn,

send him a message.

77

:

uh

78

:

thanking him for his service to the LBS community.

79

:

um And you were even a patron of the show.

80

:

And I think that's how we met uh and how you started helping me out.

81

:

um So that's really fun.

82

:

Now to have you as a guest on the show, that's like long overdue.

83

:

um And also you're doing a bunch of fun stuff now.

84

:

You just finished your PhD.

85

:

um So we're going to talk about all of the research you're doing, but yeah, maybe first,

you know, give us your origin story.

86

:

What are you doing nowadays and how did you end up working on that?

87

:

Yeah.

88

:

Yeah.

89

:

I also want to say it's, it's fun to be on this side of the podcast now.

90

:

um And it is over due, but I just wanted to wait until the scientific articles where I was

using Bayesian statistics actually were published, you know, and that

91

:

sometimes just takes a lot of time and now that's done.

92

:

So I think now is a good time to be here.

93

:

Yeah, my origin story.

94

:

yeah, it kind of meandered a little bit.

95

:

So I actually started with philosophy and economics in my bachelor degree.

96

:

Then I got more interested in the brain and the reasons why we behave how we behave.

97

:

So I studied cognitive science as a master degree.

98

:

And there in my master thesis, I did research on intermittent fasting and how it affects

our social behavior and the brains or with EEG.

99

:

um And the topic just stuck with me, but I wanted to see it more from a health

perspective.

100

:

And that's why in my PhD, I then went more into health psychology.

101

:

Yeah.

102

:

And there I again looked at intermittent fasting.

103

:

how it affects our concentrations or cognitive performance, which is very important on

daily tasks, but also to stay healthy, um to make good decisions and so on.

104

:

And also how it affects mood and sleep and those variables.

105

:

And yeah, now I finished the PhD and also kind of finished with this topic.

106

:

And I'm looking more at um self-regulation and how

107

:

or why we sometimes fail to follow our dietary goals um and how we can help people to

stick with their dietary goals despite the tempting environment that we live in.

108

:

And recently or soon, actually, I will also start working more as a data scientist, I

guess, like a clinical data scientist at a big mental health clinic in Germany.

109

:

yeah, in addition to continuing with the team here at the university of Salzburg where I

did my PhD.

110

:

Yeah, I think that's it.

111

:

many things, so many things to talk about.

112

:

Well, first, before we do that, um, you said you were using Bayesian stats in your papers

even.

113

:

you're a living proof that it is possible folks first and second.

114

:

uh

115

:

Do you remember what first drew you to Bayesian statistics and computational modeling in

general?

116

:

Yeah.

117

:

m Yeah.

118

:

Funny that you say that I feel like been proved that it's possible because sometimes it

was a bit of a struggle also with my co-authors to kind of convince them that it's

119

:

worthwhile because in psychology, unfortunately, it's not that common yet in many fields.

120

:

um But yeah, a colleague.

121

:

from the University of Auckland, David Moreau, with whom I published an article as part of

my um PhD.

122

:

He recommended the statistical rethinking from Richard McElreeve, um basically in my first

few weeks of the PhD.

123

:

And then I started reading it.

124

:

I think the first edition, I watched the video lectures, then I got the second edition,

watched the video lectures again because...

125

:

I feel like it's good to come back to this book because it's just very dense and very fun

to read.

126

:

Yeah.

127

:

And that's really my foundation for Bayesian statistics, I would say.

128

:

then I think, yeah, through the podcast, I also got to know the work of um Andrew

Gellermann, Aki Weteri, and who was the other author on regression and other stories?

129

:

yeah.

130

:

I think it's Jessica Holman you're talking about.

131

:

And also read that book and yeah, this is kind of the foundation of my patient skills.

132

:

Damn.

133

:

Pretty solid foundations.

134

:

Especially the podcast part I would say like, yeah, very solid source of information

you've got here, Christoph.

135

:

Yeah.

136

:

I think I would say from listening and working on the podcast, I mostly took away more,

less the...

137

:

practical things, but more kind of the behind the scenes, know, kind of how people share

their struggles with patient statistics and kind of the perspective on it.

138

:

think, yeah, that was good to listen to those conversations to kind of get an idea of

that.

139

:

Yeah.

140

:

Which I think is difficult in a book to transport this.

141

:

Yeah, for sure.

142

:

Yeah.

143

:

Yeah.

144

:

It's also why I'm making the podcast.

145

:

So I wasn't fishing for compliments.

146

:

was a joke, but I definitely take the compliment.

147

:

No, really.

148

:

I mean, I think I started listening.

149

:

Yeah.

150

:

At the beginning of my PhD, when all of this was still quite new.

151

:

um And then it was just good to hear people who actually develop the methods also talk

about how difficult it is for them and to just see that they're also just human.

152

:

Right.

153

:

That's really gave me more confidence that I can also use patient statistics.

154

:

Yeah.

155

:

Yeah, that's a good point.

156

:

Yeah.

157

:

That's also why I love doing that is that you often feel, you often feel alone and lost

when you start learning something or even like just when you model.

158

:

Like modeling is an inherently hard endeavor.

159

:

especially Bayesian modeling, because a lot of the things are not working out of the box.

160

:

That's kind of the idea.

161

:

And so you have to customize everything and that can be very intimidating.

162

:

And so I think having people who are regarded as experts come on the show and saying that,

no, no, you know, like I struggle too.

163

:

It's just like more advanced on the learning curve than you are, but the struggle is still

here.

164

:

Don't worry.

165

:

So this is very important, I think.

166

:

ah And talking of struggles, um do you remember actually what were the main obstacles you

were encountering?

167

:

when talking with your coauthors or even people now who I'm guessing are asking you, what

is that whole patient stats about?

168

:

Maybe even reviewers.

169

:

um What was the kind of obstacles you had and that you apparently successfully overcame?

170

:

That's a good question.

171

:

Well, regarding the reviewers, I must say I expected more obstacles, but I think the

problem there is that in health psychology, are just not many

172

:

people using patient statistics.

173

:

So I also didn't get many comments um regarding my methods, unfortunately, I would say.

174

:

So I kind of had to go to colleagues and so on to make sure that a second pair of eye also

looked at my patient methods.

175

:

um Yeah.

176

:

So that maybe is a small side remark.

177

:

m The main obstacles.

178

:

Well, yeah, with my co-authors, it just seems like more work, you know, and it's

179

:

Oftentimes on the first glance, leads to the same results.

180

:

And then if that's the case, like if the coefficient in the regression model, frequentist,

or Bayesian is more or less the same, why would you then put the extra work in?

181

:

Yeah, I think that was difficult to explain to my co-authors that it just happens to be

the case that the frequentist model also works.

182

:

But the Bayesian one is actually better and encompasses the frequentist results as well.

183

:

um With my colleagues, um so in my team, I'm more or less the only one who's really

enthusiastically using Bayesian statistics.

184

:

Some of my colleagues use it because the frequentist models fail, um because we use

complex multi-level models.

185

:

uh

186

:

And yeah, there, think the struggle is kind of with the priors for some reason.

187

:

So somehow some of my colleagues and also people I talk to are kind of, yeah, stuck with

this part um and kind of figuring out how to find reasonable priors.

188

:

Even though I would say in the beginning, that's not even so important.

189

:

I would say from my experience, right?

190

:

Yeah, No, for sure.

191

:

It's funny that it's often something people overthink and over engineer at the beginning,

because it's not that important.

192

:

And I mean, if it were, actually, think if it were, frequent these models wouldn't work

that well out of the box.

193

:

of the time, you know, because like most of the time these products are really stupid

because they're completely wide and they allow for really, really weird results a priori

194

:

and they still work.

195

:

very well most of the time.

196

:

So if priorities were that important, um, everybody would be using base actually.

197

:

So, you know, um, so basically from what you're saying, I hear that, um, the fact of

having hierarchical models and, and scarce data was something that was very important in

198

:

your motivation to

199

:

Use patient stance.

200

:

Is that correct?

201

:

I think right now it's one of the motivations I have to use it.

202

:

Yeah.

203

:

And just this ability to um look at changes and um effects within an individual since we

are doing psychological research, but then also get a good estimate of the population

204

:

effects.

205

:

Yeah.

206

:

I patient statistics is just quite nice for that, especially for the kind of individual

level predictions.

207

:

really like it because yeah, I mean, if you use frequent statistics at some point, you

would need to bootstrap anyways.

208

:

And then you might as well start with patient statistics, right?

209

:

um Yeah.

210

:

Yeah.

211

:

Yeah, for sure.

212

:

um I'm curious actually, which technical stack um are you personally using when you're

213

:

working on your, on your models and also...

214

:

you're cut out for a second there.

215

:

yeah.

216

:

Okay.

217

:

Could you repeat the question?

218

:

Yeah, no, for sure.

219

:

I'm curious actually, which technical stack you're personally using when you're uh

modeling for your papers and your studies.

220

:

And also which technical stack do you see your...

221

:

colleagues pick up most of the time and with which they have the best relationship as

beginners and why?

222

:

So my workflow is usually depending on the data, but if I need to do a lot of

pre-processing, like with the cognitive tasks that I'm using where I get data per trial

223

:

and I need to aggregate that somehow, then I usually use uh Python and

224

:

Yeah, just kind of the standard Python libraries to pre-process it and then put it into a

nice data frame.

225

:

And then I switched to R.

226

:

Yeah, I don't know whether that's a little bit hacky, but it works well.

227

:

And in R I usually use BRMS.

228

:

um Now I also, over the summer, I sat down and studied Stan a little bit more in detail.

229

:

And I think I will kind of switch to using that uh more in the future.

230

:

Um, but I usually start with BRMS and yeah, just for the standard regression models,

that's usually works quite well.

231

:

And that's also what my colleagues use because in psychology, we mostly use R and the,

what is it called?

232

:

LME R or LME 4 package for multilevel modeling.

233

:

Yeah.

234

:

LME 4 I think.

235

:

Yeah.

236

:

And yeah, that's just the...

237

:

similar syntax to BRMS.

238

:

So the switch is actually quite easy to do.

239

:

Basically, you just have to change a few letters and then you already have your Bayesian

model.

240

:

Yeah, that makes sense.

241

:

And definitely recommend BRMS for everybody.

242

:

I will put in the show notes episode 35, which was with Paul Burkner, the founder of BRMS

and Paul is in...

243

:

extremely brilliant researcher and definitely recommend checking out anything he's doing.

244

:

um Also great person.

245

:

So I will put that into the show notes.

246

:

I will also put episode 112, which was with Tomica Preto, who is the main developer of

Bambi, which is the Python equivalent of PRMS for people who want to do that, but in

247

:

Python.

248

:

which I guess is an important language right now.

249

:

I heard some people were using it from time to time, you know, but I'm not sure.

250

:

uh Actually, let's dig deeper into your research, Christophe, because you're actually

doing research on topics I personally found super interesting.

251

:

And actually I will put also an episode in the show notes with um Eric Traxler.

252

:

who is a researcher in this field who does very interesting work too.

253

:

And I see these guys using more and more Bayesian stats, think.

254

:

So this is great.

255

:

ah This definitely warranted because they usually have experiments with like 10 to 30

people.

256

:

ah So you definitely want to use Bayesian stats with solo sample sizes.

257

:

So you work a lot on...

258

:

appetite and appetite regulation and how, I don't know if you work a lot on also the

interaction with, with, uh, exercise, but, um, you, I don't know you work a lot with the

259

:

interaction with the psychology.

260

:

So you had a recent paper about appetite that looks at how dietary claims affect

cognition.

261

:

So can you talk about the study, what motivated did

262

:

and what the main findings were.

263

:

Yeah.

264

:

Yeah.

265

:

I really liked the study because it's, it's just fun to talk about this one.

266

:

So the, motivation was em maybe, you know, this saying em that breakfast is the most

important meal of the day, right?

267

:

Like basically everyone knows it.

268

:

And usually we would say, yeah, okay.

269

:

That's based on the wisdom of uh our grandparents or something like this, right?

270

:

But actually.

271

:

It's just an advertisement from 1917, I think, um invented by a colleague of Kellogg's.

272

:

So the Kellogg, the creator and inventor of the breakfast cereal, just to convince people

that they need to eat more breakfast and buy his cereals.

273

:

And when I found this out, I just found it very interesting because I think this ah

advertisement actually affected a lot of people and kind of their behavior and their

274

:

beliefs.

275

:

And I just wanted to see whether, yeah, this can also be found experimentally and not just

in their beliefs, but also in their performance, right?

276

:

So the actual ability to concentrate.

277

:

um And so what we did there was ah it was in uh collaboration with Anne Rolfs from the

Maastricht University in the Netherlands.

278

:

um And what we did was we had half of the participants um

279

:

start the experiment when they were fasted.

280

:

So they didn't eat for 16 hours.

281

:

The other half had their regular breakfast and lunch.

282

:

And then we split those two groups again so that we have four different groups.

283

:

And in the group that fasted, half of the participants who were fasted, they read a

statement saying that um fasting is good for concentration.

284

:

because it increases your seeking behavior and that's important on a lot of cognitive

tasks.

285

:

So kind of framed as a scientific finding, you know?

286

:

And the other half read a statement that fasting is bad for concentration because you're

distracted by your hunger and then you can't perform so well.

287

:

And then in the satiated groups or those that ate, half of them had the statement that

being full gives you energy so you can concentrate or that being full uh takes energy to

288

:

digest and then you can't concentrate.

289

:

And then everyone did the same psychological task, the Simon task, where you have to um

press either one button or the other one, depending on where colorful squares are on the

290

:

screen.

291

:

And it's a little bit challenging because um you have to inhibit your initial response

based on the location, but focus on the color.

292

:

And so you need to concentrate on that.

293

:

And what we found was an interaction between the two manipulations such that

294

:

If you're hungry and you expect to perform better, the performance was actually

objectively better than if you were hungry and you expected to perform worse.

295

:

And the same in the group that was full.

296

:

So if you were full and you expected to perform better, you were better.

297

:

If you were full and you thought you would perform worse, you also performed worse.

298

:

And I just find this amazing that you can show this.

299

:

The effects were very small.

300

:

Right?

301

:

So in general, participants were quite good at this task and the effects were small, but I

still find it interesting that comparing the groups, it was visible like this.

302

:

Yeah.

303

:

Yeah.

304

:

Yeah.

305

:

So it speaks to the power of the brain mainly here, right?

306

:

So it's that really the framing, the framing you gave to the participants really had an

impact albeit small on the cognitive performances they had.

307

:

a correct interpretation on my head, on my end?

308

:

Yeah, exactly.

309

:

Kind of like a placebo effect.

310

:

Right.

311

:

Yeah.

312

:

Yeah, that's fascinating.

313

:

um And so, yeah, what do you, what, so that, there are two roads I'd like to explore here.

314

:

First, um what do you make of that?

315

:

Like what are the implications of this research, especially in psychology and now that

you're going to work in a clinical environment.

316

:

You know, why, why, which implications does it have if any?

317

:

Yeah.

318

:

I would say mostly worrying implications because if we again think of advertisements um

and see that a statement like this, was really like three sentences or so that

319

:

participants read that this can already have an effect on behavior.

320

:

Yeah, I find it a bit worrying, especially if there's a topic.

321

:

so important like eating.

322

:

So I think we should be very careful with advertisement uh in the food context.

323

:

um And also I think it can also be leveraged in a positive way.

324

:

So imagine you recommend someone to do intermittent fasting for health reasons.

325

:

If you frame it positively, then maybe um just this positive framing already has a

326

:

has a positive effect on the individual as well.

327

:

um yeah, I think it can be ah helpful in the therapeutic and in the health recommendation

context as well.

328

:

Yeah.

329

:

Yeah, that makes a ton of sense.

330

:

Yeah, I actually have to listen to an episode of a podcast that ah I listen to all the

time, which is called Econ Talk.

331

:

And this week, the episode is about eating with intelligence.

332

:

Um, and the guest is someone named Julia Bellos.

333

:

And I know at the end of the episode, they were, they talk about exactly these public, um,

public policy recommendations for, uh, food health.

334

:

And so I can't tell yet what, what this is about, but that sounds very interesting and

related to what you're talking about.

335

:

And I will put these.

336

:

this episode in the show notes, because I'm sure it will be relevant to the conversation.

337

:

But something I'm curious about here is also what does the model look like for these kinds

of experiments?

338

:

know, what the technical road now, let's take that one and talk about that.

339

:

So yeah, how does it work to, you know, come up with such a study?

340

:

What does it look like for you?

341

:

Yeah.

342

:

um

343

:

So in the end, we just used the base factors to compare the four groups with an

interaction model.

344

:

um In the beginning, I actually took the time to kind of come up with a causal model in a

directed acyclic graph, ah generate simulations from it, test the model on it,

345

:

pre-register it.

346

:

um

347

:

The final model did look different.

348

:

think I just pre-registered em linear regression.

349

:

So I plan to look at the coefficients in the end.

350

:

looked at the base factors.

351

:

m But it was still, I think, good to go this extra route um and look at the potential

causal relationships there.

352

:

Wow.

353

:

Okay.

354

:

This is amazing.

355

:

This is like, yeah, you used, I can see you really, really...

356

:

Um, do your homework and listen, at least listen to the show and read Richard McElweth

because this is using all the latest best practices.

357

:

so yeah, kudos on that.

358

:

That's amazing.

359

:

um can you walk, can you walk listeners actually through your workflow here, especially

what does it look like to come up with the causal lag?

360

:

Um, which tools and packages did you use for that?

361

:

And, and also the analysis then, um,

362

:

that you had in mind before even looking at the data.

363

:

I think that part is often not talked about enough.

364

:

And so I think having an illustration of these for people, it's going to be super

interesting.

365

:

And also we definitely need to put um links to our paper.

366

:

And if you have a GitHub repo of all the analysis, put that in the show notes.

367

:

Yeah.

368

:

Yeah.

369

:

We'll make sure to do that.

370

:

Yeah.

371

:

So I'm actually recently thinking again, more about this workflow.

372

:

uh, I first heard about from Richard McElwreath.

373

:

And then I also read the books by Judea Pearl on, on DAX.

374

:

Um, but yeah, basically what we did was, so it was a relatively simple experiment, right?

375

:

Like four conditions, just one measurement.

376

:

So, um, I didn't use, Degeti or any other, um, packages, to,

377

:

come up with the deck and then to sort of say analyze it.

378

:

I just did it on the paper basically.

379

:

um And then, yeah, for the simulation, I just tried to translate the relationships into

functions and then come up with reasonable values.

380

:

um So, yeah, and then also ah kind of data sensitivity analysis.

381

:

checking the values whether that actually makes sense.

382

:

And then just ran a linear regression with an interaction term on the simulated data to

see whether it can recover the starting parameters as well.

383

:

And when I was happy with this, I pre-registered that.

384

:

um In the end, I only changed how I presented the results with the base factors, but the

model was the same, right?

385

:

Like it was still an interaction term.

386

:

But um base factors are just more understood, I think, in psychology.

387

:

So it was kind of easier to um communicate the findings to the audience with that.

388

:

um Yeah.

389

:

But I think now, if we want to talk about this, I would add one step before, right?

390

:

So because I just...

391

:

Yeah, came up with the DAG ah based on my intuition and reading the literature, but m kind

of a more recent topic is also the concepts that we use and how precise they are, right?

392

:

Because if we use imprecise concepts in our directed acyclic graphs, what we get out is

not that helpful, right?

393

:

So if my concept of, let's say, m craving, which is very important in eating behavior, if

that is not clear,

394

:

And that's the main uh outcome that I look at.

395

:

Then, yeah, we kind of have a problem, right?

396

:

thinking more about the concepts, I think that's, that's something where, uh, have

psychology and maybe research in general should go more and, you know, kind of going back

397

:

to the theories that we base our decks on, right?

398

:

I feel like sometimes, at least I am sometimes a little bit too fast, kind of reading the

literature and then coming up with the deck, maybe revising the deck.

399

:

But the single elements, I don't think enough about that, I think.

400

:

Kind of taking the time to really think about, what does this edge actually mean in the

graph?

401

:

um Yeah, I think that's an important next step.

402

:

Yeah, I'm actually very curious about how you go through the DAG process.

403

:

So this is something also amusing more and more in my own work.

404

:

For people using Python, I find that using a combination of PGMPy and NetworkX is a very

good way of coming up with the DAG and coding it up in Python, having it up in your

405

:

Jupyter Notebook.

406

:

uh So yeah, this is something I'm doing more and more.

407

:

Now that I'm going through the causal AI course from Robert Ness, will link to this

episode.

408

:

That was episode 137.

409

:

I will link to that one too.

410

:

um And I will actually make a demo of causal inference workflow in the Python invasion

framework when at some point in one of the live cohort courses that I'm starting to set

411

:

up, I know it's a popular topic.

412

:

yeah, I'm developing content around that, but yeah, can...

413

:

Verge for that workflow in Python that's usually very useful.

414

:

And the way I come up with the DAG is usually the same as you.

415

:

Once I have my literature review done, then I come up with the first DAG.

416

:

then depending on how well I know the domain, I speak with domain experts, which is what

I'm guessing you've done.

417

:

Yeah, how does it work for you in these cases?

418

:

Yeah.

419

:

I mean, to some extent, I guess I am a domain expert in the topics that I look exactly.

420

:

But em yeah, I want to formalize this more.

421

:

uh For example, now in my work, I um also work a lot with asking people about their

craving, you know, because actual food intake em is difficult to measure, right?

422

:

Because maybe you actually want a snack something, but there's no snack available.

423

:

then we don't know this, but it's still relevant.

424

:

So we asked them about their craving.

425

:

And yeah, every time I talk with other researchers on this topic, they have a different

definition of craving, right?

426

:

But we kind of all do research on it and we all say, okay, the motions are related to

craving, stress is related to craving, but it's a different kind of craving.

427

:

And so it would be nice to formally map kind of the different definitions that are out

there.

428

:

There's a nice framework called Decentralized Construct Taxonomy ah by Kjaldjorn Peters

and probably some co-authors um where you can then kind of specify your definition, how to

429

:

measure the construct with this definition.

430

:

And then you upload it and then it's visible and you can link to it.

431

:

And yeah, I think that's quite helpful in psychology.

432

:

So yeah, I would kind of have this round of asking other experts to properly define the

constructs like this before I use it then in my DAG.

433

:

And then I could link to this specific definition in my DAG, you know, and I think with

that, would be much more precise and easier to replicate.

434

:

So I think that would be ideal.

435

:

Hopefully I would, I will, I will make this.

436

:

possible in my next studies.

437

:

Yeah.

438

:

Yeah.

439

:

Yeah.

440

:

That sounds, that sounds indeed super helpful.

441

:

ah Feel free to add links to show notes about that, by the way.

442

:

um And so you also have, you also have a pre-print you told me for a Bayesian

meta-analysis on the effects of intermittent fasting on cognitive performance.

443

:

And this is a very interesting.

444

:

uh

445

:

topic that I've read a bit about.

446

:

so yeah, what can you tell us about that and what are the takeaways for listeners?

447

:

Yeah.

448

:

m Yeah, I'm quite happy I could do this study.

449

:

was the first study I started as my PhD and the last one to be published because

meta-analyses are just quite a lot of work.

450

:

This was in uh collaboration with David Moreau from the University of Auckland in New

Zealand.

451

:

And yeah, we basically, I think we scanned 17,000 experimental studies that looked at some

kind of fasting.

452

:

might be skipping breakfast.

453

:

It might be intermittent fasting, fasting for 24, 48 hours.

454

:

So very long fasting sessions as well.

455

:

And then some objective measure of performance.

456

:

So a task like the Stroop task, maybe some listeners know this.

457

:

em

458

:

similar tasks like this.

459

:

And then we put the primary research findings into a multi-level model with m several

levels in this model and um also ran this in BRMS.

460

:

uh what we found out in the end, and I think their patient statistics just really shines,

was that there was no substantial effect basically.

461

:

the

462

:

The overall em effect was negligible.

463

:

But what was interesting was that the posterior, so the m density of the posterior kind of

spanned uh quite a range.

464

:

in the end, the interpretation would be that mostly if you eat, you might expect slightly

better performance.

465

:

But there was also around 30 % of the posterior on the other side, basically, so that

better performance would be expected if you don't eat.

466

:

Which kind of shows how much heterogeneity there is between the studies and in the

research on this topic.

467

:

And then we did some follow-up analysis to explain this in more detail.

468

:

It depended on the age of participants.

469

:

Seems quite clear that if you're young, not um younger than 18, that yeah, based on those

studies, it's good to eat breakfast and to not fast.

470

:

But if you're older, then there's not much of a negative effect at least.

471

:

um We also saw that the fasting duration played a role in quite an interesting way because

for longer fasting durations,

472

:

the faster participants actually got better.

473

:

But that again, I think depends on what kind of study it was.

474

:

If it was framed as intermittent fasting study with a longer duration, then participants

also expected better performance.

475

:

And maybe that's why they performed better.

476

:

But if it was a study with a short duration, like if you skip a meal, how does your

concentration deteriorate?

477

:

Then you also expect worse performance.

478

:

Yeah.

479

:

So kind of linking back to the first study we talked about.

480

:

um Yeah.

481

:

And I was quite happy that I could use patient statistics for this one because to really

display visually this uncertainty around the estimate that is inherent in meta-analysis.

482

:

um think patient statistics is just perfect for that.

483

:

Yeah.

484

:

Yeah.

485

:

Yeah.

486

:

It makes a ton of sense.

487

:

And so like practically

488

:

What do you recommend people who are interested in intermittent fasting?

489

:

It sounds like you can pretty much do whatever you want.

490

:

It's not gonna have a huge effect.

491

:

Just do whatever works best for you.

492

:

Like if you're not hungry in the morning, well, you can do this intermittent fasting where

you don't eat before your lunchtime.

493

:

Because that's just how you're naturally wired.

494

:

If on the contrary, you're not very hungry at night.

495

:

ah

496

:

Which here, I think the science is clear on what the benefits are of not eating too much

at night.

497

:

Actually, um the positive effects it has on your sleep uh will then definitely lean into

that natural instinct you have of mostly eating a lot during the day, but not a lot at

498

:

night.

499

:

I mean, in the evening.

500

:

So like, you know, what do you usually say in a clinical context or, know, in your

studies?

501

:

Yeah.

502

:

So we actually did find this effect of the time of day.

503

:

So um if the fasting was later in the day, then the performance got worse.

504

:

So I'm kind of in line with what we already know.

505

:

um Well, I can't make any um proper recommendations, right?

506

:

Since I'm not a healthcare professional, but um basically I think if you are an adult

who's in general healthy,

507

:

I would say try it out and be open about the results.

508

:

And yeah, I don't think it would hurt because what we found was that on average there was

no effect and that can be again, positive and negative.

509

:

It can mean maybe you don't get much of a benefit for your concentration, but also your

concentration doesn't get much worse.

510

:

Right?

511

:

So if you're doing fasting for health reasons,

512

:

there's also good evidence for this, then at least you don't have to worry much about em

worse cognitive performance.

513

:

And again, that's the average, right?

514

:

So em yeah, you have to see for yourself what feels good and what makes sense for you.

515

:

Yeah.

516

:

Yeah.

517

:

Yeah.

518

:

So that's cool because that means like basically, yeah, all these claims about like the

magic power of

519

:

magic healing power of fasting or the fact that it's really bad, especially if you do it

in the morning and skip breakfast.

520

:

Well, most of that is overblown and like you can mostly lean into your natural eh instinct

and your body is actually really good at regulating it.

521

:

It's an appetite basically.

522

:

think that's, that's one of the takeaways here.

523

:

Yeah, absolutely.

524

:

Also from my personal experience, because I'm doing intermittent fasting, so I'm skipping

breakfast.

525

:

And what I found was that I'm just more in tune with my um natural hunger rhythm, I would

call it.

526

:

So I think just for that, it's already good to try it out and then, yeah, to get a better

feeling of when you actually need to eat and when not.

527

:

um Yeah.

528

:

Yeah.

529

:

It's a very good point because it's the same for me actually, like, um

530

:

I've always been told to eat mostly in the morning.

531

:

um And then not too much at night.

532

:

the not too much at night part is actually relevant.

533

:

um But the morning part, not really, um if you care about science.

534

:

so yeah, like naturally, so me naturally in the morning, I'm not, I'm not hungry.

535

:

I'll just get coffee and then I'll get hungry for lunchtime actually.

536

:

That's not like I'm not forcing myself to do that.

537

:

It's just super natural.

538

:

It's actually way harder for me to eat in the morning than not.

539

:

um like, it turns out I'm just doing intermittent fasting, you know, but it's just

naturally doing that.

540

:

without um feeling guilty because I'm not eating breakfast.

541

:

So that I think that's the most important part here.

542

:

ah What's definitely harder for me is at night.

543

:

not eating too much for dinner.

544

:

Because I do care a lot about sleep and that's something I'm optimizing a lot.

545

:

Because when you have an intellectual job, sleeping well is, I'd say, the most important

thing you should care about.

546

:

And so I'm trying to optimize a lot for that.

547

:

But that's very hard for me because in the evenings, I have lot of cravings and I want to

eat a lot.

548

:

And so that's just my natural way of doing things.

549

:

uh

550

:

So this is hard.

551

:

I don't know if you're going to do some studies about that, but if you do, do let me know

because I'll be interested in it.

552

:

Yeah, I actually recently finished a daily diary study where we also looked at this.

553

:

we asked college students or university students um to report when they were having

snacks, when they were craving, how stressed they were and whether they were in a bad or

554

:

in a good mood.

555

:

And what we found was.

556

:

that um well, overall, lot of the calories are consumed later in the day.

557

:

I think that that's already kind of an established finding and that we, it's called

grazing, right?

558

:

Like a cow, kind of just eat throughout most of our waking hours.

559

:

So this we found, but also that um negative mood had a stronger effect in the evening so

that there was more snacking in the evening if um our participants were in a bad mood.

560

:

So yeah, it's a difficult window of the day if you want to be healthy.

561

:

Yeah, but I don't think there's such a good solution.

562

:

I mean, you can try to just, yeah, not buy unhealthy snacks and so on, but then you still

have the cravings and that can also still get challenging.

563

:

yeah.

564

:

Further research should be done on that, I would say.

565

:

Yeah, yeah, yeah.

566

:

I'm going to be super interested in that for sure.

567

:

um So actually now um switching gears a bit, and I put off cost of meta-analysis

pre-printed in the show notes, but I'm also curious how you balance experimental

568

:

psychology with statistical modeling in your research, because you definitely do.

569

:

both of those.

570

:

so, yeah, how does it work for you?

571

:

you mean how I spent my time on?

572

:

Yeah, exactly.

573

:

And how, I'm guessing also there are some moments where, um, you would want to use more

complex methods, but you also have to keep in mind that you need to explain them to

574

:

people.

575

:

And so, yeah, how do you walk that?

576

:

How do you thread that needle?

577

:

Yeah.

578

:

That's a good question.

579

:

mean, I wouldn't say that I have a perfect answer for it, but, um, so I mean, what I like

is that I usually have several projects at different stages.

580

:

So in one, I'm kind of designing the study in the other one, the data collection is going

and then I have some data lying around.

581

:

So based on that, I can always kind of do everything at the same time, which is nice

because if I.

582

:

If I'm coding for two or three hours, it's nice to take a break and just to think about

the next study or to write up an article.

583

:

So that's good.

584

:

um yeah, I think ideally the experimentation and the statistics, they should be quite well

linked, right?

585

:

So that I use um statistical models that make sense for the research design that I have.

586

:

And that I design my study in a way that I can actually get the data to answer an

interesting question.

587

:

m Yeah.

588

:

So I think that's the challenge.

589

:

And if you're a researcher that you have to wear many hats um and also link the things

very well together.

590

:

um And there's just a lot to learn all the time because you have to know statistics to

some extent, you have to read a lot.

591

:

to keep up with science, you have to improve your writing skills.

592

:

um Yeah, it's a challenge, I would say, to answer your question.

593

:

yeah, I think in terms of workflow, it really helps to think of the workflow we talked

about earlier, right?

594

:

So that I kind of mentally map kind of a cycle.

595

:

And then I'm like, okay, for this study, I'm currently reading the literature for this

study.

596

:

of analyzing, so to kind of keep track of what I'm doing.

597

:

yeah, I always try to optimize my time with stuff like the, what is it called?

598

:

The Pomodoro method, right?

599

:

To kind of break up the time and the day into discrete small chunks.

600

:

I think that's very helpful.

601

:

But yeah, I mean, as a scientist, just have to be good at project management.

602

:

I think, just keeping track of a lot of different things.

603

:

While there are always too many meetings to talk about the next study or collaborations,

which are also important, right?

604

:

But still there are always too many meetings.

605

:

But I think, yeah, I would say I mostly spend the time writing coding.

606

:

And talking probably.

607

:

Yeah.

608

:

Those are the three main things I would say.

609

:

Yeah.

610

:

Yeah, that makes sense.

611

:

And what's your, what's your favorite part?

612

:

Like personally?

613

:

I must say, really like making nice graphs uh for the articles.

614

:

Like I can, I can spend a whole day on that.

615

:

Yeah, me too.

616

:

Right.

617

:

Like.

618

:

Because that's just the culmination of maybe years of work that then you just see

visually.

619

:

That's just very nice.

620

:

I also enjoy teaching because it's so different from the other things, but it's still very

related.

621

:

if I taught some classes on basic research methods and introduction to statistics, and

that just helped me actually understand uh research.

622

:

Yeah.

623

:

On a much deeper level, right?

624

:

I think once you teach something, you just understand it better yourself.

625

:

em Yeah.

626

:

So I think that's, that's always a nice change.

627

:

then, yeah, I also really enjoy just blocking a full day just to write.

628

:

em That's, it's nice to get it in a, to get in a rhythm of this as well.

629

:

Yeah.

630

:

So yeah, again, it's the mix is the nice, the nice thing about being a scientist, I would

say.

631

:

Yeah.

632

:

Right.

633

:

Yeah, yeah, no, for sure.

634

:

And definitely, definitely resonate with everything you just said.

635

:

um These are also activities that I thoroughly enjoy.

636

:

And actually, um like talking about your field in general, I'm curious in your opinion,

what do you are the most pressing, open questions at the intersection of diet, cognition

637

:

and statistical modeling?

638

:

Yeah.

639

:

um

640

:

So maybe going a bit more broadly in health psychology, which is at least in Europe, it's

a, would say relatively small field, at least in Germany, Austria, where I'm based.

641

:

um We kind of look at how to improve the health of ah people before it becomes clinically

relevant.

642

:

Right.

643

:

So a very strong focus on, yeah, on the everyday life.

644

:

um And there, I think it's,

645

:

Maybe it's because I'm currently um thinking a lot about this, but I would say em coming

up with more precise theories and terms is a pressing issue because like we have open

646

:

science now.

647

:

most researchers, would say, do pre-registrations, registered reports and all of this,

which already improved the field a lot.

648

:

um Some are using patient statistics, although I would say that in psychology it's

649

:

it will still take some time until it really arrived there.

650

:

um But we are very theory-based in what we do.

651

:

And our theories are just not precise enough for the kind of predictions we want to make.

652

:

um And then we just, we can't go forward.

653

:

We are just kind of walking on the same spot.

654

:

And yeah, I think that's just the main topic to improve.

655

:

Okay.

656

:

Yeah.

657

:

So an interaction of a lot of different factors.

658

:

ah Yeah.

659

:

Building on that and also on your teaching.

660

:

Well, actually I'm curious what are you teaching right now and are you already teaching

any Bayesian module in your teaching?

661

:

I'm guessing you do, but I'm curious.

662

:

um So as I said, I taught some basic research method classes.

663

:

But recently I gave my first workshop on patient statistics at um a conference on health

psychology.

664

:

So for researchers, um PhDs or post PhDs.

665

:

was- Thank you.

666

:

Yeah.

667

:

That was a lot of fun.

668

:

Also quite a challenge because I only had around two hours.

669

:

So it was a very short workshop, but um yeah, it was a good way to-

670

:

really focus on the most important things for health psychology researchers.

671

:

So not dwelling on the algorithms and so on and kind of giving them some tools to easily

find priors, also not dwelling too much on this, kind of giving them a workflow so they

672

:

went home with a R script that they can just adapt to their needs.

673

:

um Yeah, that was quite fun.

674

:

then now the semester started here now I'm also going to teach uh

675

:

as a lecture on, we called it mega trends in um nutrition, movement and health.

676

:

So more like a topical lecture, not so methods heavy, which will also be fun.

677

:

But I definitely enjoyed the methods workshop quite a lot.

678

:

uh

679

:

How, what were the main reactions?

680

:

um Well, I think for many, was a bit overwhelming in the moment.

681

:

um they, so my goal was also not to...

682

:

like give them the answers in two hours, right?

683

:

It was more to show them where the questions are and um to give them some things to look

at again at home.

684

:

And I think that that went quite well because from my experience, I just need more time

than you have in a workshop to really work through the examples and the code and so on.

685

:

And yeah, I think it was quite a good approach.

686

:

So yeah, the reactions were, yeah, a little bit intimidated at first, but then afterwards,

carefully optimistic, I want to call it.

687

:

Yeah.

688

:

Because I think most of us know that it would be better to use patient statistics, but

it's just like jumping into cold water sometimes.

689

:

Yeah.

690

:

And the incentives are not really here.

691

:

Right.

692

:

like, would you take so much time to learn some new method where you're not especially

going to be rewarded by it, except for in the long-term having maybe better results, more

693

:

accurate, more precise and or more predictable.

694

:

But that's something that's very much down the road.

695

:

And so in the meantime, you still have to learn the methods, takes time and it takes time

out of you writing and publishing papers, which is what researchers care most about.

696

:

um But I'm curious also, were there any main themes in the things that people, like in the

questions people had for you during and after the workshop?

697

:

Yeah.

698

:

So again, one thing were priors that they were unsure how to set them.

699

:

But I think we managed to cover that quite well.

700

:

then em yeah, the distributions you can use for modeling because in psychology we use uh

normal distribution and the logistic or Bernoulli distribution and that's it.

701

:

But at least in our learning and, but in practice, especially if you kind of ask

participants about their momentary um

702

:

craving or the momentary mood or something like this, the data is just not normally

distributed.

703

:

Right.

704

:

And then it's just, then you kind of have this box of different distributions you can

choose from.

705

:

And then it's just difficult to find the right one.

706

:

And then if you look at the formula describing it, it's a bit daunting.

707

:

um So yeah, kind of how to have a good approach for finding a fitting distribution.

708

:

was also a big topic.

709

:

Yeah.

710

:

m

711

:

Yeah, it is indeed a common topic that intimidates beginners, but it's good in a way

because it's not that hard.

712

:

you know, I think it's easy for people to pick it up, like, especially since you, at some

point you see, you how much of the, like, you don't need that many distributions and in 99

713

:

%

714

:

of the time, it can be the exponential family distributions and you don't have to care too

much about the other ones.

715

:

yeah, thankfully this is much, much easier.

716

:

Yeah.

717

:

And I think um since you mentioned this, I feel like we can also look more closely at the

distributions we are using.

718

:

So I recently also started thinking about mixed effects location scale models, if you're

familiar with that.

719

:

So basically where we just also model the standard deviation in a normal distribution.

720

:

And it's just an innovation framework.

721

:

It's such a small tweak, right?

722

:

You just basically add one link function and one linear model and that's it.

723

:

But it gives you so much more, right?

724

:

And just kind of really thinking about the distribution and what you can do with it um

already can get you very far before you have to look for some very

725

:

strange and complex distribution that not many people are using.

726

:

oh Yeah.

727

:

Yeah, yeah, exactly.

728

:

um That's indeed very, very interesting.

729

:

And feel free to add any links to that in the show notes for listeners.

730

:

And I'm also curious, what advice do you give your students who want to bring more patient

thinking into psychology?

731

:

That's a good question.

732

:

um So for example, the PhD students in our lab, some of them also think about using

patient statistics.

733

:

And I would just say it's a lot about your approach to failure and to kind of hard things,

right?

734

:

That you're fine with struggling and

735

:

that it's okay if it doesn't work on the first try on the second try and so on.

736

:

So it's, think it's less about the content itself, but more how you approach it.

737

:

Because if you have kind of a good and healthy approach to learning something difficult,

then you're going to be fine, right?

738

:

Because the resources are out there.

739

:

You just have to be able to power through and uh learn them.

740

:

Yeah.

741

:

Yeah, Yeah, it makes sense.

742

:

And also I'm curious, you know, now I'm going to stop playing this out because I've

already taken quite a bit of your time.

743

:

I still have a lot of questions, but I'm also curious.

744

:

Well, first, if you like, what's next for you now, you know, like what are, what are

exciting projects you have on the deck for, for the next few months?

745

:

Oh, many things.

746

:

So I'm looking forward to working on clinical data.

747

:

I think that's quite interesting.

748

:

um And yeah, doing something where the results are more immediately relevant for the lives

of people.

749

:

I think that's just very rewarding.

750

:

And then there's a big em kind of data set uh of people.

751

:

in the city where I'm living that was collected 10,000 individuals and I will soon get

access to that.

752

:

So kind of like the UK biobank, I think it's called, just for Austria.

753

:

Yeah, so that will be cool to just do some big data analysis kind of.

754

:

And then I'm also writing a grant um that I'm going to submit to different funding

agencies, which is also just a big part of being a scientist.

755

:

And hopefully that will get funded and then I can do what I like basically.

756

:

em Let's see.

757

:

That sounds like fun.

758

:

Actually talking about studies, if you could design your dream study, you know, without

constraints, what would it look like?

759

:

um I think it would use a lot of ambulatory physiological measurements.

760

:

know, like we have...

761

:

our activity trackers already, they are now continuous blood glucose monitors, but they

are not so good yet, right?

762

:

But it would be very cool to kind of have this data um immediately.

763

:

So in real time, and then maybe ask the study participants questions about this.

764

:

So if the blood glucose is below a certain level, how do they feel in the moment?

765

:

So really,

766

:

having like these powerful physiological measurements that you usually only really get in

the laboratory in the daily life, that would be cool.

767

:

I think in five, 10, 20 years, the study will be possible, but I think now it's, yeah,

we're not fully there yet.

768

:

Also to use this data to then in real time, ask people how they feel about it.

769

:

We're just beginning to use that properly, but yeah, that would be very cool.

770

:

Yeah, that would be super powerful.

771

:

Awesome.

772

:

Well, Christoph, before I ask you the last two questions, is there anything I forgot to

ask you or I forgot to mention that you would like to talk about?

773

:

No, I think we covered a lot of ground.

774

:

We did.

775

:

We did.

776

:

Yeah.

777

:

I'm very happy about that.

778

:

And that was super interesting.

779

:

So thank you so much.

780

:

Of course.

781

:

Now it's your time to answer the last two questions.

782

:

So if you had unlimited time and resources, which problem would you try to solve?

783

:

Yeah, I guess now I have to give a good answer since I've been listening to the podcast so

many times, hearing the answers.

784

:

um And I thought a little bit about it and I would say it's not something in my immediate

research context.

785

:

I think someone also mentioned climate change, which of course is a perfectly good answer.

786

:

But I think what I'm most worried about is kind of the trust in science um and how we can

improve this and kind of elevating the potential of science in daily life.

787

:

Because I think it just starts very early with our high school education, but then it also

goes into politics, how scientists can work with politicians together.

788

:

Because there are many smart people out there who have

789

:

very good answers to the problems we have, but people don't listen enough to them.

790

:

And scientists are also not very good at communicating this to the public sometimes.

791

:

Yeah.

792

:

So it would just take so many different changes that it would require a lot of time and

resources to, to really have this nice link between science and, and people.

793

:

So, yeah.

794

:

And I think that this would also do a lot of work.

795

:

for climate change or against climate change.

796

:

Right.

797

:

But yeah, I don't have an answer how to solve this, but I think it's a big problem that

should be addressed more.

798

:

Yeah.

799

:

No, I completely agree.

800

:

That was one of my, one of my answers actually to that question.

801

:

So yeah, I can, I can only agree with you.

802

:

Yeah.

803

:

And I do think it will have a lot of, you know, domino effect basically, because it's,

it's a meta.

804

:

It's a meta question where it's how people think about topics, not only the topics

themselves.

805

:

And so that's why I think it's so uh important.

806

:

Yeah, exactly.

807

:

And second question, if you could have dinner with any great scientific mind, dead, alive

or fictional, would it be?

808

:

Yeah, I think there I would have to say Bertrand Russell.

809

:

Oh yeah.

810

:

Yeah, because...

811

:

I just, think he would be a interesting, charismatic and fascinating person to talk to.

812

:

And what I like about him is so he uh revolutionized set theory and mathematical theories

and philosophy.

813

:

like very complex topics, but then he also wrote a book about how to be happy and how to

not be unhappy.

814

:

um So very close to people's lives.

815

:

And also written in a very accessible way.

816

:

And I just find it fascinating how you can do both.

817

:

And it would be cool to see who the person is behind such a breadth of topics.

818

:

Yeah.

819

:

Yeah.

820

:

Definitely.

821

:

That's a great answer.

822

:

I would love to join that dinner.

823

:

You're very welcome.

824

:

Yeah.

825

:

Well, amazing.

826

:

Thank you so much, Christoph.

827

:

think it's time to call it a show.

828

:

But really, really great to have you here.

829

:

First, because you're doing fascinating research and I learned a lot.

830

:

yeah, in general, thank you so much for your work.

831

:

And also thank you so much for your work in the Learn Bay Stance uh community again.

832

:

So um thank you so much, Christoph, for taking the time and being on the show.

833

:

Yeah, thank you for having me.

834

:

It was a lot of fun.

835

:

This has been another episode of Learning Bayesian Statistics.

836

:

Be sure to rate, review, and follow the show on your favorite podcatcher, and visit

learnbaystats.com for more resources about today's topics, as well as access to more

837

:

episodes to help you reach true Bayesian state of mind.

838

:

That's learnbaystats.com.

839

:

Our theme music is Good Bayesian by Baba Brinkman, fit MC Lance and Meghiraam.

840

:

Check out his awesome work at bababrinkman.com.

841

:

I'm your host.

842

:

Alex and Dora.

843

:

can follow me on Twitter at Alex underscore and Dora like the country.

844

:

You can support the show and unlock exclusive benefits by visiting Patreon.com slash

LearnBasedDance.

845

:

Thank you so much for listening and for your support.

846

:

You're truly a good Bayesian.

847

:

Change your predictions after taking information and if you're thinking I'll be less than

amazing.

848

:

Let's adjust those expectations.

849

:

Let me show you how to be a good Bayesian.

850

:

Change calculations after taking fresh data in Those predictions that your brain is making

Let's get them on a solid foundation

Chapters

Video

More from YouTube