Is Andrew Huberman Ruining Your Morning Coffee?

Today’s video is something a bit different. We thought we should do our best to test one of the most popular and widespread coffee “hacks”, this one coming from Dr Andrew Huberman.
I will note: adenosine is not my field of expertise, but if it is yours then I’d love to hear from you in the comments!

Data analysis: Susanna Pagni & Francesco Tonini

Timestamps:
00:00 Who is Huberman
05:52 Experiment outline
09:38 Shopify
10:51 Results: is Dr Huberman right?
13:47 Adenosine rant
14:45 Proposing an alternative theory
17:08 Results: caffeine, sleep, & performance
20:24 Conclusion - be skeptical
21:37 Hi Dr Huberman
21:54 Outro

Other Links:

My Books:

Things I use and like:
(*Affiliate links which may earn us a commission)

[hubermanlab]

Alas, you got it wrong. I never said to skip caffeine first thing in the morning in fact I take it right before early morning workouts. The point is that for people that experience a crash in the afternoon this is one variable they can experiment with…it relates to adenosine, caffeine sensitivity, and cortisol. This is covered in a few different HLP episodes. Clips can be a bit misleading. All the best, Andrew

[Steve-nuxt]

The only person ruining my morning coffee is me

[matteotestoni]

“Be skeptical of simple mechanisms impacting complex outcomes” amen 🙏

[RoScoHutch]

This reminds me of one of my favorite quotes.

“For every complex problem there is an answer that is clear, simple and wrong.”

H. L. Mencken

[FinnBannet]

It amazes me how society believes lies behind industry. I recently finished book called The 21 former doctor secrets by rachel morgan. She explained her career thoughts perfectly

[flightlessfish]

I’m just waiting for this to turn into Hames Joffmann morning routine “ First I make coffee then I make coffee then I make a bit more coffee”

[Neue-Johan]

20:17 Most important factor influencing my sleep; “do you have young children?” No amount of cafeïne can compensate for that.

[samuelvarga]

As a researcher, I cringe at the confidence with which Huberman gives advice on nearly every facet of life. Science is based on uncertainty, and most results ‘suggest’ or ‘indicate’ the existence of an effect, but effect sizes are important. Something that is statistically significant may be scientifically interesting but may bear no real-life significance.

[disarmedpianist]

Hi James, I'm an epidemiologist with a background in neuroscience, i run statistics for a living on health issues. Been a fan of your videos and i'm also enjoying this one immensely. I absolutely agree with you on the observation regarding adenosine. I also disagree with the univariable outlook, which seems to be the selling point by Dr. Huberman. If the world is a simple exposure-to-outcome pathway, my job would have been much simpler :) but alas, it isn't. Confounders and effect modifiers call for a more complex consideration in the real world and I am so glad that you pointed it out.

However, a few things in your video did grind my gears 😂 so here we go:
1. P-values don't mean anything if I don't know what test you ran to get them. It's not an absolute value, it's merely an indicator of how confident we are at (and you got this right) our observed outcome happening at random. In some cases, there's no need to even use p-values anymore, because if a p-value is used to show difference, then a difference only makes a difference if it makes a difference 🙃 But by looking at your first graph, i have a hard time understanding where did that p-value come from? Did you run a Chi-square? Or did you do a t-test to compare means? My guess was neither, because with 5-person sample, you would have a real hard time getting that Chi2 going, and you presented medians, so i assume you did not obtain means to compare them. I would have loved to see the means though, median is a way for me to see whether my data is skewed and helps me decide on what kind of statistical tests to run after seeing the distribution of my data. The median is not a good way to compare groups, unless you have the data skew issue.

2. All the subsequent graphs are highly problematic. Again there's the problem of not knowing what tests you ran, but i would go ahead and guess maybe spearman's r, the most basic kind of regression, was fitted. This simplified regression on its own has major issues, for the lack os capacity to control for various factors... But that aside... There's a major problem of heteroskedasticity, aka. you force-fitted a line in a cloud of data 😂 So again, while you did get a p-value<.05, it really doesn't mean much. It just means that the line where you fitted your regression looks like it's got a nice slant, and you have a difference in the slope, but your mathematical model would have a hard time estimating the effect caused by your exposure that is anywhere close to reality.

3. My guess about the cloud distribution is that you probably didnt get the definition of RCT quite right... It doesnt mean that you randomly receive treatment/placebo. It means you randomly assign people to either the treatment or placebo group, then you compare the means between the two.

Anyway, thanks for this great video! Still a major fan of yours, and i think you just gave me an idea for the next research project proposal! 😃

[kushagrachaturvedy]

As someone who deals with statistics in my day to day job, It feels so satisfying to see James, the guy who helped me discover my love for coffee, do a proper statistical experiment

[AdamDrew]

JH: He's the most...widespread science communicator on the internet.

Me: He's Hank Green?

[inkasaraswati]

As a researcher in health field, I applaud your scientific rigour of not only in your experiments and analysis but also by checking your biases in the hypotheses and results. Honestly, not all doctors and researchers put the same amount of considerations. Great work.

Edit: for all of you that mentioned about the underpowered statistics, yes I know it is underpowered, and that is not what I commented about. I'm not an expert in coffee or metabolism or anything like that, but aside from the extremely small sample size, I personally think the design is robust and I especially commend the interpretation of Huberman's premise and their results. Because good research is not just about calculating statistics, it's about interpreting what it means (and that is what most people fail at). In statistics we've also moved away from mere p values, and while there's no effect sizes or confidence intervals stated or anything, looking at the data I'm not convinced that a bigger sample would yield majorly different results. Anyways statistical significance does not equal to clinical significance, and they did a great job considering it is a YT video not a scientific paper.

[dmarti]

I’ve been practicing delaying caffeine intake for about 90 minutes after waking, and I prefer it over consuming it immediately after waking. My decision had nothing to do with afternoon crashes, but rather the following assumption: if I wait for my mind and body to wake up naturally before communing caffeine, then my mind will not associate caffeine intake with the process of waking. Hence, I take my morning coffee after I’ve gone out for a walk and then done my morning 60-minute exercise routine; and, indeed, my relationship with my morning coffee has grown healthier.

[scottcampbell]

“ We are weird and complex meat bags”. I think we have the Hames Joffmann video title.

[TheToobNube]

This video is not just about the topic in the title. It’s an excellent example of how we receive information, how to process it, how to draw conclusions.

[CarletonTorpin]

People who sell simple answers make me skeptical; truth is complicated and nuanced. The world needs good scientists, and James and his team did good science. :)

[jbkjbk]

Always good to see actual experts debunking exceptional claims made by popular hacks on the internet.

[Felpshcx]

As a scientist, it's deeply satisfying to watch such a good demonstration and explanation of what science really is. I commented on another video, and I genuinely think you would make a great scientist. You should consider collaborating with researchers! Regarding the video, I agree that 5 people are not nearly enough for anything related to humans. Even events that happen to 1 in 1, 000 people are significant in larger populations. Also, is very hard to achieve statistical significance, in humans, with a n of 5. About the correlation graphs, only the p-value was shown, which tells if the correlation is significant or not, but not if it is strong or weak. For that, you need the 'Rho' value (often called 'R') in Spearman correlations, for example. If R = 0, the line is perfectly horizontal. This means that you can have a p<0, 001 and R = 0, meaning that is "true" that this two variables don't have a correlation. So when you show the horizontal line with a high p value, what it really means is, the inexistence of correlation between those two variables is not significant. It does not mean that the line is horizontal and the pvalue is hight so the correlation is not true. Anyway, great video as always.

[TySy]

Thank GOODNESS you provide the nuance that Huberman explicitly avoids providing. As a scientist, it pains me that Huberman has as much sway that he does.

[FunkyKikuchiyo]

I’m really glad you used Layne Norton’s talking point about mechanisms. That’s a logical fallacy that he’s good at spotting, and as I’ve gotten better at it, I’ve been able to assess weird internet claims a lot better. Not just for health and biology, but in everything.