How to make better health and fitness decisions: A primer on Bayesian reasoning

Here’s something you won’t hear a lot of fitness pros admit: I’m not totally certain about all of the recommendations I give.  To be honest, I’m not totally certain about most of them. 

I’d say I have about 80-90% certainty about most of the stuff I tell you on here.  Ten years from now, I’m sure I’ll have changed my mind about a few of those things.

This is radically different from how most people form their beliefs around health and fitness.  Most people adopt one of two mentalities.

First, you have the absolutists.  These people look at (some of) the evidence, pick a side, and then put all disconfirming evidence out of their minds.  Once they’ve formed an opinion, these people are loathe to even admit any possibility they might be wrong.

This approach is encouraged by the media, which tends to report the latest study as though it both provides a definite answer and overrides all previous studies.  New study proves that eggs will make you live longer!

There’s an obvious danger here of having a false sense of certainty- remember how many times the media has flip-flopped on eggs?

The second group is the relativists.  These people look at the evidence, acknowledge that it is somewhat contradictory, and throw their hands up in the air and claim that the truth is unknowable, or possibly even unimportant. 

At their worst, these people claim that the truth is literally relative- that what’s true for me isn’t necessarily true for you, and all opinions are equally valid.  Just shoot me.  Note: we’re talking about objective truth here.  That’s different from individual variation, which we’ll talk about in a minute.

You shouldn’t be a relativist or an absolutist.  Instead, you should become what’s known in statistical and philosophical circles as a Bayesian.

What is Bayesian Reasoning?

Some of you may have heard the term Bayesian from fitness writer and former statistician Menno Henselmans, whose website is called Bayesian Bodybuilding.  Menno has said that Bayesian is essentially synonymous with “rational.”  That’s close- Bayesian reasoning certainly is rational, but that’s not quite what it means.  A more precise definition of Bayesian would be “probabilistic.” 

The term comes from the name of the 18th century statistician, philosopher and minister Thomas Bayes, who created a formula (appropriately enough called Bayes’ Theorem) for estimating the likelihood of an event occurring.  However, it was quickly realized that the same formula could be adapted to estimate the probability of a particular belief or hypothesis being true.  This practice has evolved into the field that is now known as Bayesian probability.

To an absolutist, 70% certainty is the same as 100% certainty.  To a relativist, it’s the same as 50% certainty.  To a Bayesian, 70% certainty is just 70% certainty. 

Now, you don’t actually need to learn Bayes’ Theorem, or any of the math behind this at all.  I’m not a statistician, and you don’t need to be one either.  What you do need to learn is what probabilistic reasoning looks like in action. 

Probabilistic predictions are frequently used in the fields of finance and economics, and in particular for evaluating investments.  One of the best examples of probabilistic thinking in action was the article Donald Trump is a tail-risk candidate, where author Josh Barro evaluates then-candidate Trump (this was written in early 2016) as if he were a stock that Josh was considering investing in. 

First off, look at the graphic.  Josh took everything that might happen if Trump gets elected, and plotted them along a bell curve-improbable bad outcomes on the left, likely outcomes in the middle, and improbable good outcomes on the right.  Even though he’s strongly anti-Trump, he acknowledges a 10-15% possibility that Trump will surprise him and be a good president.

Second, read his argument for focusing on the left tail of that probability curve.  Yes, nuclear war or global economic collapse are unlikely, but they would be catastrophic if they did happen.  Therefore, he argues that voters should be conservative, in the sense of being risk-averse. 

In applying Bayesian reasoning to your own decisions, you should be forming a rough mental probability graph similar to the one Josh drew, with the most likely outcome in the middle, below-expected outcomes on the left, and above-expected outcomes on the right.  Also, like Josh did, you should give serious consideration to unlikely but catastrophic outcomes.  In other words, don’t take drugs that have a 1% chance of killing you, even if the upside is good.

In a minute, I’ll show you a few examples of how I would graph out potential health and fitness choices.  First though, you need to know what kind of data you’ll need to make these choices. 

What to consider when making health and fitness choices

Okay, so now you have a basic idea of what your mental model should look like- a bell curve of the probable outcomes of any particular decision.  Now the question becomes, how do you figure out how the bell curve is shaped, what goes on it, and where?  In other words, how do you use this mental model to evaluate potential health and fitness choices?

First, you look at the evidence.  Preferably scientific research- anecdotes have value, particularly if you’re asking a question that scientists haven’t studied very much, but research always takes precedence over anecdote.

Crucially, you need to look at the sum total of the research, not just the latest study.  New studies build on old studies, but don’t replace them.  The best way to get an overview of all the research on a topic is to look at meta-analyses and narrative reviews, two types of studies which synthesize the results of many prior studies on a given topic to figure out what the research as a whole says about a given question. 

Once you’ve looked at the evidence, you ask yourself four questions:

First, what are the benefits?  What are they and how big are they?

Second, what are the risks or drawbacks?  Again, how big are they and how severe are they?

Third, how clear or certain is the evidence? Does it consistently say the same thing, or is it highly contradictory?

Fourth, how much inter-individual variability is there?  This is really two questions.  Does everybody respond the same way to whatever course of action you’re considering?  And if not, do you have any way of knowing how you’ll respond? 

I realize this is all really abstract and a bit confusing- as with most things, the best way to learn it is to see it in action, then do it yourself. 

Five examples of Bayesian reasoning in action

Here are a few examples of this thought process in action.  Since my aim here is mainly to demonstrate this thought process rather than to make definitive recommendations on any of these five things, these examples will be a little light on the citations.

Example 1: Anabolic steroids

Steroids are more popular than most people realize.  We can only get very rough estimates on how many people use them, but it’s clear that at least several million Americans have tried them at some point. 

Steroids are much more common among men than women, since they can cause women to essentially undergo a DIY sex change.  Dosages also vary quite a lot.  For the sake of this exercise, I’ll assume you’re a healthy young man considering trying a newbie-level steroid cycle of 400-600 mg/week of injectable testosterone for 10-16 weeks.

Benefits- High. You’ll gain muscle, and probably lose some fat.  Your sex drive might also go up, and your skin might look a little better.

Risks- High.  You could get acne, lose your sex drive, or start growing breast tissue, a condition called gynecomastia.  You might suffer hair loss or anger issues.  You will also suppress your body’s own testosterone production; it will recover, but it’s not clear how quickly or easily.  Steroids can also cause heart problems, though probably not at this dosage.

Certainty- High.  All of the risks and benefits I listed definitely happen, as confirmed b both studies and widespread anecdotal reports.  Certainty is basically 100% for body composition effects- you’ll definitely put on muscle.  The effects on libido and personality are less clear, and the odds of serious side effects are also less clear, but probably low at low doses.

Inter-individual variability- High.  In both studies and anecdotes, different guys respond differently to steroids, both in terms of benefits and side effects.  Hair loss is more common if you have a genetic predisposition for make-pattern baldness, and gyno is more common the higher your body fat percentage.  The other effects are hard to predict, as they depend mostly on your androgen receptor density, which you don’t really know about.

Conclusion: effective but dangerous  Steroids absolutely work, but they’re high risk, high reward.  They also have some very severe side effects.  I don’t use them and don’t think most people should.  If you do want to try them, you should at least spend over a year learning about them, use a low dose to start with, get as lean as possible first to minimize the chance of gynecomastia, monitor your blood work before, during and after your cycle, and know how to help your body recover from steroid use. 

Example 2: Meditation

Meditation is widely reported to lower stress and improve overall mental and physical health, as well as cognitive functioning.  It can be as simple as sitting in silence with your eyes closed for 2 minutes at a time.

Benefits- Moderate to high.  Meditation can reduce stress and increase quality of life, and has been suggested to have many other physical and mental benefits, like improved cognition and a stronger immune system.

Risks- low.  I can’t see any way that meditation could go horribly wrong.  Given its’ spiritual connotations, I suppose you might fall in with a creepy new-age crowd, but the only likely risk is that it won’t work, you’ll be frustrated and waste a little time.

Certainty- high.  Meditation definitely works.  The stress-reduction benefits are beyond dispute at this point.  Other benefits have varying degrees of support- there’s probably something to improved cognition, while disease prevention and life extension are more speculative. 

Inter-individual variability- moderate.  It works better for some people than others, but most people who stick with it get some result.

Conclusion- try it.  It doesn’t work for everyone, but given the substantial benefits and basically non-existent risks, there’s really no reason not to at least try meditation. 

Related article: How to start meditating in just 2 minutes a day

Example 3: Branched-Chain Amino Acids

Branched-chain amino acids include the three amino acids leucine, isoleucine, and valine.  BCAAs play a vital role in anabolic signaling, with leucine in particular having been shown to be absolutely necessary for muscle protein synthesis.  Thus, it is widely believed that BCAA supplementation will help people build more muscle, preserve muscle while fasting, or reduce muscle catabolism during workouts.

Benefits- low.  BCAAs have a strong theoretical foundation- you definitely do need them to build muscle.  However, BCAA supplementation has generally failed to improve muscle protein synthesis in studies.

Risks- low.  BCAAs don’t seem to have side effects.  I seem to recall one study where they were shown to be actively counterproductive, but mostly they do nothing. 

Certainty- moderate to high.  Studies are pretty consistent about not finding results from BCAAs, at least if they’re well-designed. Of course, supplement companies can find ways to bias their studies- such as by having the group that takes BCAAs also consume more protein than the control group.  Also, there’s always the argument that studies might not reflect real-world conditions all that well.

Inter-individual variability- low.  Muscle protein breakdown and synthesis are fundamental biological processes that don’t vary a whole lot between people.  Your diet may make a difference though- since vegetable protein has lower natural BCAA content, vegans might benefit somewhat from taking a couple of grams of BCAA with meals to boost the quality of the protein they eat.

Conclusion- waste of money.  BCAAs don’t seem to work, nor do they seem to hurt anyone.  They just waste your money.  The one exception here being the aforementioned possible minor benefit to vegans. Menno Henselmans provides a good summary of the research here.

Example 4: Vegan Diet

I don’t think I need to explain what this is.  Let’s assume your main goal is overall health, and you have no particular medical condition driving that decision.

Benefits- varies.  This depends a lot on what your basis of comparison is.  Vegan diets have consistently outperformed the standard American junk food diet, but then so does any other controlled diet.  Compared to other diets like paleo, Atkins, or the Mediterranean diet, the vegan diet usually ends up coming pretty close- there’s not a totally clear winner, at least for fat loss.  For lifespan, eating more unprocessed plant foods is good.

Risks and drawbacks- moderate.  There’s the obvious problem of missing out on a lot of foods you like.  Vegetable protein is also lower-quality than animal protein, so you need to eat more of it and even then you won’t gain muscle as easily as a meat-eater.   Vegan diets also tend to lead to a few nutrient deficiencies, particularly iron, zinc, and B-vitamins. 

It should be noted that these are long-term risks; there’s no major short-term risk to at least trying veganism. For weight loss and blood sugar control, several other diets have outperformed the vegan diet.

Certainty- high.  The vegan diet has been very well-studied and we have a pretty clear picture of its effects.  Well-controlled long-term experimental studies are still lacking, and of course it’s a heavily politicized topic, but overall we have a reasonably clear picture of the upsides and downsides.

Inter-individual variability- moderate.  Some people do respond better to it than others, but not radically so.  You might feel great or you might feel fatigued all the time, but you won’t die. 

Conclusion- probably worth a try, but might be going too far.  Going vegan for a month won’t kill you and might be interesting.  That said, remember how I said that the best diet for longevity appears to be a primarily plant-based diet? 

The benefits of the vegan diet may have more to do with adding fruits and vegetables than with subtracting meat, although that obviously depends somewhat on what kind of meat you were eating before.  Meanwhile, the drawbacks primarily come from the lack of animal protein.  You’ll likely be better off doing it halfway by eating vegan for some, but not all, of your meals. 

Example 5: Crossfit

To paraphrase Wikipedia: Promoted as both a physical exercise philosophy and also as a competitive fitness sport, CrossFit workouts incorporate elements from high-intensity interval training, Olympic weightlifting, plyometrics, powerlifting, gymnastics, girevoy sport, calisthenics, strongman, and other exercises.  Crossfit is practiced at special Crossfit gyms, which are called “boxes” instead of gyms for some reason.

Benefits- High. You can lose fat, build muscle, build cardiovascular endurance, and develop a well-rounded physique and athletic ability.  Crossfit really does work for a lot of people, so long as they avoid injury.  You can also make friends, as Crossfit does a great job of building an active and motivating social environment into its gyms.

Risks- Very high.

Also, exertional rhabdomyolysis.  And you’ll be surrounded by people who see this as some kind of silly joke.

Certainty- moderate.  Crossfit is somewhat well-researched, but for ethical reasons the workouts used in research are usually different (read: safer) than most real Crossfit workouts.  As a result, we still have to rely partly on anecdotal evidence.  It’s clear that Crossfit works for many, doesn’t work for others, and also injures a lot of people, but the anecdotes don’t provide numbers. 

According to this study, 73% of Crossfitters get injured, with 7% being injured badly enough to require surgery.  That’s much higher than traditional weightlifting, but lower than heavy contact sports like rugby.  However, you’d get very different numbers depending on which gyms- excuse me, “boxes-“ you drew your subjects from.

Inter-individual variability- high.  There’s substantial variability in how your body responds to different kinds of training.  There’s also a lot of variability between gyms, with some being much safer than others. 

Most important in my opinion is psychological variability.  Some people will enjoy the social environment of Crossfit more than others, and some people will be better able to resist the urge to push themselves too hard and get injured.

Conclusion- effective but unnecessary and dangerous.  I’d go with traditional weight training.  Crossfit does work for some people though.  My guess is, the people who do best at it would most likely be extroverts who enjoy the socializing that comes with Crossfit, and not very competitive, so they won’t be driven to overreach and get injured like the people in the video.

Related article: The most overrated everything in fitness (Crossfit combines at least 3 of them)

Level up your reasoning skills

If you’re not use to thinking in probabilistic terms, it can be a difficult skill to develop.  The best way to start thinking that way is to read the works of other people who express their thoughts in probabilistic terms.

Such people are few and far between.  Bayesian-style predictions and statements of confidence can sometimes be seen in political polling and sports reporting- particularly betting sites.  They’re otherwise uncommon in media reporting.

By far the best resource I’ve found for instilling this kind of thinking into my mind has been FiveThirtyEight.  They almost always express their predictions for the future in terms of probability, and even publish detailed projections of elections and sports tournaments.  I actually sometimes read their sports section just to absorb their way of thinking, even though I don’t care about sports. 

As for resources that are specifically about fitness, the two people who I’ve seen to most exemplify this mindset are Alan Aragon and Menno Henselmans.  I highly recommend that everyone read Menno’s blog and subscribe to Alan’s monthly research review.  Although they don’t typically say that they’re “x% sure” about their recommendations, both of them habitually speak in terms of “weight of evidence” rather than certainty. 

Regardless of who you read, start thinking in terms of how certain you are, and how probable it is that you’re correct- it’s never 100%, although it can certainly get close in some cases.  When making health decisions- deciding on a new diet, a style of workout, what supplements to buy, which toxins you need to avoid- remember to look at the weight of all available evidence, not individual studies and definitely not popular media articles.


Click here to join the discussion on Facebook