Bayesian Investor Blog

I’ve long acted as if a good diet is fairly important, and I’ve gathered lots of relevant evidence. But until recently I classified that evidence into many small topics related to specific nutrients and health problems, and never organized those ideas into an overall assessment of how important a good diet is.

Comments by Jim Babcock prompted me to investigate a broader overview.

This post will mainly focus on evaluating the importance of nutrition for adults in wealthy nations, then will summarize my guesses about how to achieve a good diet.

Nutrition has a weird mix of good consensus knowledge (e.g. fish, spinach and sweet potatoes are better than pizza, beer, and donuts), yet that knowledge does little to improve peoples’ diets. People focus instead on controversies that are more confusing, such as whether carbs are dangerous, or the relative healthiness of peanuts versus potatoes versus butter.

There are also signs of important gaps in our knowledge.

Standard Summaries

This report of the Global Burden of Disease Study seems to be the most authoritative overview that I’ve found. Alas, that mostly means that it’s the only serious-looking overview I’ve found. It looks more like educated guesses than like settled science.

Figure 2 of that paper says that “Dietary risks” (mostly cardiovascular?) are 2nd highest category of environmentally caused mortality, both for the global average and for developed countries. (For the global average, “Child and maternal malnutrition” is a bigger problem, but that’s due mainly to poor countries, and I’m focusing in this post on the health of people who read this blog). About half of the other causes of mortality are categories that appear to be influenced by diet.

In GBD 2016, poor dietary habits were the second leading risk factor at Level 2 of the hierarchy for mortality globally, accounting for nearly one in every five deaths.

I don’t see a good description of how they reached that conclusion [1], but it seems to involve more evidence from correlations than from RCTs. It looks like the RCTs mainly connect diet to biomarkers, which tend to be overstated, and often work differently in response to interventions than they did when used for evaluating the original risks. And how do they combine the harm from a low-fruit diet with the harm from a low-fiber diet? Those are hardly independent effects.

This evidence seems likely to point in the right direction, but I can easily imagine it being off by more than a factor of two. The reliance on correlations rather than causal evidence suggests that they overstate the effects – the correlation data are likely partly explained by many healthy traits correlating with good diet (traits such as good genes, good exercise habits, wealth, not smoking, etc.). See my footnote [2] for more detailed comments on this evidence than most of you will want.

But this kind of correlation data can also underestimate the importance of a good diet. Some guesses how (no single one of which looks particularly powerful):

• The categories combine good and bad foods in ways that hide the important differences. E.g. if grain-fed beef is much worse than grass-fed beef, lumping them into “red meat” hides important evidence.
• The studies could attract mainly people with moderately good diets, thereby overlooking harms to those with the least responsible diets.
• If approximately nobody in the sample population eats something that’s important for a good diet (let’s say insects), we’ll need to look elsewhere if we want to find any hints.

Other mainstream sources say similar but vaguer things:

Diet is established among the most important influences on health in modern societies. Injudicious diet figures among the leading causes of premature death and chronic disease.

The problems to which those studies refer mainly affect older people. Does that mean nutrition doesn’t matter much for the rest of the population?

It certainly matters less. Some age-related diseases develop quite gradually, and have moderately harmful effects well before death. Dementia is a clear example of this, reducing quality of life dramatically for years before death, and probably more subtly for a decade or two before death.

But dementia only affects a relatively small fraction of the population.

I suspect cardiovascular disease impairs quality of life a decade or more before death, and that’s relevant to a significant fraction of the population, but I don’t know how to find much evidence about this.

Obesity affects a large fraction of the population (39.8% by one criterion), and seems strongly affected by nutrition:

the medical cost for people who have obesity was $1,429 higher than those of normal weight.

I expect that obesity has widespread negative impacts on quality of life (e.g. via sleep apnea), but I don’t expect to find good evidence which quantifies those impacts. I expect that diets which cause obesity also cause discomfort in people’s gut, painful bowel movements, and hernias, but those impacts are even harder to quantify.

I also expect that efforts to avoid obesity cause almost as much harm as does obesity. Weight control via using willpower to eat less of what you’ve been eating will leave you feeling like food is scarce. Your subconscious will respond to that by generating fatigue and/or depression, and slowing the rate at which you burn calories. So it’s more effective at producing depression/fatigue than at controlling weight [3].

Inferences from historical patterns of nutritional progress

Another type of evidence comes from iodine supplementation. Compared to areas where iodine deficiencies were common, iodine supplementation raises IQ by an estimated 12.45 points. One summary says it “can explain roughly one decade’s worth of the upward trend in IQ in the United States (the Flynn effect)”, but also caused thyroid-related deaths in older people (deaths which could presumably have been avoided by a slower increase in iodine consumption). Income increased 11% in cohorts exposed to iodized salt.

I expect that few parts of the world are overlooking any more gains which are as cost-effective as iodine supplementation has been, but it seems unlikely that the second and third best nutrition-related interventions are an order of magnitude less cost-effective.

How many valuable interventions remain to be widely implemented?

We can get some hints about the value of the nutrition wisdom that remains to be discovered from the rate at which new insights are being found. E.g. it was less than 25 years ago that transfats were recognized as causing large problems, and the evidence about EPA/DHA ratios was published just this year.

My impression is that we’re getting important new nutritional insights at least once per decade.

Some of these discoveries result in widespread dietary improvements within a decade or two (e.g. iodine and transfats), whereas others have limited impact (e.g. fiber – maybe due to the need for larger shifts in consumer habits?).

The most recent nutritional insight that seems possibly important is this year’s news about good EPA/DHA ratios [4].

I’m guessing that the elimination of transfats was the second most valuable nutritional intervention that has been accomplished so far, avoiding an estimated 20,000 deaths per year in the U.S.

If I believed the Global Burden of Disease Study, I’d say that the following yet-to-be-solved problems are bigger than the transfat problem, causing over 2 million deaths each: Diet low in fruit, Diet low in whole grains, Diet low in nuts and seeds, and Diet high in sodium.

But it’s hard to figure out how much of that I should believe. I expect that the sodium [5] and whole grain measures are partly proxies for diets high in junk food. I’m less sure what to make of the fruit / nuts / seeds evidence – most likely a majority of those are pretty healthy, but I expect there to be a lot of variation within those categories.

But I don’t want the confusion over proxies versus real causes to obscure the fact that we’ve got lots of correlational evidence, and a moderate number of RCTs, pointing to some large effects of some dietary differences. I’m fairly sure some individual estimates are overstated, but see footnote [2] for why I suspect it underestimates the value of higher fiber diets.

Comparisons with Other Cultures

We can tell a fair amount from comparisons to hunter-gather cultures, and evidence from blue zones provides a few more hints. I also tried to get some hints from the Amish. They seem to have better health than their neighbors, but I can’t tell whether any of that is due to diet.

According to Lindeberg’s Food and Western Disease, heart attacks, stroke, type 2 diabetes, and dementia seem to not exist in primitive societies, and seem to become common shortly after societies adopt a western lifestyle [6].

The situation with cancer is less clear, but cancer appears to be much more common with a western lifestyle.

Lindeberg suggests that diet explains a good deal of those differences.

Of course, diet isn’t the only component of lifestyle that matters. Exercise, chronic stress, and infectious diseases could also cause much of those differences in disease rates. Lindeberg suggests that differences in physical activity are not enough to explain much of the difference in cardiovascular disease. I don’t know what to make of the stress and infectious disease differences – I occasionally see signs that those explain more of western diseases than does diet [7]. I get the impression that experts tend to think diet is a somewhat more plausible explanation, but I see little reason for confidence in anyone’s opinion here.

Those diseases cause a large enough fraction of mortality in wealthy nations that this tends to suggest that diet could cause more than half of all deaths in wealthy nations. It seems plausible for this estimate to be larger than the estimates from studies that look only at variations within wealthy nations, because the differences between hunter-gather diets and wealthy diets is fairly large compared to the variation within wealthy nations.

So while evidence I presented in the Standard Summary section suggested to me that diet caused 10 – 20% of deaths, the evidence in this section is enough for me to shift that estimate to 15 – 30% (and I’ll call that a 50% confidence interval).

For the impact on younger adults, obesity rates seem like the most important evidence, and imply that diet causes moderately annoying problems in 20 – 40% of the population (once again, that’s a 50% confidence interval).

Doubts and Anecdotal Evidence

Here I consider some objections that are loosely based on a comment in Jim Babcock’s post.

Why don’t I hear reports of people whose lives were improved by better nutrition?

I expect there are several reasons:

• I’ve been hearing such reports from acquaintances more than once a year recently, but until now I haven’t been putting them together into a larger picture. It hadn’t occurred to me to do so, because there was little pattern to the reports, and I was unsure how credible most reports were. Examples:
  • Vitamin D made a big difference.
  • I feel much better now that I’ve stopped being vegan.
  • I was vegan for a while, but I couldn’t get enough iron without eating kangaroo.
  • A raw food diet helped me recover from a near-fatal accident; see how much younger than my actual age I look now!
  • Pedialyte was an important part of improving my effectiveness – I was apparently deficient in potassium.
  • I had an iodine deficiency and always felt tired, now I take kelp and feel better.
  • Several reports of benefits from switching to a paleo diet, but I don’t remember those reports clearly.

  And there’s my example of probable iodine overdose from kelp. And there are also people who I’ve heard of because they’re talking about nutritional issues (Dave Asprey, Steven Gundry, Josh Whiton).

  I’ve probably forgotten some of the reports that I’ve heard, because I didn’t have any mental category for “evidence relevant to the overall importance of nutrition”. I mostly filed these reports as facts about specific people, and it took me days to recall the ones I’ve mentioned.

  These kinds of reports are highly inconclusive, but are enough for me to mostly reject the “I haven’t heard of any examples” argument as mostly due to filter bubbles or poor memory. And they weakly suggest that nutritional studies are missing some effects.

• People don’t notice effects because they are too gradual.
• Reporting biases – people who experience quick cures tend to stop talking about them fairly soon, whereas people who struggle with mediocre treatments for a long time talk more about their treatment because it stays on their mind much longer (I’ve seen research that says something resembling this, but can’t find it now).
• We don’t often hear claims like “I stopped eating junk food, and didn’t get Alzheimer’s”, for a variety of reasons that are unrelated to whether such effects happen:
  • People have been conditioned to believe that Alzheimer’s can’t be prevented.
  • Even if a good diet can prevent Alzheimer’s, it’s very unclear how to connect the timing of not eating junk food with the timing of not getting an Alzheimer’s diagnosis.
  • Even with a causal connection, I expect less than 10% of instances of “no junk food, followed by no Alzheimer’s” to reflect a cause and effect relationship.
  • Selection bias would discourage me from paying too much attention to such reports. Having Alzheimer’s reduces a person’s ability to talk with me, and makes them less able to remember how much junk food they ate as they were succumbing to Alzheimer’s.

Why don’t multivitamins show benefits?

The studies of multivitamin effects put some sort of upper bound on the importance to the average person of the nutrients in typical multivitamins, but they don’t test everything I want tested. Here are some limitations:

• Most multivitamins are missing some important ingredients of a good diet: fiber, potassium, and various omega-3’s.
• Some nutrients are reported as if they’re a single molecule, when there are multiple variants that our bodies need (e.g. vitamins E, A, and K). Multivitamins often have just one version of each, and there’s some speculation that getting plenty of one version can crowd out the other versions. I don’t have much evidence about the significance of this effect.
• Harm from overdoses may offset the benefits – this might be true for vitamin B6, vitamin A, iodine, selenium, calcium, iron (in men), manganese, and chromium. But I suspect these effects are not common enough to have much effect on a typical multivitamin study.
• Multivitamins often provide some vitamins in a synthetic form, which not everyone converts to the biologically active form, e.g.:
  • folate – the roughly 10+% of the population that most needs folate supplements seem to be somewhat harmed by the folic acid version that is commonly found in multivitamins.
  • I’ve seen a number of claims that there’s a lot of variation in how well people convert the commonly used Pyridoxine HCL into the bioactive Pyridoxal-5-Phosphate (P5P) form of B6. I haven’t found good evidence on this subject.
  • B12 comes in several forms, and there are differences in how people react to various forms. I don’t understand these issues well.
• I keep seeing claims that nutrients in pills are absorbed differently from nutrients in whole food – I’m confused about whether this is significant.
• The effects are too slow. Remember what kinds of problems I mentioned above. Obesity, cardiovascular problems, diabetes, and dementia all seem to be symptoms of metabolic problems that develop slowly and don’t go away quickly.
• Faulty theories (e.g. the antioxidant theory) mislead people about which supplements to take.

No single reason from this list seems sufficient by itself to explain the discrepancy between multivitamin research results and the theory that people are malnourished, but I think several of those reasons explain a significant fraction of the difference.

Still, the multivitamin studies seem sufficient to convince me that readers of this blog shouldn’t worry much about deficiencies of the average vitamin. And for minerals, worries should be balanced almost equally between concerns about deficiencies and overdoses.

Note that none of this says much about fiber. The evidence here isn’t as rigorous as I’d like, but the evidence seems to point relatively consistently in the direction of almost nobody getting the recommended “adequate intake”, and that “adequate intake” looks more like “we’re scared to recommend something that’s more than 2 standard deviations away from typical modern behavior” than “this is where returns on better diets become unimportant” [2].

RCTs say that fiber supplements are somewhat effective at reducing obesity.

Most fiber supplements don’t appear to behave like the naturally occurring fiber for which the nutritional benefits have been reported, and I haven’t seen widespread reporting of which fiber supplements appear to be valuable (psyllium, beta-glucan). It appears that most fiber supplements don’t work, and getting enough fiber from food seems to require large dietary changes that are somewhat inconvenient. These seem like conditions where malnutrition is relatively plausible.

So maybe there are only a small number of nutrients that are worth supplementing (such as fiber and iodine, with iodine being a mostly solved problem). Or maybe the causes of good nutrition are sufficiently diverse that what we’ve been attempting is like trying to explain intelligence via a handful of genes. There’s definitely something discouraging about the track record of “just add X to your diet” interventions, with the large exception of iodine.

Avoiding bad food is likely to be roughly as important as adding nutrients. Here are some guesses:

• smoked foods – putting smoke on foods isn’t obviously safer than putting it in your lungs.
• foods designed to be superstimuli, so that we keep eating them after we’re full (i.e. junk food).
• fats processed at high temperatures – probably has problems that are similar to transfats? Does this apply to most vegetable oils?
• mycotoxins from poorly stored seeds – and animals that were fed such seeds.
• lectins that we haven’t evolved to handle

Advice

Eat more fiber from whole foods.

Not because I have a good reason to think that fiber is the One True Nutrient that matters, but because taking fiber seriously virtually requires you to make lots of choices that are good according to many other promising nutritional criteria. Maybe junk food companies will eventually produce stuff which invalidates that heuristic, but their behavior so far suggests that it’s hard for them to do.

Ideally you should get over 40g of fiber from whole foods per 2000 calories – this heuristic almost guarantees you’ll avoid junk food (and avoid most restaurant food, and spend more time preparing food … I suspect many of you will be willing to go less than halfway to this goal). You won’t reach this goal by just adding a few high fiber foods to your diet. You’ll need to aggressively reduce the low fiber foods in your diet.

Note that many of you consume more than 2000 calories a day – use the Harris-Benedict Equation for a better estimate, and maybe try to adjust for the amount you burn through exercise.

Maybe the fiber to sugar or fiber to carb ratio is more important than the fiber to calorie ratio, so I also try to ensure that I eat a gram of fiber for every 1.5 grams of sugar.

Weight.

If you’re worried that you weigh too much, focus next on the satiety index of your foods. Only a few foods have their satiety index listed – but note the Fullness Factor algorithm near the end of that document, which enables you to estimate satiety for most foods.

Why hasn’t the satiety index or the Fullness Factor become more popular? Maybe it’s due to the moderate inconvenience of applying them to foods that aren’t listed? Maybe it’s because they don’t generate enough controversy?

Or adopt a diet that resembles one eaten by our hunter-gatherer ancestors – that will cut out almost all low-satiety foods. Honey is the main complication to that rule – it’s likely quite safe if combined with a high fiber diet and some form of intermittent fasting, but risky if combined with typical modern lifestyles. Give priority to versions of the diet that are based on knowledge of what our pre-farming ancestors ate. Sorry, I don’t have a good reference for that – see footnote [8] for my hasty summary.

Don’t try to control your weight by a diet that leaves you chronically feeling hungry (but I encourage you to try intermittent fasting of some sort).

Note that it’s somewhat hard to gain weight by deliberately eating more (see The Hungry Brain for evidence from lab animals). So there’s little risk to eating until you feel full, even if you’re obese. Feeling full may help with the willpower that you might want for switching to healthier foods.

Decent diets.

A large fraction of the population, mostly those under 50, with decent health and weight, should focus on ensuring that their diet is at least as good as average, which mostly means eating almost no junk food. I don’t have a rigorous description of junk food, but the well-known stereotypes are somewhat accurate. Here are some heuristics that seem relatively good:

• A high fiber to calorie ratio is a pretty strong sign that it is not junk food.
• Is it a whole food?
• Read the list of ingredients – how long is it? Does each one look like a whole food?
• What temperature was it cooked at? I’m suspicious of monounsaturated or polyunsaturated fats that were cooked at much more than boiling water temperatures.
• Do you continue eating it after you feel full?

Many other heuristics are moderately effective at ruling out junk food: very low carb diets, a raw food only diet, vegan organic food diets, the SCD diet, the 20 potatoes a day diet, and maybe even the UNpacked Diet (in spite of being intended as satirical). Almost anything is healthier than a diet that is heavily optimized for some combination of taste and convenience – optimizing for those always seems to end up conflicting with the goal of healthiness.

Specific foods.

My diet contains lots of sweet potato, taro root, almonds, macadamias, carrots, onions, mushrooms, cabbage, celery, peppers*, wild-caught shrimp, pasture-raised chicken and eggs, Brad’s raw kale chips*, and olive oil. I also eat moderate amounts of dark chocolate, beans (soaked and pressure cooked – Eden brand), oysters, grass-fed beef, cherries, apples, concord grapes, saskatoonberries, potatoes*, broccoli, spinach, dates, avocados, Swerve, and peanuts*.

* – temporarily not on my diet, while I try out the Plant Paradox diet.

I use potassium chloride as a partial substitute for salt – it’s probably not accomplishing much, but it’s cheap and simple, and potassium deficiencies are more common than sodium deficiencies. I’m not trying to reduce my sodium consumption – sodium is only bad over about 4.5 g/day, and that’s hard to accomplish without junk food.

I’m adding figs to my diet – chimpanzees get nearly half of their food from figs, which is another hint about what we’re adapted to eat. Plus all those biblical fig references remind us that humans have been eating figs throughout history.

Supplements

You should mostly take supplements only after getting evidence about your specific needs, but since many of you won’t get around to doing that, here’s some general advice.

Supplementing vitamins D and B12 seem relatively safe, and many people seem deficient in them, due to problems such as hiding in artificial caves and discarding the more nutritious parts of animals. I recommend blood tests to verify that you’re taking an appropriate dose, but reluctance to get blood tests probably shouldn’t stop you from supplementing these vitamins. Most likely the benefits will be too small to detect, but there’s a tiny chance they’ll do something big like prevent Alzheimer’s.

There’s speculation that vitamin D may cause hypercalcemia if you’re deficient in vitamin K2. I don’t understand this very well, but I take vitamin K2 MK-7, and I’m guessing that most people who take vitamin D should do so.

Folate (not folic acid!) supplements are a bit more likely to produce detectable benefits, mainly in those of us with the MTHFR C677T mutation. There are pesky reports of risks from taking folate if you’re deficient in B12, so pay attention to your B12 levels before supplementing with folate.

B6 supplements are potentially important for people with high homocysteine. But low doses (a few mg) probably won’t do much, and higher doses are somewhat risky unless you get B6 blood tests to determine whether you have the right dose.

The End of Alzheimer’s has some probably good advice about testing for situations where you might want to supplement other vitamins, but people whose health is reasonable should give low priority to those, since you probably get enough of them from a decent diet.

I’m planning to start taking krill oil soon, in response to the new evidence about EPA/DHA ratios, but I have low confidence in that, and I’m eating enough shrimp and oysters that fish or krill oil supplements might be redundant.

That’s about all most people should be taking as nutritional supplements. Taking a good fiber supplement is better than being deficient in fiber, but that’s unlikely to be as good as getting fiber from whole foods.

Other issues

• Refined grains are probably bad. It’s unclear how bad. Whole grains seem modestly better – they may be a relatively easy way to improve a bit on the standard American diet (SAD), but I expect a really good diet will reduce all grains.
• Animals that have been fed unnatural diets (usually grains) are less healthy to eat than those that have been fed a natural diet. It’s unclear how important this is. It’s better to avoid chicken, beef, and pork unless labeled as pasture-raised or grass-fed. Wild-caught fish/shrimp are better than farm-raised. Farming doesn’t much affect the diet of oysters and clams, and they seem better than other animals by criteria such as B12 and EPA/DHA ratios.
• Don’t worry much about how much fat, carbs, and protein you get until you’ve addressed the more important issues I’ve mentioned above. Carbs seem to be a problem mainly if you’re deficient in fiber. Worry about fiber first.
• Limit your saturated fat consumption if you’ve got an ApoeE4 allele (exception: MCT oil may still be safe). If not, saturated fat is probably good for you, especially from coconut and chocolate.
• Be suspicious of most refined fats/oils. Much like refined sugar, they’re calories without fiber and with few other nutrients.

  I avoid most vegetable oils due to concerns about whether they were damaged during processing, and whether they resemble foods we’ve evolved to eat. The main exception is extra-virgin olive oil, which seems to be safe and maybe beneficial at levels around 4 tbsp / day (widely misreported as 1 liter per week!).

  I recommend against eating much butter (which has a terrible Fullness Factor). There have been a few good studies that show that butter is healthier than some alternatives which were commonly used at the time of the study, but I don’t know whether those alternatives should be considered edible – e.g. it seems unclear how much transfats they contained.

• A bunch of things depend on whether you’re trying to get your diet up to average, or up to the top 10%, and probably also on your genes: whole grains, legumes (beans, lentils, peanuts), and dairy from grass-fed animals are likely improvements for many people, yet mostly won’t help you get an unusually safe diet. For example, the fiber/calorie ratio of brown rice is a good deal above the average American diet, yet well below mine.
• Organic probably sometimes matters, but I haven’t found much reason to worry about it.
• Glycemic index isn’t very helpful.

It often feels easier for me to follow a diet if I classify the substances I want to avoid or drastically reduce as “not food”, or in some cases “poison” or “drugs”. Whereas if I think of them as food, it feels like it takes more willpower to avoid a slippery slope that leads me to eat more of them.

Conclusion

Bad diets are likely causing health problems that are comparable to, or maybe worse than, smoking.

Shifting people away from bad diets seems roughly as hard as convincing people not to smoke – bad diets are usually more convenient and/or tastier.

Can and should bad diets be subjected to the kinds of stigma that have convinced many people not to smoke?

Unlike with smoking, it’s hard to draw a clear line between unacceptably bad diets and good diets. So in that sense, lack of exercise is a better analogy.

But at least with diets there are plenty of scapegoats (McDonalds, Pepsi, etc) to help us unite against an enemy and divert our attention from our own role in the problems.

Maybe some new technologies will fix the consequences of bad diets soon enough that it’s not worth the effort. But I’m unwilling to bet on that.

Footnotes

[1] – it’s at least partly explained in the book-length appendix 1 – page 124 has a decent overview, but I don’t have the patience to understand the full analysis.

[2] – Table 4 of the Global Burden of Disease report claims that the “Consumption of fibre between 19 g and 28 g per day” represents the “Theoretical minimum risk exposure level”. That claim looks rather implausible. I’ve only found one source that estimates the effects of fiber consumption at higher levels than 28g/day – Figure 2 of this paper(cited by GDB as one of their sources) indicates they’re 95% confident that doubling fiber consumption from 28 to 56g/day would reduce CHD by something like 10 to 30%. Figure 4 of that paper shows smaller percentage benefits for CVD, but still a clear expectation that increases in fiber consumption would continue producing benefits above 60g/day. Note that hunter-gatherer fiber consumption is often well over 50g/day, so 2 standard deviations higher than typical consumption in wealthy nations can still be abnormally low compared to an average over the past million years.

The report’s table 1 summarizes the kind of evidence that they use. For the dietary risks section, six of the risk/outcome pairs (covering 5 of the dietary measures) are supported by RCTs, but the most strongly supported one (Diet high in sugar-sweetened beverages affecting body-mass index) says that 6 out of the 10 RCTs produced null findings (but zero RCTs out of 84 produced statistically significant evidence of effects in the opposite direction). I infer from that that we should have moderately low confidence in the best evidence that this study covers.

Their table 4 shows these estimates of global (adult?) deaths from all causes attributed to diet (data in parentheses are 95% uncertainty intervals):

I suspect that at least half of these are correlations that are only loosely connected to the underlying causation. I suspect the sodium and whole grain consumption are mostly due to correlations with junk food. It’s hard to get a high sodium diet without eating lots of things that are clearly junk food. Whole grains rarely come in junk food.

[3] – I don’t know how much rigorous science has documented this; but:

• It’s supported by large amounts of anecdotal evidence.
• There’s obvious evolutionary pressure for some mechanism like this.
• There’s a pretty clear-cut mechanism (thyroid hormones signalling that there’s less energy available, mitochondria respond by burning fewer calories).

I see one weak link in this argument: the belief that brain responds to feelings of hunger (or whatever causes those feelings) rather than calories. I’ve got anecdotal evidence, and poorly articulated theories, which lead me to expect that our brains respond to predictions about how plentiful food will be, which is not the same as calories consumed.

[4] – Prior guesses about omega-3 fats led people to eat lots of salmon. It now looks like salmon has a poor EPA/DHA ratio.

Oysters have a much better EPA/DHA ratio. Oysters also look more like what our ancestors likely ate (freshwater mollusks) than do coldwater fish such as salmon [citation needed].

Oysters were already looking like one of the best foods. This news strengthens my opinion that an optimal diet is roughly bivalvegan.

Many insects would probably be almost as nutritious if fed a natural diet, but I think all the insects I can easily buy were fed a diet of grains that leaves them with mediocre types of fat.

[5] – from Effect of lower sodium intake on health: systematic review and meta-analyses:

There were insufficient randomized controlled trials to assess the effects of reduced sodium intake on mortality and morbidity. The associations in cohort studies between sodium intake and all cause mortality, incident fatal and non-fatal cardiovascular disease, and coronary heart disease were non-significant (P>0.05). Increased sodium intake was associated with an increased risk of stroke (risk ratio 1.24, 95% confidence interval 1.08 to 1.43), stroke mortality (1.63, 1.27 to 2.10), and coronary heart disease mortality (1.32, 1.13 to 1.53).

[6] – but archaeological evidence says atherosclerosis (“Atherosclerosis was common in four preindustrial populations including preagricultural hunter-gatherers”) and obesity existed in hunter-gatherers. That evidence says less about the health outcomes than do the studies of modern hunter-gatherers, but it says enough to create a fair amount of uncertainty.

[7] – e.g. for viruses: Ewald’s Plague Time, Michael Lustgarten’s book, which I haven’t read, Greg Cochran (“Alzheimer’s disease may be caused by a persistent viral infection”), this mention of HSV1 and Alzheimer’s, and “Different Brain Regions are Infected with Fungi in Alzheimer’s Disease.

For stress, there’s Why Zebras Don’t Get Ulcers.

[8] My simplistic overview of what qualifies as a paleo food:

• fruits: eat plenty of high-fiber whole fruits (raspberries, blackberries, persimmons, figs, guavas, blueberries, avocados, apricots, strawberries, pears, apples). Use fruit juices only as flavorings. Avoid fruits that come with added sugar.
• roots, leaves, stems, flowers: eat more of these. It’s hard to meet a healthy fiber target unless these are a large fraction of your diet.
• nuts: safe and probably desirable in moderate quantities, at least if raw. Defined as seeds that have a hard shell that can’t be opened with bare hands. These presumably have fewer biochemical defenses against predators. Note that peanuts and cashews don’t meet this definition.
• other seeds: the situation here is complex, and these should generally be at most a minor part of your diet. If you want a simple rule, “no grains” is a fairly good one. Humans have eaten some beans, flax, and other seeds since somewhat before the agricultural revolution, and maybe that produced some adaptation to them, but be cautious about how they’re prepared and how much of them you eat.
• animal products: it’s valuable to eat some whole, minimally processed animals that ate the kind of diet they evolved to eat. Mollusks and pasture-raised eggs are the main choices that clearly meet these criteria (for eggs, assume that other labels such as “cage-free” or “free-range” mean malnourished). Insects are great by most criteria, but the ones I get probably ate a crappy diet. Most wild-caught seafood is pretty good, but you usually miss out on important parts such as the brain and liver. It’s less clear whether typical meat qualifies as paleo.
• honey: seems to qualify as paleo, but might be risky if the rest of your diet isn’t good.
• others: assume these are bad if you’re deciding based on paleo theory: dairy, refined oils, refined sugars, alcohol, etc.