I know that this is mentioned in videos from Low Carb Down Under by Dr Mason, among others, but for visitors just browsing the comments, I'd like to quote

Emphasis mine.

It's funny that the "obviously not casual" example is margarine consumption (USA) vs divorce rate (Maine), though I can see a possible casual reason

"Carnivore cool" has to be related to having excellent quality of nutrition

More margarine is related to the demonization of animal fat, those with higher margarine probably have lower quality nutrition thus more stress, which could easily increase the divorce rate.

I wonder what the divorce rate in carnivore couples is compared to vegetarian couples

Honestly, just read the full paper directly. It's quite good and dense. I love this paper so much, its everything I ever want to say about food surveys, I'm going to pin it.

Notes:

In recent decades, many observational studies of people have associated consumption of red or processed meats with a variety of chronic diseases such as multiple types of cancer, various forms of cardiovascular disease, kidney disease, type 2 diabetes, obesity, and total mortality (Boada et al., 2016). Consider if a scientist were claiming that a new drug treated all these illnesses. The overwhelming response would be swift and certain that this was not possible. Yet, critics of meat consumption are firmly convinced that it causes multiple harms despite the softness of data supporting such claims, almost all of which are based on epidemiological associations.

Epidemiological, or observational, studies provide valuable clues to potential associations between exposure to an agent and development of a disease. Historically, that is how the sources of infectious disease were traced, such as during a cholera epidemic in London in 1854 that John Snow attributed to a contaminated well, or modern outbreaks of food poisoning like E. coli from salad greens. These relatively clear-cut sources of disease can be identified by epidemiological studies because the risk of infection in an exposed individual is several hundred times that of someone not exposed to the same contamination.

But, correlation does not equal causation. Many strongly correlated factors may have no relation, such as per capita consumption of margarine in the United States and the divorce rate in the state of Maine, which correlated at a level of 0.99 (the highest possible correlation being 1.00) over a 10-year period (Vigen, 2018).

Meat and cancer—cause and effect or only an association?

I'm just quoting whole paragraphs at this point, this is so well written and information dense, you should read it directly.

So when evidence of gene damage or oxidative stress in animals is claimed as the supporting mechanisms, but those animals do not get more cancer when fed three times the normal amount of protein combined with a calcium-deficient diet to see an effect, it is illogical to accept the mechanistic studies as confirmation of the epidemiology.

If you have to manipulate the studies to get a outcome you want, and even then you don't get it... there is a problem with the thesis your testing.

In addition, the IARC subgroup who evaluated mechanisms ignored two studies by one of its members in which bacon fed to rats actually significantly suppressed the precancerous indicators (Parnaud et al., 1998; Parnaud et al., 2000).

The working group evaluated over 800 epidemiological studies, but only 7 of 14 studies of red meat and 12 of 18 on processed meat found increased risk of colorectal cancer in people eating the most meat. There was insufficient data on meat intake and cancers at other sites in the body for the group to reach a conclusion. IARC stated that the working group was identifying hazard only—not evaluating the risk of getting colon cancer.

Bureaucratic games.....

this is circular reasoning because the risk among those eating the most meat determined whether meat consumption was deemed a hazard. Another weakness in the conclusion from IARC is that chance, bias, and confounding could not be ruled out with the same degree of confidence for the relation of red-meat consumption and colon cancer, but these three issues were considered unlikely in the linkage of processed-meat intake and colon cancer.

You can't generalize from one subgroup to another dietary context.

Nevertheless, the IARC working group concluded that for every 50 grams of processed meat eaten, the relative risk of colon cancer was increased by 18% compared with those who ate the least processed meat. How does this compare with known carcinogens? The increased relative risk of lung cancer from smoking cigarettes is 1000–3000%. The increased relative risk of liver cancer from eating moldy grains contaminated with aflatoxin is about 600%. In fields outside nutrition, the usual threshold for confidence about relative risk is in the range of 200–400%.

So even with the above games, the relative risk (not absolute risk) was 18%.

an 18% increase equals a relative risk of 1.18, and this score falls substantially below the threshold that epidemiologists in other fields generally accept as worthy of further investigation.

*The relative risk is a ratio of the disease rate in the group exposed to the highest amount divided by the rate in the group exposed to the lowest amount but this risk ratio does not reflect the absolute risk of a disease. *

The lifetime absolute risk of colon cancer in vegetarians is 4.5 out of 100; in people eating 50 grams of processed meat every day for a lifetime, the risk is 5.3 out of 100. These numbers are not statistically distinguishable in epidemiological studies

• Vegetarian 4.5 out of 100 lifetime absolute risk of colon cancer
• 50g Processed meat daily 5.3 out of 100 lifetime absolute risk of colon cancer

These numbers are not statistically distinguishable in epidemiological studies

THIS! This is why you never trust a study that only reports relative risk.

In other words, people who ate a lot of vegetables and fruit had no increased risk, no matter how much red meat they ate. These results again suggest that there are multiple other lifestyle factors that associate with dietary differences that account for the claimed differences in risk of cancer.

Yes. to me this indicates the analysis is looking at the wrong signal, perhaps industrial oils, carbohydrates, insulin resistance foods should be isolated and measured.

Those eating the most red or processed meat may be more likely to ignore other health recommendations and have multiple habits that contribute to the risk of disease.

This is exactly the healthy user bias we always speak of in epidemiology

By convention, most scientists use the 5% level of statistical significance as agreement that a group difference is likely reproducible. However, the large numbers of factors in nutrition studies—for example, 125 food items in a typical food frequency questionnaire, 40 nutrients, and 50 disease endpoints or risk factors—yield a total of 406,250 possible outcomes. Five percent of that number means 20,312 could be false positives.

Yes, typical epidemiology FFQ study has

• 406,250 possible outcomes
• 20,312 of the result pairs will be a false positive.

THIS IS WHY EPIDEMIOLOGY isn't compelling, why it can't be used as a justification for policy or personal health advice. It's just too noisy, and very open to phacking.

One example from the epidemiology literature clearly designed to show the limitations of standard statistical approaches in large studies used 10.6 million people equally divided into two cohorts—one for derivation of associations and one for validation of the relationship. The investigators searched 223 of the most common diagnoses for hospitalization in the medical records of the participants and found that 24 were statistically significant in the first cohort based on individuals’ astrological signs. Two of these associations remained statistically significant in the second cohort with relative risks of 1.15 and 1.38, numbers in the same range as the relative risk of processed meat and colon cancer (Austin et al., 2006).

This... this is the danger of taking association as causal

This entire article was pure gold and goes over every dimension of why food surveys are not sufficient to draw conclusions from. I consider this article a must read for anyone talking about, pushing, or consuming epidemiology papers.