Data vs. Intuition in Nutrition Science

Is Data Being Used in The Right Way?

We live in what should be an amazing time to be alive. We have an overwhelming amount of data at our fingertips, allowing us to be better equipped than ever to make good decisions. Artificial intelligence can analyse, summarise and synthesise information with unprecedented speed. With more data readily available, academics can build on previous research, businesses can learn more about their consumer’s behaviours and governments can make better policies. But is more data always a good thing?

Although, generally speaking, more data helps us to become wiser, not all the information we encounter will actually be factual. Many things can go wrong on the path from information to truth, and many honest mistakes will be made in the process of acquiring data. Sometimes people will inadvertently misuse data through succumbing to their biases or because they fear being wrong. In some cases, malevolent actors motivated by greed or power might intentionally manipulate data in their attempt to deceive. Sadly, bad, insufficient or misused data will result in experts misidentifying how certain diseases are caused.

We all have biases, and it’s the role of science – through the collation and analysis of data – to mitigate these biases in the quest for truth. But do most people even understand how science really works? How many of us have more than a mild comprehension of scientific methodologies? Do we really know how to analyse and use data properly? Take food, for instance. An average person consumes 32 tonnes of food in their lifetime [1], providing ample opportunity for the collection of minute data points on the frequency, volume and impact of the food we eat. Is this enough to mitigate biases on nutrition? Far from it. In fact, as I lament in Nutrition and Enlightenment Values, “few scientific disciplines invoke such an unashamed display of biases as nutrition and dietetics” [2].

We Risk Misunderstanding Risk

In a 1959 lecture called “The Two Cultures”, the scientist and writer C.P. Snow identified a phenomenon known as the “gulf of mutual incomprehension”. Snow warned that society was “being split into two polar groups”: those who understood science and those who did not [3]. And you wouldn’t be criticised for feeling that these days things are even worse. During COVID, UK Prime Minister Boris Johnson was, it’s claimed, “bamboozled” by science and “struggled with the whole concept of doubling times”. When it comes to statistics, says Sir Adrian Smith, the head of the Royal Society, Britain is “very bad” indeed [4].

But it’s not just Britain’s politicians. The majority of social media health content makers across the globe can’t grasp even fundamental concepts like absolute and relative risk when putting out their content (or, more cynically, they don’t care to understand them). This issue provides a great example of how data is commonly misused.

Absolute risk (AR) refers to the actual probability of an event occurring in a population. It gives a direct measure of risk without comparison to another group. Relative risk (RR) is the ratio of the AR in an exposed group to the AR in an unexposed (or control/baseline) group. It quantifies the strength of the association between exposure and outcome, indicating whether exposure increases, decreases or has no effect on risk.

• IE = Intervention events

• IN = Intervention non-events

• CE = Control events

• CN = Control non-events

Let’s consider the relationship between processed meat consumption and colon cancer. Suppose in a population of 1,000 people who don’t eat processed meat, five people develop colon cancer over 10 years. Their AR of colon cancer is five in 1,000, i.e. 0.5 per cent. In another population of 1,000 people who regularly consume processed meat, seven people develop colon cancer, meaning their AR is seven in 1,000, i.e. 0.7 per cent.

The RR is calculated as follows:

RR = 0.007 / 0.005 = 1.4

This means that people who consume processed meat have a 1.4 times higher risk of developing colon cancer compared to those who don’t. This translates to people who eat processed meat have a 40 per cent higher risk of developing colon cancer compared to those who don’t.

AR tells us the real-world probability (0.5 per cent vs. 0.7 per cent). RR highlights the comparative increase (40 per cent higher risk). A small AR increase may appear more alarming when presented as a large RR. It’s crucial to be aware of this when interpreting statistics relating to disease risk in nutritional epidemiology.

Quantitative Fixation

Data allows us to rely less on gut instinct. Yet academics and firms need to be aware of a different set of dangers: it’s possible to rely on data too much. During the Vietnam War, US Secretary of State for Defense, Robert McNamara, measured the success of the war through body counts, but he neglected non-measurable aspects like morale, politics and public sentiment. His actions gave rise to the concept known as the “McNamara Fallacy”, which refers to the flawed reliance on quantifiable metrics while ignoring qualitative factors in decision-making [5]. The fallacy works as follows: what can be quantified is measured (in McNamara’s assessment of the Vietnam War, enemy deaths) and what can’t be measured is ignored (strategic effectiveness). As what can’t be measured is assumed to be unimportant (such as civilian support), the conclusion is that only measurable metrics matter. This can lead to poor decisions.

The McNamara Fallacy – also known as the “Quantitative Fallacy” – is common in business, healthcare and education, where data-driven policies may overshadow critical but unquantifiable elements like creativity, ethics, culture and wellbeing. Avoiding it requires balancing quantitative analysis with qualitative judgement, recognising that not everything valuable can be measured [6]. Could it be that, at times, we’ve become so fixated on data that we’re partially blind to other aspects?

In a recent paper, researchers identify a cognitive bias related to the McNamara Fallacy that they call “quantification fixation”. The authors suggest that people put disproportionate weight on numbers because data are particularly suited to making comparisons [7]. An example could be the overemphasis nutrition scientists put on the nutrient value of food. Let’s say a hospital meal is rich in protein and fibre, contains good amounts of most vitamins, minerals and essential fats, while being low in sugar, salt and saturates. Quantitatively, the food has all the attributes of a healthy meal, firmly ticking the “nutritious” box and delivering what we’d expect any good medical institution to be providing its patients. However, what if the way it’s prepared, reheated and presented to the patient makes it unenjoyable? How patients feel psychologically and emotionally improves their outcome. Not only is food a pleasure, but an enjoyable meal is more likely to be consumed aiding the patient’s recovery. The quantitative attributes of the meal are strong, but the hospital has failed in its objective.

When endowed with more and more data, institutions risk wanting to quantify everything. But some things are vaguer than others. A meal’s enjoyment is much harder to express as a number than its nutrient value. Numbers allow for easy comparisons, but they fail to always tell the whole story. Knowledge of the nutritional value of what we eat has been, and will continue to be, incredibly useful, but we must be mindful that the food we enjoy is greater than the sum of its nutrient parts.

Data Warning

There are other risks, too. Humans bring the same cognitive biases to their analysis of numbers as they do to other decisions. Take confirmation bias, the propensity to interpret information as support for your point of view. Beliefs influence interpretation. How often do you see social media fearmongerers cherry-picking papers to back up their claims? Or, worse, citing a paper but completely misinterpreting the results?

Plenty of people struggle with basic data literacy, and too often people completely shun data. As we embrace more AI models, relying on algorithms may seem like the sensible solution. But these models draw their data only from what they know: they’re only as good as what they’ve been programmed to do. Maybe, at times, humans have an advantage. Datasets can tell us about the world as it is, but not the world as it might or should be. It’s hard to evaluate radically new innovations by looking at existing patterns. None of this is to say that our instincts should trump data. Humans have way too many biases, so to say we always make better decisions than machines is far from the truth. But we should be mindful of the limitations of data. Numbers might promise rigour, certainty and objectivity, but let’s not forget that they have flaws, too.

References:
1. Reference (2015) ‘How Much Food Does a Person Eat in a Lifetime?, 4 August. Available at: https://www.reference.com/world-view/much-food-person-eat-lifetime-1015c822087dfa41 (Accessed: 9 March 2025).
2. Collier, J. (2023) ‘Nutrition and Enlightenment Values’, Thought for Food, 28 November. Available at: https://jamescollier.substack.com/p/nutrition-and-enlightenment-values (Accessed: 9 March 2025).
3. American Physical Society (2017) ‘Archives: This Month in Physics History: May 7, 1959: C.P. Snow Gives His “Two Cultures” Lecture’ May. Available at: https://www.aps.org/archives/publications/apsnews/201705/physicshistory.cfm (Accessed: 9 March 2025).
4. The Economist (2024) ‘The Trial of Lucy Letby Has Shocked British Statisticians’, 22 August. Available at: https://www.economist.com/britain/2024/08/22/the-trial-of-lucy-letby-has-shocked-british-statisticians (Accessed: 9 March 2025); ibid (2).
5. Baskin, J. S. (2014) ‘According to U.S. Big Data, We Won the Vietnam War’, Forbes, 25 July. Available at: https://web.archive.org/web/20140911123156/https://www.forbes.com/sites/jonathansalembaskin/2014/07/25/according-to-big-data-we-won-the-vietnam-war/ (Accessed: 9 March 2025).
6. The Economist (2025) ‘Beware the Dangers of Data’, 2 January. Available at: https://www.economist.com/business/2025/01/02/beware-the-dangers-of-data (Accessed: 9 March 2025).
7. Chang, L. W. et al. (2024) ‘Does Counting Change What Counts? Quantification Fixation Biases Decision-Making’, Psychological and Cognitive Sciences, 121(46), e2400215121.