Lucy & Bo Nutrition Love leads Science follows
The Denominator

Mathematics and our pets

Trying to make sense of what the pack says

Give me a sample size and a confidence interval and I know where I stand. Give me “human grade” or “all natural” and I’m lost.

The second part was most of my experience when I started looking for the right food for my two dogs. I went through packs at the store, brand websites and reviews, and kept coming away with lines I didn’t know what to do with.

One said its food was loved by 1,000 dogs. I kept wondering, 1,000 out of how many? And how did they arrive at that number? If it came from reviews, then it is mostly the happy pet parents who got counted. The unhappy ones rarely write in, they just quietly switch brands. So even a careful count would be counting the wrong group.

Another promised a visibly shinier coat in four weeks. But a dog’s coat changes through the year anyway, with shedding and the weather, so almost any four weeks will show some change. Most of us also switch food when the coat is looking its worst, which means it was likely to look better a month later no matter what went into the bowl. There’s a name for this, regression to the mean. And it’s usually us, the pet parents, deciding whether the coat looks better, right after paying for the new food. A fairer check would be to have someone who doesn’t know which food the dog was on look at the coat.

That last part matters more than I had assumed. I came across a study where dogs with arthritis were given a dummy treatment, and their pet parents, and even their vets, often reported that the dogs had improved. Meanwhile the instruments measuring how much weight the dogs put on their legs showed no change at all1. We see what we are hoping to see. I’m a pet parent and I know I’ve done this.

Then there is “boosts immunity”, which I see everywhere, and I still don’t know what exactly is being measured when a brand says it. Immunity isn’t a single number. It is many parts working together, and a group of experts who reviewed how to measure it in nutrition research concluded that no single marker can tell you whether a diet has changed it2. A study might show that some marker in the blood went up, maybe a type of antibody, and that is a real measurement. But a higher reading in a lab report is not the same as a dog that falls sick less often. Europe’s food safety regulator draws the same line for human food. It does not accept a change in immune markers on its own as proof for an immunity claim, and wants to see an effect on infections: how often they happen, how bad they get, or how long they last. When it published that guidance in 2016, it noted that every claim of this kind it had reviewed had been turned down, mostly because the studies never measured infections at all3.

Immunity also moves with things that have nothing to do with the bowl. A review of nearly 300 studies in people found that long-running stress is linked to a broad dampening of the immune system4. In dogs the research is thinner. One small study followed ten dogs moved from their homes into a large, noisy army kennel, and within ten days the level of an antibody in their saliva had dropped sharply in every one of them5. Their food and routine changed too, so even that study can’t say exactly which part of the move did it. But it’s a reminder that a dog’s immune markers can shift for reasons no food label would mention.

These lines are written in the language we want to hear as pet parents, and they work on me as much as on anyone. What they rarely give me is something I can check.

So I started reading studies instead, and found that numbers need just as much care.

Take a made-up but fairly typical study. Twelve dogs eat two foods, a few weeks on each. Food A is a dry kibble, Food B is a fresh cooked food. On Food B the dogs produce about half the stool, and it’s easy to see how that gets reported: fresh food, half the waste.

Some of this design is good. Every dog eats both foods, so each dog is compared with itself, and that removes a lot of noise, because dogs differ from each other far more than one dog differs from one week to the next. The usual way to analyse it is to take each dog’s difference between the two foods and check whether the average difference is bigger than normal week-to-week variation would produce. A paired t-test does exactly this.

What it gives you is a p-value. It is easy to read more into it than it says. All it really tells you is whether chance alone could have produced a gap this big: a small p-value means chance is an unlikely explanation. It says nothing about how big the difference actually is, or whether it matters. For that you need the confidence interval. Think of it as the honest version of the result: not one tidy number, but a range the real answer probably falls inside, like a weather forecast saying “between 28 and 34 degrees” rather than promising exactly 31. Every study has one of these ranges. A bigger study narrows it; a smaller one leaves it wider. That width isn’t the study failing, it’s the study being honest about how precisely it can measure. A single confident-looking number would hide that, not remove it.

What really changed how I read it was looking at what was in the bowls. Kibble is roughly a tenth water. Food B is roughly three-quarters water, so in 400 grams of it about 300 grams is water and about 100 grams is food. Take the water out and the dogs on Food B were eating noticeably less actual food. Food B also had about half the fibre, and fibre is a big part of what the body doesn’t digest and passes out. Less food going in, with less fibre in it, will come out as less stool. A good share of “half the waste” was decided by what went into the bowl, before digestion had much to do with it.

That doesn’t make Food B a bad food. It may be a very good one. What the study can’t tell us is why the dogs produced less, because the two foods differ in several ways at the same time: water, fibre, fat and how they were cooked. All the statistics can be done correctly and the conclusion still won’t follow, since no test can pull apart two things the study never kept apart. The word for this is confounding, and it shows up more often than you’d expect.

Two more things took me a while to appreciate. If a study measures fifty different things and checks each against the usual five per cent cut-off (everyone’s favourite alpha (α) value in statistics), you would expect two or three to look “significant” purely by chance, even when nothing real is happening. And when a result is “not significant”, that doesn’t mean there’s no difference. Small studies often just don’t have enough dogs to pick up a real effect, so not finding something isn’t the same as it not being there.

I haven’t invented this pattern. I’ve seen it in published pet food research.

There are also places where numbers don’t help much. Some things that matter in a dog’s life are hard to measure well, and a precise-looking number from a careless method can mislead more than someone honestly saying they’re not sure. No dataset replaces a vet examining the dog in front of them.

I don’t think numbers make anyone a better pet parent on their own. Most of us are doing our best with whatever is printed on the pack, and so was I. But asking these questions, slowly and probably still in an unstructured way, has changed what I believe and what I buy. That’s what I’ll be writing about here.

If any of this sounds familiar, I’d like to hear from you. And if you have further thoughts on any of it, I’d be glad to hear them. You can write to me at letters@lucyandbo.science.

Share this piece

References

  1. Conzemius MG, Evans RB (2012). Caregiver placebo effect for dogs with lameness from osteoarthritis. Journal of the American Veterinary Medical Association 241(10):1314-1319. https://pubmed.ncbi.nlm.nih.gov/23113523/↩
  2. Albers R, Bourdet-Sicard R, Braun D, et al. (2013). Monitoring immune modulation by nutrition in the general population: identifying and substantiating effects on human health. British Journal of Nutrition 110(Suppl 2):S1-S30. https://doi.org/10.1017/S0007114513001505↩
  3. EFSA Panel on Dietetic Products, Nutrition and Allergies (2016). Guidance on the scientific requirements for health claims related to the immune system, the gastrointestinal tract and defence against pathogenic microorganisms. EFSA Journal 14(1):4369. https://doi.org/10.2903/j.efsa.2016.4369↩
  4. Segerstrom SC, Miller GE (2004). Psychological stress and the human immune system: a meta-analytic study of 30 years of inquiry. Psychological Bulletin 130(4):601-630. https://doi.org/10.1037/0033-2909.130.4.601↩
  5. Skandakumar S, Stodulski G, Hau J (1995). Salivary IgA: a possible stress marker in dogs. Animal Welfare 4:339-350. https://doi.org/10.1017/S0962728600018030↩

Corrections: none so far.  Last updated: 23 October 2026

← Back to The Denominator