How to read a dog study without getting lost

“Dogs” is a very roomy category. Dogs whose owners buy a DNA test, dogs seen at a referral hospital, insured dogs and dogs in a litter registry enter the evidence through different doors. Ask who had an opportunity to be included—and who didn’t.

In McMillan and colleagues’ 2024 longevity study, 18 organizations contributed UK records. The combined dataset contained 584,734 dogs, including 284,734 deceased dogs. It wasn’t a worldwide random sample. Read the population and methods.

For your own reading notes, keep the country, dates, setting, breed coverage and exclusions together. A very large number can make an estimate precise without making every kind of dog equally represented.

Missing records can change the story.

A Norwegian registry study of perinatal loss excluded whole litters in which every puppy died, because those litters weren’t reported to the registry. The authors say so in the methods. That matters when you interpret its percentages—or a recorded zero in a selected group. Tønnessen et al., §2.2.

“No events recorded” needs its sample size and selection rules beside it. It cannot establish that an event is impossible.

A denominator is the total underneath a fraction. It tells you what the percentage is a share of. Puppies, litters and breeds are different units, even when the paper is looking at the same event.

The Norwegian study reports both the share of puppies lost and the share of litters with at least one loss. Here’s what happens when you keep those two questions separate.

Before you compare two percentages, finish this sentence for each: “Out of all the ___, ___ had ___ during ___.” If the blanks change, pause. You may be comparing two useful findings that answer different questions.

Next, find what the number actually measures. A mean is an arithmetic average. A median is the halfway point; a survival-analysis median is the age at which estimated survival reaches 50%. A hazard ratio compares rates over time under a model. These labels aren’t interchangeable.

The longevity paper reports estimated median survival of 12.7 years for the purebred group and 12.0 for the crossbred group. It also reports 95% confidence intervals. Keep the estimate, interval and population together. McMillan et al., Survival analysis.

A 95% confidence interval comes from a method designed to cover the underlying value in 95% of repeated samples, under its assumptions. The NIST statistical handbook explains that repeated-sampling idea. It isn’t a promise about your dog, and it doesn’t account for every possible bias in recruitment or measurement. Höfler and Seaman discuss that distinction.

Don’t let a tiny p-value do all the talking.

A p-value describes how incompatible the data are with a specified statistical model; it doesn’t tell you the size or practical importance of a finding, or prove a cause. Ask “How much?” and “Compared with what?” before “Was it significant?” That’s central to the American Statistical Association’s guidance (ASA summary and principles).

And when two papers differ, compare their questions first. Teng and colleagues’ UK life-table study estimates expected remaining years at a given age. That is a different quantity from median survival. Don’t turn different measures into a trend.

Now return to the headline. Does it describe an observation, or has it quietly become a recommendation? An association can point to a worthwhile question without establishing that changing one factor will change the outcome.

Look at how the comparison was made, which other factors were considered, and what the authors could not measure. If the study didn’t test an intervention, don’t write its conclusion as an instruction to perform one.

Finally, check the publication date, corrections, funding and competing interests. A commercial connection is something to understand, not a substitute for reading the methods. Follow a useful finding to later work; a review can help you see whether several studies agree and why they might differ.

If the answer could change a health or breeding decision, take the original paper and your notes to your veterinarian or an appropriate specialist. “Here’s the result; here’s where I’m unsure it applies” gives them something concrete to work with.