Methodology & data sources

The short version

We did not copy a growth chart out of a paper. We downloaded the raw data behind the paper — 14,957,947 weight measurements from 3,475,842 dogs — and recomputed the growth curves ourselves. That is why we can tell you our error bars, publish curves for age ranges and size groups the original charts don't cover, and show you where our method and the paper's diverge.

The data

Our puppy growth curves come from the dataset released alongside:

Salt C, Morris PJ, German AJ, Wilson D, Lund EM, Cole TJ, Butterwick RF (2017). Growth standard charts for monitoring bodyweight in dogs of different sizes.PLoS ONE 12(9): e0182064. Published 5 September 2017. WALTHAM Centre for Pet Nutrition.
A correction was later published: PLOS ONE 19(11): e0314711 (2024). We note it because citing the original without the correction would be sloppy.
Data: Liverpool Data Catalogue, DOI 10.17638/datacat.liverpool.ac.uk/377, released under CC BY 4.0. Records come from Banfield Pet Hospital electronic medical records.

The dataset holds 31 breed cohorts. We verified the row count and the units directly rather than taking the readme's word for it — weights are kilograms, and ages run from birth to just under 3 years.

How we built the curves

The paper fits GAMLSS/BCPE models and publishes the result as figures — there is no table of parameters to copy, which is part of why so many sites cite this paper without using it. We took a different and simpler route that the raw data supports directly: empirical percentiles.

• Dogs are grouped by adult size category (the paper's six: under 6.5 kg, 6.5–9, 9–15, 15–30, 30–40, and over 40 kg) and by sex.
• Ages are binned weekly from week 6 to week 52, then in 4-week bins to week 156.
• Within each cell we compute the 3rd, 10th, 25th, 50th, 75th, 90th and 97th weight percentiles. Cells with fewer than 30 records are dropped.
• 8,520,272 weight records survive our filters and sit behind the published curves.

To predict adult weight we find which percentile your puppy is on today, then follow that percentile to the age its size group finishes growing. Then we apply a calibration factor — see below.

Where we depart from the paper, and why

The paper required each dog to have a predicted body condition of "normal" at a visit between ages 2 and 3, and excluded any dog ever rated thin or heavy up to age 3. We tried to apply that exactly. The problem: in the released dataset the body-condition columns are almost entirely empty — 98.6% of the predicted ratings are missing. Applying the paper's criterion literally leaves 96,505 of 3,475,842 dogs — 2.8%.

So we relaxed it: we exclude any dog ever rated thin or heavy, and use only preventive-care or healthy-pet visits, but we do not require the 2–3 year "normal" rating. Then we checked whether that mattered, by building both versions and comparing:

A few percent is small next to the ~15% spread of the prediction itself, so we went with the larger sample. You should know we made that call, which is why it is written here rather than buried.

Two things our curves do that the published charts don't

How accurate it is — measured, not asserted

Any calculator can claim accuracy. This dataset lets us test it, because we can predict a dog's adult weight from its puppy weight and then look up what it actually grew into.

We found 127,964 dogs with both a visit at 8–12 weeks and a visit near maturity. We split them in half. Calibration factors were fitted on one half (63,993 dogs) and every number below is measured on the other half, which the fitting never saw:

An 80% interval that catches the truth 79.9% of the time is the point: the range means what it says. And a median error of 15% is not us being modest — predicting an adult dog from a 9-week-old puppy is genuinely uncertain. A site quoting you one number to one decimal place has not beaten this uncertainty; it has just declined to mention it.

Accuracy by size, held out: toy 18.7%, small 14.5%, small–medium 16.3%, medium 12.4%, large 12.1%, giant 12.8% median absolute error.

What we deliberately do not claim

We do not publish breed-specific growth curves, and we want to be direct about why: the released dataset does not identify breeds by name. It ships a numeric ID, not a name. Several cohorts sit within 0.1 kg of each other on average adult weight — 34.16 kg versus 34.09 kg, for instance — so assigning those IDs to specific breed names would add precision the data does not support. It would be a coin flip dressed up as data.

So picking a breed here selects the right adult size curve and cross-checks the answer against that breed's AKC standard. It does not load a curve built only from that breed. Nobody would have noticed the difference. We would have.

The dog age calculator: a source that wasn't there

The puppy calculator was a data problem. The dog age calculator turned out to be a sourcing problem, and it's worth reading if you use any dog age calculator anywhere.

The standard chart — 15 human years for year one, +9 for year two, then +4/+5/+5.5/+6.5 per year by size — is nearly always cited to the AVMA and the AAHA Canine Life Stage Guidelines. We fetched both documents. Neither contains it. The AAHA guidelines mention "human years" exactly zero times and state that "rather than attempt to calculate age equivalents to humans", life stage should be used instead. The AVMA publishes lifespan bands by size, not aging coefficients. Different sites also use different coefficients for the same buckets. We could not trace the numbers to any primary source.

So the calculator is built from three separate things, kept separate:

• Life stage — AAHA's own definitions (JAAHA 2019;55:267–290). Senior is the last 25% of expected lifespan.
• Lifespan — McMillan et al. (2024), Scientific Reports 14:531, CC BY 4.0. 584,734 UK dogs. AAHA's definition needs a lifespan number and deliberately doesn't give one; this supplies it. We reproduce the full 155-breed median table from the paper's Supplementary Table 2 — extracted from the source .docx, not retyped from a secondary write-up, and cross-checked against ten figures quoted in the paper's own results text. We do not use the AVMA's lifespan figures: that page attributes them to an American Kennel Club expert-advice article but publishes no methodology, sample size or study, and they are pessimistic against a 584,734-dog study.
• Human-age equivalence — Wang et al. (2020), reported as Wang's result with Wang's own caveat (104 dogs, primarily Labradors, no size term). We never blend it into the stage model: that blend is the exact step AAHA declined to take.

Two honesty notes. McMillan is a UK population drawn from rehoming, insurance and vet-corporate records, so its medians are conditional on a dog entering those datasets — dogs dying very young are likely under-counted. And its size classes are categorical, taken from Kennel Club and FCI breed lists; there are no weight thresholds in the paper, which is why our size dropdown asks about the breed rather than asking you to weigh your dog.

We also nearly got the muzzle finding backwards, so it's recorded here: long-muzzled breeds (12.1 years) do not outlive medium-muzzled ones (12.8). The well-known "small long-nosed dogs live longest" line comes from an interaction (small and long-muzzled = 13.3 years), not from a main effect.

One number in the paper is internally inconsistent and we had to make a call: Miniature Dachshund is 14.0 years in the results text but 12.2 in the Fig. 3 caption. Supplementary Table 2 — the underlying data table — gives 14.0, so that's what we use. We'd previously left the breed out entirely rather than guess; pulling the source table is what resolved it.

Reproducing this

The source data is public and openly licensed. Anyone can download the 140 MB archive from the Liverpool Data Catalogue and check us. If you find an error in our derivation we want to hear about it, and we will correct the page and say what changed.