How PupCurve works, and how well

A puppy calculator is only as good as what's behind it. This page says what we used, what we did to it, how it performed when we tested it on dogs it hadn't seen, and where it breaks.

The data

The growth curves come from the dataset published with Salt et al., “Growth standard charts for monitoring bodyweight in dogs of different sizes” (PLOS ONE, 2017). It holds anonymised weights of healthy dogs seen at Banfield Pet Hospitals in the United States, released under a CC BY 4.0 licence. The paper kept dogs that attended routine or ‘healthy pet’ visits and stayed in ideal body condition into young adulthood.

The paper published charts as images for five size categories up to 40 kg, and declined to chart the giant class because breeds like the Great Dane and the Mastiff grew too differently. It didn't publish the numbers. So we went back to the underlying records and did our own analysis. That means these curves are PupCurve's analysis of the public data, not the charts in the paper, and any mistakes in them are ours.

After our own cleaning we used 1,065,961 dogs and 5,371,181 weigh-ins.

Six size classes

We use the paper's six categories, set by adult weight. Class VI (over 40 kg) is the one the paper did not chart, so ours is looser.

For each class and each week of age from 8 to 104 weeks we measured the 10th, 25th, 50th, 75th and 90th percentile of share of adult weight: the weight at that age divided by the dog's own final weight. Between classes we blend smoothly by adult weight, so a 24 kg dog isn't forced into a bucket.

From weigh-ins to a forecast

1. Each weigh-in suggests an adult weight: the weight divided by the typical share of adult weight at that age.
2. The spread of that suggestion depends on age. A weigh-in at 9 weeks is noisier than one at 30.
3. Several weigh-ins are combined, but not as if they were independent. In the data, a puppy that runs ahead of its curve at 10 weeks tends to still be ahead at 13. We measured that correlation by the gap between weigh-ins, and it limits how much each extra reading helps.
4. If you give a breed or parent weights, that enters as one more estimate, with a spread that depends on the size class.
5. We never let the range get narrower than about 6% either side. Pretending to know more than that would be dishonest.

The tool shows the 80% range. A puppy's final weight should fall inside it about four times in five.

How accurate is it?

We split the dogs in half. We fitted the size-class curves on one half, then forecast the adult weight of the other half, one weigh-in at a time, at various ages. Here is what happened.

Error is the absolute gap between the forecast and the dog's final adult weight, as a percentage of that adult weight. The ranges held slightly more than 80% of the time, so they err on the cautious side. In this test the size class was treated as known, as it is when you enter a breed or parents. With weigh-ins alone, early forecasts are a little looser, because the model first has to work out which curve a puppy is on.

Off-curve alerts

The adult weight cancels out when you compare two weigh-ins, so we can check growth rate without knowing the final size. We compare the change between two weigh-ins with what healthy dogs did over the same ages and gap. Outside the 5th to 95th percentile of that change is flagged as growing faster or slower than usual. Any weight loss in a puppy under a year old is flagged separately, because it is worth a phone call whatever the curve says.

Home scales are noisier than a clinic's, so we flag cautiously and word the alerts as prompts to check, not diagnoses.

Other calculators

• Feeding. The National Research Council (2006) growth equation, ME = 130 × BW^0.75 × 3.2 × (e^(−0.87p) − 0.1), where p is current weight over expected adult weight. Adults use resting energy requirement (70 × kg^0.75) times 1.6 for neutered and 1.8 for intact dogs. Default food energy is 380 kcal per 100 g unless you enter yours.
• Pregnancy and whelping. 63 days from ovulation, 64 to 66 days from the LH surge or progesterone rise, and a 58 to 72 day window from the first accepted breeding, from the Merck Veterinary Manual and Cornell.
• Newborn checks. Daily weights for the first two weeks, and a warning for any loss after day one, from the Merck neonatal guidance.
• Cat age. Life stages from the AAFP/AAHA feline life stage guidelines. The human-year figure is a rough rule of thumb.

Where it falls short

• The records are from one US clinic chain. Dogs elsewhere may grow a little differently.
• Breed was recorded by owners and not verified, and mixed breeds were excluded from the original curves.
• Dogs were selected for ideal adult condition, so the curves describe healthy growth, not every dog.
• Giant breeds vary too much between breeds for one curve to fit well.
• Nothing here replaces a vet who can see and handle your puppy.

Found a mistake or have a question about the method? Email [email protected].