Project · Statistics and visualisation
Dino Park Kreidetal: a season in data
The park doesn't exist and the data is simulated. That's why, for once, I know the truth and can show exactly when a number lies. The mistakes that happen here happen in real company data every day, though.
The season at a glance
Every square in the calendar is one day. On normal days 82 % of visitors are satisfied, on hot days only 68 %, and on the 4 days when a raptor briefly got into the visitor area it even drops to 41 %.
- under 60 %
- 70 to 76 %
- 82 % and more
- raptor incident
- over 32 °C
12,152 simulated visitors · aggregated per day · R
Why 3,673 answers know less than 305
The tablets stand at the main exit, but anyone who's annoyed usually leaves through the east car park and so types something in less often. A simple random sample of ticket buyers gets pretty close to the truth, while the tablet stays around 15 points too high all season.
Non-response bias · 7-day rolling mean · 95 % Wilson interval for the sample: 73.8 to 82.9 %
Try it yourself
Set how many visitors come, how satisfied they really are and who actually answers on the tablet. Your browser then simulates every visitor one by one.
error … points · computed in …
JavaScript · Bernoulli-Simulation je Besucher · Math.random
The Triceratops-Wiese was just unlucky
There are twelve paths and we run one test for each. The Triceratops-Wiese looks dangerous at first (p = 0.013), but after correcting for twelve comparisons only p = 0.16 is left. In 2,000 simulated seasons with equally safe paths, 38 % end up with a false alarm if you don't correct.
Two-proportion tests · Holm correction · 2,000 Monte Carlo seasons
Worse overall, better in every group
Dr Farn heals 66 % of her patients and Dr Moos 77 %, so at first glance Dr Moos looks better. But 82 % of Dr Farn's patients are predators and therefore the hard cases, and split by species she is better in both groups. That's what's called Simpson's paradox.
- Dr Moos
- Dr Farn
Stratified comparison · 500 simulated treatments
22 T. rex? There are 3.
The inventory list has 138 rows and 22 different spellings for just 5 species. If you simply count rows you get 22 T. rex, if you filter for the exact name you get 1. After cleaning there are actually 3, which saves 23.0 tonnes of meat a week.
- rows in the raw list
- animals after cleaning
String normalisation · ID harmonisation · deduplication
Everything is computed in R with a fixed random seed, only the simulator runs live in JavaScript in your browser. Every number on this page comes straight from the simulations, none of it is entered by hand.