The advertised stars of last week’s Dragon Week egg pool were the elusive Pokémon, Gible and Deino. Although numerous Gible hatches were observed during the first day of the event, it took until day three for a single Deino hatch to be reported in the Research Group’s verified egg project! Later in the week, anecdotal reports from travelers on the Road suggested that Deino hatch rates may have increased. Reports like these are hard to validate without large controlled data sets, but luckily the members of the Silph Research Group’s egg project were up to the task!
The graphic above shows data from over 1,100 hatches throughout Dragon Week, separated by pickup date. As more data points accumulated, it became apparent that there may have been an increase in the Deino hatch rate around August 2nd, but the low number of Deino hatches made this difficult to validate.
To assess the statistical significance of this observed step change in probability, we used a model selection metric called the Akaike Information Criterion (AIC). AIC assesses how well different models fit the data, and also penalizes each model for the number of parameters used.
In this case, we used two models:
Model 1 is a simple binomial distribution with a single probability.
Model 2 includes two different probabilities, one for the earlier low rate and one for the later high rate, as well as where the change took place (3 parameters).
Using AIC, we show that there’s strong support (0.047 relative likelihood) for an increase in Deino hatch rates around August 2nd¹. In other words, Model 2’s step change in Deino hatch rate is about 20 times more likely than Model 1’s constant Deino hatch rate. If Model 2 is correct, Deino’s hatch rate was likely less than 1% for the first three days of the event, but after the change researchers recorded a hatch rate above 3%.
Thanks for reading, travelers, and we’ll see you next time!
Analysis by Lead Researcher Blondey and Scientist Titleist
Graphic by Scientists WoodWoseWulf and Titleist
Article text by Senior Researcher DeathbyToast and Scientist Titleist
More than 40 Senior Researchers, Lead Researchers, and Scientists contributed their walking shoes, incubators, and blisters to this study.
¹Similar results were obtained using the Bayesian Information Criterion metric (BIC), with a relative likelihood of 0.088.