Method
How the score is calculated.
The Roost Score is a 0–10 number computed from a hotel’s guest rating, the number of reviews behind it, and the rating distribution of the country it sits in. Nothing about it is submitted by hotels and nothing about it is for sale.
A 9.8 from six guests is not better than a 9.2 from six thousand. Every rating is pulled toward the mean of its country in proportion to how little evidence stands behind it:
bayes = (v / (v + 50)) × rating + (50 / (v + 50)) × countryMean
where v is the review count. Fifty reviews is the point at which a hotel’s own rating carries as much weight as its country’s baseline. Below that the score is deliberately conservative.
Guests in different countries do not use the same scale. An 8.4 in Japan and an 8.4 in the United States describe different hotels. But normalizing this away entirely would erase genuine quality differences between countries, which are real and worth knowing. So we split the difference:
score = 0.70 × bayes + 0.30 × (bayes − countryMean + 7.0)
Seventy per cent of the score is the hotel on its own terms; thirty per cent is the hotel relative to its neighbours.
Not price. A cheap hotel is not penalised and an expensive one is not rewarded — price sits next to the score, never inside it.
Not star ratings. Stars are awarded by national bodies with wildly different criteria and are shown for reference only.
Not us. There is no editorial override, no promoted placement, no listing fee, and no form a hotel can fill in. If a property moves up, it is because the inputs moved.
The score inherits whatever bias exists in guest reviews, including the tendency of very small properties to collect unusually warm ones. The country baseline is computed from the hotels we hold for that country, so it moves slightly as coverage grows. Hotels with no reviews at all are not scored and do not appear in any ranking.
Every scored hotel carries its own breakdown — the raw rating, the review count, the country mean, and the weights used — on its page.