We rank the best and worst locations in this blog post. Here, we present the data and methodology for this ranking.

The availability of each location is checked periodically (currently every minute or so, every 15 minutes in the past) and, if the soonest appointment date has changed, we record the new nearest appointment. We call this an “update”.

Analyzing the past data for each location, we can determine within a given time frame how often appointments were available. We define an appointment as available if the nearest date is within a certain number of days of the time at which the update was made (max days out).

The main metric we use to rank locations is the fraction of time within a given time frame in which an appointment was available. This fraction is shown as a percentage in the tables below, which show different time frames (from January 1st to June 5th or March 1st to June 5th) as well as different max days out (30 and 90).

We then rank locations based on this fraction of time they had an available appointment: 0% being no appointments were available during this time frame and 100% being appointments were always available.

Some locations only operate periodically or were not operating during a large chunk of our time frame. To filter these locations out, we use the number of updates as a proxy for the location’s activity. To make a final list of the best or worst locations, we can define a minimum number of updates for a location to be considered active and discard locations with too few updates. This threshold number of updates will depend on the length of the time frame: longer time frames will have a higher threshold number of updates.

The plot at the top of the page shows the distribution of location probabilities of finding an appointment under 30 days away since January and filtering any locations with fewer than 190 updates. Below are the tables listing the locations, probabilities, and number of updates.