About

This website displays on-time performance by U.S. airlines and airports, as collected by the Bureau of Transportation Statistics, with data from this site. The numbers are based on historical records from one year of historical data (July 2014 through June 2015), only for flights between the busiest 100 airports during this period. This is 4,881,735 flights total.

The percentages are calculated with Bayesian smoothing: if a particular route has very few flights, like less than a thousand, we display a number that's not the raw percentage but one closer to the overall average, since there isn't much data yet to make a deterimination of the failure or delay rate. Hovering the mouse over a percentage will show the raw data. [Details: we use a simple empirical Bayes approach, calculating the posterior mean given a Beta prior whose parameters were fit on the dataset of MLEs for all route/carrier combinations. The CSV files contain "pfail_raw" and "pdelay1hr_raw" columns which are the simple percentages. Multiplyling "pfail_raw" by "n" will get the number of failed flights. Thanks to Mike Love for the suggestion.]

There are many factors that don't get taken into account. For example, these calculations aren't very good at assigning blame between the carrier, origin airport, and destination airport. And the delay and failure rates don't consider really important factors like weather or time of flight, etc.

In the meantime, the code and data is available at github.com/brendano/flightstats.

Website started by Brendan O'Connor when stuck in O'Hare.