In May, System Logic wrote about the potential effects of the sequestration on air traffic delays, as the FAA furloughed air traffic controllers. Since then we received additional data from FlightStats and conducted further analysis of the delays. We found that while more flights were delayed during the sequester, the effect was relatively small.
We analyzed daily flight delay data at 41 major commercial airports in the U.S. between January 1, 2012 and April 30, 2013. Using all 17,430 airport-day observations and a series of regression models, we examined whether delays were more common during the furlough. We controlled for airport, year, and month fixed effects to capture all (observed and unobserved) stable differences across airports as well as year- and month-specific trends common to all airports. We also controlled for the number of flights handled at each airport each day. This allowed us to estimate the net effect of the furlough.
The results, which were robust to a variety of specifications and model types, consistently indicated that the furlough itself only had a very modest effect on delays. Excessive delays (defined as 45 minutes or more) were not statistically more likely to occur during the furlough period than otherwise. Very late flights (those with a delay between 30 and 45 minutes) and late flights (those with a delay between 15 and 30 minutes) were more common during the furlough but only marginally so. The percentage of very late flights increased by less than one percentage point (from the “normal” level of 3.5% to approximately 4.1%), and the percentage of late flights increased by just slightly more than one percentage point (from the “normal” level of 7% to approximately 8.1%). Overall, the percentage of on-time flights fell by only 2-3 percentage points, from around 81% to 78-79%, as a result of the furlough.
And, accounting for the furlough in our models only explained roughly 1% additional variance in delays. By comparison, accounting for differences across months, which help account for the number of passengers and gross weather effects, explains at least 7% variance. The furlough’s effect was measurable but modest. Given that roughly 20% of flights are late even during normal times, this impact on delays was far from exceptional. That nearly one-fifth of flights routinely failed to arrive on time during the whole period is a much more dramatic conclusion about the fragility of our national infrastructure than the minor increase in moderately late flights due to the furlough.
While our models would explain even more variance if we incorporated weather effects in a more fine grained way and controlled for the operational and maintenance idiosyncrasies of different airlines, these refinements are unlikely to change our conclusion about the furlough’s impact.
Image courtesy of ZHoover123, Wikimedia Commons.
As featured on the Flight Stats blog.