Table 2.

Logistic regression models predicting onboard incidents

VariableModel 1Model 2Model 3
Dependent variableEconomy class incidentEconomy class incidentFirst class incident
DatasetAll flightsFlights with first classFlights with first class
Predictor variables
 Economy seats1.0010 (0.0012)1.0031** (0.0014)
 First class seats1.0342** (0.0139)
 Economy seat width (cm)0.9514* (0.0243)1.2175*** (0.0922)
 Economy seat pitch (cm)0.9887 (0.0101)1.0093 (0.0125)
 First class seat width (cm)0.8147 (0.1101)
 Flight distance in miles1.0004**** (0.0001)1.0004**** (0.0001)1.0003** (0.0001)
 Flight delay in hours1.1524**** (0.0151)1.1393**** (0.0157)1.0526 (0.0468)
 Cabin area (m2)1.1186** (0.0528)1.1213** (0.0610)1.4777*** (0.1969)
 International flight (1 = yes)0.6840**** (0.0681)0.7185*** (0.0720)0.8212 (0.1869)
 First class present (1 = yes)3.8431**** (0.4743)
 Boarding from front (1 = yes)2.1754*** (0.6083)11.8594** (11.8367)
McFadden’s pseudo R20.10280.05780.0675
  • Values presented are odds ratios with robust SEs. The full dataset represented ∼150–300 unique arrival and departure airports, and between 500 and 1,000 unique flight routes. SEs are adjusted clusters based on plane route (i.e., the specific departure airport and arrival airport combination). All models include fixed effects for flight regions (suppressed for space but included in SI Methods). Observations were dropped because they were in a flight region that had no incidents. Flights with first class present are ∼46.1% of the population of flights. No flights without first class boarded from the middle of the plane. *P < 0.10, **P < 0.05, ***P < 0.01, ****P < 0.0001.

  • Seat pitch data are not available because many first class seats had their own pods/beds.