A common hypothesis is that an increase in road supply increase car use and that an increased supply of transit and trains reduce it. The purpose of this paper is to analyze how increased road supply as well as increased rail and transit supply impact car use, while controlling for self-selection effects by following individuals over time.

The earlier literature on how road supply impact car use has been using cross-sectional data making it impossible to control for self-selection effects (spatial sorting). Selection bias would imply that individuals who do have preferences against car travelling tend to settle in locations with high transit or slow mode accessibility (often characterized by dense land use). One exception is Duranton and Turner (2011), finding that car use is significantly increased by increased lane kilometres of roads. They use however aggregate data on the level of city in the US, and thus do not acknowledge that it is not the kilometres of roads in the city that determines the accessibility by roads for trips in the city, but the accessibility it creates. Moreover, Duranton and Turner disregard the potential endogeneity problem arising because it is more likely that roads are built in states where road use and thus congestion increases.

We use panel register micro data on the full population of Mälardalen from 1998 and 2005. These data include car use (vehicle kilometers per year from car inspections) and socioeconomics. We match this to historical transport systems from 1998 and 2005 (including road and rail supply and transit provision) provided by Trafikverket. Locations of residences and establishments are geocoded on a grid which is 1000 by 1000 meters in rural areas and 250 by 250 meters in urban areas.

We find that the ratio of changes in accessibility by car and public transport indeed have a positive impact on driving distances. However, the impact on car is very small relative to changes in incomes, population size and fuel prices.