Research Evidence

Previous studies have sought to estimate how changes in lane miles have affected vehicle-travel. These studies result in estimates of elasticities of vehicle-miles of travel with respect to lane miles of road capacity. The majority of these studies use aggregate data on vehicle-miles of travel (VMT) and lane miles and typically use yearly data, that is all the VMT over the course of a year and all the existing lane miles in a given year summed over an entire state, county or metropolitan statistical area (MSA), i.e., a city or region. Much of this work extended the seminal studies of Hansen et al. (1993) and Hansen & Huang (1997) who analyzed data for California and showed a statistical association between lane-miles and VMT. Critics claimed that the results of these early studies did not demonstrate a causal effect, that is, that growth in lane miles led to growth in travel, but merely showed contemporaneous growth in VMT and lane miles.

Since 2000, a variety of different studies have all estimated empirical models that account for some degree of “causal impact”, and have led to more reliable estimates of the induced travel effect. Noland (2001) estimated state-level elasticities for multiple road types (controlling for cqapacity in prior periods) while Noland & Cowart (2000) estimated MSA-level elasticities and attempted an instrumental variable model, one of the main techniques to control for causality in statistics. Fulton et al. (2000) estimated a sequential model that demonstrated a causal impact of increasing lane miles using county-level data for three states. Cervero and Hansen (2002) and Duranton and Turner (2011) estimated models using instrumental variables. Hymel, Small, and Van Dender (2010) use state-level data and control for simultaneous effects, that is, how vehicle travel itself may affect the building of new capacity. Rentziou, Gkritza & Souleyrette (2012) use a different modeling approach and find similar effects. Graham et al. (2014) use data on urbanized areas to estimate a mixed model propensity score estimator that is sensitive to the “dose”, or amount of lane mileage. Hymel (2019) also estimates a model with instrumental variables and comes to the conclusion that “capacity expansion is not a viable long-term solution to urban traffic congestion” (p65). The modeling in these studies control for causal effects and demonstrate a linkage between road expansion and increases in VMT. All have found statistically significant impacts, but the elasticities generated are quite different.

Cervero (2003) also estimated a model using facility-specific data, in this case measuring road capacity for selected projects rather than aggregate changes in lane miles. Again, he found statistically significant results in this case using a structured equation model, also known as a “path” model. Cervero (2003) found that 40% of VMT growth can be attributed to capacity improvements (the remainder being due to other effects). This was substantially more than Noland’s (2001) estimate that attributed 28% to capacity improvements. Surprisingly, Cervero (2003) interpreted this as not being a large contributor to VMT growth, while Noland’s study was interpreted as showing capacity expansion was the major influence. If anything, this shows how modeling results can easily be misinterpreted and in all cases, these models controlled for other factors that affect VMT growth.

At least three reviews of the induced travel literature have been done over the last 25 years. Noland & Lem (2002) summarized existing work up through about 2002. Noland & Hanson (2013) also produced a review that included work conducted in the 2010’s including some simulation and modeling studies and the influential work of Duranton & Turner (2011). These, and more recent work, are summarized in the table below. More recently, Volker & Handy (2022) published a review as part of their development of the California Induced Travel Calculator. These elasticities in the table below show a range of short-term and long-term elasticities and some variation between studies. Most research suggests that larger and generally more congested roads (for example, interstate highways in urban areas) will have larger elasticity values. These results are used in the calculator as specified in the data section.