1. QUESTIONS
Induced travel occurs when new road capacity is built, and this is usually assumed to be more likely when capacity is added to a congested highway. The new capacity, e.g. a new lane, will lead to reductions in travel time which will lead to increases in travel, especially over time as individuals adjust their schedules, route, choice of mode, trip destinations, trip frequency, and where they live and work. In the long run, a new equilibrium will occur between the amount of travel demanded and the supply of highways (Noland and Hanson 2013).
Research over decades has shown relationships between increases in lane-kms and increases in vehicle-km of travel (VKT). This includes causal studies based on instrumental variable estimation (Duranton and Turner 2011; Hymel 2019) with one study using a detailed propensity matching technique (Graham, McCoy, and Stephens 2014), among others. Volker and Handy (2022) provide the most up to date review of induced travel studies. All of these studies estimate elasticities for higher functional road classes, usually for interstates, motorways, and principal arterials. To the best of my knowledge, no research on induced travel associated with local roads has been published.
The question to be answered here is: Can trip generation associated with a new development provide a measure of induced travel elasticities for local roads?
2. METHODS
Data for this analysis is drawn from the New Jersey Department of Transportation (2024) and the Maryland Department of Transportation (2024). This data includes daily vehicle-miles of travel (VMT) and lane-miles for a variety of road classifications. The New Jersey data includes center-line mileage, while Maryland provides lane-mile data. For New Jersey, the center-line mileage is multiplied by two, as almost all local roads have one lane in each direction. Only data on local roads is used; this data includes both rural and urban roads. “Urban roads” are defined by the Federal Highway Administration as those roads in urbanized areas (based on Census definitions) with a population of 5,000 or more. The daily VMT data is annualized (times 365 days) and all data is converted to metric units and reported as such here. A limitation is that data reported by states on local roads may be of lower quality than what is reported for higher functional classifications, especially estimates of VMT.
The approach used to estimate induced travel elasticities is based on a simple calculation of trips generated from households in a theoretical new residential development with a small network of local roads. Assume a new development with an area of about 17.5 hectares. This can hold 80 lots of about 0.22 hectares each (about a typical ½ acre lot in Imperial units), a somewhat typical exurban development pattern. This estimate does not include space dedicated to roads. If we imagine a small network as shown in Figure 1, with each lot having about a 30 m frontage and 60 m in depth, this leads to eight houses along a 120 m stretch of road and we can assume six roads at 120 m intervals. In my estimate of road length, I have not counted space for the road, such as intersection size. This network now has 2.16 km of center-line road length and 4.32 lane-kms. Most exurban developments have curvilinear roads, a simple network, however, allows for simpler calculations.
.jpg){kind=icon}
Household trip generation rates are the initial input. Based on the US National Household Travel Survey (NHTS) (Bricka et al. 2024) for 2017, there were 1865 vehicle trips/year per household (Table 3-6 in Bricka et al.). This comes to 5.1 trips per household per day. Similarly there were 1231 trips/year per person, or 3.37 trips per person per day (Table 4-3 in Bricka et al.). This would come to 8.43 trips per household per day assuming 2.5 people per household. These estimates likely include only trips generated from the home so may be an underestimate, as they do not account for trips to the home. Results from the 2022 NHTS are lower, but this could have been reduced due to the pandemic. An older study conducted in Virginia found rates of 9.2-10.8 trips per house and included commercial trips associated with each house, something likely to have increased substantially since their study was done (Ulmer et al. 2003).
Given this range of trip generation estimates, I calculate VKT estimates within the development for 2, 5, and 10 trips per household per day. Referring to Figure 1, the distance for each lot from the Main St., which leads to the development exit is 30, 60, 90, and 120 m. In total this would be 300 m of driving distance, or 600 m for the entire road counting the lots on both sides of Main St. Assume 120 m to exit the development plus 120 m for each subsequent block along Main St. up to 720 m for the last road (6th St.). The amount of travel for all houses on each street segment to exit the development is shown in Table 1 and the total VKT for the development based on one trip per household is 39.6 km. Assume that each trip includes a return and 2 trips per house per day this comes to 158.4 km. The annual total VKT generated by the development is then 57,816 km. Table 2 shows estimates for different trip generation rates that are used for the induced travel elasticity calculations.
3. FINDINGS
Thus, for 4.32 lane-kms of new local road, between 57,816 and 289,080 km/year are generated, depending on the number of daily trips from each household. This calculation assumes that those living in the houses built on the new roads are newly relocated to the region. Duranton and Turner (2012) and Garcia-López, Gomez-Hernandez, and Sanchis-Guarner (2024) provide evidence for how new roads can lead to population growth, so this assumption seems reasonable, although both studies focused on major roads. Garcia-López, Gomez-Hernandez, and Sanchis-Guarner 2024 estimated an elasticity of population with respect to lane-kms of 0.723 and for employment of 0.698 in Great Britain, while Duranton and Turner (2012), using US data, estimated a much smaller elasticity of 0.13. Some of the residents may have relocated within the region, but newcomers would have moved to where they previously resided especially given that in most areas, housing is in high demand, so the net impact could be an increase in newly generated trips and VKT. I also assume that these developments are not walkable, that is, there is no easy and safe way to walk to a final destination outside the development.
The key question is what is the elasticity of travel associated with newly built local roads? To determine the elasticities associated with the new roads I use both New Jersey and Maryland county-level data and calculate the “implied elasticity”, that is, what is the elasticity implied by the increase in VKT from the associated increase in lane-km (allowing for assumptions on how many trips are generated)? The formula for this is as follows:
¯ε=VKTnewVKT⋅LKmNewLKm
Where: = implied elasticity of VKT with respect to lane-km for a given road type
VKTnew = new VKT on the corresponding road type
VKT = the current vehicle-km of travel in the county on the corresponding road type
NewLKm = the newly added lane km for the county on the corresponding road type
LKm = the current lane-km in the county on the corresponding road type
Implied elasticity estimates for all local roads are shown in Table 3 by county for New Jersey and in Table 4 for Maryland assuming 2, 5 and 10 trips generated per household per day. Additional results for both rural and urban roads separately are shown in the supplemental material. Implied elasticity calculations for the combination of rural and urban roads fall in between those for each separately. All these results show substantial variation in elasticity estimates, likely due to variation in initial VKT levels and the extent of the local road network. Most studies do not consider this variance, one recent exception being Chang, Indra, and Maiti (2023) using the same data as Duranton and Turner (2011) to estimate the heterogeneity of elasticities between different MSAs (ranging from about 0.8 to 1.46).
Induced travel elasticities are generally estimated for specific road types, for example a lane-km elasticity associated with an arterial road is based on regressions linking VKT on an arterial to the lane-kms of the arterial. Similar relationships apply for other road types, e.g. Noland (2001) has estimates for multiple functional classifications, but not local roads. To my knowledge all induced travel estimates are based on VKT and lane-kms for specific road types for a designated area (e.g. a county, metro area, or state), with no cross-elasticities between different road types. My implied elasticity calculations make this same assumption, as the VKT is only that generated from the houses on the new roads, but the implied elasticity is calculated from all the local roads in each county.
Depending on the specific context, there may be additional travel on local roads to access final destinations away from home (such as work, shopping, and recreational destinations); this travel is not included in the calculations and if considered would increase the value of the elasticities.
In any case, how do these calculations of induced VKT elasticities compare to estimates done for higher functional road classifications? Induced travel estimates in the most recent research have been found to be 1.0 or slightly higher, implying that any increase in capacity is quickly filled to the average level of traffic as before the expansion (Duranton and Turner 2011; Hymel 2019). The calculations for local roads are smaller. Taking the average and standard deviation across all the counties for both states for each level of trips generated by the development shows there is substantial variation in the local road induced travel elasticities. These are shown in Table 5, by trip generation rate with the average ranging from 0.095 to 0.611 with fairly large standard deviations. When rural and urban roads are combined, the average elasticities are between the values for each separately and standard deviations are similar.
Another way to evaluate the implied elasticities is to base them off the total of all local roads within each state. Table 6 displays calculations for all local roads in both states plus combining across both states. The magnitude of the elasticity is lower, as would be expected when extending the impact over a larger area.