Induced demand: elasticity estimates and how to account for them in a traffic impact study

1. The problem in one sentence

A conventional traffic impact study (TIS) usually treats demand as fixed: it forecasts design-year volumes, then sizes the improvement to carry them. When the improvement is added capacity on a congested network, demand is not fixed. Lower travel times generate new and longer trips, and the forecast the design was based on no longer holds.

This article covers four things:

• what the empirical literature says about the size of that effect;
• how the effect is expressed as an elasticity;
• a transparent method for carrying it into a TIS or corridor study;
• where the method breaks down.

2. Definitions

Induced travel is the increase in vehicle miles travelled (VMT) caused by a capacity increase, net of traffic simply diverted from other routes. Its sources are:

• new trips;
• longer trips (more distant destinations);
• mode shift to driving;
• shifts in time of day;
• over the longer run, changes in land use and location.

Elasticity of VMT with respect to lane-miles (ε) is defined as:

ε = (ΔVMT / VMT) ÷ (ΔLM / LM)

Most studies estimate it in log–log form, so the lane-mile coefficient is the elasticity. Read it like this:

• ε = 1.0: a 10 % increase in lane-miles produces a 10 % increase in VMT on that facility class. The added capacity fills up.
• ε = 0.4: 40 % of the proportional capacity gain is consumed by new travel.

Studies distinguish two horizons:

• Short-run elasticities capture behavioural response within roughly 1–2 years.
• Long-run elasticities (lags of about 4 years or more) also capture land-use and location effects.

This convention is described in CARB’s technical background document [3].

Elasticities with respect to lane-miles are not the same as elasticities with respect to travel time, which some model post-processors use. Don’t mix the two.

3. What the evidence says

Source	Scope	Estimate	Note
Duranton & Turner (2011), American Economic Review [1]	US metro areas, 1983–2003, instrumental-variables design	ε ≈ 1.0 for interstate VKT vs interstate lane-km: OLS 0.86, first-difference 1.00, IV 1.03 (the authors’ preferred estimate); 228 MSAs, HPMS 1983/1993/2003	“VKT increases proportionately to roadway lane kilometers for interstate highways.” Major urban roads 0.67–0.89 (OLS only, suggestive).
Cervero (2003), California freeway projects, summarised by CARB [3]	Project-level	Short-run 0.10, long-run 0.39	0.10 = 0.42 × 0.24 (speed-vs-capacity × VMT-vs-speed); the long-run value adds development effects; 7–8-year lag. Cervero attributes the low values to not all speed gains being due to capacity.
Cervero & Hansen (2002), J. Transport Economics and Policy, cited by NCST [2]	California urbanized counties, FHWA classes 1–3	5-year elasticity 0.79	County-level, simultaneous-equation model
NCST Induced Travel Calculator (UC Davis) [2]	California, MSA and county level	1.0 for interstate expansions; 0.75 for other freeways, expressways and major arterials (FHWA functional classes 2–3)	Long-run (≈ 3–10 years after opening); rough 95 % interval ±20 %. Draws on Duranton & Turner, Cervero & Hansen (2002), Handy & Boarnet (2014) and later work.

Three conclusions follow:

1. Regional-scale long-run elasticities for urban limited-access highways cluster near 1.0; project-level estimates are lower. On congested metro networks, a widening is largely absorbed by new travel within a decade.
2. Short-run effects are smaller, so a year-of-opening “before/after” count understates the eventual impact.
3. Project-level and region-level estimates differ. Raw corridor counts that include traffic diverted from parallel routes overstate induction; project-level studies that net diversion out (e.g. Cervero) give lower values than regional estimates. Choose the elasticity that matches the geography you’re forecasting.

4. The method (elasticity approach)

The NCST calculator formalises the simplest defensible approach [2]:

Induced VMT = (ΔLM / LM) × VMT₀ × ε

where:

• ΔLM is the lane-miles the project adds;
• LM is the existing lane-miles of the same facility class in the same geography (MSA for interstates, county for classes 2–3 in the calculator);
• VMT₀ is existing VMT on that class and geography;
• ε is the elasticity for that class.

The denominator matters. Dividing by corridor lane-miles instead of regional lane-miles would massively overstate the effect. The elasticities were estimated at the regional scale, so they must be applied at the regional scale.

How this enters a traffic impact study

A TIS works with link and intersection volumes, not regional VMT, so the regional result has to be carried down to those volumes:

1. Compute regional induced VMT with the formula above, for the design horizon.
2. Allocate it to the study corridor. Use a calibrated travel-demand model where one exists, by adding the induced VMT as extra trips or by running the model with a feedback loop (assignment ↔ distribution/mode choice). Without a model, a documented proportional share is the fallback: for example, the corridor’s share of class VMT after the widening, which is an assumption to state explicitly.
3. Convert to link volumes. Divide the allocated VMT by the corridor length to get added vehicles per day, then apply the study’s K and D factors for peak-hour directional volumes. Dividing by length gives an average section volume. Applying the base K factor is conservative, because part of the induced travel occurs off-peak (peak spreading).
4. Re-run the operational analysis (HCM methods, or micro-simulation for oversaturated conditions) with the adjusted volumes. Report LOS, delay and queues for the “with induced demand” case alongside the conventional case.
5. Report both, plus a sensitivity band. Present ε at the low (short-run), central and high (long-run) values, and the ±20 % interval.

5. Worked example (illustrative numbers)

Inputs (hypothetical metro area):

• Interstate lane-miles in the MSA: LM = 1,600
• Proposed widening: one lane each direction over 10 miles, so ΔLM = 20 lane-miles
• Interstate VMT₀ in the MSA: 30,000,000 vehicle-miles per day
• Study corridor length: 10 miles

Step 1: proportional capacity change. ΔLM / LM = 20 / 1,600 = 1.25 %

Step 2: regional induced VMT (long-run, ε = 1.0). 0.0125 × 30,000,000 × 1.0 = 375,000 VMT/day (±20 % → 300,000 to 450,000)

Step 3: sensitivity on ε.

ε	Basis	Induced VMT/day	If 40 % falls on the corridor: added vehicles/day on the 10-mile section
0.10	Short-run, project level [3]	37,500	≈ 1,500
0.39	Long-run, project level [3]	146,250	≈ 5,850
0.75	Class 2–3 value [2]	281,250	≈ 11,250
1.00	Interstate, long-run [1][2]	375,000	≈ 15,000

The 40 % corridor share is an assumption chosen to make the conversion explicit. It is far above the corridor’s share of class VMT (10 mi × 110,000 / 30,000,000 ≈ 4 % here), and it implicitly includes traffic diverted from parallel routes, which the regional elasticity nets out. A real study derives the share from a model with distribution/mode-choice feedback, or documents a corridor-level basis. On its own, the regional formula says little about corridor volumes.

Step 4: interpretation.

• If the base design-year forecast for the widened section is, say, 110,000 vehicles/day, the long-run case adds about 15,000/day (≈ +14 %).
• That can be enough to return the peak hours to the congestion level the project was meant to relieve.
• The benefit of the widening is real but temporary, unless demand is managed (pricing, managed lanes, transit, land-use policy).

6. Regulatory context

• California has moved project screening from LOS to VMT under SB 743 and CEQA Guidelines §15064.3 (adopted Dec 2018, mandatory statewide from 1 July 2020; see also OPR’s Technical Advisory, Dec 2018). Caltrans’ Transportation Analysis Framework (2nd ed., Sept 2025) and its Transportation Analysis under CEQA identify the NCST calculator as a standard method for capacity-increasing projects. Travel-demand models are also allowed under TAF §4.4 conditions [4].
• Most other US states and agencies don’t require induced-demand analysis in a TIS. Including it is good practice, and the topic is routinely raised in NEPA comments.

7. Limitations

• Transferability. The elasticities are averages from US metro data. Use local evidence where it exists, and widen the sensitivity band for small or uncongested regions, where ε is likely lower.
• Capacity reductions and lane-type conversions (road diets, general-purpose→HOV/HOT) cannot be estimated with the calculator. HOV/HOT lane additions are covered, and pure toll lanes need supplemental analysis [2].
• Time profile. A long-run ε describes year 3–10 outcomes. Opening-year counts will understate it.
• Double counting. If the travel-demand model already has full feedback between assignment and the earlier steps, don’t add elasticity-based VMT on top. Use the elasticity as a check on the model instead.
• Emissions and safety. Induced VMT changes greenhouse-gas and crash-exposure estimates. Carry it through to those analyses consistently.

8. Checklist for the study report

• ☐ Facility class and geography stated; ε chosen to match the class and the geography
• ☐ ΔLM and LM from the same data source and year (e.g. HPMS)
• ☐ Short-run and long-run cases, with sensitivity band
• ☐ Corridor allocation method documented
• ☐ Operational results reported with and without induced demand
• ☐ Model feedback checked to avoid double counting
• ☐ Limitations and data vintage disclosed

Sources

1. Duranton, G. & Turner, M. A. (2011). “The Fundamental Law of Road Congestion: Evidence from US Cities.” American Economic Review 101(6): 2616–2652. PDF (Brown University) · TRID record
2. National Center for Sustainable Transportation, UC Davis. Induced Travel Calculator: Background on Induced Travel. travelcalculator.ncst.ucdavis.edu · Webinar
3. Handy, S. & Boarnet, M. G. (30 Sept 2014), prepared for the California Air Resources Board. Impact of Highway Capacity and Induced Travel on Passenger Vehicle Use and Greenhouse Gas Emissions: Technical Background Document. PDF
4. Caltrans. Transportation Analysis Framework & Transportation Analysis under CEQA, 2nd ed. (Sept 2025). PDF