Transportation Economics/Revenue/The Choice Between Taxes and Tolls



This section describes some recent empirical research investigating the question of why some US states employ road tolls to a greater degree than others[1]. The use of toll roads across US states as a revenue source is rather uneven. Several states receive no revenue from tolls, while others (particularly New York and New Jersey), receive a sizable share of their highway revenues from tolls.

From a political economy perspective, tolls may seen by governments as a way to pass the costs of financing transportation links on to non-residents. For example, when road links in a jurisdiction serve a large share of "pass-through" traffic (trips with neither an origin nor destination in the jurisdiction), the jurisdiction may employ tolls in order to capture revenue from non-residents.

A classic example of this phenomenon can be seen in the case of the Delaware Turnpike. Situated along Interstate 95 between Baltimore and Philadelphia, the Turnpike is only 11.2 miles long and carries a large amount of traffic bound for destinations outside the state of Delaware. It is noted that this section of toll road has the highest per-mile tolls (35.7 cents per mile) of any toll road in the US, a sign that it is perhaps exploiting a position of market power in order to extract rents from out-of-state travelers.

A related observation is that some states with a relatively large share of non-resident workers ("labor importers") make more extensive use of tolls. For example, New York City has a large number of workers that commute from northern New Jersey daily. Those who drive or carpool are likely to encounter bridge or tunnel tolls at one of several crossing points along the Hudson River. It is worth noting in this case, however, that some of these facilities are owned and operated by the port authority that serves the two neighboring states. By analogy, these types of tolling policies may be seen as a sort of beggar-thy-neighbor policy in which the tolls function similarly to tariffs and the tolling jurisdiction seeks to improve its economic position at the expense of its neighbors.

Research Questions


States have the right to impose tolls, yet not all of them do, and among those that do there is significant variability in the degree to which tolls are employed. This section investigates three related research questions empirically using data from the 50 states.

  • What are the explanatory factors that influence levels of toll use?
  • How significant are they?
  • What would happen if transportation powers were decentralized to metro areas or counties?



Several variables are hypothesized to affect the share of state highway revenues accounted for by tolls. One variable that is thought to be central to the preference for tolls is the share of non-resident workers in a state. This variable is expected to positively influence the use of toll in a given state. In other words, a state that imports much of its labor from a neighboring state is more likely to adopt tolls in order to capture revenue from non-residents. Since non-residents cannot vote, the use tolls allows a state to, in a sense, "tax foreigners living abroad". Conversely, a state with fewer non-resident workers is more likely to get most of its revenue from taxes. Since more travel is likely to be "local" in the sense that origins and destinations lie within the state's borders, there is less opportunity for free riding by non-residents. The use of taxes rather than tolls also generally implies lower losses due to collection costs.

A state's use of tolls also is anticipated to depend on the policy decisions of neighboring states regarding the use of taxes or tolls. For example, if a state exports some of its labor to neighboring states (i.e. has residents working in neighboring states), those workers may end up paying tolls to the neighboring state(s). The home state of those workers may retaliate by adopting tolls itself in order to recapture some of the revenue. The choice of neighboring states to adopt tolls would be expected to have a positive influence on a state's use of tolls, though this effect would be smaller relative to the effect of non-local trips (non-resident workers). This is due to the fact that tolls in a labor-exporting state will disproportionately affect its own residents.

These incentives are illustrated in the figure and table below. The figure shows a stylized representation of the problem in terms of the residence and workplace locations of citizens in neighboring states. In the figure there is a single road link connecting a state to its neighbors. The table shows a matrix of possible outcomes in terms of incentives to toll or not toll the road, depending on the combination of residential and workplace locations of residents. The matrix shows possible residential locations corresponding to points A and C in the figure, and workplace locations corresponding to points B and D. If the residents live at C and work at D, a case in which residence and workplace are in the same state, then there is no incentive to toll since most trips are local. If the residents live at C, but some work at B (out of state), a situation corresponding to some exporting of labor, then there is a small incentive for the state to toll in order to recapture some of the lost revenue. Next, consider the situation where some workers live outside the state at A, but work at D (non-resident workers). In this case there is a moderate incentive to toll in order to capture the revenue from non-resident workers. Lastly, there is the case where residents live at A, but work at B. In this case, where both the residence and workplace are located outside the state, but are served by the road link running through it, there is a large incentive to toll in order to capture revenue from non-local travelers (recall the Delaware Turnpike example).


' ' Workplace '
In (D) Out (B)
Residence In (C) Local (Resident Worker) Exported Labor
No incentive to toll Small incentive to toll
Out (A) Imported Labor (Non-Resident Worker) Through
Medium incentive to toll Large incentive to toll

Population of a state is hypothesized to have a positive influence on the share of revenues from tolls as well. States that are more populous may have more congestion and higher costs for building and maintaining highways. They may thus require more revenues from tolls, especially if they are smaller in terms of land area. Also, states with large cities may be more likely to import workers than states that are more rural.

An important control variable in the empirical model is the historical use of tolls. It is observed that much of the toll road mileage in the US is concentrated in the northeastern states. These states historically have contained many of the centers of population in the US. The imperative of connecting these population centers with high-performance highways in response to the growth of auto and truck fleets led to the development of many toll-financed highways prior to the advent of the Interstate Highway System. Many of these highways were grandfathered into the Interstate System, but remained as toll facilities. Thus, the share of revenues from tolls in many states depends on the extent of toll roads during the period prior to the commencement of major work on the Interstate network, a sort of "historical inertia" effect. In order to measure the effect of historical toll miles, a variable is constructed by calculating the ratio of toll miles in 1963 (when the Interstate program took full effect) to miles of limited access highways in 1995. The variable measuring historical toll miles is expected to have a large and positive effect on the share of revenue derived from tolls.

  • Share of Toll Revenue Can be Explained by:
  • Non-Resident Workers (+)
  • Neighboring States’ Policies (+)
  • Historical Factors (Miles of Toll Road before Interstate Era) (+)
  • Perceptions of Toll Incidence
  • Toll Incentives

Data by State


The data used to fit the empirical model represent a cross-section of 49 US states for the year 1995. Alaska is excluded from the analysis, because although its data identify the Alaska Marine Highway as a highway toll, the service is actually a ferry. Thus, retaining data from Alaska would potentially introduce a source of bias into the analysis. The District of Columbia is also excluded, since it does not have independent authority over its own roads.

The major variables in the data set are reproduced in the table below, along with a correlation matrix of the variables. Note that the data include a variable measuring the percentage of land in a state that is federally owned ("Federal Land"). This variable is intended to capture the effect of states designated as "public lands" states under the Interstate Act. States with such a designation were eligible for 95 percent federal financing of its sections of interstate highway, as opposed to the 90 percent threshold applied to other states. Given that the Interstate Act required federally funded roads to be free of tolls, the "public lands" states had additional incentive to use non-toll sources of revenue to finance their highways. Since this variable was found to be statistically insignificant in the estimation of the model, it was dropped from the final model specification.

' Percentage ' ' ' Miles '
STATE Revenue from Tolls (S) Workers Who Live Out of State (O) Residents Who Work Out of State Federal Land Toll Roads in 1963 Freeways, Expwy, 1995
Alabama 0 2.4 3.6 3.3 0 925
Arizona 0 1.1 1.6 41.5 0 1250
Arkansas 0 4 3.2 8.3 0 646
California 2.1 0.5 0.4 44.6 0 3750
Colorado 0.3 0.8 1 36 17 1170
Connecticut 0 4.6 4.7 0.2 194 542
Delaware 25.3 13.8 9.5 2.2 11 51
Florida 7.8 0.8 1 7.6 207 1861
Georgia 0.4 2.8 2.4 3.9 11 1413
Hawaii 0 1 0.5 8.5 0 77
Idaho 0 2.6 4 60.6 0 613
Illinois 9.3 2.8 2.9 1.3 185 2245
Indiana 4.3 3.3 4.8 1.7 157 1303
Iowa 0.1 3.7 4.3 0.2 0 781
Kansas 6.5 7.1 7.6 0.5 241 1008
Kentucky 0.8 6.3 6.7 4.2 205 855
Louisiana 2.9 2.1 1.9 2.8 0 929
Maine 10.5 2.1 3.1 0.9 112 383
Maryland 7 7 17.3 3.1 42 711
Massachusetts 10.4 5 3.1 1.2 124 762
Michigan 0.7 0.8 1.5 10.1 0 1458
Minnesota 0 2.3 1.8 3.1 0 1042
Mississippi 0 3.1 5.9 4.3 0 726
Missouri 0.1 7.2 4.8 3.8 0 1460
Montana 0 0.8 1.2 27.5 0 1190
Nebraska 0.2 4.3 2.3 1.2 0 497
Nevada 0 4.3 1.2 77.1 0 586
New Hampshire 11.8 8.5 16.8 12.8 77 266
New Jersey 27.3 7 11.7 3.3 309 728
New Mexico 0 1.9 2.5 33.9 0 1003
New York 33.2 5.1 2.4 0.7 629 2328
North Carolina 0.1 2.2 1.8 6.9 0 1237
North Dakota 0 5.9 3.7 4 0 570
Ohio 3.3 2.8 2.2 1.1 241 1937
Oklahoma 7.6 1.1 2.9 1.5 174 1064
Oregon 0.5 3.7 2.1 51.8 0 780
Pennsylvania 11.7 3.4 4.3 2.2 469 2087
Rhode Island 3.7 7.6 11.9 0.7 0 137
South Carolina 0 2.1 1.8 3.8 0 894
South Dakota 0 3 4 5.5 0 681
Tennessee 0 4.6 3.3 5.7 0 1176
Texas 2.5 0.9 0.8 1.4 30 4474
Utah 0.1 1 1.3 63.1 0 948
Vermont 0 4.9 5.8 6.4 0 1329
Virginia 4.7 6.5 9.3 9.4 35 339
Washington 4.1 1.5 2.7 24.1 0 1079
West Virginia 6.6 8.3 9.7 7 86 560
Wisconsin 0 1.4 3.2 5.3 0 830
Wyoming 0 2.6 2 48.5 0 916
Correlations Matrix
' ' ' ' ' ' ' ' '
Toll Share (S) Population (P) Toll Mile 1963 Imported Workers Neighbor Effect (N) Land Density Federal Land (%)
Toll Share (S) 1
Population (P) 0.27 1
Toll Mile 1963 0.71 0.39 1
Imported Workers (O) 0.49 -0.27 0.16 1
Neighbor Effect (N) 0.61 -0.03 0.36 0.54 1
Land -0.32 0.36 -0.22 -0.5 -0.45 1
Density 0.53 0.18 0.39 0.39 0.7 -0.48 1
Federal Land (%) -0.28 -0.07 -0.32 -0.29 -0.22 0.45 -0.32 1



The results of the estimation of two different models are reproduced below. One model contains the variable N, the "neighbor state" or "neighbor toll" effect, the other does not. The neighbor state variable measures the share of revenue from tolls in each of a state's neighbors weighted by the share of that state's residents commuting to those neighbors. As the fit of the two models suggests, this variable is a statistically significant predictor of the share of a state's revenue derived from tolls, though its explanatory power is not very large. Both models have adjusted R-squared values of around 0.60, indicating that the models explain about 60 percent of the variance in the dependent variable (share of revenue from tolls).

All of the variables in the models are statistically significant at the p < 0.10 level or less. Since the independent and dependent variables are structured as percentages, their coefficients can be interpreted as elasticities. For example, the coefficient on the variable for imported workers (non-resident workers) in Model 1 is about 0.85, indicating that each one percent increase in a state's share of non-resident workers is associated with a 0.85 percent increase in the state's share of revenues derived from tolls. An exception to this interpretation is seen with the neighbor state variable (N), which has an intrinsically non-linear structure.

In terms of practical significance, the historical toll miles variable (ratio of 1963 toll miles to 1995 limited access highway miles) is the most important predictor, followed by state population size, non-resident workers, and the variable capturing the neighbor state effect.

' Model 1 ' Model 2 '
Coefficients t Stat Coefficients t Stat
Intercept -0.03 -2.42 ** -0.036 -2.51 **
Population (P) (millions) 0.00383 3.20*** 0.00386 3.15***
Mile Ratio (M) 0.3 4.13*** 0.35 5.25***
Imported Workers (O) 0.85 2.92 ** 0.84 2.85 **
Neighbor Effect (N) 89877 1.71 *
Regression Statistics
Multiple R 0.81 0.79
R Square 0.65 0.63
Adjusted R Square 0.62 0.6
Standard Error 0.04 0.05
Observations 49 49
F 20.87 25.76
Significance F 0 0

The model does a fairly good job of predicting the shares of revenue derived from tolls in each of the states. As a test of the policy implications of the model, it is also applied to counties and metropolitan areas in California to see what might happen under a regime of more decentralized highway finance and governance. Model 2, the model excluding the neighbor effect variable, is applied to data on counties and metropolitan areas. The effect of toll bridges in the San Francisco Bay Area are not counted as historical toll road mileage, and so this variable essentially takes on a value of zero. The counties and metro areas therefore vary primarily in terms of population size and cross-border flows.

The results show that decentralizing authority over highways to metropolitan areas has little effect on the adoption of tolling. Only one metropolitan area, Stockton, has a predicted share of revenues from tolls of over 10 percent. However, once the unit of analysis is reduced to the county level, the results are more dramatic. Several counties have predicted shares of toll revenue exceeding 20 percent, with a maximum value of 34.5 percent observed in San Francisco. This prediction for San Francisco matches reality fairly well, as there are toll facilities at two of the entrance points to the city (the Golden Gate and Bay Bridges), as well as fairly expensive, city-owned parking in the central business district -- another possible way to collect revenue from non-residents. It should be noted though, that the Golden Gate bridge is owned by an independent authority and the Bay Bridge is managed by the Bay Area Toll Authority and maintained by Caltrans. Another observation is that Orange County, the county with the highest predicted share of revenues from tolls in the Los Angeles region under the county-level simulations, is a county that makes extensive use of tolls as a revenue source. The county owns the SR-91 Express Lanes on the Riverside Freeway and also participates in the Transportation Corridor Agencies, a joint-powers authority that oversees the planning and financing of the county's three other toll roads.



The empirical model describing states' use of tolls as a source of highway revenue shows that toll use depends positively on cross-border flows, neighboring state tolls, the historical use of tolls to finance limited access highways, and population size. Applying this model to smaller jurisdictions, as was done with the simulations for California counties and metropolitan areas, reveals that devolving authority over highways to metropolitan areas is insufficient to induce significant adoption of direct road pricing in the form of tolling. However, under more radical forms of decentralization, such as devolving authority to counties, the use of tolls may become more widespread. This is due primarily to the effect of creating a greater degree of cross-border flows by bringing jurisdictional control down to a lower level of government.


  1. Levinson, D.M. (2005). "Why States Toll: An Empirical Model of Finance Choice". Journal of Transport Economics and Policy. 35 (2): 223–237.