Such duly named “self-help” county taxes amount to a quasi-natural experiment because they have been enacted at various times by various counties primarily because public officials have successfully addressed various political issues, rather than seizing on economic factors relevant to economic growth. Keith Dunn, the executive director of California’s Self-Help Counties Coalition, notes that for the past twenty years the passage of any proposed self-help tax legislation has required the support of at least 67 percent of eligible voters. Building such support requires skillful political leaders who are willing to conduct sufficient outreach and can craft legislation that embodies a successful compromise among several competing interests, independent of economic conditions in the county.2 Accordingly, our identification strategy is to use the additional modest highway spending that is funded by self-help tax legislation as a valid instrument to determine the causal effect of highway congestion on measures of economic performance.
It could be argued that there must be some degree to which traffic conditions motivate efforts to raise a tax and motivate voters to support it. However, as discussed in detail below, self-help county taxes amount to a broad tax covering all travel modes and infrastructure, not just automobile transportation. In fact, roads get a modest share of the money—a share that must accumulate for decades before it generates reductions in congestion. Thus the causal path from road congestion to the passage of a self-help county tax that is expected to result in reasonably prompt reductions in congestion is far from clear a priori, and we find no evidence to suggest that such a path exists.
The Model
Traditional efficiency analyses of highway congestion measure the delay costs that motorists who travel during peak travel periods impose on other motorists.3 This study goes beyond the external costs to other motorists by using panel data to estimate the effect that highway congestion has on the economic performance of urban areas in California, as measured by their GDP, employment, labor-earnings, and trade-flow growth rates.4 The model begins with the demand for transportation, measured by traffic volume (V), and the supply of transportation, measured by infrastructure capacity (W). An equilibrium where transportation demand is sufficiently greater than transportation supply results in road congestion (C), so we adopt a formulation used by many authors, in which congestion rises as a power of the volume-capacity ratio:
where α is a constant that, for example, takes on a value of 2.5 for urban arterials and 4.0 for urban expressways (Small, Winston, and Evans 1989).
Given the preceding conceptual discussion and following previous work, our general model is a reduced form that relates congestion, which tends to grow over time because capacity cannot keep up with traffic volume, to economic growth and other controls. It can be described as
where Git is the growth rate of an economic performance variable in geographic unit i during year t, Cit is the level of congestion, Xit is an array of controls, and εit is a random error term.
In our empirical work, we use a log-linear specification, so our model can be summarized as
where γ is the causal effect of congestion level C on the growth rate, Xβ is an array of controls and coefficients, ϕt is the year dummy, ci is the urban-area dummy, and ε is the random error term.
Below, we summarize the available data to measure congestion and the growth-rate performance variables. We then describe and provide extensive justification for the instrument for congestion, discuss the data used to measure it, and summarize the final sample used for estimation.
Congestion
Congestion is measured using estimates of annual hours of delay per auto commuter from the Texas A&M Transportation Institute (TTI). Data are provided for the years 1982–2011 for all urban areas with more than 500,000 people. Auto commuters are defined as people who make trips by car during morning (6:00–10:00 a.m.) and evening (3:00–7:00 p.m.) peak periods. The numbers of auto commuters are estimated using data from the National Household Travel Survey, conducted by the Federal Highway Administration (FHWA). The Texas Transportation Institute adds measurements of peak-period delays to measurements of travel delay during nonpeak hours to estimate the total annual delay experienced by auto commuters.
To compute congestion-induced delays during both peak and nonpeak periods, TTI estimates two speeds for a given roadway segment: the free-flow speed, or the average speed observed during light traffic periods of the day (for example, 10:00 p.m. until 5:00 a.m. the next morning), and the actual speed observed during a given time interval of the day. By comparing actual and free-flow speeds, TTI is able to estimate congestion-induced speed reductions for different hours of the day. It then scales up those speed reductions using traffic-volume data to compute the total amount of time lost to traffic congestion.
In recent years, travel-speed data have come from INRIX, a private company that monitors travel times on most major roads in the United States. (The institute has made considerable efforts to align earlier data with INRIX data.) Traffic-volume data come from the FHWA’s Highway Performance Monitoring System. The INRIX speed data are recorded in fifteen-minute intervals for every day of the year, allowing TTI to account for both daily and hourly variations in congestion levels.5 For the sample of California urban areas from 1982 to 2011, discussed later, annual delay per auto commuter ranged from two hours to eighty-nine hours, with a mean delay of thirty-four hours per year.
Economic-Performance Measures
County-level economic performance measures are used that include real GDP, wages, employment, and originating freight traffic transported by truck. Real GDP, wages, and employment data for the period 1982–2011 were provided by the Brookings Institution’s Metropolitan Policy Program, using data from Moody’s Analytics.6 Freight flows, measured as thousands of tons of commodities transported by truck across California counties, were obtained from the California Statewide Freight Forecasting Model, which combines 2007 data from the FHWA’s Freight Analysis Framework with demographic data to forecast flows for 2010.7 The forecast flows are not adjusted to account for any unanticipated changes in congestion.
GDP, wages, and employment are expressed in terms of annual growth rates as
where
Using Self-Help County Taxes as an Instrument for Highway Congestion
There are two fundamental challenges to estimating the effect of congestion on an economic performance measure (for example, employment): omitted variables and reverse causality. Omitted variables most likely arise because some variables that affect both congestion and an economic performance measure, such as certain types of weather (Sweet 2014), are omitted from the model because they are difficult to quantify. Reverse causality most likely occurs because an economic performance measure is closely related to the volume of passenger and freight traffic on the road and thus will affect congestion. The standard approach to minimizing the bias from omitted variables and reverse causality is to use an appropriate instrumental variable that is correlated with the explanatory variable of interest (in this case, highway congestion) but is not