rural highway expansion and economic development: impacts
TRANSCRIPT
Rural Highway Expansion and Economic Development: Impacts on Private1
Earnings and Employment2
Michael Iacono (*corresponding author)3
Research Fellow4
University of Minnesota5
Department of Civil Engineering6
500 Pillsbury Drive SE7
Minneapolis, MN, 55455 USA8
E-mail: [email protected]
Phone: +01-(612) 626-002410
David Levinson11
RP Braun-CTS Chair of Transportation Engineering12
University of Minnesota13
Department of Civil Engineering14
500 Pillsbury Drive SE15
Minneapolis, MN, 55455 USA16
E-mail: [email protected]
Phone: +01-(612) 625-635418
http://nexus.umn.edu19
July 30, 201220
Word Count: 6,814 words + 5 tables + 1 figure21
1
Abstract1
With the interstate system substantially complete, the majority of new2
investment in highways is likely to take the form of selective capacity ex-3
pansion projects in urban areas, along with incremental expansions and up-4
grades to expressway or freeway standards of existing intercity highway cor-5
ridors. This paper focuses specifically on the latter type of project, rural6
highway expansions designed to connect smaller outstate cities and towns,7
and examines their effects on industry-level private earnings and local em-8
ployment. We examine three case study projects in rural Minnesota and use9
panel data on local earnings and employment to estimate the impacts of the10
improvements. Our results indicate that none of the projects studied gener-11
ated statistically significant increases in earnings or employment, a finding12
we attribute to the relatively small time savings associated with the projects13
and the maturity of the highway network. We suggest that for rural high-14
way expansion projects, as with other types of transportation projects, user15
benefits should be a primary evaluation criterion rather than employment16
impacts.17
Keywords: Highways; Economic Development; Employment; Panel data18
2
Introduction1
There remains a large amount of interest at state and local levels in using trans-2
portation investment as a means to promote economic development. Cities and3
regions that are growing slowly or not at all view improvements to infrastruc-4
ture networks, especially transportation networks, as a potential way to stimulate5
growth by lowering the costs of local firms and making their location a more at-6
tractive place for private investment and expansion. Transportation investment7
programs often become more attractive when coupled with the offer of grants8
from higher levels of government. They also benefit from the reputation of infras-9
tructure projects as a “safe” type of investment during periods of lower growth.10
This has been seen most recently with the United States government’s promotion11
of the American Recovery and Reinvestment Act, where infrastructure spending12
became emblematic of the bill’s efforts to promote employment, despite being a13
relatively small portion of the overall spending. Yet, as fewer resources have be-14
come available for such projects at the state and local levels in recent years, state15
departments of transportation and other public works organizations have begun16
to sharpen their focus to determine where and how such resources should be de-17
ployed in order to yield the greatest returns. This study evaluates the potential18
of transportation investment to generate increases in private economic activity by19
empirically examining a recent set of case studies of highway expansion projects20
in rural Minnesota.21
With the interstate system substantially complete, the majority of new invest-22
ment in highways is likely to take the form of selective capacity expansion projects23
in urban areas, along with incremental expansions and upgrades to expressway or24
freeway standards of existing intercity highway corridors. These types of projects25
are less likely to generate the kinds of larger-scale growth and relocation impacts26
associated with the interstates [16, 4], yet their effects may still be measurable.27
This paper focuses specifically on the latter type of project, rural highway expan-28
sions designed to connect smaller outstate cities and towns, and examines their29
effects on industry-level private earnings and local employment. For both types30
of measures of economic activity, we construct panel data sets and attempt to es-31
timate the impact of the improved highway as a fixed, unobserved increment to32
growth that is phased in after the completion of construction. Our analysis of33
employment, which uses city-level data, also attempts to account for spatial dif-34
ferences in growth effects at a sub-county level which might be brought about by35
the highway improvement.36
The paper is organized as follows: the next section provides a brief overview37
3
of the types of methodological approaches to studying the relationship between1
highways and economic development and their results. The third section provides2
more detailed descriptions of the three highway expansion projects that are the3
focus of this study. The fourth and fifth sections document the design and results4
of analyses of private earnings and employment for the case study projects. The5
final section summarizes the results of the findings and suggests what they might6
imply for the evaluation of rural highway projects.7
Measurement of the Economic Impacts of Highways8
The measurement of the relationship between highway networks and economic9
development is marked by a heterogeneous set of methodological approaches,10
often with different aims and conducted at different scales, which have led to11
generally mixed results on the question of magnitude and direction. Perhaps the12
most well-known among these studies is the literature on public capital stocks and13
economic growth, motivated largely by attempts to explain the US productivity14
slowdown of the 1970s and 1980s. Despite early published results which claimed15
exceptionally large returns to infrastructure investment, including highway capital16
stocks [1, 17], subsequent studies which relied on more disaggregate data and17
corrected for certain econometric issues tended to show more modest results [12,18
10, 18]. These types of studies were generally carried out at rather aggregate levels19
and considered only the size of the capital stock (as opposed to its composition),20
and thus were not able to offer much in the way of policy-relevant direction on21
investment at more local levels.22
A second line of empirical studies into the relationship between highways and23
economic development can be grouped into what we might call growth regres-24
sions. This term is used to capture a range of empirical approaches that are less25
explicitly tied to economic theory, but which systematically study the relation-26
ship between transportation and economic development. These include structural27
models which consider highway networks as an exogenous input to equilibrium28
models of population and employment [3, 5] or population, income and employ-29
ment [6]. Others model highway networks as endogenously determined within the30
scope of labor and housing markets [22]. Most of these studies show highway net-31
works as having small, but positive impacts on local growth, typically measured32
at the county level.33
The class of growth regression studies can also include studies which have34
investigated the issue of causality between highways and economic development35
4
by means of temporal precedence, such as vector autoregression analyses [15, 13].1
Most such studies employ highway data aggregated to the state level, as these tend2
to be more readily available, though there are examples of county-level studies as3
well [21]. These studies tend to find evidence of mutual causation between eco-4
nomic development and the growth of highway networks, but the largest effects5
of highways on development seem to be concentrated in urban regions.6
A third class of studies which merit attention are those which use quasi-7
experimental techniques to more directly isolate and estimate the causal effect8
of highway improvements on economic development. One study by Rephann9
and Isserman [19] used a quasi-experimental matching framework to try to iden-10
tify “treatment” and “control” counties that could be used in an evaluation of the11
development effects the Appalachian Development Highway System. Another12
study by Chandra and Thompson [4] used the development of interstate highways13
through rural areas as a natural experiment (interstates were originally planned14
to primarily connect urban areas) to examine their effects on the growth of rural15
areas. Both of these studies found that the primary effects of the highways under16
study were to redistribute growth, either from neighboring counties to counties17
adjacent to the highways, or from rural counties to those closer to existing urban18
areas.19
Due to the difficulty of finding such experimental cases, researchers must often20
settle for other, less direct methods of evaluating the effects of highway improve-21
ments. Longitudinal methods in general, and especially those which can account22
for unobserved heterogeneity such as various panel data techniques, offer an ac-23
ceptable substitute. These methods can be more useful when they are coupled24
with spatially disaggregate data. These are the methods employed in the present25
study, which is targeted specifically toward rural highway expansion projects.26
Case Study Projects27
Our empirical analysis proceeds in two phases, focusing alternately on private28
earnings and employment, and takes as case studies three rural highway expansion29
projects:30
• The incremental expansion of Minnesota Trunk Highway (TH) 371 along31
a roughly 30-mile segment between Little Falls and Brainerd, Minnesota.32
The project involved expansion of a two-lane highway to a four-lane divided33
highway and included the construction of a new bypass around the city of34
Brainerd.35
5
• The completion of a four-lane bypass along US Highway 71 around the city1
of Willmar, Minnesota. This project was initiated in the mid-1980s, with2
much of the grading work being done then, but the construction of the full,3
freeway-grade four-lane bypass with interchanges was not completed until4
the early 2000s.5
• The expansion of US Highway 53 north of Virginia, Minnesota. This project6
involved expansion of an 11-mile segment of rural highway from a two-lane7
to a four-lane divided highway on a new alignment.8
Each is project is described in more detail below.9
TH 371 Expansion10
The first case study, described briefly above, is the expansion of Minnesota TH11
371. This project was completed in various phases between August 1998 and12
October 2005. The first phase involved completion of a bypass around the city13
of Brainerd, Minnesota on TH 371 and a new interchange at the junction of the14
new TH 371 alignment and the old alignment, now a business route for 371 which15
serves the city directly. The remainder of the project expanded the roughly 30-16
mile segment of 371 between Little Falls (in Morrison County) and Brainerd (in17
Crow Wing County) in two phases, the first upgrading the highway to four lanes18
between the junction with US Highway 10 near Little Falls and CSAH 48 near19
Camp Ripley, and the second phase completing the new four-lane segment be-20
tween Camp Ripley and the new Brainerd bypass. These latter two phases in-21
cluded the construction of two new interchanges and were completed between22
May 2003 and October 2005. Taken together, the improvements to TH 371, in-23
cluding the right-of-way and construction costs, had a combined cost of around24
$60 million.25
US 71/TH 23 Expansion26
The primary objective of this project was to construct a four-lane, freeway-grade27
bypass of the city of Willmar, Minnesota which had a reported population of just28
under 20,000 as of the 2010 census. Grading work for the new highway was begun29
during the 1980s, along with with expansion of a 5-mile segment of US 71 and30
Minnesota TH 23 just north of Willmar from two to four lanes. Due to funding31
constraints, the bypass of Willmar was not completed as a four-lane highway until32
6
the fall of 2001. Shortly after the bypass was completed, an additional segment of1
TH 23 northeast of Willmar (from the junction of US 71 and TH 23 north of Will-2
mar to New London, MN) was expanded to four lanes, completing a continuous3
four-lane section of TH 23 from New London, MN to the south end of Willmar.4
This latter project was completed in the spring of 2003. The two projects together5
were largely completed over the period from 1999 to 2003 at a combined cost of6
around $60 million.7
US 53 Expansion8
This project was undertaken to expand a segment of US Highway 53 from two9
to four lanes on a new alignment between the towns of Virginia and Cook in St.10
Louis County, Minnesota. The expanded segment of highway was 11 miles long,11
with construction being completed in phases between February 2007 and October12
2009. Total project cost was approximately $30 million.13
County and Industry-Level Earnings14
Data15
Our analysis of aggregate, county-level impacts of specific projects uses industry16
earnings as a key measure of economic impact. The data on earnings are made17
available by the Bureau of Economic Analysis (BEA) as part of its Regional Eco-18
nomic Accounts. The term “earnings” specifically refers to three components19
of personal income: wage and salary disbursements, supplements to wages and20
salaries, and proprietors’ income.21
An important consideration in using the BEA data on earnings is that, over22
the time period that will be evaluated in the county-level analysis (1991 to 2009),23
the industrial classification system that is used to aggregate data to various in-24
dustry levels was converted from the former Standard Industrial Classification25
(SIC) system to the currently-used North American Industrial Classification Sys-26
tem (NAICS). One impact of this reclassification was that the number of “one-27
digit” industries (the highest level of industry aggregation in the data) included28
in the data set increased from nine to 20. Another related impact was that many29
of the former SIC one-digit industries were split up, or in some cases recom-30
bined, to form new industries. For example, the one-digit industry classified as31
“Transportation and Public Utilities” (SIC code 500) under the SIC system was32
7
reclassified into two new one-digit industries under the NAICS system, “Utilities”1
(NAICS code 300) and “Transportation and Warehousing” (NAICS code 800).2
While there is some overlap between the two classification systems, including3
a few years for which data were reported in both systems, data are generally avail-4
able for download from BEA using the SIC system for the years 1969 through5
2000, while the NAICS system is used for more recent years leading up to 2009.6
Another issue that arises with the use of the industry-level data is data suppres-7
sion. BEA may suppress data at any level of industry aggregation if 1) the total8
annual earnings for a given year are less than $50,000 or 2) publication of the9
figure would disclose confidential information (for example, an industry in which10
a single firm greatly influences aggregate trends at a county level). In either case,11
the figures are included in any higher-level totals aggregated by industry or geog-12
raphy. Fortunately, there are few instances of missing or suppressed data among13
the counties at the one-digit industry level. The few observations that contained14
missing or suppressed data were removed from consideration.15
Specification16
We now turn to the empirical analysis of a pair of case studies of road network17
improvements in Minnesota using aggregate (county-level) data as units of obser-18
vation. Our analysis focuses on earnings in specific industries which have been19
identified by previous research [8] as using transportation as an input more in-20
tensively. These industries include construction, manufacturing, wholesale, retail,21
and trucking.22
Our analysis of these projects takes as the period of study the 19-year period23
between 1991 and 2009. This represents a long enough time period to examine the24
changes in economic activity which occurred both before and after the completion25
of the case study projects. Our focus is on the changes in private earnings in the26
four industries described above. Data sets for each of the study locations are con-27
structed from the county (or counties) in which the project is located, along with28
all neighboring counties. We model the earnings in each industry at the county29
level as a function of national and state-level economic trends, along with changes30
in population. The effect of a highway project is represented as a series of indi-31
cator variables identifying the county in which the project is located, along with32
a specific time period, either before, during, or after completion of construction,33
when the improved highway is opened to traffic. In each case, this series of indi-34
cator variables breaks the study period into three distinct phases and attempts to35
measure changes in industry-level earnings in the county receiving the highway36
8
improvement over time relative to neighboring counties. Formally, we can write1
the model predicting county-level earnings in a given industry as:2
lnyit = α+β1lnGDPt+β2lnStateEarnt+β3lnPopit+3∑
j=1
γjCountyj+εit (1)
where:3
lnyit = natural log of earnings in a given industry in county i at time t4
lnGDPt = natural log of real GDP (in 2009 dollars) at time t5
lnStateEarnt = natural log of state-level earnings in a given industry at time t6
lnPopit = natural log of population in county i at time t7
Countyj = indicator variable identifying the county (or counties) in which the8
highway improvement was located during a specific period, j9
εit is an error term, and10
α, β1, β2, β3, andγj are parameters to be estimated11
12
The model represented by Equation 1 includes controls for three exogenous13
factors. The first is national output, which may be seen as influencing the demand14
across all sectors. The second is state-level earnings in a given industry. This15
variable is taken to represent industry-level demand shifts which are unrelated to16
broader economic growth [11, 4]. The third factor is population, which we assume17
to be exogenous for the purpose of this analysis, and which is assumed to influence18
the local demand for goods in each industry.19
The series of county-time period indicator variables, Countyj , provide the20
interpretation of the economic impact of the project on earnings in a given in-21
dustry. If the difference in the estimated coefficients of these variables in the22
pre-construction period (period 1) and the post-construction period (period 2) is23
statistically significant, this would provide possible evidence of an increase in24
earnings attributable to the project under study.25
Estimation and Results26
TH 371 Expansion27
It will be useful for the purposes of this analysis to consider the projects related to28
the expansion of TH 371 together as a complete set of improvements. There are29
9
practical reasons for doing so, since the projects were considered part of an inte-1
grated strategy for improving TH 371 as an inter-regional highway corridor. More2
importantly, considering them together allows us to define a single construction3
period for the improvements which fits the modeling framework described above.4
We define the construction period as covering the years from 1998 through 2005.5
The years prior to 1998 are considered part of the “pre-construction” period, while6
the years after 2005 are considered “post-construction”.7
We fit the model described in Equation 1 to four data sets representing earnings8
data for the construction, manufacturing, retail and wholesale industries. Descrip-9
tive statistics for these data are reproduced in Table 1.10
(place Table 1 about here)11
12
The county/time indicator variables were defined for both Crow Wing and13
Morrison counties, since each of these counties contained a substantial share of14
the improved section of TH 371. The other neighboring counties included in15
the data were Aitkin, Cass, and Mille Lacs counties. Each of the industry-level16
regressions had 95 observations, with the exception of the “wholesale” data set,17
which was missing one observation due to data suppression.18
We estimate the models using ordinary least squares with panel-corrected stan-19
dard errors. This technique allows us to fully exploit the panel structure of the20
data, accounting for correlation across panels within the data set, while also cor-21
recting for serial correlation among the residuals in the model, and has been shown22
via simulation to generate efficient parameter estimates [2]. Our estimates are gen-23
erated using the “xtpcse” procedure in Stata (version 10), with the Prais-Winsten24
method for correcting for serial correlation and an AR(1) structure assumed for25
correlation among the residuals. The estimation results for the four industry-level26
earnings regressions are provided below in Table 2.27
(place Table 2 about here)28
29
The use of the panel correction procedure results in very high R2 values for30
each of the models due to the correction for autocorrelation among the residuals31
and cross-panel correlation among observations. The Wald χ2 statistic provides32
a test of the null hypothesis that the parameters of all nine independent variables33
are jointly equal to zero. The large test statistic for each of the models allows this34
hypothesis to be rejected at any reasonable level of significance. Table 4.1 also35
provides the estimated ρ autocorrelation parameter for each of the models.36
The models appear to provide a good fit to the data. The population and state-37
level industry earnings variables are positive and highly statistically significant38
10
for nearly every industry, with one exception being the population variable in the1
manufacturing regression. Of interest, a negative coefficient is observed for the2
GDP variable in three of the four regressions (the exception being manufacturing),3
indicating that growth in national output adds little to earnings growth in these4
industries once state-level industry output is controlled for.5
Examining the county-time indicator variables for the two counties where the6
improvement occurred, we see little evidence of a statistically significant effect of7
the completion of the highway on earnings in each of the four industries. Again,8
these variables capture any unobserved differences in earnings between the coun-9
ties where the highway improvements occurred and neighboring counties, ob-10
served at three points in time (before, during, and after completion of construc-11
tion). For example, looking at the manufacturing regression we see a small in-12
crease in earnings in Crow Wing County between time periods 1 and 3 (before13
and after construction) relative to neighboring counties. However, the difference14
in parameter values between these two periods is less than one standard error for15
either of the parameter estimates, indicating that this differences is not statisti-16
cally significant, or rather that the change in parameter values from the pre- to17
post-construction period is simply the result of chance variation. Comparing the18
estimates for the county-time indicators in the other models reveals largely the19
same result. Where there is any evidence of an increase in earnings, it fails to be20
large enough to rise to the level of statistical significance.21
US 71/TH 23 Expansion22
Similar to the case study of the TH 371 expansion, we identify periods before,23
during, and after completion of construction to examine changes in earnings. The24
analysis is simplified slightly in this case, since all of the highway improvements25
are contained within Kandiyohi County. Accordingly, the number of explanatory26
variables in our model is reduced from nine to six. The data set is comprised of27
six counties, Kandiyohi plus five neighboring counties (Chippewa, Meeker, Pope,28
Renville and Swift). The data sets for the manufacturing and retail industries con-29
tain a full set of 114 observations, while the construction and wholesale industries30
are missing four and three observations, respectively, due to data suppression. The31
models are fitted using the same techniques as were applied to the TH 371 case32
study. Descriptive statistics for this case study are included in Table 2, while a33
summary of the model results is presented below in Table 3.34
(place Table 3 about here)35
The results of the earnings regressions for the US 71/TH 23 study area are36
11
similar to those observed in the previous case study. The population and statewide1
industry earnings variables are uniformly positive and statistically significant at2
the p < 0.05 level. The national GDP variable still has a negative coefficient3
in two of the four industry regressions, though one of the two is not statistically4
significant. National output is positive and significant for the manufacturing and5
wholesale industry regressions.6
Looking at the county-time period indicators, there is again little evidence of7
statistically significant changes in earnings for each of the four industries exam-8
ined. Comparing the coefficients for the pre- and post-construction periods, there9
is only a small change for the construction and manufacturing industries, nei-10
ther rising to any level of significance. There is a slightly larger change for the11
county-time indicators in the retail and wholesale industry regressions. However,12
the magnitude of the difference in each case is only equal to roughly one standard13
error, indicating that the difference is not statistically significant. Moreover, in the14
case of the wholesale industry the coefficients decline in value over time indicat-15
ing that, all else equal, earnings in Kandiyohi County were declining relative to16
those in neighboring counties.17
Analysis of Local Employment18
In the previous section, we examined the impacts of highway improvements in19
two of the case study locations by estimating their effects on private earnings20
in several sectors of the economy. While no evidence of statistically significant21
were found, we noted the possibility that some impacts might exist at a smaller22
geographic scale due to relocation or changes in the location of new economic23
activity induced by the highway improvements. In this section, we examine this24
possibility by using more disaggregate data on employment levels from the Quar-25
terly Census of Employment and Wages (QCEW), aggregated to the minor civil26
division for analysis by staff at the Minnesota Department of Employment and27
Economic Development.28
Data29
As was done with the county-level earnings analysis, we assemble a panel data30
set to analyze the changes in a cross-section of locations through time in response31
to the case study projects. The use of disaggregate employment data from the32
QCEW allows us to construct models that are more spatially explicit, accounting33
12
for the likelihood that locations that are directly served by the improved highway1
will benefit more than those not served.2
Our unit of analysis for each the case studies is the municipal level, includ-3
ing all incorporated cities for which at least total annual employment figures are4
reported. Many cities also have industry-level annual totals, but the lack of con-5
sistency in reporting due to suppression (particularly among smaller towns) leads6
us to focus our attention on total employment in order to avoid a potential source7
of bias. In analyzing the US71/TH23 and US 53 case studies, we restrict the sam-8
ple of cities included in the analysis to those within the county where the project9
is located. In the case of the improvements to TH 371 we include cities in both10
Crow Wing and Morrison Counties, since the expanded section of highway spans11
parts of both counties and connects the largest cities in each county (Brainerd and12
Little Falls).13
The QCEW employment data are available annually dating back to 2000, with14
the most recent year of complete data being 2010 as of the time of this writing15
(though some quarterly data are available for 2011). While the QCEW data con-16
tain relatively detailed information about employment at the minor civil division17
(MCD) level, it is somewhat difficult to find other variables that are measured at18
the same geographic scale and that are available on an annual basis. Similar to our19
analysis of county-level earnings, we focus on predicting employment levels as a20
function of a few, relatively straightforward demographic and economic factors.21
Specification22
City population and per capita incomes are the primary statistical controls em-23
ployed in our model of local employment. The former variable helps to control24
for the cross-sectional heterogeneity in the sample of employment data and rep-25
resents an indicator of market size, since larger cities are likely to be served by26
a larger base of service industries. A variable representing income levels is in-27
cluded, both to account for differences in consumption levels (holding population28
constant) and as a way to account for some of the macroeconomic trends present29
during the study period, primarily the onset of the recession toward the end of the30
decade. Since no measure of municipality-level incomes is available on an annual31
basis, we use as a proxy measure per capita wages. Wages represent one of the32
major components of personal income and correlate reasonably well with other33
measures of total income. There are, however, a couple of notable limitations34
to their use as a proxy for incomes. The first is that they ignore other poten-35
tially important sources of income, such as proprietors’ income, transfers, and36
13
other sources of non-wage income. The second is that, as economists have noted,1
wages tend to be “sticky” in the downward direction during recessions. That is,2
they do not fall as fast as output during a recession, due to previous contractual3
commitments and imperfect information with regard to price changes.4
In order to account for the possible differential effects on city-level employ-5
ment due to the highway improvements in each case study, we introduce into each6
model a set of time- and location-specific indicator variables. These variables7
combine the time element, which identifies whether the observation took place8
before or after the highway improvement was completed, with a spatial indica-9
tor to identify the location of the city or town relative to the improved highway.10
Spatially, the observations are split into three groups:11
• Cities that are located directly on the improved segment of highway12
• Cities that are located upstream or downstream from the improved segment13
of highway14
• Cities that are located neither on nor upstream or downstream from the im-15
proved segment of highway16
We hypothesize that, all else equal, the cities located along the improved seg-17
ment of highway will experience more employment growth than those not near the18
highway (“non-highway” cities). Cities located upstream or downstream from the19
improved segment of highway may also see some growth effects, since they may20
also be beneficiaries of the improved highway, though not as much as those cities21
directly effected. These cities are expected to experience employment growth22
greater than the non-highway cities, but not as great as the cities directly affected.23
The general spatial relationships between the location of the observed cities and24
the improved highway are depicted in Figure 1.25
(place Figure 1 about here)26
27
These space and time-specific characteristics are combined into a set of indi-28
cator variables which capture unobserved differences in employment levels (those29
not attributable to the population and income variables previously mentioned)30
across locations and before and after the highway improvements of interest. Since31
there are three location classifications and two time periods, there are a total of32
six combinations of these variables. We omit one of these time-location com-33
binations, namely the one corresponding to non-highway cities observed before34
the highway improvement, in order to use it as a reference case against which35
14
to evaluate the impacts in other locations and time periods. The use of variables1
representing “before” and “after” periods allows their comparison to check for2
statistically significant changes over time.3
The formal specification of the empirical model is described in Equation 2:4
ln(eit) = α + β1ln(Pit) + β2ln(Iit) +5∑
j=1
γi(Highwayi) + εit (2)
where:5
ln(eit) = natural log of total private sector employment in city i at time t6
ln(Pit) = natural log of population in county i at time t7
ln(Iit) = natural log of real per capita income (in 2009 dollars) in city i at time t8
Highwayi = indicator variables representing location and time-varying character-9
istics of city i10
εit is an error term, and11
α, β1, β2, andγi are parameters to be estimated12
13
As noted, the continuous variables in the model are transformed into their14
natural logarithms to allow for direct elasticity estimates. The set of time and15
location-specific indicator variables are collectively referred to asHighwayi. These16
variables will be described in more detail in the next section. The descriptive17
statistics for the three case study projects are listed in Table 4.18
(place Table 4 about here)19
20
Estimation and Results21
Table 5 displays the results of the basic employment regressions, predicting to-22
tal private sector employment for each of the cities in the three case study areas.23
The models are fit using ordinary least squares estimation. The time and location-24
specific indicator variables are given names which correspond to their location.25
For example, the “H” variable represents cities located along the improved seg-26
ment of highway. The “before” and “after” subscripts denote the time aspect of27
these variables, and correspond to the periods before and after completion of the28
highway improvement. For the variables in each of the estimated equations, the29
table contains the parameter estimates, standard errors, and associated t-statistics.30
We discuss the results of each of the case studies individually.31
(place Table 5 about here)32
15
US71/TH23 Expansion1
We first examine the case of the set of improvements to US Highway 71 and Min-2
nesota Trunk Highway 23, including the Willmar Bypass, in Kandiyohi County.3
The summary results of the model in the first few columns of Table 5 indicate4
that the model provides a good overall fit to the panel data on total employment.5
As should be expected, employment appears to scale with population, with a one6
percent increase in population being associated with greater than a 0.9 percent in-7
crease in employment. Local per capita incomes are also strongly associated with8
employment levels, though the effect is somewhat smaller.9
Looking at the variables representing the time and location, there appears to10
be no evidence of the completion of the highway improvement on employment11
in non-highway cities. The Oafter variable, which measures the residual effect12
of being located in a non-highway (or “off-highway”) city after the completion13
of the highway improvement (relative to being in the same location before the14
improvement), has a very small coefficient which is not statistically different from15
zero at any reasonable level of significance.16
Likewise, there seems to be no statistically significant difference in the vari-17
ables representing cities located on the improved highway segment before and18
after the improvement. The Hbefore variable, representing the residual difference19
in employment levels between cities located on the improved highway segment20
prior to completion of the improvement and non-highway cities during the same21
period, is fairly large, indicating that the cities located along the improved high-22
way already had high employment levels relative to their populations prior to the23
highway improvement. The cities in the “highway” class include the three largest24
cities in the county (Willmar, New London and Spicer), which likely serve as25
higher-order trade centers, accounting for much of this difference. However, it26
is important to note that the difference between the coefficients of the Hbefore27
and Hafter variables, which measure the change in the residual difference be-28
tween highway cities (both before and after the improvement) and the baseline29
case (non-highway cities observed before the improvement), is very small and not30
statistically significant, indicating that the effect of location did not improve as a31
result of the highway expansion. Stated differently, the advantage of being located32
on the improved highway segment (relative to location in a non-highway city) did33
not change as a result of the completion of the highway improvement.34
Lastly, we also consider the variables representing cities located upstream or35
downstream from the improved segment of highway. We include in this class all36
cities located on either US 71 or TH 23 that are not within the improved segment37
16
of highway (the Willmar Bypass or the expanded stretch of TH 23). The difference1
between the Ubefore and Uafter coefficients is about 9 percentage points and in the2
positive direction, indicating that there may be some change in employment for3
cities in this class relative to those in the non-highway class due to the highway4
improvement. We can formally test for equality of the two coefficients using5
Welch’s t-test to determine whether or not the observed difference in coefficients6
is likely due to chance variation. The test statistic generated for the null hypothesis7
of equality of the two coefficients was -1.56, which corresponds roughly to a p <8
.12 level of significance. This is not particularly strong evidence against the null9
hypothesis that the two coefficients are equal, hence we cannot confidently state10
that the observed difference resulted from the effect of the highway improvement,11
as opposed to simply chance variation in the data.12
TH 371 Expansion13
The second case study we consider is the the expansion of TH 371 between Little14
Falls in Morrison County and the Brainerd/Baxter area of Crow Wing County,15
including the Brainerd Bypass. As described previously, this project expanded TH16
371 between the two locations to four lanes in each direction, adding several new17
interchanges and the bypass around the central business district of Brainerd. This18
case study contains the largest sample of the three case studies fully analyzed in19
this section. This is due to the fact that we include both Crow Wing and Morrison20
Counties in the sample, as the improved highway traverses significant parts of21
both counties. We estimate the impacts of this project in the same manner that we22
did the US71/TH23 case study.23
The results in Table 5 again indicate a good overall fit for the model, with the24
population and income variables both having the expected sign and level of sig-25
nificance. The coefficient of the population variable is very similar in magnitude26
to the one estimated in the US71/TH23 case, though the coefficient of the income27
variable is somewhat larger. The variable representing non-highway cities shows28
no indication of a change in employment as a result of the completion of the high-29
way improvement. The coefficient for the variable representing the cities located30
on the improved highway (Baxter, Brainerd, Camp Ripley and Little Falls) after31
completion of the improvements is slightly larger in magnitude than the variable32
for the same set of cities observed before completion of the highway improve-33
ment. However, the magnitude of this difference is small relative to the estimated34
standard errors of the coefficients, indicating that this difference is well within35
the error bands of the two coefficients and unlikely to represent a statistically sig-36
17
nificant effect. The same is true of the variables representing the cities located1
upstream and downstream from the improved segment of TH 371. The difference2
in the estimated coefficients is small, both in absolute terms and, more impor-3
tantly, relative to their respective standard errors. Therefore we have no evidence4
from this case study that the highway improvement had a statistically significant5
effect on employment levels.6
US 53 Expansion7
The context for our third case study, the expansion of US Highway 53 north of8
Virginia, Minnesota in St. Louis County, is slightly different than the previous9
two and thus requires a modification to our empirical approach. The improve-10
ments to US Highway 53 that are the focus of this case study are an expansion11
to four lanes of an 11-mile segment of the highway between the towns of Vir-12
ginia and Cook. Unlike the previous two case studies, the improved segment of13
US 53 in this case does not pass through any cities. Thus, we cannot define the14
location characteristics of the cities in the sample in the same way that we treated15
the previous cases. As an alternative, we retain the variables representing cities16
upstream and downstream from the improved highway (in this case including all17
cities in St. Louis County along US 53) and add a classification for cities not18
located immediately along the highway, but within 10 miles of it. These are the19
variables labeled A in Tables 4 and 5 to indicate their adjacency to the improved20
highway. The definition of non-highway cities is adjusted accordingly to include21
all cities not within 10 miles of US 53. The definition of cities in the “adjacent”22
class allows us to account for the unique geographic clustering of several mining23
towns near US 53 in the Iron Range region of the county.24
The results of the model for the US 53 improvements in Table 5 show similar25
findings to those in the previous two case studies. Population and income vari-26
ables have similar signs and magnitudes to those included in the other models.27
Again, employment in non-highway cities do not appear to have been affected28
by the completion of the improvements. The coefficient for the Uafter variable,29
indicating the effect on employment in cities located on Highway 53 after the30
completion of the expansion project, is smaller than the corresponding variable31
for the pre-improvement period, though again this difference is too small to reg-32
ister as statistically significant. The same is true for cities located within 10 miles33
of the highway, as indicated by the adjacency variables. Overall, the model results34
appear to reject the notion of statistically significant employment impacts due to35
the expansion of the highway.36
18
Conclusions1
The findings of no statistically significant impacts on earnings and employment2
from the two sets of empirical analyses at the core of this study, while consistent,3
do require some caveats. First, we note that in both of the empirical applications4
there were relatively few years in the time period following the construction of the5
highway improvements in each of the case studies. This was especially true for the6
analysis of the employment data, where only 11 years were available, and thus the7
period of analysis was somewhat compressed. Likewise, in both cases the latter8
years in the sample corresponded to a fairly deep depression, from which most9
US states are continuing to recover. Thus, it may be useful to continue to track the10
economic growth paths of the case study locations for several years following the11
completion of this study and, if necessary, to update the analysis.12
We note that our findings are in general agreement with a number of other13
recent studies of highway improvements, including those focusing on rural re-14
gions, which have found modest to insignificant economic impacts of network15
expansions [4, 20, 14]. These findings are broadly attributed to the maturity of16
the highway network [9, 8] and the relatively modest improvements in travel time17
brought about by the projects in question. In a similar vein, others have noted18
the importance of distinguishing between average and marginal net rates of return19
[7] in studies where transportation infrastructure is treated in a more aggregate20
fashion and longer time series are used.21
We do not interpret the results of our analysis as necessarily providing prima22
facie evidence that the projects in question were not cost-effective or economically23
viable. Rather, we suggest that projects like these should continue to be evaluated24
based on their ability to deliver benefits to users. Travel time savings, safety ben-25
efits, and other types of measurable social benefits, such as reductions in pollutant26
emissions, are the core justification for undertaking improvements to highways27
and other transportation networks. Moreover, focusing only on jobs created or28
local income effects may cause analysts to lose sight of the fact that for highway29
improvement projects like those covered in this study, many of the benefits may30
be diffuse and accrue to non-local users of the highway. The focus on user ben-31
efits seems particularly appropriate in light of the fact that there are few projects32
that could not pass a standard cost-benefit test, yet also be expected to provide33
sizable economic development benefits. We believe this observation applies not34
only to the rural highway expansion projects examined here, but more broadly to35
the majority of transportation projects under consideration at state DOTs and local36
planning organizations.37
19
References1
[1] Aschauer, D. A. (1989). Is public expenditure productive? Journal of Mone-2
tary Economics 23, 177–200.3
[2] Beck, N. and J. Katz (1995, September). What to do (and not to do) with4
time-series cross-section data. American Political Science Review 89(3), 634–5
647.6
[3] Carlino, G. A. and E. S. Mills (1987). The determinants of county growth.7
Journal of Regional Science 27(1), 39–54.8
[4] Chandra, A. and E. Thompson (2000). Does public infrastructure affect eco-9
nomic activity? Evidence from the rural interstate highway system. Regional10
Science and Urban Economics 30(4), 457–90.11
[5] Clark, D. E. and C. A. Murphy (1996). Countywide employment and pop-12
ulation growth: an analysis of the 1980s. Journal of Regional Science 36(2),13
235–256.14
[6] Deller, S. C., T.-H. Tsai, D. W. Marcouiller, and D. B. English (2001, May).15
The role of amenities and quality of life in rural economic growth. American16
Journal of Agricultural Economics 83(2), 352–365.17
[7] Eberts, R. W. (2009). Understanding the Contribution of Highway Investment18
to National Economic Growth: Comments on Mamuneas’s Study. Technical19
report, W.E. Upjohn Institute. prepared for the Federal Highway Administra-20
tion, US Department of Transportation.21
[8] Fernald, J. (1999). Roads to prosperity? Assessing the link between public22
capital and productivity. American Economic Review 89, 619–638.23
[9] Forkenbrock, D. J. and N. S. Foster (1996, August). Highways and business24
location decisions. Economic Development Quarterly 10(3), 239–248.25
[10] Garcia-Mila, T., T. J. McGuire, and R. H. Porter (1996, February). The effect26
of public capital in state-level production functions reconsidered. Review of27
Economics and Statistics 78(1), 177–180.28
[11] Glaeser, E. L., H. D. Kallal, J. A. Scheinkman, and A. Shleifer (1992, De-29
cember). Growth in cities. Journal of Political Economy 100(6), 1126–1152.30
20
[12] Holtz-Eakin, D. (1994). Public sector capital and the productivity puzzle.1
Review of Economics and Statistics 76, 12–21.2
[13] Jiwattanakulpaisarn, P., R. B. Noland, and D. J. Graham (2010). Causal3
linkages between highways and sector-level employment. Transportation Re-4
search, Part A 44, 265–280. Dynamic panel VAR with spatial spillover.5
[14] Jiwattanakulpaisarn, P., R. B. Noland, and D. J. Graham (2012, September).6
Marginal productivity of expanding highway capacity. Journal of Transport7
Economics and Policy 46(3), 333–347.8
[15] Jiwattanakulpaisarn, P., R. B. Noland, D. J. Graham, and J. W. Polak (2009,9
March). Highway infrastructure and state-level employment: a causal spatial10
analysis. Papers in Regional Science 88(1), 133–159.11
[16] Keeler, T. E. and J. S. Ying (1988). Measuring the benefits of a large public12
investment: the case of the U.S. Federal-Aid Highway System. Journal of13
Public Economics 36, 69–85.14
[17] Munnell, A. H. (1990, January/February). Why has productivity growth15
declined? Productivity and public investment. New England Economic Review,16
3–22.17
[18] Nadiri, M. and T. Mamuneas (1998). Contribution of highway capital to18
output and productivity growth in the US economy and industries. Federal19
Highway Administration.20
[19] Rephann, T. J. and A. M. Isserman (1994). New highways as economic de-21
velopment tools: an evaluation using quasi-experimental control group match-22
ing methods. Regional Science and Urban Economics 24, 723–751.23
[20] Shirley, C. and C. Winston (2004). Firm inventory behavior and the returns24
from highway infrastructure investments. Journal of Urban Economics 55,25
398–415.26
[21] Stephanedes, Y. J. (1990). Distributional effects of state highway investment27
on local and regional development. Transportation Research Record: Journal28
of the Transportation Research Board 1274, 156–164.29
[22] Wu, J. and M. Gopinath (2008, May). What causes spatial variations in30
economic development in the United States? American Journal of Agricultural31
Economics 90(2), 392–408.32
21
Tabl
e1:
Des
crip
tive
stat
istic
sfo
rTH
371
and
US7
1/T
H23
proj
ectc
ase
stud
ies
TH
371
proj
ectv
aria
bles
Con
stru
ctio
nM
anuf
actu
ring
Ret
ail
Who
lesa
leV
aria
ble
Mea
nS.
D.
Min
Max
Mea
nS.
D.
Min
Max
Mea
nS.
D.
Min
Max
Mea
nS.
D.
Min
Max
lnG
DP t
30.0
90.
1729
.80
30.3
130
.09
0.17
29.8
030
.31
30.0
90.
1729
.80
30.3
130
.09
0.16
29.8
030
.31
lnSt
ateE
arn t
15.9
80.
2515
.53
16.2
817
.03
0.08
16.8
817
.21
16.1
90.
1216
.03
16.4
916
.12
0.14
15.8
216
.27
lnPo
p it
10.2
10.
449.
4411
.05
10.2
10.
449.
4411
.05
10.2
10.
449.
4411
.05
10.2
10.
449.
4411
.05
Cro
wW
ing 1
0.07
01
0.07
01
0.07
01
0.07
01
Cro
wW
ing 2
0.08
01
0.08
01
0.08
01
0.08
01
Cro
wW
ing 3
0.04
01
0.04
01
0.04
01
0.04
01
Mor
riso
n 10.
070
10.
070
10.
070
10.
070
1M
orri
son 2
0.08
01
0.08
01
0.08
01
0.07
01
Mor
riso
n 30.
040
10.
040
10.
040
10.
040
1N
9595
9594
1
US7
1/T
H23
proj
ectv
aria
bles
Con
stru
ctio
nM
anuf
actu
ring
Ret
ail
Who
lesa
leV
aria
ble
Mea
nS.
D.
Min
Max
Mea
nS.
D.
Min
Max
Mea
nS.
D.
Min
Max
Mea
nS.
D.
Min
Max
lnG
DP t
30.0
90.
1729
.80
30.3
130
.09
0.17
29.8
030
.31
30.0
90.
1729
.80
30.3
130
.09
0.16
29.8
030
.31
lnSt
ateE
arn t
15.9
80.
2615
.53
16.2
817
.03
0.08
16.8
817
.21
16.1
90.
1216
.03
16.4
916
.12
0.14
15.8
216
.27
lnPo
p it
9.76
0.46
9.25
10.6
39.
750.
469.
2510
.63
9.75
0.46
9.25
10.6
39.
750.
479.
2510
.63
Kan
diyo
hi1
0.07
01
0.07
01
0.07
01
0.07
01
Kan
diyo
hi2
0.04
01
0.04
01
0.04
01
0.04
01
Kan
diyo
hi3
0.05
01
0.05
01
0.05
01
0.05
01
N11
011
40
114
111
22
Tabl
e2:
Indu
stry
-lev
elea
rnin
gsre
gres
sion
sfo
rTH
371
impr
ovem
ents
Con
stru
ctio
nM
anuf
actu
ring
Ret
ail
Who
lesa
leV
aria
ble
Coe
ff.
S.E
.t-
valu
eC
oeff
.S.
E.
t-va
lue
Coe
ff.
S.E
.t-
valu
eC
oeff
.S.
E.
t-va
lue
lnG
DP t
-0.6
480.
303
-2.1
40.
654
0.23
92.
73-0
.390
0.13
1-2
.97
-1.7
131.
048
-1.6
3ln
Stat
eEar
n t1.
084
0.17
46.
240.
871
0.23
33.
730.
839
0.07
610
.99
2.19
20.
853
2.57
lnPo
p it
0.74
30.
155
4.79
-0.0
910.
179
-0.5
11.
063
0.12
28.
700.
231
0.14
51.
59C
row
Win
g 10.
685
0.17
93.
822.
194
0.18
811
.67
0.51
50.
144
3.59
1.18
10.
137
8.64
Cro
wW
ing 2
0.69
80.
179
3.90
2.13
90.
188
11.3
80.
510
0.14
43.
551.
174
0.13
38.
81C
row
Win
g 30.
797
0.18
44.
342.
013
0.19
210
.46
0.53
00.
148
3.57
1.11
00.
143
7.76
Mor
riso
n 10.
211
0.09
92.
131.
504
0.13
311
.28
-0.0
300.
098
-0.3
10.
779
0.13
85.
66M
orri
son 2
0.11
60.
091
1.28
1.53
70.
124
12.4
1-0
.019
0.09
2-0
.20
0.70
40.
134
5.28
Mor
riso
n 30.
183
0.09
31.
961.
433
0.13
110
.97
0.04
00.
094
0.43
0.94
80.
173
5.47
Con
stan
t11
.839
6.86
31.
73-1
0.22
04.
510
-2.2
7-1
0.24
71.
685
-6.0
85.
307
8.67
70.
61ρ
0.68
80.
723
0.74
60.
540
Wal
dχ2(9)
1699
.83
2359
.04
2047
.87
912.
80A
djus
tedR
20.
997
0.96
10.
997
0.97
1N
9595
9594
23
Tabl
e3:
Indu
stry
-lev
elea
rnin
gsre
gres
sion
sfo
rUS
71im
prov
emen
ts
Con
stru
ctio
nM
anuf
actu
ring
Ret
ail
Who
lesa
leV
aria
ble
Coe
ff.
S.E
.t-
valu
eC
oeff
.S.
E.
t-va
lue
Coe
ff.
S.E
.t-
valu
eC
oeff
.S.
E.
t-va
lue
lnG
DP t
-0.3
140.
291
-1.0
81.
196
0.12
59.
57-0
.348
0.15
1-2
.30
0.64
00.
219
2.92
lnSt
ateE
arn t
0.70
60.
175
4.03
0.37
10.
155
2.40
0.34
90.
082
4.27
0.68
40.
204
3.35
lnPo
p it
1.59
00.
205
7.77
1.21
20.
116
10.4
20.
826
0.11
47.
270.
425
0.20
92.
03K
andi
yohi
1-0
.230
0.23
5-0
.98
-0.1
210.
119
-1.0
10.
715
0.13
95.
130.
951
0.22
34.
27K
andi
yohi
2-0
.233
0.23
6-0
.99
-0.1
400.
120
-1.1
70.
796
0.13
95.
710.
760
0.22
53.
38K
andi
yohi
3-0
.206
0.23
8-0
.86
-0.1
630.
122
-1.3
40.
864
0.14
06.
160.
727
0.22
63.
21C
onst
ant
-7.2
037.
039
-1.0
2-4
3.27
24.
654
-9.3
011
.114
4.66
62.
38-2
4.60
34.
956
-4.9
6ρ
0.75
90.
608
0.86
20.
701
Wal
dχ2(6)
1035
.86
990.
1514
57.0
410
64.7
8A
djus
tedR
20.
979
0.98
30.
998
0.97
9N
110
114
114
111
24
Tabl
e4:
Des
crip
tive
stat
istic
sfo
rem
ploy
men
tana
lysi
sca
sest
udie
s
US
71/T
H23
TH
371
US
53V
aria
ble
Mea
nS.
D.
Min
Max
Mea
nS.
D.
Min
Max
Mea
nS.
D.
Min
Max
lne i
t5.
401.
672.
839.
525.
361.
791.
619.
286.
121.
892.
6410
.83
lnPit
6.50
1.29
4.61
7.03
6.39
1.48
3.37
9.53
7.25
1.71
4.44
11.3
7lnI i
t9.
150.
857.
6810
.89
9.02
0.91
5.99
11.2
08.
990.
697.
1410
.66
Oafter
0.35
01
0.33
01
0.08
01
Hbefore
0.10
01
0.08
01
Hafter
0.17
01
0.07
01
Ubefore
0.07
01
0.08
01
0.22
01
Uafter
0.12
01
0.07
01
0.05
01
Abefore
0.23
01
Aafter
0.06
01
N12
129
524
7
25
Tabl
e5:
Priv
ate-
sect
orem
ploy
men
treg
ress
ions
forU
S71
/TH
23,T
H37
1,an
dU
S53
impr
ovem
ents
US7
1/T
H23
TH
371
US5
3V
aria
ble
Coe
ff.
S.E
.t-
valu
eC
oeff
.S.
E.
t-va
lue
Coe
ff.
S.E
.t-
valu
elnPit
0.92
60.
011
84.8
80.
921
0.01
278
.44
0.91
20.
008
115.
35lnI i
t0.
660
0.01
349
.24
0.82
20.
019
42.4
20.
710
0.02
924
.64
Oafter
0.00
20.
028
0.07
-0.0
140.
036
-0.3
8-0
.042
0.07
5-0
.56
Hbefore
0.52
80.
045
11.6
00.
126
0.06
71.
89H
after
0.52
40.
041
12.9
00.
152
0.07
02.
16Ubefore
-0.2
670.
045
-5.9
10.
036
0.06
00.
590.
155
0.03
84.
09Uafter
-0.1
760.
038
-4.6
9-0
.006
0.06
5-0
.09
0.13
30.
055
2.43
Abefore
-0.1
550.
040
-3.8
3A
after
-0.1
130.
072
-1.5
7C
onst
ant
-6.7
690.
140
-48.
48-7
.959
0.18
1-4
3.98
-6.8
590.
237
-28.
92A
djus
tedR
20.
995
0.97
90.
987
N12
129
524
7
26
Figure 1: Location of cities relative to improved highway segment
Improved highway
segment
Upstream /
downstream cities
Non-highway city Non-highway city
City on improved
highway segment
27