Vous êtes sur la page 1sur 20

The Location Decisions of Biodiesel Refineries

T. Randall Fortenbery, Steven C. Dlier, and Lindsay Amiel


ABSTRACT. We examine the community characteristics that lead to locating a biodiesel plant. Utilizing
data that includes all biodiesel plants in the United
States, we employ spatial econometrics to evaluate
those characteristics that lead to a plant siting. While
public policies associated with biofuels may be endogenous to location decisions, we find them unimportant except in the case of consumption mandates.
Despite evidence of plant clusters, it does not appear
that operating plants are critical to the siting of a new
plant. Results provide insights for community leaders
relative to actions that can increase the probability a
plant will be built. (JEL 018, Q42)

I. INTRODUCTION
Despite significant growth in U.S. biofuels
production capacity, controversy exists concerning the impacts biorefineries have on individual communities. As a result there are a
range of reasons some plants face little local
resistance and others fail to get local approval
(Fortenbery and Dlier 2008; Low and Isserman 2009). The earliest domestic biorefineries focused on the production of ethanol, with
biodiesel production lagging substantially. In
2009, for example, the United States produced
10.75 billion gallons of ethanol from plants
that had a total capacity of about 13.5 billion
gallons (Renewable Fuels Association 2010),
while total U.S. biodiesel production was only
490 million gallons from plants with a total
capacity of about 2.7 billion gallons (National
Biodiesel Board 2010).
Due to ethanol's domination in the U.S.
biofuels market, there have been several studies examining the relationships between ethanol plants and their host communities (see,
e.g., McNew and Griffith 2005; Swenson
2008; Fortenbery, Turnquist, and Foltz 2008;
Flora 2008). The works of Sarmiento and Wil-

Land Economics February 2013 89 (1): 118-136


ISSN 0023-7639; E-ISSN 1543-8325
2013 by tbe Board of Regents of tbe
University of Wisconsin System

son (2007), Haddad, Taylor, and Owusu


(2010), Lambert and his colleagues (2008),
and Low and Isserman (2009) are the most
germane to this study in that they explicitly
model the location decisions of ethanol plants
in the United States. Because of much lower
market penetration, however, little work has
focused on issues associated with biodiesel
plants. Such work is important for two reasons. First, the current federal Renewable Fuels Standard (RFS) explicitly allows for biodiesel to be counted as an advanced biofuel,
meaning it can count toward the RFS. This
provides significant opportunity for biodiesel
to enter the fuel stream for commercial trucks.
More importantly, understanding the location
criteria surrounding biodiesel plants could
provide important insight in understanding
next-generation biorefineries. Since cornbased ethanol has essentially reached its maximum allowable contribution to the RFS,
plants focused on cellulosic ethanol production and other advanced biofuels (bio-jet fuel
etc.) will need to be developed. Because these
plants will be developed as part of a comprehensive energy policy, not as a rural development initiative, the issues surrounding plant
location may be much more similar to those
associated with biodiesel plants, as opposed
to earlier-generation ethanol plants.
Initially one might presume that impacts
would not be all that different: both generate
employment and other benefits from the production of renewable transportation fuels, and
the positives and negatives associated with
biofuel manufacturing might be similar regardless of whether the fuel produced is a gasoline or a diesel fuel substitute. From a
The authors are, respectively, professor and Tom
Mick Endowed Chair, Scbool of Economic Sciences,
Wasbington State University, Pullman; professor. Department of Agricultural and Applied Economics,
University of Wisconsin-Madison; and graduate researcb assistant. Department of Agricultural and Applied Economics, University of Wisconsin-Madison.

89(1)

Fortenbery, Deller, and Amiel: Location of Biodiesel Refineries

broader perspective it is important to note that


the initial promotion of ethanol production
was widely seen as a rural development policy
aimed at helping corn farmers and mral communities capture more value-added dollars.
The promotion of biodiesel, on the other hand,
has been viewed more as an energy policy.
There are, however, considerable differences
in the location decisions of ethanol versus biodiesel plants. The greatest percentage of ethanol producing capacity has been focused in
the upper Midwest, where the dominant feedstock (corn) is readily available (Haddad, Taylor, andOwusu2010).
The location of biodiesel plants has been
more heterogeneous than that of ethanol
plants, both in terms of geographical location
and the size of communities hosting plants.
The purpose of this study is to understand
what drives this difference. It could be due to
a variety of factors, including the ability of
most biodiesel plants to utilize various feedstocks ranging from recycled cooking grease
to soybean oils, and the specific chemistry of
biodiesel. Biodiesel can be introduced at any
point along the fuel marketing chain, making
it less critical where plants actually locate.
Perhaps most importantly, biodiesel has not
been primarily viewed through the lenses of
mral economic development policy and increased agricultural value added at the community level. Commercial biodiesel has historically been developed by entrepreneurs that
do not necessarily have agricultural backgrounds. As such, the strong ties to production
agriculture that dominated early ethanol plant
development do not apply to biodiesel, and
more importantly may not apply to next-generation biofuels.
IL U.S. HISTORY OF BIODIESEL
PRODUCTION
The potential for a biodiesel fuel goes back
to at least 1900, when German engineer Rudolf Diesel used peanut oil to power one of
his engines at the Paris Exposition (Wu 1998).
As noted above, however, growth in U.S. biodiesel production has lagged that of ethanol
production, in terms of both total output and
as a percentage of petroleum based fuels.
There was effectively no commercially pro-

119

duced biodiesel before 2001 in the United


States, and by 2005 total biodiesel production
represented only 0.21% of the U.S. diesel fuel
pool (U.S. Energy Information Administration 2007). In 2007 total U.S. biodiesel production increased to 500 million gallons, and
increased another 200 million gallons in 2008
(Weber 2009). Production in 2008 represented
about 1.2% of total U.S. distillate fuel consumption. While annual percentage increases
in U.S. biodiesel production have been impressive, measured in both total volume and
as a percentage of diesel consumption, they
pale in comparison to the relationship between ethanol and gasoline.
Despite its slow start, the past several years
have seen increased interest in alternative diesel fuels to both control emissions and contribute to a reduction of petroleum dependency
(Durbin 2007). Current renewable fuels legislation mandates a total of 36 billion gallons of
renewable fuel in the U.S. transportation fuel
stream by 2022. Only 15 billion gallons ofthat
can come from com-based ethanol. Further, the
Environmental Protection Agency (EPA) has
mandated that specific portions of the 36 billion gallon renewable fuel total required by
2022 be satisfied by biodiesel. In 2011, for example, the total biodiesel contribution was required to be 0.8 billion gallons (U.S. Environmental Protection Agency 2010). This is
expected to increase over the next several
years. In addition, biodiesel can count toward
other advanced biofuels mandates under
RFS2. This leaves considerable opportunity
for biodiesel to increase its contribution to the
overall domestic biofuel supply.
Biodiesel is made by a process called esterification, in which an alcohol (usually
methanol) is mixed with a catalyst such as
vegetable oil to produce a methyl ester
(Nobbe 1995). While there are many different
types of vegetable oil, most U.S. biodiesel is
made from soybean oil. This is due to both its
availability and a $1 per gallon soybean oil
blending credit (Williamson 2002). In addition to vegetable oil, waste grease, like the
yellow grease from deep fryers and restaurants' trap greases, can be used for biodiesel,
but these recycled feedstocks tend to produce
lower yields (Fortenbery 2004).

120

Land Economics

In addition to feedstock variety, the chemical properties of biodiesel and ethanol are
fundamentally different and affect where they
can be introduced into the fuel stream. Both
can be splash-blended, meaning the biofuel is
simply mixed into the petroleum fuel by
dumping it in, but there are some issues associated with both the transport and distribution of gasoline mixed with ethanol that are
not constraints for a biodiesel/petroleum diesel mix. For example, neither ethanol nor an
ethanol/gasoline blend can be moved through
the current pipeline distribution system. As a
result, ethanol is usually splash-blended with
gasoline into a tanker truck just prior to delivery to a retail location. Since the blending of
biodiesel with petroleum diesel does not result
in a product incompatible with pipeline delivery, biodiesel blending can occur anywhere
along the fuel distribution network. Thus,
from both input and output market perspectives, biodiesel plants are more flexible relative to ethanol plants in terms of where they
locate.
Given the potential for biodiesel production to grow in response to the RFS, a more
complete understanding of what contributes to
a plant's location decision and ultimate success is warranted. To that end, the objectives
of this study are to first identify potential community characteristics that might contribute to
an attractive location environment from a
plant's perspective, and then to estimate the
relative importance of each of the identified
characteristics. This will be useful in assessing the potential for a given community to attract a plant as production capacity increases,
and to understand the community dynamics
that lead to economic success.
Before turning to the theory of firm location, our model and the empirical results consider the simple spatial location of biodiesel
plants (Figure 1). Unlike ethanol plants,
which are highly concentrated or clustered in
the Com Belt of the Midwest, biodiesel plants
are more scattered across the United States.
Indeed, commercial biodiesel plants are located at all four comers of the United States,
from San Diego, California, to Bellingham,
Washington, to Maine and southem Florida
and the very southern tip of Texas. Other than
a concentration in Houston, Texas, commer-

February 2013

cial biodiesel plants appear to be evenly distributed across much of the United States.
III. SPATIAL DEPENDENCY IN
BIODIESEL PLANT LOCATION
To test for spatial dependency in the location of the biodiesel plants identified in Figure
1, we computed the Moran's / statistic, which
is a standard global statistic of spatial dependency (Anselin 1988).' The test statistic explicitly measures whether the pattem of plant
locations is spatially random or not. Spatial
dependency, or spatial autocorrelation, is the
spatial equivalent of autocorrelation in time
series data. For the biodiesel data, 4he global
Moran's / is 0.0143 with a Z score of 9.563.
This provides strong evidence that there does
exist spatial dependency in the plant location
data at a greater than 99.9% level of statistical
confidence. Thus, the pattern observed in Figure 1 is not spatially random. This dependency strongly suggests that aspatial statistical modeling approaches will lead to
significant model misspecification and biased
parameter estimates (Anselin 1988).
Given that biodiesel plant locations are not
distributed randomly across U.S. counties, the
issue of spatial concentrations or clusters
arises. As previously observed, there is a
strong spatial clustering of ethanol plants, but
from Figure 1 it is not clear whether there are
similar clusters for biodiesel plants. To test for
spatial clustering we calculated the Anselin
local Moran's / statistic (Anselin 1995) along
with the Getis-Ord Gi* statistic (Getis and
Ord 1992; Ord and Getis 1995). In general,
we identify several spatial clusters. The specific test results and a description of the clusters are presented in the Appendix.

'The

Moran's

is

defined

as

/ =-

-, where wj is a spatial weight matrix defining the spatial proximity of X; to Xj. Generally the
weight matrix is dened on adjacency where viy takes on a
positive value for counties that are adjacent and zero if nonadjacent.

89(1)

Fortenbery, Deller, and Amiel: Location of Biodiesel Refineries

121

I 1 Dot = 1
number

FIGURE 1
Location of Biodiesel Plants

Modeling the Location of Biodiesel Plants


A general problem facing a biodiesel plant
relative to its location decision is that potential
customers and feedstock suppliers are scattered across a broad landscape. The firm faces
a location choice that involves several considerations that will impact its ability to maximize profits. A primary issue is the transport
costs associated with shipping inputs to the
firm and final product to consumer markets.
In addition the firm's location decision will be
impacted by such issues as land and labor
costs, taxes and quality of public services, and
the local regulatory environment, among others. The final site selected need not be the lowest transportation cost site if lower land and
labor costs, for example, are sufficient to offset higher transportation costs. The location
decision is often discussed as a two-step process: step one identifies areas where transportation costs are minimized, and step two involves identifying a specific plant site.
Following Gabszewicz and Thisse (1986),
McCann (2002), and Shaffer, Dlier, and Marcouiller (2004), the firm's problem can be formally stated as

[1]
i= 1

1=1

where II is profit; P,- is price charged at market, / = 1 . . . m; /(P,) is demand for the firm's
product at market i= I . . . m; s' is spatial
location of market = 1 . . . m; t{s,s') is the
cost of transporting one unit of output from
firm location s to market location .';/is fixed
costs facing the firm to produce output; v is
the constant marginal cost of producing one
unit of output; x, is production inputs from
market i=l . . . n; d(s,s') is the cost of transporting one unit of input x from market location s' to firm location s; and ^(jc,) is the
output level of the firm.
In a spaceless world, t(s,s') = d{s,s') = 0,
and the biodiesel firm has one decision, what
price (Pi) to charge in each of its separate markets. Once a specific market's price has been
established, say P*, the amount sold in that
market is determined by its respective demand
function Di(P*). In a spatial world where
'
and d(s,s')^O, the selection of loca-

122

Land Economics

February 2013

EIGURE 2
Theoretical Location Decision Problem

tion has a direct impact on the effective price


of output. Here the consumer must endure
higher prices because of the cost of transportation of the output: the greater the distance
the output is shipped the higher the effective
price of the output. This transportation cost
will result in lower quantity demanded and
total revenue for the firm.
This presents two problems. The first centers on selecting a location {s) that minimizes
the cost of shipping raw materials, such as
soybean oil, to the plant. The second is minimizing output transport costs. From the point
of view of a biodiesel plant, the output market
is where biodiesel is blended with petroleum
diesel. Where the blended biodiesel is shipped
is irrelevant to the firm. The plant's revenue
is maximized by offering the lowest delivery

or effective price possible, inclusive of transportation costs and local factors that influence
production costs.
To better understand the problem, consider
a situation where a biodiesel firm is looking
at three output markets (m = 3) as well as three
input markets ( = 3), and the markets overlap
(Figure 2). The firm is selecting a location
{s*), somewhere between the three markets
(5'',.2 ,53-)^ tj^at \yjll maximize profits. In this
simple example, the firm is shipping feedstock
from three markets {d{s,s')) to a centrally located facility and then shipping product back
to the same three markets {t(s,s')). Clearly the
firm will locate within the triangle defined by
the three hypothetical markets.
In the second stage of the location decision,
communities often have some influence on the

89(1)

Fortenbery, Deller, and Amiel: Location of Biodiesel Refineries

final outcome. Communities can offer highquality infrastrticture, skilled labor, building
locations, and generally high quality of life
characteristics and not only offer low-cost alternatives for the firm but, more importantly,
offer a viable comparative advantage over
other locations. This indicates communities
must be sensitive to the total cost of production in one community compared with other
communities. Thus the community seeks to
keep transportation rates low or to offset
higher transportation costs by reducing the
nontransportation costs through things like
lower wages, inexpensive land, and tax concessions. Indeed, much of the empirical firm
location literature is aimed at attempting to
gain a better understanding of how these secondary factors influence firm decisions.
Empirically measuring the firm's decision
criteria outlined in equation [1] is difficult if
not impossible without firm-specific data that
is generally not available to researchers. As
an altemative, Guimaraes, Figueiredo, and
Woodward (2004) note that researchers have
turned to McFadden's (1974) random utility
framework to provide both a theoretical and
empirical foundation for their work. Here the
firm decomposes latent or unobserved profits
(n,>,) of location altemative / for firm n into
systematic (IT,-,,) and stochastic (,) components. The systematic profit is a function of k
separate characteristics describing a particular
spatial site (X,^), which have individual
weights {ik). The relationship can be expressed as Il/;, = n,-(-,, where n ,
- Zjk= iPin-^ink-

The stochastic elements of profit (,) are


random variables that capture noise in the location decision-making process, such as a lack
of complete information of all potential sites
and/or the ability to process, including incomplete information. As outlined by Hunt, Boots,
and Kanaroglou (2004) this framework allows
researchers to specify a probability that a firm
will select an altemative from the choice set.
The probability that firm n selects altemative
site / from a set of C alternatives equals the
probability that the latent profit of site / is
greater than or equal to the best latent profit
of all other alternatives;

p{i) = P{n,,+e,, >nj

123

c.

[2]

Note that the set of relevant altematives is defined by the spatial boundaries of the relevant
markets, or the triangle determined by the
three markets in Figure 2.
In order to estimate the relationship outlined in equation [2] we must make some assumptions about the form of the stochastic
element of profits (e,). If we assume that
these are independently and identically distributed, the multinomial logit model of
McFadden (1974) results. This model can be
expressed as
[3]

2
./=!

which can be directly estimated. Carlton


(1979, 1983) showed that the conditional logit
model fits the random utility model, or in our
case random profit model, of McFadden for
firm location.
As we saw in the spatial analysis of biodiesel plant locations (Figure 1 and the Appendix), there exists prima facie evidence of
some clustering among plants. The spatial
concentrations, through the Anselin local
Moran's / and Getis-Ord Gi* estimation, however, do not ensure that these biodiesel plants
are benefiting from agglomeration economies
as outlined by Porter (1990, 1995, 2000) and
others. Simply because firms are in close spatial proximity does not ensure that they are
behaving like an economic cluster. What we
can conclude, however, is that there is significant spatial dependency in the location of biodiesel plants.
Estimation Methods
Because there is prima facie evidence that
there is spatial dependency, traditional methods of estimating location decision and/or
concentration patterns are suspect. Unfortunately, as noted by Robertson, Nelson, and De
Pinto (2009), most researchers elect to either
ignore the potential estimation problems or
employ simplistic "step-around" methods.
More recently, however, two approaches that

124

February 2013

Land Economics

directly address spatial dependency have received considerable attention. The first is a recursive-importance-sampling (RIS) estimator
suggested by Beron, Murdoch, and Vijverberg
(2003) and Beron and Vijverberg (2004), and
the second a Bayesian strategy of Markov
Chain Monte Carlo (MCMC) developed by
LeSage (1999, 2000) and documented by
LeSage and Pace (2009).^ For this study we
have elected to use the Bayesian strategy offered by LeSage. To test for consistency of our
results we estimate a spatial logit along with
a spatial tobit model of spatial concentration.
In the first model, F, is the binary dependent variable for county /, which takes the
value of one if a biodiesel plant is located in
the county and zero otherwise. Thus, the spatial logit or probit model is the most appropriate. To estimate the spatial logit model, we
rely on the MCMC sampling method. This
method was chosen over others since it can
easily accommodate heteroskedastic disturbances, and the logit model can be estimated
relatively easily by changing the degrees of
freedom. In order to employ the spatial Bayesian approach, however, a distribution for Y
must be established.
The value of the dependent variable proxies the profitability of a biodiesel plant located
in one county relative to another county. It is
assumed to be correlated with a latent, countylevel unobserved utility associated with the
existence, or nonexistence, of a biodiesel
plant in a county. Because more profitable
plants are expected to lead to more stable employment prospects, a higher and more stable
tax base to support the provision of local services (education for example), and a more secure business environment for other busi- As noted by Franzese and Hays (2009), there are other
approaches that have been suggested, such as a two-step
generalized method of moments estimator (Pinkse and Slade
1998; Kelijian and Prucha 1998; Klier and McMillen 2005).
More recently Schnier and Felthoven (2011) suggest that a
method offered by Train (2003) that utilizes a Cholesky matrix within a multinormial framework that they then apply
to land use decisions. Indeed, much of the spatial discrete
choice econometrics has developed in models of land use
(e.g., McMillen 1989: Bockstael 1996: Bell and Irwin 2002;
Irwin and Bockstael 2002; Wang and Kockelman 2009) and
the fisheries literature (e.g.. Hicks and Schnier 2010; Smith
2010: Schnier and Felthoven 2011). However, the RIS and
Bayesian are the most commonly used in applications.

nesses interacting with a biodiesel plant, a


county's utility associated with a plant's existence is assumed proportional to a plant's
profitability. If the value of the dependent
variable is equal to one, then the profitability
of the biodiesel plant in that county is greater
than in another county, thus the county's utility is higher. The converse is tme if the value
of the dependent variable is equal to zero. If
Yi = 0, then the profitability of the plant in the
county is lower than in another county. More

formally,

let

y = Un UQI

\/i=\,...,n,

where Un is the utility if a biodiesel plant is


located in county / and t/g/ is the utility if a
biodiesel plant is not located in county . If
y,* > 0, then a biodiesel plant would be located
in the county and F,- = 1. If the biodiesel plant
selects to locate in county i, then that is a
profit maximizing location. If >'/<0, then a
biodiesel plant would not be located in the
county and y = 0. That is to say,
yi=\

if y*>0,

y, = 0 if

y*<Q,

and P(yi = 1) = P(Uu > UQ) = P(y > 0).

As a consequence, the conditional distribution


for the latent variable takes the following
form:
TMVN{fi,Q.) s.t. //= (I -

and

subject to the restriction that


a<y*<b,
where the bounds of the linear inequality
depend upon the observed values of y. If
>>, = 0, then y * is bounded from above by zero;
if >>,= 1, then >;* is bounded from below by
zero. Using this definition ofy* instead of the
binary variables, it is possible to estimate the
model parameters and p using the conditional posterior distributions that arise for the
spatial Bayesian autoregressive model with
continuous dependent variables (LeSage and
Pace 2009).
For ease of notation, let A = (I - pW).
Then the joint posterior distribution of the parameters 6= {p, p} is given by the following:

89(1)

Fortenbery, Deller, andAmiel: Location of Biodiesel Refineries

lAlexp]

-X)\\y'

where n{) is the prior distribution for and


7t{p) is the prior distribution for the p parameter. We will assign a normal prior distribution to . Note that in both the nonspatial as
well as spatial logit models, and CT| cannot
both be identified. To solve this problem a\
is generally set equal to one (LeSage and
Pace 2009). Thus, the prior distribution for
is assigned to be ~N{c,T)
such that
c = ( X ' X ) - ' X ' A > ' ' and r = ( X ' X ) " ^ The
prior distribution for the parameter p is a uniform distribution on the interval [I/A,I/A]
such that X is the minimurn_eigenvalue of the
spatial weight matrix and A is the maximum
eigenvalue.
The parameters specified above will be estimated using the Gibbs sampling method.
There are three main steps that must be completed during each iteration of the procedure.
First, can be sampled from the following
conditional distribution: p(\ Po,y*) =^ N{c*,T*)
such that the mean is given by c* =(X'X-^
To ' ) ^ ' (X ' A j * + TQ^ ' Co) and the covariance
matrix is given by J*
^{X'X-hT^)'K
TQ , Co, and po are initial values for T, c, and
p and can be set arbitrarily. In our model, the
initial values for co and TQ are zero and
10~ '^, respectively. Next, the values obtained
for are then used to estimate the value of
p. The conditional distribution for p is given
by

y* = + z (LeSage and Pace 2009). A comprehensive explanation of the approach is detailed by Geweke (1991) and LeSage and
Pace (2009).
The estimates fromp()| pQ,y*),p{p\ \,y*),
and p(y*\pi,i)
constitute one iteration
through the MCMC sampler. The estimates
from this iteration are used as the initial values
in the following pass. The Gibbs sampling
procedure must be repeated until the values of
the estimates converge.
In addition to a simple dummy variable for
the presence of a biodiesel plant, we also have
data on the number of biodiesel plants in a
given county. Thus, in the second model, 7,is the number of biodiesel plants in county i
and takes the value of zero if there are no biodiesel plants in the county. The data is truncated from below by zero; however, while
there are simple attempts to develop spatial
Poisson or ordered probit estimators (e.g.,
Wang and Kockelman 2009), there are no spatial zero-inflated Poisson or ordered probit estimators that we are aware of. Specifically,
most U.S. counties do not have a biodiesel
plant, resulting in a spike on zero in the distribution. To take advantage of the additional
information in the count data as well as to
explore the stability of the spatial logit models, we employ the spatial tobit model. The
spatial tobit model is very similar to the spatial logit model described above.
In the spatial tobit model.
y 1 = 1 if y^o,
yi = O if y'<0,
and Piyi = 1 ) = P( [/j , >

Since the conditional distribution does not


take a known form, it must be evaluated using
the Metropolis-Hastings algorithm (Holloway, Shankar, and Rahmanb 2002).
Finally, a sample from y* must be drawn.
According to Geweke (1991), sampling from
the truncated multivariate normal distribution
specified above is the same as sampling from
the n-variate normal distribution z~N{O,Q)
s.t. a- n<z<b . z'" is determined by using the Gibbs sampling technique on the distribution of Zi, conditional on the values of
z - . The sample for y * is obtained by setting

125

/)

= P(y > 0).

The main difference is in the sampling from


y* for the latent variable z. All else follows
from the logit/probit estimator described
above.
Model Specification and Data
To our knowledge no one has attempted to
model the location decisions of biodiesel
plants, and as a result there is little direct empirical literature to draw upon in specifying
our model. There are, however, a handful of
studies that examine the location decisions of

126

Land Economics

ethanol plants (e.g.. Sarmiento and Wilson


2007; Lambert et al. 2008; Low and Isserman
2009; Haddad, Taylor, and Owusu 2010). Unfortunately, the technology that is involved in
ethanol production and distribution is fundamentally different than that for biodiesel. In
addition, ethanol production, as noted earlier,
was looked to as a vehicle for rural development in areas where the actual feedstock was
produced. This facilitated a clustering of ethanol plants in com producing regions.
Biodiesel plants have a much wider variety
of potential feedstocks, implying that even in
the case of contributing to economic development in feedstock rich areas, the spatial
clustering could be quite different. Thus, in a
location model it is not clear how t{s,s') versus
dis,s') should play out for biodiesel plants.
Second, as noted earlier, the chemistry of ethanol as an additive is such that ethanol blending in gasoline is more difficult than biodiesel
blending. These two conditions suggest that
biodiesel plants may be more fiexible in their
location decisions than are ethanol plants.-^
In order to examine the drivers of biodiesel
plant location decisions, we employ countylevel data for the year 2005. Much of the data
are drawn from the 2002 Census of Agriculture (U.S. Department of Agriculture 2004),
the County and City Data Book published by
the U.S. Census Bureau (U.S. Census Bureau
2005), or the Bureau of Economic Analysis's
Regional Economic Information System'*
(BEA-REIS). The plant location data is derived from County Business Patterns, U.S.

3 Another problem with relying on the ethanol plant location literature as a foundation for our work is the manner
in which spatial dependency has been handled. Low and
Isserman (2009) make similar observations about ethanol
input-output markets coupled with simple mapping analysis
to draw their conclusions. Lambert et al. (2008) acknowledge serious spatial dependency in their data and address it
by including a range of spatially descriptive variables as
controls. Haddad, Taylor, and Owusu (2010) ignore the issue, and Sarmiento and Wilson (2007) address a very simple
problem; how does spatial proximity between ethanol plants
influence future location decisions? Thus, through their basic research. Sarmiento and Wilson (2007) address spatial
dependency, but in their empirical work this dependency is
treated simplistically.
"* U.S. Departtnent of Commerce, Bureau of Economic
Analysis, Regional Economic Information System, available
at www.bea.gov/regional/docs/reis2006dvd.cfm.

February 2013

Census Bureau,-^ and is 2007 data. The variables hypothesized to infiuence biodiesel
plant location can be grouped into five blocks:
input markets, output markets, local market
socioeconomic characteristics, transportation
infrastructure, and policy options used to promote biofuels.
To measure the impacts of input variables,
the following proxies are employed:

Number of soybean (or other oilseed) crush


plants per 10,000 persons
Number of restaurants and food service firms
per 10,000 persons
Average farm sales
Acres of crops harvested
Farm share of total county iticome

The crush plant concentrations are aimed at


capturing the potential for oil-bearing crops
being processed within the county. The number of restaurants and food service firms captures the ability to supply the plant with recycled oils and grease from food preparation.
The three measures of farm size are aimed at
capturing the potential input supply derived
from raw agricultural products. It is expected
that the higher the concentration of each of
these potential input sources the greater the
likelihood a biodiesel plant locates.
The output markets are measured by the
following:
Number of tank farms per 10,000 persons
Number of trucking and busing firms per
10,000 persons
Number of fuel pipeline firms per 10,000 persons
Because biodiesel is splash-blended, it can be
added to diesel at any point along the marketing chain, but most blending occurs at tank
farms. Trucking and busing firms are important consumers of diesel and thus represent
significant potential local biodiesel demand.
Although biodiesel tends to be blended at tank
farms, technology exists to add it directly into
the pipeline distribution network. Hence the
number of pipeline firms is designed to proxy
this potential.

^ U.S. Department of Commerce, County Business Patterns, available at www.census.gov/econ/cbp/.

89(1)

Fortenbery, Deller, and Amiel: Location of Biodiesel Refineries

We also include a range of local socioeconomic variables to capture not only potential
market demand but also the potential likelihood of support for biodiesel as an alternative
fuel source. These variables include the following:
Population in 2005
Houses owner-occupied (percent)
Bachelor's degree or higher (percent)
Workers who drove to work alone (percent)
Votes cast for president 2004 Republican (percent)
Persons 5 years and over residing in same house
in 1995 and 2000 (percent)
Net nonfartn btisiness job growth 2000-2004
Local government total revenue per capita
Local government revenue from the property
tax (percent)

Population captures the scale of the potential


local market, and percentage of houses owneroccupied is a proxy of wealth. The latter also
is a very simple measure of the NIMBY, or
"not in my back yard," phenomenon. In the
case of ethanol production there have been
several cases where local communities actively campaigned against the operation of the
plant. Concems over odors and demands
placed on local water supplies are often referenced. Although the technology for biodiesel is different from that for ethanol, misperceptions may result in homeowners
discouraging such firms. Home stability is a
complementary measure of homeownership
and reflects the stability of the local community. Education level is again a proxy for the
potential willingness to embrace biodiesel as
an altemative fuel source.^ The percentage of

* We do not have a strong prior relative to the sign on


the education variable. If education is positively related to
income, and higher-income individuals disproportionally
represent the driving population of alternative fuel vehicles
(they tend to be more expensive and are newer additions to
the private automobile fleet), then a positive relationship
might be expected. Another possibility is that education correlates to a heightened awareness of environmental issues.
To the extent that biodiesel is viewed as a "greener" fuel
than petroleum diesel, this might also lead to a positive relationship. Conversely, however, lower education levels
likely correlate to a higher percentage of blue collar workers.
Since most of the jobs created by a biorefinery are blue collar
in nature, the job creation aspect of a new plant may result

127

drivers who commute alone is also a potential


measure of demand. The percentage of total
votes cast for President Bush in 2004 is a simple measure of political leanings, which may
again reflect the demand for altemative fuels
such as biodiesel. We expect the coefficient
on this variable to be negatively related to
plant location for two reasons: (1) the current
democratic administration has been more aggressive in pushing bioenergy as a federal policy than the previous administration, which
had strong ties to the oil industry; and (2) concems over potential climate change, and the
development of bio-based fuels as a potential
response to CO2 emissions, tends to be more
prominent in the liberal as opposed to conservative political rhetoric and agenda. Net business growth captures the local economic performance of the county. We expect that
higher-growth counties will reflect a growing
potential market for biodiesel. The two local
public finance measures are designed to capture the role of taxes and tax burdens, in particular the property tax, on firm formation.
There are two measures of the transportation infrastructure:
Distance to roads (meters)
Distance to rail lines (meters)
The distance of each county to major roads
and rail lines was calculated in ArcMap (ESRI
2006) by finding the distance from each polygon (each county) to the nearest line (either a
rail line or major road). Due to the scale of
the data, some small railroads and closely parallel tracks are not included in the analysis.
Major roads are defined as interstates, as well
as major highways.'' Higher values of both reflect lower concentration of transportation infrastructure and should be associated with
higher transportation costs.
There are a significant number of public
policy initiatives at both the federal and state
levels to encourage the formation of alternative fuel sources including biodiesel. We ex-

in it being preferred in areas with lower education levels


compared to communities with higher average education.
' See the National Atlas web site at www.nationalatlas.
gov/maplayers.html?openChapters = chptrans#chptrans for
the shape files and metadata on the road and rail data used
in the analysis.

128

Land Economics

pect that federal policies will have limited impact at the local level but promote overall
industry growth. In addition, because federal
policies are the same across all U.S. counties,
there are no spatial variations to capture.
Hence we do not consider such policies. Several states, however, have been very aggressive in promoting alternative fuels and biodiesel in particular. The problem is that local
and state policies may be endogenous relative
to the plant siting decision. This makes introducing explicit policy variables problematic.
To explore the potential impact of state-level
biofuel promotional policies we introduce two
dummy variables to proxy policy. The first is
one if there are local or state-level consumption mandates for biofuels, zero otherwise.
The second is one if there are local or state
production incentives, zero otherwise.
IV. MODELING RESULTS
Results of estimating the Bayesian spatial
logit models described above are presented in
Table 1, while the spatial tobit models are provided in Table 2. For completeness we also
report the aspatial logit and tobit estimated
models. The models are specified in two different ways; (1) with the variables that account for public policies aimed at supporting
(both consumption mandates and producer incentives) biodiesel production, and (2) without the public policy variables. This lets us
test whether currently available policy tools,
including supply and demand-side incentives,
make a general contribution to our understanding of biodiesel plant location decisions.
Before tuming to specific results, there are
some general observations about the overall
results that warrant discussion. First, the spatial p parameter is not statistically significant
in any of the four models estimated using spatial methods.^ Despite the evidence from the
* The weights (W) matrix is constructed based on rowstochastic weighting, which is in essence a simple adjacency
matrix. We have tested alternative specifications of the W
matrix and find that our results are generally insensitive to
the W matrix specification. This is consistent with the work
of Lesage and Pace (2010). They show that model estimates
and inferences are not significantly sensitive to different
specifications of the spatial weights matrix when the marginal effects in the model are properly calculated.

February 2013

global and local Moran's / and the Getis-Ord


Gi* statistic, the level of spatial dependency
does not appear to be sufficient to affect the
reliability of the logit or tobit estimators. It
may be that the spatial dependency and clusters identified above are narrowly focused in
too few regions (e.g., Houston, Texas), or the
access to rail transportation adequately captures spatial dependencies. This supports the
second observation that there is stability in individual parameter results across the spatial
and aspatial estimators. Third, there is stability in results across the logit and tobit models,
lending credence to our results.
In general the results indicate that neither
the number of crush plants nor the number of
restaurants and food service firms in a county
impacts the location of a biodiesel plant. Surprisingly, the number of tank farms also does
not encourage a siting and, in fact, reveals a
negative relationship with plant location.
Thus, unlike ethanol plants, it appears that
biodiesel plants are not sensitive to being located next to the market where their feedstocks are produced, nor next to primary petroleum distribution centers. This may make
sense since the splash-blending chemistry of
biodiesel provides more flexibility than ethanol in terms of where and how biodiesel enters
the fuel stream, thus the reliance on traditional
petroleum distribution centers for blending
may not be important. Further, the export market for unblended biodiesel is more significant
than for ethanol (thus a significant amount of
production does not need to go to a tank farm
or any other U.S. facility for blending).
Population is statistically significant and
positive, implying a tendency for biodiesel
plants to locate in areas with relatively large
populations. This challenges the idea that biodiesel plants can or should be promoted as a
pure rural development strategy in the same
way as ethanol plants. Interestingly, however,
both owner-occupied housing and education
are significant but with negative coefficients.
Thus, greater wealth, as proxied by home ownership, reduces the likelihood of a plant locating in a county. This may be a result of the
NIMBY effect, with location constraints being
greatest around privately owned, single-family
dwellings, as opposed to counties with a higher
percentage of rental properties. The negative

89(1)

129

Fortenbery, Deller, and Amiel: Location of Biodiesel Refineries


TABLE 1
Biodiesel Locational Analysis via Logit
Aspatial Logit

Spatial Logit

Base

Policy

Base

Policy

-3.147603
(0.0221)
0.979318
(0.1473)
-0.333416*
(0.0675)
- 0.004460
(0.7147)
0.006943
(0.6778)
0.010157
(0.9818)
0.000002*
(0.0000)
-0.036966*
(0.0076)
-0.031008*
(0.0181)
0.066326*
(0.0000)
-0.014028*
(0.0536)
-0.014702
(0.3161)

-3.751473
(0.0080)
0.869525
(0.2074)
-0.308620*
(0.0898)
-0.004501
(0.7183)
0.000230
(0.9893)
0.034694
(0.9372)
0.000002*
(0.0000)
-0.043842*
(0.0020)
-0.031214*
(0.0178)
0.071126*
(0.0000)
-0.010128
(0.1772)
- 0.007764
(0.6061)

-1.513060
(0.0118)
0.451941
(0.1167)
-0.160140*
(0.0161)
- 0.002580
(0.3406)
0.002857
(0.3584)
- 0.037700
(0.4896)
0.000001*
(0.0000)
-0.019310*
(0.0031)
-0.016520*
(0.0071)
0.033478*
(0.0000)
- 0.005940*
(0.0469)
- 0.008320
(0.1278)

-1.874203
(0.0049)
0.387683
(0.1584)
-0.144110*
(0.0443)
-0.003291
(0.3047)
- 0.000464
(0.4859)
- 0.039544
(0.4776)
0.000001*
(0.0000)
- 0.022886*
(0.0016)
-0.016000*
(0.0086)
0.036272*
(0.0000)
- 0.004046
(0.1353)
- 0.004094
(0.2882)

Distance to roads
(meters)
Distance to rail lines
(meters)
Policy: mandates present

0.000001
(0.9508)
-0.000116
(0.1090)
0.001791*
(0.0000)
0.000001*
(0.0874)
0.000002*
(0.0004)
0.007032
(0.9937)
-0.000031
(0.2704)
-0.000020
(0.1774)

0.000000
(0.4796)
- 0.000056*
(0.0514)
0.001009
(0.3331)
0.000000*
(0.0569)
0.000001*
(0.0002)
-0.011290
(0.4914)
-0.000012
(0.1506)
- 0.000009
(0.0937)

Policy: incentives present

Spatial parameter p

0.000000
(0.9977)
-0.000139*
(0.0671)
0.003117
(0.5205)
0.000001*
(0.0853)
0.000002*
(0.0073)
-0.018915
(0.9831)
-0.000032
(0.2486)
-0.000021
(0.1706)
0.535533*
(0.0054)
0.126378
(0.5126)

0.000000
(0.5002)
- 0.000069*
(0.0320)
0.001424
(0.2800)
0.000000*
(0.0529)
0.000001*
(0.0022)
-0.033275
(0.4696)
-0.000013
(0.1206)
-0.000010
(0.0686)
0.277018*
(0.0027)
0.037696
(0.3335)
0.057756
(0.2451)

Intercept
Number of crush plants per 10,000 persons
Number of tanker farms per 10,000 persons
Number of restaurants and food service firms per
10,000 persons
Number of trucking and busing firms per 10,(KX)
persons
Number of pipeline transportation of crude oil firms per
10,000 persons
Population 2005
Houses owner-occupied
(percent)
Bachelor's degree or higher
(percent)
Workers who drove to work alone
(percent)
Votes cast for president 2(X)4 Republican
(percent)
Persons 5 years and over residing in same house in
1995 and 2000
(percent)
Net nonfarm business job growth 2000-2004
Local govemment total revenue per capita
Local govemment revenue from the property tax
(percent)
Average farm sales
Acres of crops harvested
Farm share of total county income

0.076654
(0.1647)

Note: Number in parentheses is the marginal signifcanee level.


* Significance at the 10% level.

sign on the college education variable could


indicate several things. One is another angle
on the NIMBY effect, where more highly educated communities (counties) are less willing

to promote or give access to manufacturing


such as biodiesel. Another might be related to
the labor markets biodiesel producers are seeking. A higher level of education is likely related

130

Land Economics

February 2013

TABLE 2
Biodiesel Locational Analysis via Tobit
Aspatial Logit

Intercept
Number of crush plants per 10,000 persons
Number of tanker farms per 10,000 persons
Number of restaurants and food service firms per
10,000 persons
Number of trucking and busitig firms per 10,000
persons
Number of pipeline transportation of crude oil firms per
10,000 persons
Population 2005
Houses owner-occupied (percent)
Bachelor's degree or higher (percent)
Workers who drove to work alone (percent)
Votes cast for president 2004 Republican (percent)
Persons 5 years and over residing in same house in
1995 and 2000 (percent)
Net nonfarm business job growth 2000-2004
Local government total revenue per capita
Local government revenue from the property tax
(percent)
Average farm sales
Acres of crops harvested
Farm share of total county income
Distance to roads (meters)
Distance to rail lines (meters)
Policy: mandates present

Spatial Logit

Base

Policy

Base

Policy

- 2.927779
(0.0320)
0.914394
(0.2121)
-0.312298*
(0.0594)
-0.013009
(0.2578)
- 0.000073
(0.9964)
0.004024
(0.9925)
0.000001*
(0.0000)
-0.042537*
(0.0017)
-0.012753
(0.2944)
0.062900*
(0.0000)
-0.013282*
(0.0532)
-0.012895
(0.3615)
0.000008
(0.1975)
- 0.000089
(0.2184)
0.002042
(0.6547)
0.000001*
(0.0793)
0.000002*
(0.0004)
- 0.200794
(0.8121)
-0.000021
(0.3225)
-0.000019
(0.1541)

-3.421892
(0.0144)
0.818732
(0.2694)
-0.293539*
(0.0760)
-0.013648
(0.2445)
- 0.006006
(0.7184)
0.030724
(0.9422)
0.000001*
(0.0000)
-0.048801*
(0.0004)
-0.012320
(0.3110)
0.067605*
(0.0000)
-0.009812
(0.1622)
- 0.006774
(0.6376)
0.000009
(0.1850)
-0.000113
(0.1302)
0.003058
(0.5135)
0.000001*
(0.0931)
0.000002*
(0.0066)
-0.170611
(0.8409)
- 0.000022
(0.3102)
-0.000019
(0.1463)
0.504963*
(0.0096)
0.074762
(0.6784)

-3.312318
(0.0188)
0.955980
(0.1523)
-0.391880*
(0.0144)
-0.013854
(0.1583)
0.001718
(0.4612)
-0.166159
(0.4146)
0.000001*
(0.0000)
- 0.051795*
(0.0012)
-0.017160
(0.1167)
0.076155*
(0.0000)
-0.015467*
(0.0337)
-0.015019
(0.1916)
0.000011
(0.0857)
-0.000111
(0.1004)
0.002314
(0.3354)
0.000001*
(0.0446)
0.000002*
(0.0001)
-0.281747
(0.3877)
- 0.000030
(0.1157)
- 0.000026*
(0.0504)

-4.011502
(0.0074)
0.843613
(0.1755)
- 0.368772*
(0.0277)
-0.016011
(0.1245)
-0.005913
(0.3822)
-0.14U91
(0.4360)
0.000001*
(0.0000)
-0.059111*
(0.0003)
-0.016051
(0.1345)
0.082327*
(0.0000)
-0.011521
(0.0818)
- 0.007263
(0.3367)
0.000011
(0.0772)
-0.000140*
(0.0566)
0.003615
(0.2572)
0.000001*
(0.0485)
0.000002*
(0.0029)
^0.218296
(0.4110)
- 0.000032
(0.1106)
- 0.000025*
(0.0555)
0.589585*
(0.0045)
0.093404
(0.3351)
0.036229
(0.2140)

Policy: incentives present

Spatial parameter p

0.043299
(0.1754)

Note: Number in parentheses is the marginal significance level.


* Significance at the 10% level.

to a higher average wage scale, placing upward


pressure on the costs of production. From our
theoretical model of location choices, the geographic location (step one in the location process) of a particular county may be ideal, but

the spatial isocost curves appear to rise (higher


costs) with higher levels of education. In addition, the labor demands of biodiesel plants,
including technical jobs, do not require a college degree, but perhaps skills more associated

89(1)

Fortenbery, Deller, and Amiel: Location of Biodiesel Refineries

with technical schools. One way to test this


latter interpretation might be to replace our
measure of labor force education with a measure of the supply of labor with a technical degree.
The percentage of drivers that drive to
work alone is significant and positive, suggesting that it does capture some measure of
demand potential and affects the location decisions of biodiesel plants.^ Political affiliation/attitudes do not seem to be important.
Likewise, the age distribution and nonfarm
job growth do not appear to be important determinates in locating a plant.
Interestingly, the amount of local government revenue per capita does matter, and the
sign of the coefficient is negative. This is consistent with theory that counties with higher
local taxes causes the spatial isocost curve to
move up, making the site less attractive. It does
not seem to matter whether local governments
derive the majority of their funding from property taxes or other sources of funding (such as
state aids or the sales tax). While the location
of oilseed processing facilities was not important, both average farm sales and crop acres
positively impact on location decisions. This
may simply indicate a greater willingness of
rural, farm-based communities to accept
value-added businesses that could potentially
use farm products as inputs, rather than some
location decision that revolves around feedstock procurement. The farm share of total
county income does not seem important.
Distance to roads measured by access to
major highways does not matter, but rail access is important, and the greater the distance
to a rail siding the less likely a plant is to
locate. This makes sense for two reasons.
First, if plants do not locate next to oilseed
processing facilities, they are likely taking delivery of vegetable oil by rail. This is by far
the cheapest way to move oil, and the ability
to utilize unit trains can greatly reduce the

^ It should be noted that most cars in the United States


are not diesel, thus care must be taken in interpreting this
result. But as consumers look for more highly fuel efficient
vehicles, the potential future demand for diesel-powered
cars, and biodiesel in turn, could be significant in the near
future. The popularity of small-displacement, high-efficiency diesel cars in Europe may transfer to U.S. markets.

131

costs of transporting feedstocks (Fortenbery


2004). Second, since biodiesel can be moved
by rail in the same vessels used to haul petroleum products, including diesel fuel, access to
rail services is critical in minimizing distribution costs of the refined product. This result
complements the results on education and
home ownership rates discussed above. More
highly educated people and those that own
their home may be less supporting of freight
rail service in their communities; another
form of the NIMBY effect may be at play.'
The two policy variables, simple dummies
if the state has implemented any incentive
programs or renewable fuel mandates, suggest
that incentive programs do not infiuence the
location decisions of biodiesel plants. The
presence of mandates, however, is positively
associated with the presence and/or concentration of biodiesel plants. This result speaks
to the viability of incentives, generally in the
form of tax incentives or subsidies, to encourage business activity: despite the popularity of such incentives, they seldom have the
desired results.
V. CONCLUSIONS
As we move toward developing the capacity to satisfy the biofuels objectives contained
in RFS2, additional production capacity for
biodiesel will be necessary. In this study we
provide the first look at the important characteristics that lead to an attractive environment for locating a biodiesel plant. As such,
we provide insight to both communities and
investors relative to the attractiveness of specific communities (U.S. counties) where
plants may be located.
In general, we find that there is some spatial dependency associated with plant location. In other words there is some evidence
that biodiesel plants do tend to spatially cluster together in certain geographic locations.
While we account for this spatial clustering in
our formal modeling we cannot draw any inferences concerning whether these spatial
clusters are true "economic clusters" in the

'" This would reflect a desire to be insulated from the


noise and traffic congestion associated with active rail lines.

132

Land Economics

spirit of Porter (1990, 1995, 2000). This is the


next step in our line of inquiry.
We do find evidence that biodiesel plants
are not attracted to communities with higher
education levels and home ownership rates,
but are attracted to ones that have access to
rail service and rely on agricultural production
as a significant enterprise. At the same time,
however, biodiesel plants tend to be drawn to
more populous counties. Neither the processing of feedstock (i.e., presence of a crush
plant) nor the presence of tank farms appears
to be particularly important in a plant's siting
decision. The results taken in tandem seem to
suggest that biodiesel plants are attracted to
larger output markets (consumers) rather than
input supplies, but output markets that are less
wealthy. We think that the very last result on
wealth may reflect NIMBY effects.
The results also provide some insight relative to policy initiatives a community may
want to pursue in terms of attracting a plant.
While endogeneity issues suggest some caution in interpretation, it does appear that consumption mandates are more likely to impact
positively on plant location. Further, investment in feedstock processing is less important
than development of transportation infrastructure (rail). Providing public support to these
sectors appears to be more important to a
plant's location decision than direct public investment in biodiesel production. These findings are in contrast to the often-proposed local
policy actions and suggest that communities
need to more carefully consider what type
policy initiatives will have the greatest impact
on a plant's location decision.

Febmary 2013

Consider first the Anselin local Moran's / (Eigure


A l ) , " then the Getis-Ord Gi* (Eigure A2).'2 There
appear to be several spatial clusters based on the Local Moran's /, including parts of northern Illinois and
southem Wisconsin, parts of Iowa along with Houston, Texas, and several locations in Califomia as well
as the Seattle, Washington, region. The dominant
clusters appear to be in Houston, the southern half of
California, and the Seattle region. There appear to be
smaller clusters in northem Colorado and southem
Wyoming, as well as the Atlantic Coast from New
York City to southern New Hampshire. In addition,
the other areas with biodiesel plants, such as much of
the Appalachia region, parts of Missouri, Kansas, and
Oklahoma, along with the upper parts of Minnesota
and North Dakota, appear to be spatially isolated.
The results of the Getis-Ord Gi* (Eigure A2) generally complement and reaffirm the results of the Anselin local Moran's /. Here the spatial concentration
is most evident in the Gulf region of Texas, which
includes Houston, much of Califomia, particularly the
region from San Erancisco to San Diego, and the Seattle region. There is also a weaker cluster from New
York City north through Vermont. Notice, however,
that the northern Illinois, southem Wisconsin, and
Iowa cluster identified by the local Moran's / is not
confirmed by the Getis-Ord Gi* statistic. Also note
that there is a statistically significant lack of biodiesel
activity in the westem parts of the Great Plains along
with parts of the northem Rocky Mountain range
through Montana.
If we examine the simple plotting of plant locations
(Eigure 2) coupled with the two spatial concentration
or clustering statistics (Anselin local Moran's /, Eigure Al, and the Getis-Ord Gi*, Eigure A2) it is clear
that biodiesel plants appear to be much more footloose in their locations than ethanol plants. In addition, there appear to be perhaps five spatial clusters:
Houston, Los Angles, San Erancisco, Seattle, New
York City, and southem New England, and perhaps
the Chicago area.
' ' The Anselin local Moran's / is computed as

APPENDIX: TEST RESULTS FOR


SPATIAL CLUSTERING OF BIODIESEL
PLANTS
We employ two tests to see if the characteristics of
a particular observation (i.e., number of biodiesel
plants in a given county) are similar or dissimilar to
neighboring observations. The null hypothesis is that
there is no association between the value observed at
a location and values observed at nearby sites. The
altemative is that nearby sites have either similar
(large positive values of the statistic) or dissimilar values (large negative values).

n
X X

"

/,=-4 2

^'-^

yfij{Xi-JQ, where s} =-'-^^^

x"^,

SI J=ljti
n- 1
X, is the number of biodiesel plants within county /, and w,j
consists of spatial weight matrix elements identifying adjacent counties.
2 The Getis-Ord Gi* is computed as G* =

andXand

w are the same as above.

Fortenbery, Deller, and Amiel: Location of Biodiesel Refineries

89(1)

Anselin Local Moran's I


Z Score
^|<-1.96
I
I -1.95-1.96

>1.96
FIGURE Al
Spatial Clustering of Biodiesel Plants via Anselin Local Moran's /

FIGURE A2
Spatial Clustering of Biodiesel Plants via Getis-Ord Gi*

133

134

Land Economics

Acknowledgments
Support for this work was provided in part by the
Wisconsin Agricultural Experiment Station under the
Hatch Act.

References
Anselin, Luc. 1988. Spatial Econometrics: Methods
and Models. Dordrecht, The Netherlands: Kluwer
Academic Publishers.
. 1995. "Local Indicators of Spatial AssociationLISA." Geographical Analysis 27 (2): 9 3 115.
Bell, Kathleen P, and Elena G. Irwin. 2002. "Spatially Explicit Micro-Level Modeling of Land-Use
Change at the Rural-Urban Interface." Agricultural Economics 27 (3): 217-32.
Beron, Kurt J., James C. Murdoch, and Wim P. M.
Vijverberg. 2003. "Why Cooperate? Public
Goods, Economic Power, and the Montreal Protocol." Review of Economics and Statistics 85 (2):
286-97.
Beron, Kurt J., and Wim P. M. Vijverberg. 2004.
"Probit in a Spatial Context: A Monte Carlo Analysis." In Advances in Spatial Econometrics: Methodology, Tools and Applications, ed. Luc Anselin,
Raymond J. G. M. Elorax, and Serge J. Rey. Berlin: Springer-Verlag.
Bockstael, Nancy E. 1996. "Modeling Economics and
Ecology: The Importance of a Spatial Perspective." American Journal of Agricultural Economics 1?, {5): 1168-80.
Carlton, Dennis W. 1979."Why New Eirms Locate
Where They Do: An Econometric Model." In Interregional Movements and Regional Growth, ed.
William Wheaton, 13-50. Washington, DC: Urban
Itistitute.
. 1983. "The Location and Employment
Choices of New Eirms: An Econometric Model
with Discrete and Continuous Endogenous Variables." Review of Economics and Statistics 65 (3):
440-49.
Durbin, Dick. 2007. U.S. Senator Dick Durbin, comments. http://durbin.senate.gov/showRelease.cfm
?releaseld = 276868, (accessed March 8, 2010,
source no longer available).
ESRI. 2006. ArcGIS: Release 9.2. Redlands, CA: Environmental Systems Research Institute.
Elora, Cornelia B. 2008. "The Impact of Ethanol
Plants on Rural Communities." Paper presented at
the 2008 Bio-Economy Conference, Ames, Iowa,
September 9-10. Available at www.bioeconomy
conference.org/08%20Presentations%20
approved/Breakouts/Human,%20Social%20and%
20Community%20Impacts/Elora,%20CorneIia.
pdf (accessed March 8, 2010).

February 2013

Fortenbery, T. Randall. 2004. "Biodiesel Feasibility


Study: An Evaluation of Biodiesel Eeasibility in
Wisconsin." Staff Paper No. 481, Department of
Agricultural and Applied Economics, University
of Wisconsin-Madison.
Eortenbery, T. Randall, and Steven C. Deller. 2008.
"Understanding Community Impacts: A Tool for
Evaluating Economic Impacts from Bio-Euels
Production." Journal of Extension 46 (6): article
6EEA2. Available at www.joe.org/joe/2008december/pdf/JOE_v46_6fea2.pdf
Eortenbery, T. Randall, Allen Tumquist, and Jeremy
Eoltz. 2008. "Progress or Devastation: The Effects
of Ethanol Plant Location on Local Land Use."
Paper presented at the 2008 Bio-Economy
Conference, Ames, Iowa, September 9-10. Available at www.bioeconomyconference.org/08%20
Presentations%20approved/Breakouts/Human,%
20Social%20and%20Community%20Impacts/
Eortenbery,%20Randy.pdf (accessed March 8,
2010).
Eranzese, Robert J., and Jude C. Hays. 2009. "A
Comparison of Small-Sample Properties of Several Estimators for Spatial-Lag Models." Unpublished working paper. University of Illinois,
Champaign-Urbana.
Gabszewicz, Jean Jaskold, and Jaques-Eranois
Thisse. 1986. "Spatial Competition and the Location of Eirms." In Location Theory., ed. Jean
Jaskold Gabszewicz, Jaques Eranois Thisse, Masahisa Eujita, and Urs Schweizer. New York: Harwood Academic Publishers.
Geweke, John. 1991. "Efficient Simulation from the
Multivariate Normal and Student-i Distributions
Subject to Linear Constraints." In Computer Science and Statistics: Proceedings of the TwentyThird Symposium on the Interface, 571-78. Alexandria, VA: American Statistical Association.
Getis, Arthur, and J. Keith Ord. 1992. "The Analysis
of Spatial Association by Distance Statistics."
Geographical Analysis 24 (3): 189-206.
Guimaraes, Paulo, Octavio Eigueiredo, and Douglas
Woodward. 2004. "Industrial Location Modeling:
Extending the Random Utility Framework." Journal of Regional Science 44 (1): 1-20.
Haddad, Monica A., Gary Taylor, and Franicis
Owusu. 2010. "Location Choices of the Ethanol
Industry in the Midwest Com Belt." Economic
Development Quarterly 24 (1): 74-86.
Hicks, Robert L., and Kurt E. Schnier. 2010. "Spatial
Regulation and Endogenous Consideration Sets in
Eisheries." Resource and Energy Economics 32
(2): 117-34.
Holloway, Garth, Bhavani Shankar, and Sanzidur
Rahmanb. 2002. "Bayesian Spatial Probit Estimation: A Primer and an Application to HYV Rice

89(1)

Fortenbery, Deller, andAmiel: Location of Biodiesel Refineries

Adoption." Agricultural Economics 27 (3): 383402.


Hunt, Len M., Barry Boots, and Pavios S. Kanaroglou. 2004. "Spatial Choice Modelling: New Opportunities to Incorporate Space into Substitution
Patterns." Progress in Human Geography 28 (6):
746-66.
Irwin, Elena G., and Nancy E. Bockstael. 2002. "Interacting Agents, Spatial Extemalities and the
Evolution of Residential Land-Use Pattems."
Journal of Economic Geography 2 (1): 31-54.
Kelijian, Harry H., and Ingmar R. Prucha. 1998. "A
Generalized Spatial Two-Stage Least Squares Procedure for Estimating a Spatial Autoregressive
Model with Autoregressive Disturbances." Journal of Real Estate Finance and Economics 17(1):
99-121.
Klier, Thomas, and Daniel P McMillen. 2005. "Clustering of Auto Supplier Plants in the U.S.: GMM
Spatial Logit for Large Samples." Working Paper
2005-18. Chicago: Federal Reserve Bank.
Lambert, Dayton, Michaek M. Wilcox, A. English,
and Lance A. Stewart. 2008. "Ethanol Plant Location Determinants and County Comparative Advantage." Journal of Agricultural and Applied
Economics 40 (1): 117-35.
LeSage, James P. 1999. Applied Econometrics Using
Matlab. Toledo, OH: Department of Economics,
University of Toledo.
. 2000. "Bayesian Estimation of Limited Dependent Variable Spatial Autoregressive Models."
Geographical Analysis 32 (1): 19-35.
LeSage, James P, and R. Kelley Pace. 2009. Introduction to Spatial Econometrics. Boca Raton, FL:
Taylor and Francis.
. 2010. "The Biggest Myth in Spatial Econometrics." Social Science Research Network. Available at http://ssrn.com/abstract = 1725503 (accessed December 22, 2011).
Low, Sarah A., and Andrew M. Isserman. 2009."Ethanol and the Local Economy: Industry Trends, Location Factors, Economic Impacts and Risks."
Economic Development Quarterly 23 (1): 71-88.
McCann, Philip. 2002. "Classical and Neo-Classical
Location-Production Models." In Industrial Location Economics, ed. Philip McCann. North
Hampton, MA: Edward Elgar.
McEadden, Daniel. 1974. "Conditional Logit Analysis of Qualitative Choice Behavior." In Frontiers
in Econometrics, ed. Paul Zarembka, 105^2.
New York: Academic Press.
McMillen, Daniel P. 1989. "An Empirical Model of
Urban Fringe Land-Use." Lafid Economics 65 (2):
138^5.
McNew, Kevin, and Duanne Griffith. 2005. "Measuring the Impact of Ethanol Plants on Local Grain

135

Markets." Review of Agricultural Economics 27


(2): 164-80.
National Biodiesel Board. 2010. Biodiesel: America's
Advanced Biofuel. Available at www.biodiesel.org
(accessed March 8, 2010).
Nobbe, Gary. 1995. "Canola Eill Up." Omni 17 (6).
Ord, J. Keith, and Arthur Getis. 1995. "Local Spatial
Autocorrelation Statistics: Distributional Issues
and an Application." Geographical Analysis 27
(4): 286-306.
Pinkse, Joris, and Margaret E. Slade. 1998. "Contracting in Space: An Application of Spatial Statistics
to Discrete-Choice Models." Journal of Econometrics 85 (1): 125-54.
Porter, Michael E. 1990. The Competitive Advantage
of Nations. New York: Free Press.
. 1995. "The Competitive Advantage of the
Inner City." Harvard Business Review (May/
June): 55-71.
-. 2000. "Location, Competition and Economic
Development." Economic Development Quarterly
14(1): 23-34.
Renewable Fuels Association. 2010. Industry Resources. Available at www.ethanolrfa.org/pages/
industry-resources (accessed March 8, 2010).
Robertson, Richard D., Gerald C. Nelson, and Alessandro De Pinto. 2009. "Investigating the Predictive Capabilities of Discrete Choice Models in the
Presence of Spatial Effects." Papers in Regional
Science 88 (2): 367-88.
Sarmiento, Camilo, and William W. Wilson. 2007.
"Spatial Competition and Ethanol Plant Location
Decisions." Department of Agribusiness and Applied Economics, Report No. 604 (May). Fargo:
North Dakota State University.
Schnier, Kurt E., and Ronald G. Felthoven. 2011.
"Accounting for Spatial Heterogeneity and Autocorrelation in Spatial Discrete Choice Models: Implications for Behavioral Predictions." Land Economics. 87 (3): 382-401.
Shaffer, Ron, Steven C. Dlier, and David Marcouiller. 2004. Community Economics: Linking
Theory and Practice. Oxford: Blackwell Professional Publishing.
Smith, Martin D. 2010. "Toward an Econometric
Foundation for Marine Ecosystem-Based Management." Bulletin of Marine Science 86 (2): 461-77.
Swenson, David. 2008. "Ethanol's Impact on Rural
Communities." AgMRC Renewable Energy Newsletter. Available at www.agmrc.org/renewable_
energy/renewable_energy/ethanols_economic_
impact_on_rural_communities.cfm
(accessed
March 8, 2010).
Train, Kenneth. 2003. Discrete Choice Methods with
Simulation. New York: Cambridge University
Press.

136

Land Economics

U.S. Census Bureau. County and City Data Book:


2005. 12th ed. Washington, DC: U.S. Department
of Commerce, Economics and Statistics Administration.
U.S. Department of Agriculture. 2004. 2002 Census
of Agriculture, United States Summary and State
Data, Vol. 1. Geographic Area Series Part 51, AC02-A-51. Washington, DC: U.S. Department of
Agriculture. Available at www.agcensus.usda.gov/
Publications/2002/index.php.
U.S. Energy Information Administration. 2007. Biofuets in the U.S. Transportation Sector. Available
at www.eia.doe.gov/oiaf/analysispaper/biomass.
html (accessed March 8, 2010).
U.S. Environmental Protection Agency. 2010. EPA
Finalizes Regulations for the National Renewable
Fuel Standard Program for 2010 and Beyond.

February 2013

EPA-420-F-10-007. Washington, DC: U.S. Environmental Protection Agency.


Wang, Xiaokun, and Kara M. Kockelman. 2009. "Application of the Dynamic Spatial Ordered Probit
Model: Patterns of Land Development Change in
Austin, Texas." Papers in Regional Science 88 (2):
345-65.
Weber, J. Alan. 2009. Feedstock Supplies for U.S.
Biodiesel Production. Columbia, MO: MARC-IV
Consulting, Inc.
Williamson, Dave. 2002. "The Ecology Center's
Trucks No Longer Run on Petrol." 17 (2): 32-33.
Wu, Corinna. 1998. "Fill 'Er Up . . . with Veggie Oil:
Vegetable Oils Are Moving from the Kitchen Table to the Car Engine." Science News 154 (23):
364-66.

Copyright of Land Economics is the property of University of Wisconsin Press and its content may not be
copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written
permission. However, users may print, download, or email articles for individual use.

Vous aimerez peut-être aussi