Content uploaded by Raja Chakir
Author content
All content in this area was uploaded by Raja Chakir on Jun 03, 2023
Content may be subject to copyright.
Spatial Autocorrelation in Econometric
Land Use Models: An Overview
Raja Chakir and Julie Le Gallo
Abstract This chapter provides an overview of the literature on econometric land
use models including spatial autocorrelation. These models are useful to analyze the
determinants of land use changes and to study their implications for the environment
(carbon stocks, water quality, biodiversity, ecosystem services). Recent methodolog-
ical advances in spatial econometrics have improved the quality of econometric mod-
els allowing them to identify more precisely the determinants of land use changes and
make more accurate land use predictions. We review the current state of the literature
on studies which account explicitly for spatial autocorrelation in econometric land
use models or in the environmental impacts of land use.
1 Introduction
Land use plays a vital role in many major societal issues: food security (Verburg
et al. 2013), preservation of biodiversity and ecosystem services (Foley 2005), climate
change mitigation (Lal 2004) and the achievement of many Sustainable Development
Goals (Gao and Bryan 2017). Land use choices are the result of complex decision-
making processes related to the local and global biophysical and socioeconomic
drivers. The researcher faces two central and related questions: “what drives land
use change?” and “what are the (environmental and socioeconomic) impacts of land
use change on stakeholders and the whole society?”. The answers to these questions
are crucial for the design of public policies related to how to feed the growing world
population and avoid unwanted land use effects on the environment.
R. Chakir (B
)
Université Paris-Saclay, INRAE, AgroParisTech, Economie Publique, 78850 Thiverval-Grignon,
France
e-mail: raja.chakir@inrae.fr
J. Le Gallo
CESAER UMR1041, Agrosup Dijon, INRAE, Université de Bourgogne Franche-Comté, 21000
Dijon, France
e-mail: julie.le-gallo@agrosupdijon.fr
© Springer Nature Switzerland AG 2021
A. Daouia and A. Ruiz-Gazen (eds.), Advances in Contemporary Statistics
and Econometrics,https://doi.org/10.1007/978-3- 030-73249- 3_18
339
340 R. Chakir and J. Le Gallo
Various disciplines (economics, statistics, geography, land use science) have
developed a range of empirical land use modeling approaches, using either aggre-
gate or individual data. However, most of this work pays little attention to spatial
autocorrelation (SA) in modeling land use although spatial interdependence is preva-
lent in all economic decisions in general and in land use decisions in particular. As
a result, “standard” statistical and econometric methods, which assume indepen-
dent observations, are inappropriate. More generally, taking account of the spatial
dimension in econometric models involves two effects: spatial heterogeneity and
SA. Spatial heterogeneity is the spatial differentiation of variables and behaviors in
space and usually does not require specific econometric methods. Switching models,
semi-parametric modeling of coordinates or clustered robust inference can handle
this effect appropriately. Conversely, SA refers to the lack of independence among
geographic observations. It measures the degree of similarity between an attribute in
one location and the same attribute in neighboring locations (Anselin 1988). Unlike
temporal autocorrelation, SA is multidimensional requiring a specialized set of tech-
niques, which are not simple extensions of two-dimensional time series methods. In
this chapter, we focus on SA in econometric land use models.
There is a growing body of work on econometric modeling of land use. These
studies address the determinants of land use and land use change and their impacts
on water quality (Bockstael 1996), deforestation (Chomitz and Gray 1996), car-
bon sequestration costs (Lubowski et al. 2006), and habitat fragmentation (Lewis
and Plantinga 2007). Before the 1990s, econometric land use studies that explicitly
introduced SA of observations were relatively rare as the presence of SA makes
discrete choice models analytically intractable and requires use of computationally
expensive Bayesian techniques or simulation estimation methods (Fleming 2004).
Thus, most land use studies and especially those based on individual data avoid
thorough treatment of spatial effects or use ad hoc procedures aimed at reducing the
negative consequences of ignoring them.1
Although land use studies taking explicit account of SA have increased (Brady
and Irwin 2011), they remain relatively scarce (Ay et al. 2017; Chakir and Le Gallo
2013;Lietal.2013; Sidharthan and Bhat 2012; Ferdous and Bhat 2012; Chakir and
Parent 2009). Most econometric land use models in papers published in high quality
journals still tend either to ignore SA, or use ad hoc methods to deal with it (e.g., Irwin
et al. 2003; Carrion-Flores and Irwin 2004; Lubowski et al. 2008; Fezzi and Bateman
2011). This is because SA raises several issues related to econometric estimation,
hypothesis testing, and prediction—especially in the case of discrete choice models
(Billé and Arbia 2019).
Then, the aim of this chapter is to present the state of the art in the literature
on econometric land use models and to show how methodological developments in
spatial econometrics have been introduced into these models. We point out that this
is not an exhaustive review; rather the objective is to highlight the main contributions
to econometric land use models and their methodological advances. Our literature
reviews depart from those provided by Brady and Irwin (2011), which summarize
1Ignoring spatial effects can result in biased and/or inefficient parameter estimates or assessment
of statistical significance (Anselin 1988).
Spatial Autocorrelation in Econometric Land Use Models: An Overview 341
the econometric challenges of spatial models in land use and hedonic model context,
Plantinga (2015), who focuses on methods for integrating economic land use and
biophysical models and Chakir (2015), who reviews methodological developments
in spatial econometrics that have been introduced into land use models. The main
goal of this literature review is to summarize the studies which include SA explicitly
in land use models or in models of the environmental impacts of land use.2
The remainder of the chapter is organized as follows. First, we provide some
general considerations related to the econometric modeling of land use (Sect. 2).
Then we focus, respectively, on linear (Sect. 3) and discrete choice econometric land
use models (Sect. 4) models. Section 5shows how SA enhances models that focus
on the impact of land use on various environmental outcomes. Section6concludes
and highlights some directions for further research.
2 Econometric Land Use Models
Most econometric land use studies are based on the classical theory which consid-
ers that land use activities are chosen to maximize land rents and that rents vary
with land characteristics, in particular soil fertility (Ricardo 1817) and location (von
Thunen 1875). Yet, other factors might influence land use decisions for a given land
parcel: socioeconomic factors (input and output prices) and policy variables (taxes
and subsidies). The extent and significance of these determinants are analyzed in
two broad categories of models: aggregate land use models which use aggregate
(county level, state level, etc.) data, and individual land use models which are based
on parcel-level or sample plot data. Table1presents a summary of some papers that
provide econometric modeling of land use employing both aggregate and individual
data.
Aggregate and individual land use models are complementary and provide dif-
ferent insights into the determinants of land use and land use changes, and their
environmental effects. The choice between an aggregate and individual land use
model often depends on data availability and the objective of the study. If the objec-
tive is to make land use predictions at the scale of one or a group of countries (such as
European countries), an aggregate data model is required. If the objective is to study
the effects of land use on biodiversity or water quality, a model based on individual
data is more relevant. Both approaches have drawbacks.
On the one hand, aggregate data limits the capacity to explain the effects of het-
erogeneous physical characteristics such as soil quality on land use choices. Because
the data are aggregated to units such as the county, intra-county variations in soil
quality are ignored. Moreover, while aggregate data can be useful to study global
issues (changes in land use shares within a region), the results are of limited use for
policy making related to the spatial organization of land use in a region, or local
issues related to biodiversity, water quality, or urbanization.
2We did a literature search for articles adopting an explicit spatial econometric approach to land
use issues. Then, among these articles, we chose those that we considered the most important either
from a methodological point of view or from the point of view of the environmental impacts of land
uses.
342 R. Chakir and J. Le Gallo
Tabl e 1 Example of econometric individual and aggregate, spatial and aspatial land use studies.
(AER: American Economic Review, AJAE: American Journal of Agricultural Economics, ARER:
Agriculture and Resource Econ. Review, EE: Ecological economics, GA: Geographical Analysis,
FS: Forest Science, JARE: J. of Agri and Resource Econ. JEEM: Journal of Environmental Eco-
nomics and Management, JGS: Journal of Geographical Systems, JRS: Journal of Regional Science,
LE: Land Economics, LUP: Land Use Policy, PIRS: Papers in regional Science, RSUE: Regional
Science and Urban Economics, SEA: Spatial Economic Analysis, WBER: World Bank Economic
Review)
Paper Land use
categories
Model Spatial Journal
Aggregate land use share studies
Alig (1986)Crops, 3 types de
forest,
pasture and urban
Land-use share No FS
Lichtenberg
(1989)
7 crops Land-use share No AJAE
Stavins and Jaffe
(1990)
Crops, forest Land-use share No AER
Wu and Segerson
(1995)
6cultures Land-use share No AJAE
Plantinga (1996)Agriculture to
forest
Land-use share No AJAE
Plantinga et al.
(1999)
Agriculture,
forest and
urban/other land
use
Land-use share No AJAE
Hardie and Parks
(1997)
Agriculture,
forest,
urban/other use
Land-use share No AJAE
Plantinga and
Ahn (2002)
Crops, forest Land-use share No JAR E
Chakir and
Le Gallo (2013)
Agriculture,
forest,
urban and other
use
Land-use share Yes EE
Marcos-Martinez
et al. (2017)
19 land use
categories
Land-use share Yes LUP
Chakir and
Lungarska (2017)
Agriculture,
forest,
urban and other
use
Land-use share Yes SEA
Marcos-Martinez
et al. (2017)
Extensive
grazing, pastures,
cereals, annuals,
perennials
Land-use share Yes LUP
Amin et al.
(2019)
Deforestation Deforestation
area
Yes JEEM
(continued)
Spatial Autocorrelation in Econometric Land Use Models: An Overview 343
Tabl e 1 (continued)
Paper Land use
categories
Model Spatial Journal
Individual discrete choice studies
McMillen (1989)Farm, residential Multinomial logit No LE
Bockstael (1996) Urbanization Probit No AJAE
Chomitz and
Gray (1996)
Deforestation Multinomial logit No WBER
Claassen and
Tegene (1999)
Culture, pasture Probit No ARER
Carrion-Flores
and Irwin (2004)
Urbanization Probit No AJAE
Lubowski et al.
(2006)
Crops, pasture,
forest,
urban, range and
CRP
Nested logit No JEEM
Chakir and Parent
(2009)
Agriculture,
forest, urban
and other uses
Multinomial
probit
Yes PIRS
Wan g a n d
Kockelman
(2009)
4levelsof
urbanization
Ordered probit Yes PIRS
Ferdous and Bhat
(2012)
4levelsof
urbanization
Ordered probit Yes JGS
Sidharthan and
Bhat (2012)
Urban,
commercial,
industrial
and
non-developed
Multinomial
probit
Yes GA
Li et al. (2013)Farm, forest,
grass,
water, urban,
unused
Multinomial
probit
Yes LE
Bhat et al. (2015)Commercial,
industrial,
residential,
underdeveloped
Multiple discrete-
continuous
probit
Yes JRS
Carrión-Flores
et al. (2018)
Commercial,
industrial,
residential,
parks, agriculture
Multinomial logit Yes RSUE
On the other hand, one of the frequent difficulties related to modeling land use
at the individual level is the lack of “good” explanatory variables or their scale
incompatibilities. Although geophysical explanatory variables such as slope, altitude
and soil quality are increasingly available at very fine resolution, economic variables
344 R. Chakir and J. Le Gallo
(rents, conversion costs, and prices) are either not available or observable only at
aggregate scales. To compensate for this lack of data, empirical models often use
proxies for rents at more or less aggregated scales. Another difficulty of individual-
level land use models is related to the complexity involved in estimating discrete
choice models in the multinomial case. This difficulty is accentuated if SA is included
in the specification.
In relation to this latter issue, SA in land use choices tends not to be included in
theoretical frameworks but added ex post in the empirical specification. In land use
modeling, SA can stem from two sources. First, it can arise from spillovers among
the error terms due to omitted spatial variables affecting land use decisions such
as weather or soil quality. A spatial error model or spatial robust inference allows
to control for these omitted variables provided that they are not correlated with the
observables. Second, it can arise from spillovers among land use decisions or spatial
interaction relationships in the land use choices. This might be due, for example, to
the neighboring plots being owned by the same landowner, or to shared information
which induces forest or agricultural clustering and landowners adopting the same
technology based on shared learning. In this case, a spatial autoregressive model
would account for these spatial interactions.
In the case of aggregate data, logarithmic transformation on land use shares implies
linear equations that can easily be estimated. Therefore, SA in the case of land use
models can be estimated using spatial models in the linear case (Sect. 3). Conversely,
in most cases of individual data (Sect. 4), the presence of SA tends to make discrete
choice models analytically intractable and requires use of simulation estimation
methods or Bayesian techniques (Smith and LeSage 2004). Other estimation pro-
cedures have been proposed in the literature: the expectation-maximization method
(McMillen 1992), the generalized method of moments (GMM) (Pinkse and Slade
1998), and the composite maximum likelihood method (Sidharthan and Bhat 2012;
Ferdous and Bhat 2012). For a detailed review of the inclusion of SA in discrete
choice models see Fleming (2004), Smirnov (2010), Billé and Arbia (2019).
Considering SA also sheds new light on the issue of prediction. Comparing indi-
vidual and aggregate models with respect to their predictive accuracy is an ongo-
ing and still open issue with mixed evidence. The seminal paper by Grunfeld and
Griliches (1960) examined the relative power of individual (micro) and aggregate
(macro) models for explaining aggregate outcomes and found that an aggregate model
often performs better. In the context of land use models, Wu and Adams (2002)show
that even in the case of linear models, the choice between the micro- and macro-scales
to make aggregate predictions cannot generally be resolved by a priori reasoning. Ay
et al. (2017) show that introducing SA in aggregate land use models provides better
predictions than using individual aspatial models with higher numbers of observa-
tions. This suggests that there might be little to be gained from using individual land
use data if the sole objective is to predict land use at the aggregate spatial resolution.
Some studies choose none of these modeling approaches and resort instead to ad
hoc methods to circumvent the problems related to estimating discrete choice models
in the presence of SA (De Pinto and Nelson 2007). These models are summarized
below:
Spatial Autocorrelation in Econometric Land Use Models: An Overview 345
•Spatial sampling: Most early studies in the land use literature simply purge the
data of SA using a spatial sampling technique which allows construction of a
data sample without neighbors. This is a fairly widespread practice: Nelson and
Hellerstein (1997), Carrion-Flores and Irwin (2004), Irwin et al. (2003), Irwin and
Bockstael (2004), Lewis and Plantinga (2007), Lubowski et al. (2008), De Pinto
and Nelson (2009), Fezzi et al. (2015)
•Introduction of latitude and longitude as explanatory variables:Nelsonetal.
(2001), Muller and Zeller (2002) claim to account for SA by using two additional
explanatory variables representing the latitude and longitude of each observation.
While this type of correction is likely to be useful if the spatial effect is caused by
an unobserved variable which varies linearly between regions, it captures spatial
heterogeneity rather than capturing the SA as claimed by the authors;
•The introduction of spatially shifted geophysical variables:Nelsonetal.(2001),
Munroe et al. (2002) use spatial shifts (i.e., weighted averages of values in neigh-
boring locations) of geophysical variables such as soil type, slope, and vegetation
index as exogenous variables. A possible justification for using these types of vari-
ables is that they account for the direct influence of the environment on land use
decisions in a particular location.
While useful, these methods cannot control for substantive SA, an issue to which
we turn in the next two sections.
3 Linear Land Use Models
The objective of most studies using aggregate data is to identify the determinants of
land use shares. Most U.S. econometric studies use the county scale and land use
data derived generally from federal sources such as agricultural census. The most
common method is to specify county land use shares as a logistic function. Examples
of studies that use this method include Lichtenberg (1989), Plantinga (1996), Hardie
and Parks (1997). While the authors attempt to explain the factors that influence the
share of land allocated to a particular land use, other aggregate data studies try to
explain changes in land use shares in an area (Stavins and Jaffe 1990; Plantinga and
Ahn 2002). All these studies ignore SA. More recent studies taking explicit account
of SA have been conducted at the French level by Chakir and Le Gallo (2013), Ay
et al. (2017), Chakir and Lungarska (2017) who estimate aggregate land use share
models at the department level, 12 ×12 km and 8 ×8 km grid cells, respectively.
3.1 Land Use Share Models
Although all econometric studies are based on the same economic theory, several
variants of theoretical land allocation models have been proposed (Lichtenberg 1989;
Stavins and Jaffe 1990; Plantinga 1996; Hardie and Parks 1997). We present here a
346 R. Chakir and J. Le Gallo
fairly simple version of these models based on Wu and Segerson (1995)’s static model
where the landowner ni(ni=1,...N) in the region i(i=1,...,I)isassumedtobe
risk neutral and maximizes his expected profit from the use k(k=1,...K) on quality
land j(j=1,...,J), at time t(t=1,...T), denoted πjk(x(t,ni), ajk(t,ni), ni),
where x(t,ni)is a vector of the exogenous variables such as prices, costs and other
economic variables and ajk(t,ni)is the area of land of quality jallocated to use k.For
each quality of land, the landowner chooses the area ajk(t,ni)≥0 that maximizes
his total profit:
K
k=0
πjk(x(t,ni), ajk(t,ni), ni)subject to
K
k=0
ajk(t,ni)=Aj(t,ni)(1)
where Aj(t,ni)is the total surface of available quality land j. The resolution of the
optimization program (1) gives the optimal area a∗
jk(x(t,ni), Aj(t,ni), ni)allocated
to each use kfor each quality of the land jat time t. The optimal share of land allocated
to the use kis
sk(x(t,ni), t,ni)=1
Aj(t,ni)
j
a∗
jk(t,ni)(2)
The optimal uses derived from the theoretical model for each owner should be
aggregated to match the scale of the observed data. In practice, the available data are
the shares of land uses at an aggregate resolution (county, region, municipality). The
land use share k(k=1,...,K) in the region iat time tis written as
sikt =pikt +εikt =eβ
kXit
K
j=1eβ
jXit
+εikt ∀i=1,...,I,∀k=1,...,Kand ∀t=1,...,T
(3)
where pikt is the expected share of land allocated for use kin the iregion at time
t. The observed land use share at time t,sikt may differ from the optimal land use
share due to possible hazards such as climate or policy shocks. These elements, of
zero average, are captured by the error term εikt.Xit are the explanatory variables
and β
kare the associated coefficients.
As in Wu and Segerson (1995), Plantinga et al. (1999), most aggregate stud-
ies specify land use shares in the logistic functional form for three reasons: first,
this functional form allows predicted land use shares to stay between 0 and 1, sec-
ond, this specification is parsimonious in terms of parameters, and third, logarithmic
transformation allows use of linear equations which are easily estimated. This trans-
formation3has been proposed by Zellner and Lee (1965) and, applied to land use
choices, it allows to write the logarithm of each use share normalized by a given
share as follows:
3This transformation corresponds to the additive log ratio (ALR) transformation in the literature on
composition data in statistics, see Aitchison (1986) for more details.
Spatial Autocorrelation in Econometric Land Use Models: An Overview 347
yikt=ln(sikt/sikt
)=β
kXit +uikt ∀i=1,...,I,∀k=1,...,Ket ∀t=1,...,T
(4)
where uikt is the transformed error term. Model (4) has K−1 equations that are
seemingly unrelated regressions (SUR) and can be estimated by methods accounting
for correlations between the error terms associated with each equation.
3.2 Spatial Autocorrelation in Linear Models
In linear specification models, SA is handled by the inclusion of spatially lagged
variables, that is, weighted averages of the observations of “neighbors” of a given
location. These spatially lagged variables can be used as the dependent variable (spa-
tial autoregressive SAR models), explanatory variables (spatial cross regressive SLX
models), or the error terms (SEM) or any combination of these options which results
in a range of spatial models (Elhorst 2010). For instance, the spatial autoregressive
combined (SARAR) model accounts simultaneously for autocorrelation in the error
term and for spatial associations of the dependent variable. The spatial Durbin model
(SDM) is a combination of SAR and SLX and can be reduced to SEM (LeSage and
Pace 2009), while the spatial Durbin error model (SDEM) integrates all the elements
of the SLX and the SEM. Finally, the general nesting spatial (GNS) model combines
the SARAR and the SLX models (see Table 2). Until the early 2000s, most empirical
spatial econometric studies were interested mainly in two specifications: SAR and
SEM. Specifications accounting for richer and combined forms of SA are now more
commonly estimated. For more details on the taxonomy of linear SA models for
cross-sectional data see Elhorst (2014).
The choice of the best spatial specification can be made based on theory or by
applying statistical tests to different models. The literature proposes several strate-
gies, the most common being either the so-called classical strategy starting from the
simplest “specific to general” model, the most general model going from “general
to specific”. Florax et al. (2003) compare these strategies and show that the classical
approach gives the best results in terms of identifying the best specification and most
precisely estimated parameters but LeSage and Pace (2009) argue that the choice of
the best specification should start with the SDM. Elhorst (2010) proposes a mix of
these two strategies.
3.3 Example of Spatial Land Studies with Linear Models
This section provides some examples of aggregate land use studies which take
account of SA.
Some works include SA in order to improve the specification and understanding
of what drives land use change. For instance, Meyfroidt and Lambin (2008) analyze
the causes of reforestation in Vietnam during the 1990s on a national scale, and
test emerging forest transition theories on the same scale. They build a reforestation
spatial lag regression model using census and geographic data at a fine level of
348 R. Chakir and J. Le Gallo
Tabl e 2 Summary table of the estimated linear land use (LU) spatial model specifications (Chakir
and Lungarska 2017). ρis the spatial autoregressive coefficient, λthe SA coefficient, γand βrep-
resent a vector of unknown parameters to be estimated. Wis a nonnegative n×nmatrix describing
the spatial configuration or arrangement of the units in the sample
Model Model Interpretation
SEM y=Xβ+εand ε=λWε+uUnobserved omitted variables
follow a spatial pattern, data
measurement errors
SAR y=ρWy+Xβ+εLU for one location is
determined jointly with that of
neighbors
SLX y=Xβ+WXγ+εLU for one location is
determined by the explanatory
variables of neighbors
SDM y=ρWy+Xβ+WXγ+εA combination of SLX and
SAR and can be reduced to
SEM
SARAR y=ρWy+Xβ+εand
ε=λWε+u
A combination of SEM and
SAR
SDEM y=Xβ+WXγ+εand
ε=λWε+u
A combination of SEM and
SLX
GNS y=ρWy+Xβ+WXγ+ε
and ε=λWε+u
A combination of SLX and
SARAR
aggregation for the whole country. Their results show that forest land distribution
affects forests not just in the focal district but also in neighboring districts. This
observation can be interpreted in terms of a diffusion process: early and successful
implementation of the policy in some districts may have facilitated its rapid adoption
by neighboring districts.
Marcos-Martinez et al. (2017) estimate the determinants of land use in Australia’s
intensive agricultural region during the period 1992–2010. They estimate land use
shares with spatial error and random effects combined with variance decomposi-
tion analysis to identify the statistical significance, direction and magnitude of the
observed associations between land-uses and their drivers. Their results show that
improved transportation infrastructure, zoning regulations and mechanisms to reduce
exposure to farm debt and climate variability risks have significant impacts on the
configuration of the Australian agricultural landscape.
Amin et al. (2019) analyze whether protected areas are efficient instruments to
fight deforestation in Brazilian Amazonia. They estimate a dynamic SDM and assess
the impact of different types of protected areas (integral protected areas, sustainable
protected areas, indigenous lands) on deforestation. The results differ according to
the type of protected area: (i) integral protected areas and indigenous lands reduce
deforestation; (ii) sustainable use areas do not contribute to reducing deforestation;
Spatial Autocorrelation in Econometric Land Use Models: An Overview 349
and (iii) the spillover effects generated by integral protected areas and indigenous
lands lead to a reduction in the deforestation in their vicinities.
Two studies focus on prediction in spatial land use share models. Chakir and
Le Gallo (2013) make a methodological contribution to the literature by control-
ling for both unobservable individual heterogeneity and SA in an aggregate land use
model. Their study was conducted on a panel of land use data at the French depart-
ments NUTS3 scale, observed between 1992 and 2003. The authors were interested
in the relationship between four land uses (agriculture, forest, urban, and other) and
their potential economic and demographic determinants. The econometric model
consists of a system of three equations with a panel dimension and SA in the errors
associated to each equation. Thus, their econometric model is a SUR model with
random individual effects and autoregressive spatial structure of the error term. The
model was estimated using the feasible generalized least square (FGLS) estimation
method proposed by Baltagi and Pirotte (2011) for SUR-SEM-RE (Seemingly Unre-
lated Regressions-Spatial Error Model-Random Effects) model estimations. Their
results are of three orders: first, controlling for both unobservable individual het-
erogeneity and SA yields the best predictions relative to any other specification in
which SA and/or individual heterogeneity are ignored. Second, taking into account
the correlations between the error terms in the different equations does not seem to
improve prediction performance. Third, ignoring individual heterogeneity introduces
substantial loss of prediction accuracy.
Chakir and Lungarska (2017) estimate land use share models for France at a homo-
geneous (8 ×8 km) grid scale for five land use classes—agriculture, pasture, forest,
urban, and other. They investigate the determinants of land use shares using eco-
nomic, physical and demographic explanatory variables. They model SA between
grid cells and compare prediction accuracy and estimated elasticities for the dif-
ferent spatial model specifications (ordinary least square (OLS), SLX, SEM, SAR,
SDM, SDEM, SARAR, GNS). They compare these spatial specifications using three
rent proxies: farmers’ revenues, land prices, and shadow land prices derived from
a mathematical programming model. Their comparison is based on several criteria:
quality of economic explanation (significance of agricultural rents and their marginal
impacts), prediction quality (NRMSE), specification tests (LM tests), and goodness
of fit (log-likelihood, R2, AIC). The test results show that the SDM, SDEM, SARAR,
and GNS models should be considered. According to the goodness of fit (pseudo-R2,
log-likelihood and AIC) and prediction quality criteria, GNS is the specification that
best fits their data. In a context of aggregate land use, the existence of autocorrelation
is due mainly to spatially correlated errors—essentially a data measurement prob-
lem. This applies especially to their case since they use artificially constructed grids,
and different scales for the explanatory variables and land use data. Their results
show also that including SA in land use share models improves the quality of the
predictions which confirms the results in the previous aggregate land use literature.
350 R. Chakir and J. Le Gallo
4 Discrete Choice Land Use Models
When using individual (parcel or plot) data, the land use variable is generally a
categorical variable so that estimating land use patterns on individual data usually
requires a discrete choice framework. Discrete choice models are based on McFadden
(1974)’s random utility theory which states that the landowner decides to switch from
one use to another if the expected net revenues exceed the revenues from the original
use.
4.1 Individual Choice Land Use Model
This section presents the theoretical land use model based on individual data as in
Lubowski et al. (2008). We assume that the landowner chooses the land use of a plot
based on the costs and benefits associated with each possible use. For example, the
landowner chooses land use kat time tif:
Rkt −rCjkt >Rjt ∀j,k=1,...,Kand ∀t=1,...,T(5)
where Rjt and Rkt represent the discounted expected net benefits at time tof a unit
of land for uses jand k, respectively, Cjkt is the marginal cost of converting a unit
of land from use jto use kat time t(Cjjt =0) and ris the discount rate.
In order to estimate the determinants of land use econometrically, the theoretical
model suggests comparing the benefits and costs of converting land from one use
to another at each date. To move to the econometric specification, land use conver-
sion revenues and costs are rewritten as functions of the observed and unobserved
variables. Thus, the utility Uikt of the owner of parcel iwith land use kat time tis
written as follows:
Uikt =βxikt +ikt ∀i=1,...N,∀k=1,..., Kand ∀t=1,...,T,(6)
where xikt are the observed explanatory variables, βthe vector of parameters to be
estimated and ikt are the error terms which take account of the unobserved variables
that might affect the landowner’s utility.
We assume that the owner has a choice between Kland use categories for each
parcel at each date. The landowner chooses the optimal land use for his or her plot
by comparing the utilities associated to each land use category. If yit =1,2,... K;
is the landowner’s land use choice for the parcel iat time t, we obtain
yit =k,if Uikt ≥max Uijt ∀i=1,...N,∀j,k=1,...,Kand ∀t=1,...,T,
(7)
Thus, the probability that the parcel iis allocated to the use kat the time tis
written as
Spatial Autocorrelation in Econometric Land Use Models: An Overview 351
P(yikt =1)=Pr[Uikt ≥maxUijt](8)
for all j=1,...,Kwith yikt =1ifkis the observed use and 0 otherwise; Uikt is
the utility associated with land use k.
Since estimation of discrete choice models in the multinomial case is dimension-
ally constrained, some studies are limited to two use categories and use a probit
model in the binary case (Bockstael 1996; Kline and Alig 1999; Irwin and Bockstael
2002). Other studies estimate a multinomial logit model because of its computational
simplicity (Chomitz and Gray 1996; Nelson and Hellerstein 1997;Nelsonetal.2001)
which involves the questionable assumption of independence of irrelevant alterna-
tives (IIA). Finally, a nested logit model could be a good alternative if the alternatives
can be partitioned into several subsets.
4.2 Spatial Autocorrelation in Discrete Choice Models
SA can be accounted for in the discrete choice land use model (Eqs. (6)–(8)), by
including the spatially lagged variables or the error terms. Simplifying the notations
and removing the subscripts, the general nonlinear nesting model (GNNM) can be
written as follows:
U=ρWU +Xβ+WXγ+ε, and ε=λWε+u(9)
where WU is the shifted utility function for the weight matrix W,ρis the autoregres-
sive spatial parameter which indicates the magnitude of the interaction between the
latent variables U,γ, like β, is a vector of the unknown parameters to be estimated, λ
is the parameter of the intensity of the SA between the residuals, and uis a classical
error term such as u∼iid(0,σ2I). The GNNM model presented in Eq.(9) becomes
a SEM model if ρ=0 and γ=0, and becomes a SAR model if λ=0 and γ=0.
In contrast to the linear case, the spatially lagged variable in the SAR model is not
observable. For example, in the case of a land use model, it is the utility associated
to the profitability of neighboring plots and not the observed land use which should
define the utility function of the landowner (Anselin and Cho 2002).
In the case where the error terms εfollow a normal distribution, estimation of the
probit-SAR model raises two problems. On the one hand, heteroskedasticity makes
the classical estimators inconsistent. On the other hand, estimation of a probit-SAR
requires computation of a likelihood function with N−1 (where Nis the number
of observations) integrals which makes maximum likelihood estimation impossible.
This second difficulty applies also to the logit model case (Anselin and Cho 2002).
Several approaches have been proposed in the literature to deal with these estimation
problems, including simulation estimation (Geweke et al. 1994) or Bayesian (LeSage
2000) methods able to deal with the computation of multidimensional integrals of the
likelihood function. Other estimation procedures have been proposed to cope with the
problems associated to the introduction of SA in the case of discrete choice models.
These include the expectation-maximization method (McMillen 1992), the GMM
352 R. Chakir and J. Le Gallo
Tabl e 3 Summary table of the estimated spatial discrete choice models
Model Estimation method Example
Spatial autoregressive logit Bayesian Blackman et al. (2008)
Ordered probit Bayesian Wang and Kockelman (2009)
Multinomial probit Bayesian Chakir and Parent (2009)
Random parameter logit Max simulated likelihood Lewis et al. (2011)
Multinomial probit Max approximate CML Sidharthan and Bhat (2012)
Ordered probit Max CML Ferdous and Bhat (2012)
Multinomial logit GMM Li et al. (2013)
Multiple discrete-continuous
probit
Max CML Bhat et al. (2015)
Conditional parametric probit Max Locally Weighted
log-Likelihood estimator
McMillen and Soppelsa (2015)
Multinomial logit GMM Carrión-Flores et al. (2018)
(Pinkse and Slade 1998), the maximum pseudo-likelihood method (Smirnov 2010),
and finally the method of maximum approximate composite marginal likelihood
(CML) (Sidharthan and Bhat 2012). For detailed reviews of SA in discrete choice
models see Fleming (2004), Smirnov (2010). Simulation estimation and Bayesian
methods have been employed only recently to deal with the computational problems
associated to considering SA in discrete choice models. Because these methods are
still relatively expensive to implement, their use in the land use literature remains
limited. Table 3provides an overview of these studies.
4.3 Examples of Spatial Land Use Studies with Discrete
Choice Models
To tackle the complexities induced by SA in discrete choice models for land use,
some papers resort to Bayesian methods, for example, Wang and Kockelman (2009)
who estimate an ordered probit spatial dynamic model using satellite land cover
data. Chakir and Parent (2009) also use a Bayesian approach to estimate land use
determinants in a multinomial probit econometric model which accounts for both
unobservable individual heterogeneity and SA in errors. They analyze the determi-
nants of land based on a panel of 3,130 points in the Rhône department in France
between 1992 and 2003. It appears that land use changes are indeed influenced by
unobserved factors in neighboring plots. Finally, Blackman et al. (2008) estimate a
bayesian heteroskedastic SAR logit model of land cover for a shade-grown coffee
region in southern Mexico. Their results show that all other things being equal plots
close to large cities are less likely to be cleared which contrasts to the pattern usu-
ally observed in natural forests. They also find that belonging to a coffee-marketing
cooperative, farm size, and certain soil types are associated to tree cover while prox-
Spatial Autocorrelation in Econometric Land Use Models: An Overview 353
imity to a small town center is associated to forest clearing. This study is extended
in Blackman et al. (2012) who estimate a SAR probit model.
Other papers use variants of maximum simulated likelihood. Lewis et al. (2011)4
estimate a random parameter logit model to take account of the non-observed space-
time components of the willingness to pay. This specification makes it possible to
take account of spatial heterogeneity rather than SA and also allows consideration
of heteroskedasticity via a block variance-covariance matrix with individual effects
which depend on space. It is a kind of SA but with no spatial structure and with a
matrix of weights as in spatial models. In the spatial econometrics literature, CML
has become a popular approach for estimating spatial probit models and has been
used to model land use. For instance, Ferdous and Bhat (2012) analyze changes in
the intensity of urban land use taking account of both the spatial dimension and
temporal dynamics. Their econometric model is an ordered probit estimated using
CML. The results show that ignoring the presence of spatial autocorrelation and spa-
tial heterogeneity introduces important bias and that ignoring spatial heterogeneity
is more serious than ignoring lagged spatial dynamics. Sidharthan and Bhat (2012)
use maximum approximate CML (MACML) to estimate a multinomial probit-type
land use model with SA between plots and spatial heterogeneity.
Finally, rather than tackling the spatial autoregressive coefficient directly as in
the previous papers, McMillen and Soppelsa (2015) estimate a conditional paramet-
ric spatial probit model imposing far less structure on the data than conventional
parametric models. They illustrate the approach using data on 474,170 individual
lots in the City of Chicago. Their results suggest that simple functional forms are
not appropriate for explaining the spatial variation in residential land use across the
entire city. Similarly, Carrión-Flores et al. (2018) propose a GMM spatial estimator
for a multinomial logit model with spatial lag dependence. The model is linearized
to avoid the repeated matrix inversion required for the full GMM estimation. The
linearization breaks up the estimation procedure into two simple steps: a standard
multinomial logit model with no SA followed by a two-stage least squares (TSLS)
estimation of the linearized model which accounts for SA. This model is applied to
estimate land use conversion in the rural-urban fringe for four different land uses
(agricultural, residential, industrial and commercial). The results show a positive
SA of about 0.36—a result consistent with the widely-accepted idea that land use
conversion is a spatial process.
5 Land Use and Its Impacts on the Environment
Land use is considered as one of the main drivers of global changes to nature, which
endanger numerous species or cause their extinction and compromise the supply
of ecosystem services (ES) which are important for humans (Millenium Ecosystem
Assessment 2005). The protection of ES is emerging as a major concern alongside
4Several attempts in the literature introduce spatial dependence in multinomial models but, except
for Lewis et al. (2011) to the best of our knowledge, they havenot been used in the land use literature.
354 R. Chakir and J. Le Gallo
climate change issues (IPCC 2019) and biodiversity conservation (IPBES 2019).
This has resulted in land use becoming a growing concern for policy-makers as
means of protecting ecosystems (Bateman et al. 2013). There is a large literature
estimating the effects of land use on various ES: water quality (Fezzi et al. 2015),
carbon sequestration (Lubowski et al. 2006), and biodiversity (Polasky et al. 2008).
In this context, accounting for SA when studying the impacts of land use on ES
is a major issue. Research shows that including SA in species distribution models
improves model fit and prediction accuracy (Record et al. 2013) and that ignoring
SA can produce inaccurate results (Kühn 2007). Below, we review a selection of
those studies that model SA explicitly to estimate the impacts of land use on the
environment.
5.1 Land Use and ES
The relationships between land use and ES is complex. For instance, some land uses
such as intensive agriculture could have negative impacts on ecosystems while others
could contribute to the provision of many ES. For example, tropical forests are an
example of a supplier of ES at various scales. At the local scale, these services include
wood, secondary forest products, pollination, etc. More generally, they sequester
large amounts of carbon which regulates the global climate (IPCC 2019). In addition,
the productivity of some land uses such as agriculture is dependent on ecosystems
such as biological pest control, soil fertility, and pollination. Thus, degradation of
these ecosystems constitutes a serious threat to the long-term agricultural productivity
growth. Below, we provide two examples of spatial studies dealing with this link.
Chen et al. (2020) employ an integrated spatial panel approach to examine the geo-
graphic variations and spatial determinants of the ES balance in the middle reaches
of the Yangtze River urban agglomerations (MRYRUA) in China. They analyze the
spatio-temporal evolution features of landscape patterns and the supply of demand for
and balance among ES and landscape pattern metrics for the period 1995–2015. The
results indicate that construction land in the MRYRUA has increased continuously,
while farmland has decreased. Counties with higher ES supply and balance indices
are concentrated primarily in mountainous areas, while the indices of ES demand
in the three smaller urban agglomerations, plains areas, counties surrounding major
cities, and along major traffic routes are higher. SA and spatial spillover effects of
the ES balance index are observed in the MRYRUA. Population density and road
density are negatively associated to an ES balance. Landscape pattern metrics are
also statistically significant, either positive or negative. The findings suggest that both
drivers and spillover effects should be accounted for when considering integrative
ecosystem management and land use sustainability measures in urban agglomer-
ations. Both have important implications for urban planning and decision-making
related to development and ES.
Klemick (2011) uses cross-sectional farm survey data to estimate the value of
fallow ES in shifting cultivation in one region in the Brazilian Amazon. The objec-
tive is to test whether it provides economically significant local externalities which
Spatial Autocorrelation in Econometric Land Use Models: An Overview 355
might justify forest conservation from a local perspective. The author estimates a
production function to determine the contributions to agricultural income of on-farm
and off-farm forest fallow. Soil quality controls, instrumental variables and spatial
econometric approaches help address issues of endogeneity and variation in unob-
servable factors over space. The results suggest that Bragantina farmers generally
allocate land between cultivation and fallow efficiently taking account of beneficial
spillovers. This finding does not necessarily imply that farmers intentionally internal-
ize the value of these services but might suggest that private land tenure plays a role
in promoting sustainable land management given the different findings from other
studies of shifting cultivation in common property tenure regimes which identify
overexploitation of fallow biomass.
5.2 Land Use and Water Quality
There is a large literature on the effects of land use on water quality and freshwater
biodiversity. Most of these papers ignore SA. Here, we provide some examples of
studies that model SA explicitly in a study of land use and water quality.
Most studies show that forest areas have a positive impact on water quality com-
pared to intensive agriculture, livestock, and urban areas. For example, Abildtrup
et al. (2015) analyze the economic impacts of land use on the cost of drinking water
supply, taking account of both the organizational choice of water supply and spatial
factors in the same model. They estimate a model for the choice of management type
and for the price of water, accounting for the potential dependence of the error terms
between equations, as well as between neighboring water services. They estimate
a sample selection model adapted to a spatial context, that is, allowing for spatial
lags and spatial error processes. The model is applied to data from the French Vos-
ges department. The results show spatial interactions related to the characteristics of
neighboring water services but no SA of the error terms in the management choice
equation, or in prices. They show that forest land cover significantly reduces water
supply costs at the large but not the local scale.
Induced land use adaptation on freshwater biodiversity is analyzed by Bayramoglu
et al. (2020). They study the links between land use (agriculture, pasture, forest, and
urban environment) and the fish-based index (FBI) an indicator of the ecological
state of surface water measured for various French rivers observed between 2001
and 2013. They estimate two models: a spatial econometric model of land use and a
spatial panel statistical model of the FBI. Their results indicate that adapting land use
to climate change is reducing the biodiversity of freshwater in France. Furthermore,
rivers located in regions with intensive agriculture and pastures are associated to
lower freshwater biodiversity than those in forest regions. Simulations show that
climate change will exacerbate these negative impacts through changes to land use.
They show how two policies for regulating the level of fertilizers in agriculture and
carrying capacity in grasslands could help improve freshwater biodiversity and cope
with the adverse effects of land use and climate change.
356 R. Chakir and J. Le Gallo
5.3 Land Use and Climate Change
The interactions between land use and climate are complex (IPCC 2019). First, land
use and land practices affect the global concentration of greenhouse gases (Houghton
2003). Second, while land use change is an important driver of climate change, a
changing climate can lead to changes in land use. For example, farmers might convert
pasture to crop land which has higher economic returns under changing climatic
conditions. Third, spatially heterogeneous land use activities have important impacts
on local weather (Feddema et al. 2005). Fourth, land use changes could play an
important role in mitigating climate change either by increasing carbon sequestration
or by reducing greenhouse gas emissions. This could be achieved by adopting land
uses such as afforestation or preservation of permanent pasture (Pielke 2005).
Land use adaptation to climate change could exacerbate the adverse impacts of
land use on the environment. For example, Lungarska and Chakir (2018) show that
in France, climate change will reduce forest areas which could increase greenhouse
gas emissions. They estimate a spatial econometric land use model and simulate the
impacts of two IPCC climate change scenarios (A2 and B1, horizon 2100) and a mit-
igation policy in the form of a tax on greenhouse gas emissions (0–200 euros/tCO2)
aimed at reducing agricultural greenhouse gas emissions. They show that both cli-
mate change scenarios lead to an increase in agricultural area at the expense of forests.
Greenhouse gas mitigation policies reduce expansion of agriculture, and therefore
could counteract the consequences of climate change on land use. Taking account of
land use adaptations to climate change makes it possible to reduce abatement costs
in the agricultural sector.
6 Conclusion
The objective of this review was to summarize the literature on econometric land
use modeling and show how SA can be accounted for in these models. Despite the
recent advances in econometric land use models, several research directions remain
to be explored and several issues need to be addressed concerning data, theories, and
empirical models.
First, there is a frequent lack of data to construct relevant explanatory variables
implied by theoretical models. In particular, land rents are described in the theoretical
model as among the main decision variables related to land use or land use change
but are unobservable in the case of agricultural or urban use. In the case of forestry
use, these rents are even more difficult to calculate. More research is needed along
these lines, and especially to investigate the question of the links between land price
and land rent, drawing on the work of Randall and Castle (1985), Goodwin et al.
(2003).
Second, more investigation is needed into scale issues in land use studies. For
example, most economic variables refer to administrative units rather than grids
which makes it easier to estimate econometric models at the same administrative
Spatial Autocorrelation in Econometric Land Use Models: An Overview 357
scale (such as department, municipality, or small agricultural regions in the French
case). However, a land use model with aggregate spatial resolution is less relevant for
assessing the local ecological effects of land uses. Ecological issues such as habitat
quality or dispersion of species operate on fine scales. Ecological conditions vary
considerably within each administrative unit, introducing additional uncertainty for
ecological assessments.
Third, in addition to the spatial dimension, it would be interesting to incorporate
the dynamic dimension explicitly in econometric land use models (Epanchin-Niell
et al. 2017). Methodological advances in the specification and estimation of spatio-
temporal panel models are one of the difficulties related to spatial econometrics
as noted in Arbia (2011). The estimation methods developed by Ferdous and Bhat
(2012), Sidharthan and Bhat (2012) seem promising as alternatives to the computa-
tionally intensive Bayesian or simulation methods.
Fourth, all the models presented here assume implicitly that the spatial weight
matrix is exogenous. If spatial units refer to individual landowners making land use
choices, these choices might be influenced substantially by the choices of peers with
whom they choose to be linked in which case the weight matrix becomes endoge-
nous. Identification of endogenous peer effects and how to disentangle them from
exogenous effects and correlated effects in networks has been studied extensively.5
The way landowners form networks and how these affect land use decisions are of
considerable interest to understand the drivers of these decisions.
Finally and related to this issue, structural models should be further developed
to study the links between land use and land use changes, and their effects on the
environment for example on GHG emissions and biodiversity. The advantage of a
structural approach is that it makes more explicit assumptions about observable and
unobservable variables. The structural approach also makes it possible to unambigu-
ously account for the endogeneity of prices and the feedbacks that determine the
market equilibrium (Timmins and Schlenker 2009). The aim is to propose a theo-
retical economic model which includes the farmer’s decisions about crop rotations,
choice of inputs (fertilizers), land allocation between agricultural and grassland uses,
and herd size and composition. This would be quite challenging and would force a
limited focus on a subset of these decisions (Kaminski et al. 2013).
Addressing these issues would help to improve the quality of econometric land
use models. Developing accurate models is important for policy making to allow
for more accurate predictions about land use and future changes and more accurate
measurement of the effects of these changes on natural resources (biodiversity, water
quality, soil quality, and air quality).
5See Hsieh et al. (2019) for a recent paper on the specification and estimation of network formation
and network interaction and applications of this literature to land use issues can be found in Isaac
and Matous (2017), Baird et al. (2016).
358 R. Chakir and J. Le Gallo
Acknowledgements We dedicate this contribution to Christine Thomas-Agnan, a friend and col-
league for many years. We wish her a nice, sweet, and pleasant retirement. We are grateful for
helpful comments from the editors and the two anonymous reviewers. Raja Chakir acknowledges
the support from the French state aid managed by the Agence Nationale de la Recherche as part
of the “Investments d’Avenir” Programme within STIMUL (Scenarios Towards integrating multi-
scale land use tools) flagship project (LabEx BASC; ANR-11- LABX-0034) and Cland Institut de
convergence (ANR-16-CONV-0003).
References
Abildtrup, J., Garcia, S., & Kere, E. N. (2015). Land use and drinking water supply: A sample
selection model with spatial dependence. Revue d’Économie Régionale et Urbaine,2015(1–2),
321–342.
Aitchison, J. (1986). The statistical analysis of compositional data. London: Chapman and Hall.
Alig, R. (1986). Econometric analysis of the factors influencing forest acreage trends in the south-
east. Forest Science,32(1), 119–134.
Amin, A., Choumert-Nkolo, J., Combes, J.-L., Motel, P. C., Kéré, E. N., Ongono-Olinga, J.-G.,
et al. (2019). Neighborhood effects in the brazilian amazônia: Protected areas and deforestation.
Journal of Environmental Economics and Management,93, 272–288.
Anselin, L. (1988). Spatial econometrics: Methods and models. Dordrecht: Kluwer Academic Pub-
lishers.
Anselin, L., & Cho, W. (2002). Spatial effect and ecological inference. Political Analysis,10, 276–
297.
Arbia, G. (2011). A lustrum of sea: Recent research trends following the creation of the spatial
econometrics association (2007–2011). Spatial Economic Analysis,6(4), 377–395.
Ay, J.-S., Chakir, R., & Gallo, J. L. (2017). Aggregated versus individual land-use models: Modeling
spatial autocorrelation to increase predictive accuracy. Environmental Modeling & Assessment,
22, 129–145.
Baird, J., Jollineau, M., Plummer, R., & Valenti, J. (2016). Exploring agricultural advice net-
works, beneficial management practices and water quality on the landscape: A geospatial social-
ecological systems analysis. Land Use Policy,51, 236–243.
Baltagi, B. H., & Pirotte, A. (2011). Seemingly unrelated regressions with spatial error components.
Empirical Economics,40, 5–49.
Bateman, I. J., Harwood, A. R., Mace, G. M., Watson, R. T., Abson, D. J., Andrews, B., et al. (2013).
Bringing ecosystem services into economic decision-making: Land use in the United Kingdom.
Science,341, 45–50.
Bayramoglu, B., Chakir, R., & Lungarska, A. (2020). Impacts of land use and climate change on
freshwater ecosystems in France. Environmental Modeling & Assessment,25(2), 147–172.
Bhat, C. R., Dubey, S. K., Alam, M. J. B., & Khushefati, W. H. (2015). A new spatial multiple
discrete-continuous modeling approach to land use change analysis. Journal of Regional Science,
55(5), 801–841.
Billé, A. G., & Arbia, G. (2019). Spatial limited dependent variable models: A review focused
on specification, estimation, and health economics applications. Journal of Economic Surveys,
33(5), 1531–1554.
Blackman, A., Albers, H. J., Avalos-Sartorio, B., & Murphy, L. C. (2008). Land cover in a managed
forest ecosystem: Mexican shade coffee. American Journal of Agricultural Economics,90(1),
216–231.
Blackman, A., Avalos-Sartorio, B., & Chow, J. (2012). Land cover change in agroforestry: Shade
coffee in El Salvador. Land Economics,88(1), 75–101.
Spatial Autocorrelation in Econometric Land Use Models: An Overview 359
Bockstael, N. E. (1996). Modeling economics and ecology: The importance of a spatial perspective.
American Journal of Agricultural Economics,78(5), 1168–1180.
Brady, M., & Irwin, E. (2011). Accounting for spatial effects in economic models of land use: Recent
developments and challenges ahead. Environmental & Resource Economics,48(3), 487–509.
Carrion-Flores, C., & Irwin, E. G. (2004). Determinants of residential land-use conversion and
sprawl at the rural-urban fringe. American Journal of Agricultural Economics,86(4), 889–904.
Carrión-Flores, C. E., Flores-Lagunes, A., & Guci, L. (2018). An estimator for discrete-choice mod-
els with spatial lag dependence using large samples, with an application to land-use conversions.
Regional Science and Urban Economics,69, 77–93.
Chakir, R. (2015). L’espace dans les modèles économétriques d’utilisation des sols: enjeux
méthodologiques et applications empiriques. Revue d’Economie Regionale Urbaine,1/2(1), 59–
82.
Chakir, R., & Le Gallo, J. (2013). Predicting land use allocation in France: A spatial panel data
analysis. Ecological Economics,92, 114–125.
Chakir, R., & Lungarska, A. (2017). Agricultural rent in land-use models: Comparison of frequently
used proxies. Spatial Economic Analysis,12(2–3), 279–303.
Chakir, R., Parent, O. (2009). Determinants of land use changes: A spatial multinomial probit
approach. Papers in Regional Science,88(2), 327–344, 06.
Chen, W., Chi, G., & Li, J. (2020). The spatial aspect of ecosystem services balance and its deter-
minants. Land Use Policy,90, 104263 (2020)
Chomitz, K. M., & Gray, D. A. (1996). Roads, land use, and deforestation: A spatial model applied
to Belize. World Bank Economic Review,10(3), 487–512.
Claassen, R., & Tegene, A. (1999). Agricultural land use choice: A discrete choice approach.
Agricultural and Resource Economics Review,28(1), 26–36.
De Pinto, A., & Nelson, G. C. (2007). Modelling deforestation and land-use change: Sparse data
environments. Journal of Agricultural Economics,58(3), 502–516.
De Pinto, A., & Nelson, G. C. (2009). Land use change with spatially explicit data: A dynamic
approach. Environmental and Resource Economics,43, 209–229.
Elhorst, J. P. (2010). Applied spatial econometrics: Raising the bar. Spatial Economic Analysis,
5(1), 9–28.
Elhorst, J. P. (2014). Spatial econometrics: From cross-sectional data to spatial panels. Berlin:
Springer.
Epanchin-Niell, R., Kuwayama, Y., & Walls, M. (2017). Spatial-dynamic complexities of climate
challenge for rural areas: Integrating resource and regional economic insights. American Journal
of Agricultural Economics,99(2), 447–463.
Feddema, J. J., Oleson, K. W., Bonan, G. B., Mearns, L. O., Buja, L. E., Meehl, G. A., et al.
(2005). The importance of land-cover change in simulating future climates. Science,310(5754),
1674–1678.
Ferdous, N., & Bhat, C. R. (2012). A spatial panel ordered-response model with application to the
analysis of urban land-use development intensity patterns. Journal of Geographical Systems,15,
1–29.
Fezzi, C., & Bateman, I. J. (2011). Structural agricultural land use modeling for spatial agro-
environmental policy analysis. American Journal of Agricultural Economics,93(4), 1168–1188.
Fezzi, C., Harwood, A. R., Lovett, A. A., & Bateman, I. J. (2015). The environmental impact of
climate change adaptation on land use and water quality. Nature Climate Change,5(3), 255–260.
Fleming, M. M. (2004). Techniques for estimating spatially dependent discrete choice models.
In R. Florax & L. Anselin (Eds.), Advances in spatial econometrics: Methodology, tools, and
applications (pp. 145–167). Berlin: Springer.
Florax, R. J. G. M., Folmer, H., & Rey, S. J. (2003). Specification searches in spatial econometrics:
The relevance of Hendry’s methodology. Regional Science and Urban Economics,33, 557–579.
Foley, J. A. (2005). Global consequences of land use. Science,309(5734), 570–574.
Gao, L., & Bryan, B. A. (2017). Finding pathways to national-scale land-sector sustainability.
Nature,544(7649), 217–222.
360 R. Chakir and J. Le Gallo
Geweke, J., Keane, M., & Runkle, D. (1994). Alternative computational approaches to inference in
the multinomial probit model. The Review of Economics and Statistics,76(4), 609–632.
Goodwin, B. K., Mishra, A. K., & Ortalo-Magne, F. N. (2003). What’s wrong with our models of
agricultural land values? American Journal of Agricultural Economics,85(3), 744–752.
Grunfeld, Y., & Griliches, Z. (1960). Is aggregation necessarily bad? The Review of Economics and
Statistics,42(1), 1–13.
Hardie, I. W., & Parks, P. J. (1997). Land use with heterogeneous land quality: An application of
an area base model. American Journal of Agricultural Economics,79(2), 299–310.
Houghton, R. A. (2003). Revised estimates of the annual net flux of carbon to the atmosphere from
changes in land use and land management 1850–2000. Tellu s B ,55(2), 378–390.
Hsieh, C.-S., Lee, L.-F., Boucher, V. (2019). Specification and estimation of network formation and
network interaction models with exponential probability distribution. SSRN.
IPBES. (2019). Global assessment report on biodiversity and ecosystem services. Technical report,
IPBES.
IPCC. (2019). IPCC special report on climate change, desertification, land degradation, sustainable
land management, food security, and greenhouse gas fluxes in terrestrial ecosystems. Technical
report, IPCC.
Irwin, E., & Bockstael, N. (2002). Interacting agents, spatial externalities and the evolution of
residential land use patterns. Journal of Economic Geography,2, 331–54.
Irwin, E. G., Bell, K. P., & Geoghegan, J. (2003). Modeling and managing urban growth at the rural-
urban fringe: A parcel-level model of residential land use change. Agricultural and Resource
Economics Review,32(1), 83–102.
Irwin, E. G., & Bockstael, N. E. (2004). Land use externalities, open space preservation, and urban
sprawl. Regional Science and Urban Economics,34(6), 705–725.
Isaac, M., & Matous, P. (2017). Social network ties predict land use diversity and land use change:
A case study in Ghana. Regional Environmental Change,17, 1823–1833.
Kaminski, J., Kan, I., & Fleischer, A. (2013). A structural land-use analysis of agricultural adaptation
to climate change: A proactive approach. American Journal of Agricultural Economics,95(1),
70–93.
Klemick, H. (2011). Shifting cultivation, forest fallow, and externalities in ecosystem services:
Evidence from the eastern amazon. Journal of Environmental Economics and Management,61(1),
95–106.
Kline, J. D., & Alig, R. J. (1999). Does land use planning slow the conversion of forest and farm
lands? Growth and Change,30, 3–22.
Kühn, I. (2007). Incorporating spatial autocorrelation may invert observed patterns. Biodiversity
Letter,13, 66–69.
Lal, R. (2004). Soil carbon sequestration impacts on global climate change and food security.
Science,304(5677), 1623–1627.
LeSage, J., & Pace, R. (2009). Introduction to spatial econometrics. Boca Raton: CRC Press.
LeSage, J. P. (2000). Bayesian estimation of limited dependent variable spatial autoregressive model.
Geographical Analysis,32, 19–35.
Lewis, D. J., Plantinga, A. J., Nelson, E., & Polasky, S. (2011). The efficiency of voluntary incentive
policies for preventing biodiversity loss. Ressource and Energy Economics,33(01), 192 – 211.
Lewis, D. J., & Plantinga, A. J. (2007). Policies for habitat fragmentation: Combining econometrics
with GIS-based landscape simulations. Land Economics,83(19), 109–127.
Li, M., Wu, J., & Deng, X. (2013). Identifying drivers of land use change in China: A spatial
multinomial logit model analysis. Land Economics,89, 632–654.
Lichtenberg, E. (1989). Land quality, irrigation development, and cropping patterns in the northern
high plains. American Journal of Agricultural Economics,71(1), 187–194.
Lubowski, R., Plantinga, A., & Stavins, R. (2008). What drives land-use change in the united states?
A national analysis of landowner decisions. Land Economics,84(4), 551–572.
Spatial Autocorrelation in Econometric Land Use Models: An Overview 361
Lubowski, R. N., Plantinga, A. J., & Stavins, R. N. (2006). Land-use change and carbon sinks:
Econometric estimation of the carbon sequestration supply function. Journal of Environmental
Economics and Management,51, 135–152.
Lungarska, A., & Chakir, R. (2018). Climate-induced land use change in France: Impacts of agri-
cultural adaptation and climate change mitigation. Ecological Economics,147, 134–154.
Marcos-Martinez, R., Bryan, B. A., Connor, J. D., & King, D. (2017). Agricultural land-use dynam-
ics: Assessing the relative importance of socioeconomic and biophysical drivers for more targeted
policy. Land Use Policy,63, 53–66.
McFadden, D. (1974). Conditional logit analysis of qualitative choice behavior. In Frontiers in
Econometrics, Chap. 2. New York: Academic.
McMillen, D., & Soppelsa, M. E. (2015). A conditionally parametric probit model of microdata
land use in Chicago. Journal of Regional Science,55(3), 391–415.
McMillen, D. P. (1989). An empirical model of urban fringe land use. Land Economics,65(2),
138–145.
McMillen, D. P. (1992). Probit with spatial autocorrelation. Journal of Regional Science,32(3),
335–348.
Meyfroidt, P., & Lambin, E. F. (2008). The causes of the reforestation in Vietnam. Land Use Policy,
25(2), 182–197.
Millenium Ecosystem Assessment. (2005). Ecosystems and human well-being: Synthesis. Technical
report, World Resources Institute. Washington, DC: Island Press.
Muller, D., & Zeller, M. (2002). Land use dynamics in the central highlands of Vietnam: A spatial
model combining village survey data with satellite imagery interpretation. Agricultural Eco-
nomics,27(3), 333–354.
Munroe, D. K., Southworth, J., & Tucker, C. M. (2002). The dynamics of land-cover change in
western Honduras: Exploring spatial and temporal complexity. Agricultural Economics,27(3),
355–369.
Nelson, G. C., Harris, V., & Stone, S. W. (2001). Deforestation, land use, and property rights:
Empirical evidence from Darien, Panama. Land Economics,77(2), 187–205.
Nelson, G. C., & Hellerstein, D. (1997). Do roads cause deforestation? Using satellite images in
econometric analysis of land use. American Journal of Agricultural Economics,79(1), 80–88.
Feb.
Pielke, R. A. (2005). Land use and climate change. Science,310(5754), 1625–1626.
Pinkse, J., & Slade, M. E. (1998). Contracting in space: An application of spatial statistics to
discrete-choice models. Journal of Econometrics,85(1), 125–154.
Plantinga, A., Mauldin, T., & Miller, D. (1999). An econometric analysis of the costs of sequestering
carbon in forests. American Journal of Agricultural Economics,81, 812–24.
Plantinga, A. J. (1996). The effect of agricultural policies on land use and environmental quality.
American Journal of Agricultural Economics,78(4), 1082–1091.
Plantinga, A. J. (2015). Integrating economic land-use and biophysical models. The Annual Review
of Resource Economics,7(1), 233–249.
Plantinga, A. J., & Ahn, S. (2002). Efficient policies for environmental protection: An econometric
analysis of incentives for land conversion and retention. Journal of Agricultural and Resource
Economics,27(1), 128–145.
Polasky, S., Nelson, E., Camm, J., Csuti, B., Fackler, P., Lonsdorf, E., et al. (2008). Where to
put things? Spatial land management to sustain biodiversity and economic returns. Biological
Conservation,141, 1505–1524.
Randall, A., & Castle, E. N. (1985). Land resources and land markets. In A. V. Kneese & J. L.
Sweeney (Eds.), Handbook of natural resource and energy economics (Vol. 2, Chap. 13, pp.
571–620). Amsterdam: Elsevier.
Record, S., Fitzpatrick, M. C., Finley, A. O., Veloz, S., & Ellison, A. M. (2013). Should species
distribution models account for spatial autocorrelation? A test of model projections across eight
millennia of climate change. Global Ecology and Biogeography,22, 760–771.
362 R. Chakir and J. Le Gallo
Ricardo, D. (1817). The principles of political economy and taxation. London: John Murray,
Albemarle-Street.
Sidharthan, R., & Bhat, C. R. (2012). Incorporating spatial dynamics and temporal dependency in
land use change models. Geographical Analysis,44(4), 321–349.
Smirnov, O. A. (2010). Modeling spatial discrete choice. Regional Science and Urban Economics,
40, 292–298.
Smith, T. E., & LeSage, J. P. (2004). A Bayesian probit model with spatial dependencies. Advances
in Econometrics,18.
Stavins, R. N., & Jaffe, A. B. (1990). Unintended impacts of public investments on private decisions:
The depletion of forested wetlands. American Economic Review,80(3), 337–352.
Timmins, C., & Schlenker, W. (2009). Reduced-form versus structural modeling in environmental
and resource economics. Annual Review of Resource Economics,1(1), 351–380.
Verburg, P. H., Mertz, O., Erb, K.-H., Haberl, H., & Wu, W. (2013). Land system change and
food security: Towards multi-scale land system solutions. Current Opinion in Environmental
Sustainability,5(5), 494–502.
von Thunen, J. H. (1875). Der isolierte staat in beziehung auf landwirthschaft und nationalokonomie;
hrsg. von h. schumacher-zarchlin.
Wang, X., & Kockelman, K. M. (2009). Application of the dynamic spatial ordered probit model:
Patterns of land development change in Austin, Texas. Papers in Regional Science,88(2), 345–
365.
Wu, J.,& Adams, R.-M. (2002). Micro versus macro acreage response models: Does site-specific
information matter? Journal-of-Agricultural-and-Resource-Economics,27(1), 40–60.
Wu, J., & Segerson, K. (1995). The impact of policies and land characteristics on potential ground-
water pollution in Wisconsin. American Journal of Agricultural Economics,77(4), 1033–1047.
Zellner, A., & Lee, T. (1965). Joint estimation of relationships involving discrete random variables.
Econometrica,33, 382–94.
