Professional Documents
Culture Documents
Business Establishment Growth
Business Establishment Growth
Business establishment growth in the Appalachian region (2000–2007) was regressed on industry
sector composition controlling for demographic, physical, and economic determinants. We test
the hypothesis that local response to growth determinants is geographically heterogeneous using
Smooth Transition spatial process models. This class of models exhibiting endogenous regime
switching behavior provides another tool for exploring the spatially heterogeneous effects of
local determinants on economic growth.
Key Words: Appalachia, business establishment growth, smooth transition models, spatial
processes
Models explaining geographic heterogeneity of decisions using discrete spatial regimes, finding
economic growth are ubiquitous. For example, that the importance of local determinants varied
Partridge et al. (2008) found that the effects of depending on how counties were classified as
fiscal policies and other local characteristics on metropolitan, micropolitan, or noncore. Wojan,
growth varied considerably across rural areas in McGranahan, and Lambert (2010) allowed
the United States using Geographically Weighted parameters to vary across metropolitan–
Regression (GWR). Lambert and McNamara micropolitan–noncore counties and three re-
(2009) explained food manufacturer location source amenity categories, finding that the
interaction effects between individuals in creative
class occupations and entrepreneurs on economic
Wan Xu is a graduate research assistant, Department of indicators were heterogeneous across regimes.
Agricultural and Resource Economics, University of
Arbia, Basile, and Piras (2005) applied a non-
Tennessee, Knoxville, Tennessee. Dayton M. Lambert
is an assistant professor, Department of Agricultural parametric regression estimator to examine geo-
and Resource Economics, University of Tennessee, graphic nonlinearities of factors explaining the
Knoxville, Tennessee. regional heterogeneity of growth in Italy. Other
This research was supported by the Appalachian
examples of spatial econometric models admit-
Regional Commission and the University of Tennessee
Institute of Agriculture. ting individual or group-specific responses are
The article was greatly improved by comments numerous, including spatial adaptive filters (SAF)
from anonymous reviewers. We also express our (Foster and Gorr, 1986), quantile regression
gratitude to Raymond Florax and Val Pede for their
(Lambert et al., 2007), Quant’s (1958) regime
insights about smooth transition process models, and
to Tim Ezzel for his collaboration. switching regression, Casetti’s (1972) spatial ex-
The views expressed are those of the authors. pansion model, multilevel hierarchical modeling
310 Journal of Agricultural and Applied Economics, August 2011
(Voss, White, and Hammer, 2004), and random economic asymmetries (e.g., Fujita, Krugman, and
coefficient models (Anselin, Wendy, and Cho, Venables, 1999; Brakman et al., 2001). To the
2002). Regional studies using these approaches extent that the STAR’s autocatalytic function sorts
typically regress some economic indicator on spatial units along a continuous hierarchy, the
local factors hypothesized to explain growth smooth transition model also lends itself to iden-
according to parameters or functional forms tifying endogenous break points across space
unique to spatial units. The underlying tenets of resulting from (for example) differential trade
these methods are generally consistent with costs; access advantage to urban economies; job
the conventional idea that the constraints, op- and people migration; and ‘‘catastrophic agglom-
portunities, and politics guiding growth are eration’’, the superconcentration of industries into
ultimately context-dependent and that devel- a few regions (Fujita and Thisse, 2002; Baldwin
opment strategies are more likely to succeed et al., 2003). The endogenous sorting process
when tailored to local conditions (Irwin et al., so applied complements other modeling efforts
2010). Global solutions implied by models that that explain rural–urban growth interactions
restrict responses to local determinants to be resulting from hierarchical access advantages
the same everywhere may also understate im- (e.g., Partridge et al., 2008).
portant sources of heterogeneity that could This article demonstrates the capabilities of
provide insight about connections to wider the STAR family of spatial process models by
economies and specific solutions to regionwide examining the factors associated with business
resource allocation problems. establishment growth in the Appalachian re-
This research applies a relatively new spatial gion, 2000–2007. Despite significant economic
econometric model to explain business establish- improvement over the last four decades, nearly
ment growth in the Appalachian region from one fourth of the counties in the region con-
2000–2007, the Smooth Transition Regression tinue to struggle in terms of key economic in-
(STAR) model of Pede (2010) and Pede, Florax, dicators (Keefe, 2009). Improvement in relative
and Holt (2009). Like the GWR and SAF economic well-being is evident in many Ap-
methods, parameter estimates of the STAR model palachian communities from 1960–2011 but
assume different values at different locations. spatial inequities persist. Like many rural areas,
However, GWR and SAF models are extreme the challenges facing Appalachian communi-
examples of the incidental parameter problem ties are often reduced to tradeoffs between
because each spatial unit has its own vector of scale economies, transportation costs, external
coefficients; probabilistic statements about the market demand, and resource endowments and
coefficients are impossible, and interpretation is how these factors influence business location
restricted to the idiosyncrasies of the sample and job creation (Irwin et al., 2010).
(Anselin, 1988). In addition, the calibration of the This article has two objectives. The first
GWR model is sensitive to outliers, hetero- objective is to motivate a relatively new class
skedasticity, and possibly spatial error dependence of spatial regression models—STAR models—
(Cho, Lambert, and Chen, 2010). Poor calibration which allow for endogenous sorting of spatial
of GWR models may also lead to data over- units into growth regimes. The approach is useful
smoothing. An advantage of the STAR model is for modeling the effects of access advantage and
that the incidental parameter problem is circum- transport costs on local growth as a data-driven
vented and the usual robust covariance estimators process, bypassing the need for artificial di-
can be applied to make inferential statements. chotomies such as ‘‘rural’’ or ‘‘urban’’ county
Unlike the SAF or GWR models, nonlinear re- classification schemes. The second objective
lationships across space are modeled using ‘‘au- is to identify which counties in Appalachia
tocatalytic’’ switching functions under the STAR were positioned to withstand challenging eco-
specification. From a theoretical perspective, the nomic times in terms of sustained business es-
notion of endogenous regimes is also consistent tablishment growth given initial levels of existing
with the ‘‘New Economic Geography’’ results that infrastructure, demographic attributes, and in-
focus on the causes and consequences of regional dustry structure. The 2000–2007 period includes
Xu and Lambert: Business Establishment Growth in Appalachia 311
the low point of a brief recession (in 2001) and policy in terms of business retention and in-
the economic recovery that lasted through dustry recruitment.
2007. Some industry sectors may have posi-
tively contributed to business establishment Empirical Model
growth during this period. However, in other
locations, different industry sectors may have Local factors hypothesized to influence busi-
no effect, or even a negative impact, on employ- ness establishment growth include demographic
ment and income growth (Feser, Renski, and characteristics, settlement patterns, growth mo-
Goldstein, 2008; Spencer et al., 2010). Identifying mentum, infrastructure, human and social capital,
which industry sectors contributed to business physical and natural amenities, and industry
establishment expansion during this period structure. The baseline log-linear model in re-
may inform regional economic development duced form is:
Table 1. Summary Statistics of Growth Indicators, Industry Sector Shares, and Local Determinants
Data Standard
Variable Description Source Mean Deviation
lnestden00 Log establishments/area, 2000 CBP/1 0.546 1.338
distmet Distance to metro county, 1993 ERS/2 33.466 28.860
percomm Percent commute outside county, 2000 2000 Census/3 39.963
emprt Percent employment rate, 2000 REIS/4 95.394 1.580
lnmedhhi Log median household income, 2000 2000 Census 10.444 0.246
perestab20 Percent of firms with < 20 employees, 2000 CBP 88.030
perestab100 Percent of firms with > 100 employees, 2000 CBP 2.193
Dpop9000 D Population, 1990–2000 2000 Census 0.103 0.140
Demp9000 D Employment, 1990–2000 2000 Census 0.122 0.125
Destabs9000 D Establishments, 1990–2000 2000 Census 0.181 0.182
perblk2000 Percent black, 2000 2000 Census 16.885
peramind2000 Percent American Indian, 2000 2000 Census 0.428
perhsp2000 Percent Hispanic, 2000 2000 Census 2.220
pctpop2064 Percent Pop. 20–64 years old, 2000 2000 Census 58.717
c00p65ov Percent Pop. 641 years old, 2000 2000 Census 13.434
hsdip2000 Percent high school diploma, 2000 2000 Census 73.099
pctcc Percent population creative occupations, 2000 ERS 16.941
amenity Natural amenity index 2000 Census –0.200 1.178
pubpct Percent public land 2000 Census 7.410
interstate Interstate (51) ESRI/5 0.469
adhs Appalachian Development Highway (51) ARC/6 0.136
Industry sectors (% Establishments)
ag Percent agriculture, forestry CBP 1.60
mining Percent mining CBP 0.60
manu Percent manufacturing CBP 5.70
retail Percent retail trade CBP 20.2
infor Percent information CBP 1.50
prof Percent professional services CBP 6.30
art Percent arts, entertainment, and recreation CBP 1.30
accfood Percent accommodation and food services CBP 7.40
1. CBP, County Business Pattern.
2. ERS, Economic Research Service.
3. 2000 Census, U.S. Census, 2000.
4. REIS, Regional Economic Information System.
5. ESRI, ESRI ArcView GIS.
6. ARC, Appalachian Region Commission.
by the percent of the population with bachelor’s influence business growth (Jacobs, 1965). We
degrees (perhsdip) and the percent of persons include the percent of persons working in
working in creative occupations (percc). In so-called creative occupations (Wojan and
previous decades, rural areas with low educa- McGranahan, 2007) to proxy the stock of local
tion attainment attracted employers offering talent and intellectual capacity.
low-skill, low-wage jobs, but many of these firms Natural amenities and public land availabil-
relocated operations abroad or adopted new ity may influence business establishment growth
technologies requiring higher skilled labor in the by attracting new firms and people to locations
1990s (Johnson, 2001). Demand markets may with open space, wilderness, or scenic environ-
also harbor relatively more creative individuals ments (Deller et al., 2001; McGranahan, 2008).
capable of solving market logistic problems or In low-amenity places, growth may depend pri-
combining old ideas in new ways, which may marily on changes in demand for producer
Xu and Lambert: Business Establishment Growth in Appalachia 313
services driven by expansion of basic economic growth. Feser, Renski, and Goldstein (2008)
sectors (Wojan, McGranahan, and Lambert, found that clustering did not guarantee em-
2010). However, high-amenity areas may also ployment growth but was associated with new
be remote and difficult to access. A natural business formation from 1998–2002 in the
amenity index (amenity) was included to mea- Appalachian region. Therefore, we maintain no
sure the relationship between economic growth priors on the expected relationships specific sec-
and locations abundant in natural amenities tors might have on business establishment growth.
(McGranahan, 1999). The variable is an aggre-
gate index of sunlight, humidity, temperature, Spatial Processes, Regional Adjustment, and
topography, and water resources. The percent of Endogenous Growth Regimes
the county in public land was included to control
for the effects of public access to unbuilt areas The family of STAR models developed extends
on business establishment growth (landpub). Pede (2010) and Pede et al.’s (2010) previous
Dummy variables indicating the presence of work on STAR process models to the regional
a national interstate (interstate) or an Appala- adjustment model. In addition to endogenous
chian Development Highway (adhs) in a county stratification of counties into separate growth
were included in the regression to control for regimes, we hypothesize that business estab-
the influence of transportation infrastructure on lishment growth is simultaneously determined
business establishment growth. The expected sign by business establishment growth in neighbor-
P
is generally ambiguous. Good roads may be at- ing counties, for example, nj 5 1,i6¼j wij yj , where
tractive to prospective firms, which would in- W denotes spatial connectivity. Local growth
crease the likelihood of attracting new investment. may be influenced by information spillovers,
However, good roads may encourage out- thick labor markets, or forward–backward link-
commuting, which could encourage growth else- ages to other spatial units (Anselin, 2002; Moreno
where, thereby offsetting growth in local retail or et al., 2004). Most studies incorporating spatial
service sectors (Kahn, Orazem, and Otto, 2001). dependence typically use a spatial process model
Industry structure is measured by the per- attributable to Whittle (1954) whereby an en-
centage of manufacturing establishments with dogenous variable specifies interaction between
less than ten employees (perestab20) and the per- spatial units plus a disturbance term. Anselin and
centage of manufacturing establishments with Florax (1995) called this a spatial lag autore-
more than 100 employees (perestab100). Both gressive (SAR) model. The SAR model with
variables intend to capture effects attributable to autoregressive disturbances of order (1,1) (ARAR)
agglomeration economies and economies of scale (Anselin and Florax, 1995) contains a spatially
internal to firms (Lambert, Brown, and Florax, lagged endogenous variable (Wy) and spatially
2010). The relationship between industry sectors dependent disturbances: y 5 rWy 1 Xb 1 e,
and business establishment growth was measured e 5 lWe 1 u, where u is independently and
as sector kshares at the two-digit NAICS level; identically distributed with mean zero and co-
E
Ski2000 5 Ei2000
tot , where Eki2000 is the number of variance W, and W is a matrix defining re-
i2000
business establishments in sector k, county i, and lationships between spatial units. When the
Etot
i2000 is the total number of business estab- weights are defined as contiguous neighbors or
lishments in a county. There were eight industry groups of observations bounded by some distance
sectors considered: agriculture and forestry metric, local shocks are transmitted to all other
(NAICS 11), mining (NAICS 21), manufacturing locations with the intensity of the shocks de-
(NAICS 31–33), retail (NAICS 44), information creasing over space. The reduced form of the
services (NAICS 51), professional services ARAR model is y 5 A21Xb 1 A21B21u, with
(NAICS 54), arts and entertainment (NAICS 71), (respectively) A 5 (I – rW) lag autoregressive
and food and accommodations (NAICS 72) and B 5 (I – lW) error autocorrelation spatial
(Table 1). Of interest is the extent to which the filters. The inverted matrices A21 and B21 are
initial level (or ‘‘stock’’) of an industry sector was spatial multipliers that relay feedback/feed-
associated with aggregate business establishment forward effects of shocks between locations
314 Journal of Agricultural and Applied Economics, August 2011
(Fingleton, 2008), thereby distinguishing this class 2006) and the biological sciences (Schabenberger
of models from other econometric models. and Pierce, 2002), but less so in the spatial
We use a ‘‘gravity weight’’ specification to econometric literature. A spatial analogue of
model spillover potential between counties. the STAR model was introduced by Gress
Gravity weight specifications are hypothesized (2004), Basile and Gress (2005), and Basile
to proxy market potential across regions (Fujita, (2008). Pede, Florax, and Holt (2009) and Pede
Krugman, and Venables, 1999; Fingleton, 2008). (2010) modified Lebreton’s (2005) spatial ver-
Each element in W is sion of the time-series STAR model by including
a spatially lagged variable in the transition func-
pop98 98
i popj
wij 5 tion. Pede, Florax, and Holt also modified the
dij
spatial STAR model to accommodate spatial lag
where popi is the population of county i in 1998 and error multiplier effects. The approach applied
and dij is the network distance between counties here is parametric and extends their work.
i and j. The elements of W are presumed to be The economic geography literature gener-
exogenous; hence, the 2-year lag of population. ally predicts that friction caused by heteroge-
The spatial weight matrix was row-standardized, neous transport costs may induce ‘‘catastrophic
rendering it scale-neutral (Anselin, 1988). agglomeration’’ such that the distribution of
The reduced form of the SAR-type models firms and jobs across space concentrates into
suggests that estimation of the marginal effects one or a few regions (Brakman, Garretsen, and
is more complicated than the typical marginal van Marrewijk, 2001). The endogenous sorting
effects of log-linear models. There are a variety of regions admitted by the smooth transition
of methods whereby the marginal effects as- model are hypothesized to identify these so-called
sociated with SAR-type models can be calcu- ‘‘bifurcations’’ that could emerge in regional
lated (e.g., LeSage and Pace, 2009, pp. 34–39). economies resulting from uneven trade costs.
In this application, the influence of the lag mul- The STAR model provides a direct test for
tiplier is approximated as a geometric series (see identifying these structural breaks and the extent
also Anselin and Lozano-Gracia, 2008). For ex- to which access to agglomeration economies af-
ample, the ‘‘total effect’’ of a covariate k is the fects the relationship between local resource
global impact of that variable on a given spatial constraints and growth. Let G(g, c; v) be an au-
unit: A-1(Inbk) 5 [In 1 rW 1 r2W 2 1 r3W 3 1 tocatalytic function (Schabenberger and Pierce,
r4W 41 r5W 5 1. . . rqW q]bk, where the order q 2002) such as the logistic function, [1 1 exp(–g[v
refers to the location itself (q 5 0), the impact of – c]/sv)]21, with (respectively) slope and location
the neighbors (q 5 1), the impact of the neigh- parameters g and c and a transition variable v. The
bors of the neighbors (q 5 2), etc. In the limit, parameters are approximately scale-neutral when
A21 tends to (1 2 r)21 and the ‘‘total’’ marginal they are normalized by the standard deviation
effect can be written as bTotal k 5 bk ð1 rÞ1 . of the transition variable (sv). The adjustment
The ‘‘indirect effect’’ is the difference between model with regime-switching potential is
the total and direct effect or the impact neigh-
(2) Dy 5 GZb1 1 ð1GÞZb2 1 u,
boring locations (on average) has on a given
spatial unit given an incremental change in the where ‘‘’’ is the Hadamer product operator, Z
covariate at that location ðbIndirect
k 5 1rr
bk Þ . a matrix of covariates, and (b1, b2) coefficients
Provided a consistent covariance estimator, stan- corresponding with regimes 1 and 2. Equation 2
dard errors of the total and indirect effects may be can be rearranged accordingly (Maddala, 1983):
estimated using the delta method (Greene, 2000).
Dy 5 Zb2 1 G Z ðb2 b1 Þ 1 u, ! Dy 5 Zb
Smooth Transition Regime Model (3)
1 G Zd 1 u,
Smooth transition models are well developed in with the interaction between the transition function
time-series analysis (Terasvirta and Anderson, and the covariates permitting nonlinear parameter
1992; Van Dijk and Franses, 2000; Holt and Craig, variation among spatial units. As g increases,
Xu and Lambert: Business Establishment Growth in Appalachia 315
spatial units are sorted into more distinct Management and Budget) as the transition vari-
groups. Intermediate values of g identify spatial able (distmet). A number of alternative transition
units along a continuum that are ‘‘in transition’’ variables are conceivable (e.g., Pede, 2010), but
as determined by the transition variable, v (for using the distance to the nearest metropolitan
example, Figure 1). The parameter c is a loca- county is appealing to the extent that 1) the
tion parameter that determines the inflection geographic effects of trade costs on business es-
point on the regime splitting curve according to tablishment growth are hypothesized to be non-
the transition variable (Figure 1). For larger linear, possibly causing bifurcations in regional
values of g (e.g., greater than 100), spatial units growth trajectories (e.g., Fujita and Thisse, 2002);
split into distinct regimes with the interaction and 2) that the urban–rural hierarchy is important
coefficients (d) the difference from the refer- with respect to firm location decisions and eco-
ence group mean response to local deter- nomic growth (Partridge et al., 2008; Lambert
minants (the b’s) and the alternative regime. and McNamara, 2009).
Rejection of the null hypothesis d 5 0 suggests
a nonlinear relationship between local cova-
riates and business establishment growth. For Spatial Regimes and Spatial Process Models
large values of g, (3) behaves ‘‘as if’’ counties
were categorized using dummy variables (e.g., The basic smooth transition model is more
‘‘metropolitan’’ or ‘‘nonmetropolitan’’) and then complex when local spillovers between counties
interacted with every explanatory variable. and regime splitting potential are possible. For
There are no regimes when d 5 0 and the effects example, combining the STAR with the ARAR
of the covariates are geographically invariant. spatial process model suggests the following
Thus, when there are regimes, the location-spe- reduced form specification:
cific marginal effects (ME) of the basic STAR ARAR-STAR: Dy 5 A1 Zb 1 A1 GZd
model are MEi 5 b 1 Gi d. (4) 1 A1 B1 u ! Dy 5 rWDy 1 Zb 1 GZd
Of particular importance is the choice of the
1 B1 u.
transition variable (v), which is hypothesized to
drive the sorting process. Ideally, v conveys in- This specification suggests the following
formation about connectivity between spatial hypotheses with respect to a baseline a-spatial
units and is also exogenous. We use the road model that could be estimated using Ordinary
network distance of a county to the nearest Least Squares (OLS) or the usual spatial error
metropolitan county (defined by the Office of (SEM) and spatial lag (SAR) process models:
Figure 1. Example of the Transition Function G(g, c; v) and Different Levels of the Smoothing
Parameter, g. (left panel) Note That Two Distinct Regimes Emerge When g 5 100, Whereas There
Are No Regimes Identified When g 5 0. The Parameter c Functions as a Location Parameter; the
Inflection of the Transition Function is Centered on c. Transition Function of the Regime-Splitting
Variable, G(g, c; distmet), for Destabs2000–2007 (right panel)
316 Journal of Agricultural and Applied Economics, August 2011
Pede (2010) and Pede et al. (2010) outline the Instrumental Variable Estimation of Smooth
estimation procedures for recovering the STAR Transition Regression Spatial Process Models
spatial process model parameters using maxi-
mum likelihood (ML). We relax the distribu- Estimation of the SAR-STAR model with in-
tional assumption of normality maintained under strumental variables applies the same principle as
Xu and Lambert: Business Establishment Growth in Appalachia 317
the procedure typically used to estimate the SAR Similar steps may be applied to estimate the
model with instruments (e.g., Anselin, 1988, ARAR–STAR with the instruments defined
p. 86; Kelejian and Prucha, 1999). Anselin (2006) following Kelejian and Prucha (2010) (K&P)
surveyed a variety of instruments that could be general moments procedure with some minor
used to generate predicted values of the endoge- modifications. For instance, an iterative procedure
nous, spatially lagged dependent variable: Wy ~ 5 is applied to estimate a heteroskedastic-robust
~ ~
PWy where P is symmetric, positive definite, and version of the error autoregressive parameter (l).
idempotent projection matrix. Replacing Wy with The algorithm used in this application to estimate
its predicted values, the outcome variable is that ARAR–STAR version follows.
regressed on Z~ 5 [X, Wy],
~ yielding the SAR-
1. Estimate the STAR model, yielding G;
IV estimator; d0 5 (Z~0 Z) ~ Z~0 y. Standard errors
2. Given G, construct a residual vector with the
for the estimator are adjusted for the ‘‘first-stage’’
1 IV estimator based on ZG; Z~G .
regression such that AsyCovðd 0 Þ 5 s2IV Z~0 Z~
3. Find the error autoregressive parameter fol-
(assuming homoskedastic
errors)
2 with variance
P lowing K&P’s procedure for estimating the
s2IV 5 1n ni 5 1 yi d 0 0 Z i , where Z includes
ARAR process model with autoregressive
the original data (Greene, 2000). A hetero-
and heteroskedastic disturbances;
skedastic-robust version could be estimated as
4. Detrend the outcome and design matrix vari-
AsyCovðd ÞHET ables with the Cochran–Orcutt transformation
(17) n 1 1 as y* 5 y – lWy; Z G 5 Z G lWZ G .
5 Z~0 Z~ Z~0 WZ~ Z~0 Z~ ,
nk 5. Update the STAR parameters (g, c) given
(y*, Z G ); and
with W denoting the diagonal matrix of the 6. Return to step 1, and iterate until conver-
squared residuals and the sample size divided gence (e.g., 0.000001, in this application).
by the degrees of freedom a small sample
correction factor. Examples of instrumental Standard errors of the ARAR–STAR parameters
variables (IVs) typically used in the applied are estimated using the asymptotic covariance
literature include Q0 5 [X, WX, W2X] (e.g., matrix suggested by K&P (p. 60).
Kelejian and Prucha, 1999). An alternative set of The stepwise iterative procedure used for
instruments, which is adopted here, includes Lee’s the ARAR–STAR may be extended to cases
(2003) ‘‘best’’ set of IVs such that QBEST 5 where only error autocorrelation and spatial
[X,WðI r ^ WÞ1 X ^ ^ ,r
b0 ], with (b ^ 0 ) obtained nonlinearities are considered, as in the case of
0 0
from the IV regression with instruments Q0. the SEM with endogenous regimes (SEM–
Modification of the SAR–IV to the SAR–STAR STAR). In this case, the IV matrix is an identity
IV estimator is straightforward. and Wy is omitted from the design matrix (Z).
Standard errors may be estimated using an
1. Replace Wy by its predicted values in the appropriate heteroskedastic-robust covariance
design matrix Z (as previously shown). matrix as previously.
2. Find good starting values of the shape (g) and If maximum likelihood were used to estimate
location (c) parameters of the transition func- the STAR-type models, then a stepwise ‘‘specific-
tion G(g 0, c0; v) using a grid search. to-general’’ (Florax, Folmer, and Rey, 2003)
3. Given reasonable starting values, use a con- specification search could be applied using the
strained nonlinear optimization routine to Lagrange Multiplier tests developed by Pede,
minimize the objective: Florax, and Holt. In this application, a ‘‘general-to-
1 0 specific’’ approach (Hendry, 2006; Larch and
minðg ,cÞ y0 1 Z G Z~0G Z~G Z~G y, Walde, 2008) was considered. Therefore, hypoth-
eses about spatial nonlinearity, lag, error, ARAR
f
where ZG 5 [X, GX, Wy], and Z~G 5 [X,G X,Wy]. processes, and their combinations (H1–H8) were
tested by calculating Wald statistics based on the
4. Estimate standard errors using a heteroskedastic- robust covariance matrix of the full ARAR–STAR
robust covariance matrix (e.g., Equation 17). model (Equation 4, see K&P, p. 60).
318 Journal of Agricultural and Applied Economics, August 2011
Percent population 641 years old, 2000 –0.0933E–02 0.61 –0.0113E–01 0.61 –0.0709E–02 0.85 –0.0859E–02 0.85
Percent high school diploma, 2000 0.0820E–02 0.37 0.0993E–02 0.38 0.0959E–02 0.64 0.0116E–01 0.64
Percent population creative occupations, 2000 1% 0.00 1a% 0.00 –1% 0.12 –1% 0.12
Natural amenity index 0.02 0.00 0.02a 0.00 –0.01 0.37 –1% 0.37
Percent public land 0.0593E–02 0.15 0.0718E–02 0.15 –0.0756E–02 0.23 –0.0915E–02 0.23
Interstate –0.01 0.22 –0.01 0.22 0.01 0.51 0.01 0.51
ADHS 0.01 0.50 0.01 0.51 0.01 0.62 0.01 0.62
r 0.17 0.00
g 100.00 0.70
c 50.91 0.00
Sq. Corr. 0.63
Industry Sector Establishments
Percent agriculture –0.47 0.05 –0.57 0.05 0.92 0.01 1.11 0.01
319
320 Journal of Agricultural and Applied Economics, August 2011
Note: an ‘‘a’’ in the Total Effect column indicates the indirect effect (total effect – direct effect) is significant at the 5% level. Total effects were estimated as (1- r)21 b. Standard errors were
P Value
0.83
0.89
0.01
0.50
0.18
0.38
0.26
closely with the diameter of most counties in
Total Effect
the region. The counties in the top percentile
(the counties where G 5 1) are generally as-
Coefficient sociated with nonmetropolitan counties, and
counties in the bottom of the hierarchy (e.g., those
where G 5 0) are associated with metropolitan
–0.15
–0.06
1.12
–0.79
0.94
–1.23
0.52
Farthest
www.arc.gov/research/MapsofAppalachia.asp).
In Table 3, the ‘‘closest’’ regime (G 5 0)
corresponds with counties located in or near
P Value
0.48
0.27
0.12
0.34
0.01
0.39
0.13
manufacturing
Conclusions
Variable
Percent
Percent
Percent
Percent
Percent
Percent
Percent
Figure 2. Spatial Distribution of the Transition Variable G(g, c; distmet) for Destabs2000–2007 (left
panel) and the ARC Economic Distress Index (right panel)
Transition model of Pede, Florax, and Holt, to a incidental parameter problem commonly asso-
regional growth model; and 2) analyze the re- ciated with local regression methods like GWR.
lationship between local demographic and phys- The STAR model also provides a method for
ical attributes, industry sectors, and business testing some of the fundamental theories es-
establishment growth in the Appalachian region poused by the economic geography literature
using this method. The change in business es- that predict the effects of transport costs and
tablishment was regressed on industry sector agglomeration economies on regional growth.
composition controlling for local demographic, To the extent that the STAR family of spatial
physical, and economic determinants. The hy- models is able to identify structural breaks
pothesis that the relationship between industry across space, the method provides an interesting
sectors on growth was geographically heteroge- avenue for future research testing theories about
neous was tested using a relatively new spatial labor migration, wage earnings, and firm loca-
econometric approach, a STAR model. Findings tion patterns as related to localization and urban
suggest the relationship between the local deter- agglomeration economies.
minants and growth was nonlinear across the re- Much work remains to be done in terms of
gion, and the association between local attributes comparing estimation procedures suitable for
and business establishment growth could be modeling smooth transition growth processes.
characterized into two distinct growth regimes. Although the advantages and disadvantages of
The spatial distribution of the regime members ML and GMM estimation are well known, the
was remarkably similar to the distribution of the performance of the STAR model and its spatial
ARC’s ‘‘economic distress’’ index. process variants under different experimental
The STAR family of models is relatively circumstances remains to be documented. The
new to the spatial economic toolbox. The performance of diagnostics used to specify
method contributes in terms of its ability to STAR-class models should also be thoroughly
identify regimes using a data-driven approach investigated. In this application, a ‘‘general-to-
as opposed to one in which regimes are identi- specific’’ approach was taken to specify the re-
fied using conventions potentially fraught with gression model. How this specification search
definitional problems (i.e., ‘‘urban’’ vs. ‘‘rural’’ compares with a ‘‘specific-to-general’’ approach
dichotomies). The STAR model also relaxes the could provide information regarding which types
assumption that a single parameter completely of tests should be used under different assump-
summarizes an entire region but bypasses the tions (e.g., Lagrange Multiplier tests, assuming
322 Journal of Agricultural and Applied Economics, August 2011
a normal distribution vs. Wald tests in which Europe.’’ Région et Dévéloppement 21(2005):
distributional assumptions are relaxed). 93–118.
Boots, B., and C. Dufournaud. ‘‘A Programming
Approach to Minimizing and Maximizing Spa-
References tial Autocorrelation Statistics.’’ Geographical
Analysis 26(1994):54–66.
Aldstadt, J., and A. Getis. ‘‘Using AMOEBA to Brakman, S., H. Garretsen, and C.V. Marrewijk.
Create a Spatial Weights Matrix and Identify An Introduction to Geographical Economics:
Spatial Clusters.’’ Geographical Analysis 38(2006): Trade, Location and Growth. Cambridge, UK:
327–43. Cambridge University Press, 2001.
Anselin, L. Spatial Econometrics: Methods and Bureau of Economic Analysis. Internet site:
Models. Dordrecht: Kluwer Academic Pub- http://www.bea.gov/ (Accessed 2007).
lishers, 1988. Casetti, E. ‘‘Generating Models by the Expansion
———. ‘‘Under the Hood: Issues in the Specifi- Method: Applications to Geographical Re-
cation and Interpretation of Spatial Regression search.’’ Geographical Analysis 4(1972):81–91.
Models.’’ Agricultural Economics 27(2002): Cho, S., D.M. Lambert, and Z. Chen. ‘‘Geo-
247–67. graphically Weighted Regression Bandwidth
———. ‘‘Spatial Regression.’’ Working Paper, Selection and Spatial Autocorrelation: An
Spatial Analysis Laboratory, Department of Ge- Empirical Example Using Chinese Agriculture
ography and National Center for Supercomputing Data.’’ Applied Economics Letters 17(2010):
Applications, Urbana–Champaign, 2006. 767–72.
Anselin, L., and R. Florax. Small Sample Properties Clark, C.D., D.M. Lambert, W.M. Park, and M.D.
of Tests for Spatial Dependence in Regression Wilcox. ‘‘Willingness to Fund Public Education in
Models. Berlin, Germany: New Directions in a Rural Retirement Destination County.’’ Journal
Spatial Econometrics, 1995. of Research in Rural Education 24(2009):1–16.
Anselin, L., R.J.G.M. Florax, and S.J. Rey. Ad- Deller, S.C., T.-H. Tsai, D.W. Marcouiller, and
vances in Spatial Econometrics: Methodology, D.B.K. English. ‘‘The Role of Amenities and
Tools and Applications. Berlin, Germany: Quality of Life in Rural Economic Growth.’’
Springer, 2004. American Journal of Agricultural Economics
Anselin, L., and N. Lozano-Gracia. ‘‘Error in 83(2001):352–65.
Variables and Spatial Effects in Hedonic House Feser, E., H. Renski, and H. Goldstein. ‘‘Clusters
Price Models of Ambient Air Quality.’’ Em- and Economic Development Outcomes: An
pirical Economics 34(2008):5–34. Analysis of the Link between Clustering and
Anselin, L., K. Wendy, and T. Cho. ‘‘Spatial Ef- Industry Growth.’’ Economic Development
fects and Ecological Inference.’’ Political Quarterly 22(2008):324–44.
Analysis 10(2002):276–97. Fingleton, B. ‘‘A Generalized Method of Moments
Arbia, G., R. Basile, and G. Piras. ‘‘Using Spatial Estimator for a Spatial Model with Moving
Panel Data in Modeling Regional Growth and Average Errors, with Application to Real Estate
Convergence.’’ Working paper, Institute for Prices.’’ Empirical Economics 34(2008):35–57.
Studies and Economic Analyses, Rome, Italy, Florax, R.J.G.M., H. Folmer, and S.J. Rey. ‘‘Speci-
2005. fication Searches in Spatial Econometrics: The
Arraiz, I., D.M. Drukker, H.H. Kelejian, and I.R. Relevance of Hendry’s Methodology.’’ Regional
Prucha. ‘‘A Spatial Cliff-Ord Type Model with Science and Urban Economics 33(2003):557–79.
Heteroskedastic Innovations: Small and Large Foster, S.A., and W.L. Gorr. ‘‘An Adaptive Filter
Sample Results.’’ Journal of Regional Science for Estimating Spatially-Varying Parameters:
50(2010):592–614. Application to Modeling Police Hours Spent in
Baldwin, R., R. Forslid, P. Martin, G. Ottaviano, Response to Calls for Service.’’ Management
and F. Robert-Nicoud. Economic Geography Science 32(1986):878–89.
and Public Policy. Princeton, NJ: Princeton Fritsch, M., and P. Mueller. ‘‘Regional Growth Re-
University Press, 2003. gimes Revisited—The Case of West Germany.’’
Basile, R. ‘‘Regional Economic Growth in Europe: Discussion paper, Germany. Max Planck Institute
A Semi-parametric Spatial Dependence Ap- of Economics, Entrepreneurship, Growth and
proach.’’ Regional Science 87(2008):527–44. Public Policy Group, March 2004.
Basile, R., and B. Gress. ‘‘Semi-Parametric Spatial Fujita, M., P. Krugman, and A.J. Venables. The Spa-
Auto-Covariance Models of Regional Growth in tial Economy. Cambridge, MA: MIT Press, 1999.
Xu and Lambert: Business Establishment Growth in Appalachia 323
Fujita, M., and J.-F. Thisse. Economics of Ag- Empirical Look at Southeastern Non-Metro-
glomeration: Cities, Industrial Location, and politan Counties, 2000–2004.’’ The Review of
Regional Growth. Cambridge, UK: Cambridge Regional Studies 37(2007):251–78.
University Press, 2002. Lambert, D.M., and K.T. McNamara. ‘‘Location
Getis, A., and J. Aldstadt. ‘‘Constructing the Spa- Determinants of Food Manufacturers in the
tial Weights Matrix Using a Local Statistic.’’ U.S., 2000–2004: Are Nonmetropolitan Counties
Geographical Analysis 36(2004):90–104. Competitive?’’ Agricultural Economics 40(2009):
Greene, W.H. Econometric Analysis. Upper Sad- 617–30.
dle River, NJ: Prentice Hall, 2000. Larch, M., and J. Walde. ‘‘Lag or Error? Detecting
Gress, B. ‘‘Semiparametric Spatial Autocovar- the Nature of Spatial Correlation.’’ Data Analysis,
iance Models.’’ PhD Dissertation, University of Machine Learning and Applications. C. Preisach,
California, Riverside, 2004. H. Burkhardt, L. Schmidit-Thieme, and R.
Hendry, D.F. ‘‘A Comment on Specification Decker, eds. Berlin, Germany: Springer, 2008.
Searches in Spatial Econometrics: The Relevance Lebreton, J.-D. ‘‘Dynamical and Statistical
of Hendry’s Methodology.’’ Regional Science and Models for Exploited Populations.’’ Australian
Urban Economics 36(2006):309–12. and New Zealand Journal of Statistics 47
Holt, T.M., and L.A. Craig. ‘‘Nonlinear Dynamic (2005):49–63.
and Structural Change in the U.S. Hog-Corn Lee, L.-F. ‘‘Best Spatial Two-Stage Least Squares
Cycle: A Time-Varying STAR Approach.’’ Estimators for a Spatial Autoregressive Model
American Journal of Agricultural Economics with Autoregressive Disturbances.’’ Econometric
88(2006):215–33. Reviews 22(2003):307–35.
Irwin, E., A.M. Isserman, M. Kilkenny, and M.D. LeGallo, J., and C. Ertur. ‘‘Exploratory Spatial
Partridge. ‘‘A Century of Research on Rural Data Analysis of the Distribution of Regional
Development and Regional Issues.’’ American per Capita GDP in Europe, 1980–1995.’’
Journal of Agricultural Economics 92(2010): Regional Science 82(2003):175–201.
522–53. LeSage, J.P., and K. Pace. Introduction to Spatial
Jacobs, J. The Death and Life of Great American Econometrics. Boca Raton, FL: Taylor &
Cites. London, UK: Penguin Books, 1965. Francis Group LLC, 2009.
Johnson, T.G. ‘‘The Rural Economy in a New Maddala, G.S. Limited-Dependent and Qualita-
Century.’’ International Review of Regional tive Variables in Econometrics. Cambridge,
Science 24(2001):21–37. UK: Cambridge University Press, 1983.
Kahn, R., P.F. Orazem, and D.M. Otto. ‘‘Deriving McGranahan, D.A. Natural Amenities Drive Ru-
Empirical Definitions of Spatial Labor Markets: ral Population Change. Agricultural Economic
The Roles of Competing versus Complementary Report No. 781. Washington, DC: U.S. De-
Growth.’’ Journal of Regional Science 41(2001): partment of Agriculture, 1999.
735–56. ———. ‘‘Landscape Influence on Recent Rural
Keefe, S.E. Participatory Development in Appalachia: Migration in the U.S.’’ Landscape and Urban
Cultural Identity, Community, and Sustainability. Planning 85(2008):228–40.
Knoxville, TN: University of Tennessee Press, Moreno, R., E. Lopez-Bazo, E. Vaya, and M.
2009. Artis. ‘‘External Effects and Cost of Pro-
Kelejian, H.H., and I.R. Prucha. ‘‘A Generalized duction.’’ Advances in Spatial Econometrics:
Moments Estimator for the Autoregressive Methodology, Tools and Applications. Berlin,
Parameter in a Spatial Model.’’ International Germany: Springer, 2004.
Economic Review 40(1999):509–33. Partridge, M.D., D.S. Rickman, K. Ali, and M.R.
———. ‘‘Specification and Estimation of Spatial Olfert. ‘‘The Geographic Diversity of U.S.
Autoregressive Models with Autoregressive Nonmetropolitan Growth Dynamics: A Geo-
and Heteroskedastic Disturbances.’’ Journal of graphically Weighted Regression Approach.’’
Econometrics 157(2010):53–67. Land Economics 84(2008):241–66.
Kooijman, S.A.L.M. ‘‘Some Remarks on the Sta- Pede, V.O. ‘‘Spatial Dimensions of Economic
tistical Analysis of Grids Especially with Respect Growth: Technological Leadership and Club
to Ecology.’’ Annals of Systems Research Convergence.’’ Ph.D. Dissertation, Purdue
5(1976):113–32. University, 2010.
Lambert, D.M., C.D. Clark, M.D. Wilcox, and Pede, O.V., R.J.G.M. Florax, and M.T. Holt. ‘‘A
W.M. Park. ‘‘Do Migrating Retirees Affect Spatial Econometric STAR Model with an Ap-
Business Establishment and Job Growth? An plication to U.S. County Economic Growth,
324 Journal of Agricultural and Applied Economics, August 2011