A Tale of Two Borders: Use‐Value 
Assessment, Land Development, and 
Irrigation Investment
Daniel P. Bigelow, Todd Kuethe
Bigelow, D. P., & Kuethe, T. (2020). A tale of two borders: use‐value assessment, land 
development, and irrigation investment. American Journal of Agricultural Economics, 102(5), 
1404-1424.
10.1002/ajae.12086
This is a pre-copyedited, author-produced PDF of an article accepted for publication in American 
Journal of Agricultural Economics following peer review. The version of record [A Tale of Two 
Borders: Use‐Value Assessment, Land Development, and Irrigation Investment. American Journal 
of Agricultural Economics 102, 5 p1404-1424 (2020)] is available online at: https://doi.org/10.1002/
ajae.12086.
Made available through Montana State University’s ScholarWorks 
scholarworks.montana.edu 
A Tale of Two Borders: Use-Value Assessment, Land
Development, and Irrigation Investment
Daniel P. Bigelow Todd Kuethe
February 18, 2020
Abstract
Since 1960, all 50 states in the US have adopted some form of preferential tax
treatment for farmland. These provisions often take the form of use-value assessment,
where farmland is taxed on the basis of its value in agricultural production, as opposed
to its full market value. While the main goal of use-value assessment is to slow the
conversion of farmland to non-agricultural uses, other channels of influence are also
possible, such as those stemming from reinvestment of foregone tax expenses. Despite
its widespread nature, there is little empirical evidence pertaining to the influence
of use-value assessment on land-use or investment decisions. Using a postmatching
difference-in-differences framework, we exploit the temporal and spatial discontinuities
surrounding the adoption of use-value assessment in Kansas in 1989 to measure how
use-value assessment affected plot-level land development and irrigation investment de-
cisions. The results of our analysis indicate that, as intended, the use-value assessment
policy curtailed development in the Kansas City metropolitan area. Evidence regarding
the potential investment-spurring effects of use-value assessment is more mixed, sug-
gesting that farmers may have increased irrigation in some areas because of use-value
assessment-induced tax savings.
JEL Codes: Q15, R14, Q24
Keywords: property tax, irrigation, urban development, land-use policy and regulation
1
Running head: Use-value assessment
Acknowledgements: Daniel P. Bigelow is an assistant professor in the Department of
Agricultural Economics and Economics at Montana State University. Todd Kuethe is an
associate professor and the Schrader Endowed Chair in Farmland Economics in the Depart-
ment of Agricultural Economics at Purdue University. We thank two anonymous reviewers,
Terry Hurley, and seminar participants at Montana State University, University of Georgia,
and the 2018 Northeastern Agricultural and Resource Economics Association annual meeting
for useful suggestions on earlier versions of the paper. Special thanks go to Patrick Flanagan
of USDA-Natural Resources Conservation Service and Vince Breneman and Ryan Williams
of USDA-Economic Research Service for providing us with the NRI data and supplemental
NRI variables.
2
All 50 states in the US provide some form of preferential tax treatment of agricultural land
(Anderson and England, 2014).1 Between 1960 and 1995, each state adopted a preferential
taxation policy (Anderson and England, 2014). Preferential farmland tax treatment lowers
the tax burden of farmland owners in an effort to reduce the conversion of agricultural
land to alternative, generally developed, uses with higher private economic returns. In this
respect, preferential taxation is borne of the same motivations as other policy tools designed
to mitigate the negative side effects of excessive development (or “urban sprawl”), such as
greenbelts, urban growth boundaries, and development impact fees (Nechyba and Walsh,
2004; Quigley and Rosenthal, 2005).
Preferential farmland tax programs take a variety of forms, but the most common type
of policy is use-value assessment (UVA).2 Under a pure UVA program farmland is taxed
according to its value in perpetual agricultural use, as opposed to its full market value.
This can have a significant effect on the values used in tax assessments, since the true
market value of a farmland parcel may include capitalized future development returns that
have yet to be realized by the landowner (Plantinga, Lubowski, and Stavins, 2002). By
removing capitalized future non-agricultural use components from the land values used in
tax assessments, landowners will, in theory, face less financial pressure to convert their land.
In more remote rural areas where land is less susceptible to development pressure, UVA
will have minimal effect on the likelihood of a farmland parcel being developed. In some
states, however, UVA programs are explicitly designed to reduce assessed values beyond
the amount solely attributable to capitalized future non-agricultural returns (Anderson and
England, 2014), and may also decrease the volatility of property tax expenses for farmland
owners, which some have suggested may lead to increases in farm investment (Conklin and
Lesher, 1977).
This paper examines the degree to which UVA affects land development and farm in-
vestment. Using plot-level panel data on land use, we exploit variation in the timing of
UVA implementation across states to study the effects of UVA adoption in Kansas in 1989.
Kansas was a relatively late adopter of UVA and by 1989 all surrounding states had had a
3
preferential farmland tax policy in place for at least 14 years (Anderson, Giertz, and Shimul,
2015). Importantly, our data on development and irrigation decisions spans 1982-2012, cov-
ering the period before and after Kansas adopted UVA. In addition, we show that the UVA
program adopted in Kansas reduced tax expenses in both the rural and more urbanized
parts of the state, suggesting the policy may have had an effect on the investment behavior
of landowners whose land did not have development potential. To identify the effects of
UVA, we rely on variation in micro-level decision outcomes across state borders before and
after Kansas’s policy was implemented. We focus on how the policy affected two separate
decisions of farmland owners. First, we analyze how UVA adoption impacts farmland devel-
opment by focusing on the Kansas City metropolitan area, located along Kansas’s eastern
border. Second, we investigate how UVA adoption affects irrigation decisions at the ex-
tensive margin, specifically through groundwater irrigation along Kansas’s western border
overlying the Ogallala (or High Plains) Aquifer.
The existing literature provides little in terms of rigorous empirical evaluation of the
effects of UVA and related policies. A significant share of prior research was conducted
in the 1970s, when a large number of US states adopted preferential farmland tax policies
in response to rising farmland prices (Blase and Staub, 1971; Schwartz, Hansen, and Foin,
1975, 1976; Conklin and Lesher, 1977; Coughlin, Berry, and Plaut, 1978). In a study ex-
plicitly designed to investigate the degree to which preferential tax treatment achieves its
intended effect of reducing farmland conversion, Morris (1998) examines county-level trends
in farmland retention before and after policy adoption and finds that reduced tax burdens
for farmland owners are associated with a dampening of farmland loss. Polyakov and Zhang
(2008) also show that differential property taxation slows the pace of development, and fur-
ther demonstrate that land-use decisions are relatively inelastic with respect to incremental
changes in property tax rates. Butsic, Lewis, and Ludwig (2011) show that rural property
tax credits administered through a “circuit breaker” mechanism have a relatively weak in-
fluence on reducing development. In related research, preferential farmland taxation has
been shown to reduce overall tax revenues surrounding metropolitan areas (Anderson and
4
Griffing, 2000) and shift the property tax burden to the owners of other types of property
(Dunford and Marousek, 1981; Chicoine and Hendricks, 1985). A recent study by Dinterman
and Katchova (2019) suggests that the financial impacts of changes in use-value calculations
are shared by both farmland tenants and landlords. The impact of UVA of farm investment
and production decisions is relatively unexplored, but was first suggested by Conklin and
Lesher (1977), who theorized that preferential taxation policies have the potential to spur
farm investment (or, equivalently, slow disinvestment) in urbanizing areas by reducing uncer-
tainty over tax expenses. To our knowledge, there are no recent studies of how preferential
tax policy affects irrigation or other investment decisions of farmland owners.
Our study also contributes to the broader literature on the effects of land-use policy on
land-use decision-making, which includes a number of recent studies of the effects of various
growth controls on development outcomes (such as Geshkov and DeSalvo, 2012; Dempsey
and Plantinga, 2013; Wrenn and Irwin, 2015; Towe, Klaiber, and Wrenn, 2017; Dempsey
et al., 2017; Zhang, Wrenn, and Irwin, 2017). This growing body of literature is the direct
result of increased data availability and the rise of quasi-experimental methods in applied
economic research. A quasi-experimental research design is crucial to the evaluation of
land-use regulations, given that the placement and implementation of such policies are gen-
erally endogenously related to land market outcomes (see, e.g., Butsic, Lewis, and Ludwig,
2011). As a result, credible estimates of the effects of land-use regulations must exploit
some plausibly-exogenous source of identifying variation, typically over time, to accurately
capture the impact of policies on the rate, amount, or location of development. We are un-
aware of any existing empirical studies that focus on how farmland property tax relief might
affect irrigation decisions, though Bigelow et al. (2017) show that the relative stringency of
growth containment policies can influence the amount of irrigated land that remains in a
city’s environs.
This study makes several contributions to the literature on preferential farmland taxa-
tion. First, by using a panel of plot-level outcomes, our results provide micro-scale empirical
evidence of UVA reducing land development, which, to our knowledge, has yet to be doc-
5
umented in existing studies. We find that development becomes significantly less likely,
by roughly 1.8-3.4%, in Kansas due to the adoption of UVA. Moreover, we find that the
bulk of the development effects manifest 13 years after the policy was adopted, with the
effect tapering off thereafter, which suggests there are large option values associated with
changes in expectations regarding development potential. These results also accord with the
conventional wisdom regarding the temporary nature of how UVA disincentivizes farmland
conversion (Anderson and England, 2014). Second, our results provide suggestive evidence
that UVA adoption may have spurred an expansion of irrigated land in western Kansas. The
results vary from state to state, with immediate increases ranging from 3.1-7.5% for land in
Kansas counties bordering Colorado, but generally negative or null effects found for lands
located along other border segments (Nebraska and Oklahoma). If UVA affected irrigation
outcomes, the policy potentially exacerbated the open-access groundwater extraction prob-
lem in the Ogallala Aquifer, an unintended environmental consequence similar to results
found in other public policy contexts (e.g., Stavins and Jaffe, 1990). However, our results
also indicate that the changes along the Colorado border may have been partly driven by
differences in farmland characteristics which made irrigation more likely on the Kansas side
of the border.
Preferential Farmland Assessment in Kansas and Surrounding States
Kansas, the focal point of our study, adopted its UVA policy in 1989, making it one of the
last states to implement a preferential taxation program.3 Although Kansas’s policy was
formally adopted in 1989, the legislation that eventually led to the policy was passed in
1985 (Kansas Department of Revenue, 2018). Prior to 1989, agricultural land in Kansas was
assessed on the basis of 30% of fair market value.4 Since adopting UVA, assessed values used
for taxation purposes have been “based on the productive potential directly attributable to
the natural capabilities of the land, not fair market value” (Anderson and England, 2014, p.
44). Variation in assessed farmland values, which are a function of the imputed eight-year
rolling average of a hypothetical landlord’s share of net farm income, stems from three main
sources: (1) land use (i.e., cropland vs. grassland/pasture), (2) soil quality, and (3) county-
6
level variation in property tax rates. Under Kansas’s UVA program, the assessment ratio, or
the fraction of the assessed value actually used to compute tax expenses, remained at 30%.5
The four states surrounding Kansas all adopted some form of preferential farmland tax
assessment well before Kansas: Colorado (1970), Nebraska (1972), Oklahoma (1974), and
Missouri (1975). As described in more detail below, differences in UVA adoption timing
between Kansas and its neighbors are critical to our identification strategy. However, agri-
cultural property taxation is not static, and the methods adopted in individual states have
shifted to varying degrees over time.
To isolate the potential effects of UVA on land development, we focus on the border
Kansas shares with Missouri. The Kansas-Missouri border is straddled by the Kansas City
metropolitan area, a large urban area that provides an ideal setting to assess the influence of
the UVA policy on land development (figure 1). Missouri adopted its use-value assessment
legislation in 1975 (Lapping, Bevins, and Herbers, 1977), but the process by which use values
were to be determined was not formalized until 1985. Starting in 1985, Missouri farmland
was formally assessed using a net income measure, which was subsequently switched to a
measure based on cash rents, before reverting back to a net income measure in 2007 (Food
and Agricultural Policy Research Institute, 2007).6
To study the impacts of UVA on irrigation decisions, we study (bi-directional) dryland-
irrigated farmland conversions along the border Kansas shares with Colorado, Nebraska,
and Oklahoma. The region is home to the Ogallala Aquifer, a major source of irrigation
water supply, which spans a portion of all three border segments (figure 1). Irrigation in
the region, primarily from groundwater, became widespread during the post-World War II
era as farmers adopted new pumping and irrigation technologies that increased access to the
aquifer and allowed corn to be grown in the area (Hornbeck and Keskin, 2014).7
Colorado, which shares the bulk of Kansas’s border overlying the Ogallala Aquifer,
adopted its UVA program in 1970. In Colorado, use-value estimates are derived from a
net income approach, where assessed values vary based on a moving ten-year average of crop
yields, the share of crops typically received by landlords, and typical landlord expenses (Col-
7
orado Assessors Association, 2019). Net income is converted to use value using a statutory
capitalization rate of 13%, and property taxes are then computed using an assessment ratio
of 29%.8 Nebraska passed legislation to allow for use-value assessment in 1972, but the pro-
gram has evolved over time due to legal conflicts.9 Since 1992, farmland property has been
assessed under a classified-use system. Under this system, taxes in Nebraska were initially
based on 80% of the full market value of the land, with the fraction reduced to 75% in 2007.
As a result, Nebraska has by far the least generous farmland property tax system (from a
landowner’s standpoint) in our study area (Midwestern Office of the Council of State Gov-
ernments, 2012). Lastly, Oklahoma began implementing use-value assessment in 1974 using
an income capitalization approach based on prevailing cash rental rates (Hapgood, 2010;
Anderson, Giertz, and Shimul, 2015), adjusted for soil quality. Starting in 1981, Oklahoma
switched from a rent-based method to one based on a classification scheme of "productivity
points" which are a function land use, soil type, and the corresponding soil productivity
index. With the exception of a few minor adjustments to update the soils data used in de-
termining assessed values, the Oklahoma UVA methodology has remained unchanged since
1981.
Of three states neighboring the western portion of Kansas, Colorado and Kansas have
the most similar UVA programs, especially given the comparable assessment ratios used in
both states. Oklahoma’s program is also comparable in spirit, though based on its own
idiosyncratic productivity index which implicitly places a heavy discount on the true use
value of agricultural land. Further, the tax policies in both Colorado and Oklahoma have
changed little since their initial implementation in the 1970s. Nebraska, by contrast, features
a program that currently bears little resemblance to those in the other three states and,
furthermore, one that has changed significantly over the study period.
Supporting Evidence
Of primary importance to our identification strategy is the late adoption of preferential farm-
land taxation in Kansas relative to its neighboring states. Although the intent of Kansas’s
use-value assessment policy was to reduce property taxes on farmland with development
8
potential, this may not necessarily hold for land located in more remote areas, where use
values could conceivably be larger than market values. Moreover, the farmland assessment
ratio of 30%, which remained constant before and after the UVA policy was implemented, is
substantially higher than those presently used for other land classes in Kansas, such as the
11.5% rate imposed on residential land (Hildreth, 2018), which makes it unclear if the policy
change led to a meaningful reduction in property taxes for any farmland owners. What
is key, then, is whether or not the computed use values are lower than the corresponding
market values. To show that the UVA policy did indeed result in a windfall gain for farm-
land owners throughout the state, we estimate a set of county-level regression models using
reported farmland property tax expenses for 1987 and 1992.10 We focus on 1987 and 1992 to
match the quinquennial frequency of our main data source described below. The estimation
dataset includes all counties in the five states in the study area (Kansas, Missouri, Colorado,
Oklahoma, and Nebraska) that reported farmland tax expenses in both years. The tax
expense measures are adjusted for inflation using the GDP implicit price deflator available
from the Bureau of Economic Analysis. While the UVA legislation in Kansas was passed
in 1985, the actual policy did not go into effect until 1989, and thus the 1987 tax expenses
in Kansas were still administered under the previous taxation framework based on market
value. We do note, however, that we can not rule out that the 1987 tax expenses may be
influenced by actions taken in anticipation of the eventual implementation of the policy. If
anything, however, this would bias these first-stage estimates upward, making the estimates
conservative.
To generate the effects of interest, we regress logged farmland tax expenses on a set of
county fixed effects and allow for differential 1992 (post-UVA) effects for Kansas. Standard
errors are clustered by USDA agricultural district, of which there are 41 in the entire five-
state region. The results, with marginal effects in brackets, are shown in table 2. Using
all counties in the study area, the post-UVA regression coefficient for Kansas amounts to a
marginal effect of -12.49%, indicating a clear reduction in tax expenses for the population
of Kansan landowners. This suggests that use values were 12.49% lower than the previous
9
market values upon which tax assessments were made. If the UVA policy induced any short-
term value-increasing investment behavior, this should be interpreted as a lower bound on
true magnitude of the difference between use and market value. To show that the policy
affected both rural and periurban tax expenses, we split the sample up based on the 75th
percentile of the population density distribution for 1987.11 Strong negative coefficients of
-13.52% and -8.94% emerge for both the rural and more urban segments of the dataset,
respectively. Similar results are obtained if we use alternative property tax variables, such
as the effective farmland property tax rate (tax expense divided by land value), property
taxes per farm, and property taxes per farmland acre. The only coefficient estimate with a
p-value greater than 0.1 is that for the effective property tax rate in urban counties. The lack
of precision is likely attributable to the relatively small sample of urban counties. We show
in Appendix A that these regression results are insensitive to the use of farmland acreage
weights.
In the appendix, we also present historical state-level trends in the same farm property
tax variables from U.S. Department of Agriculture (2014). The state-level data predate
the availability of the Census data, and thus provide a longer, though coarser, depiction of
how the UVA policy affected Kansas relative to its neighbors. For all of the same variables,
whether taking the change over 1982-1992 or 1987-1992, Kansas has either the biggest decline
or smallest gain. The preceding analysis provides firm evidence that Kansas’s adoption of
the UVA policy in 1989 led to meaningfully different farmland property taxation rates in
both urban and rural areas, which we use as motivation for the subsequent analysis.
Model
In this section we motivate our empirical application by providing a conceptual overview of
the relationships use-value assessment shares with land development and irrigation decisions.
We then present our empirical model and discuss key aspects of our identification strategy.
Conceptual Model
The relationship between UVA and land development decisions was formalized by Anderson
(1993), which we briefly review here to illustrate the mechanisms at work in our empirical
10
application. Following prior work on urban economic theory, Anderson’s (1993) model posits
that the per-acre value of a parcel of land is determined by the sum of two discounted streams
of future returns: (1) the net income stream generated by farming, f(t), which is pertinent
until period D, when (irreversible) development occurs, and (2) the subsequent discounted
net income stream accruing from development. The landowner, therefore, selects the optimal
time of development, D, to maximize the parcel’s value, V (t,D).
When land is taxed at its market value, the tax expenses owed by farmland owners may
be written as:
{ ∫ D ∫ ∞ }
(1) τδFV F (t,D) = τδF f(u)e−(r+τ)(u−t)du+ e−(r+τ)(D−t) h(u,D)e−(r+τd)(u−D)du
t D
where r is a constant discount rate, τ is the property tax rate (or millage rate), and δF is
the assessment ratio (Anderson, 1993). The assessment ratio is bound between 0 and 1 and
represents the fraction of the assessed value that is used in computing property taxes. This
contrasts with the property tax expenses landowners face under use-value assessment:
{ ∫ ∞ }
(2) τδUV U(t,D) = τδU f(u)e−(r+τ)(u−t)du .
t
Intuitively, the land development effect brought about by UVA centers on the presump-
tion that τδUV U(t,D) ≤ τδFV F (t,D). If this holds, either through a lower assessed value
(V U(t,D)) or assessment ratio (δU) under UVA, then UVA reduces property tax payments
for farmland owners and disincentivizes development by increasing the opportunity cost of
converting their land. In practice, one would expect V F (t,D) to converge to V U(t,D) the
further one moves away from the rural-urban fringe. The large disparity between V F (t,D)
and V U(t,D) has been demonstrated in numerous empirical applications (e.g., Zhang and
Nickerson, 2015). As a result, UVA, in its purest form (i.e., where δU = 1), should have
minimal effect on the rate of farmland development in rural areas.
As it relates to our study area, one important farm investment decision that could be
affected by UVA is the use of irrigated, rather than dryland, production. Following Negri,
11
Gollehon, and Aillery (2005) and Chambers and Just (1989), the annual net return to agri-
cultural production in (1) and (2), f(·), can be expressed as multi-output indirect profit
function. Let Y be a vector of agricultural commodity outputs and P be a vector of exoge-
nous output prices. Farms employ a vector of variable inputs X (e.g., water, labor, fertilizer,
energy, etc.) based on an associated vector of exogenous input prices W. Farms operate
a quantity of fixed, allocable land, represented by the scalar N , that is subject to tax rate
τ and assessment ratio δ. The discrete choice of irrigation is represented by the scalar S,
which is associated with a lump-sum annualized capital cost of irrigation ω. Indirect profits
are also subject to exogenous physical characteristics Θ, such as soil conditions and climate,
which affect production.12 The indirect profit function takes the form:
(3) f j(P,W,ω, δj, τ, N,Θ) = max[P′Y −W′X− ωS − τδjNV j(D) | Y ∈ Y (X,Θ, N,S)]
X,S
where Y (X,Θ, N,S) is the restricted production possibilities set and j = F,U indexes the
method of property tax assessment.
Irrigation adoption is assumed to be discrete and dichotomous in equation (3), with SI
and SD representing production technology for irrigation and dryland production, respec-
tively. As formulated, our model is static and we implicitly assume that the necessary water
conveyance structures and water rights are already in place and available to the farmer.13
Separate indirect profit functions associated with irrigated and dryland production can be
written as:
(4) f jI (P,W,ω, δj, τ, N,Θ) = max[P′Y −W′X−ωS j j
I − τδ NV (D) | Y ∈ Y (X,Θ, N,SI)]
X
and
(4’) f jD(P,W,δ, τ, N,Θ) = max[P′Y −W′X− τδjNV j(D) | Y ∈ Y (X,Θ, N,SD)].
X
The profit-maximizing farm operator adoptions irrigation when the net returns from doing
12
so (4) exceed the returns from dryland farming (4’). Adoption of use-value assessment will
create an incentive to invest in irrigation if fUI (P,W,ω, δU , τ, N,Θ) > fFD (P,W,δF , τ, N,Θ).
In theory, use value should deviate minimally from market values in rural areas. As shown
in the previous section, however, even with a constant assessment ratio, rural landowners
in Kansas experienced a 13.5% reduction in property tax expenses after UVA was passed,
suggesting that use values are lower than corresponding market values. Additionally, the
formulaic determination of assessed values under UVA may provide more stable and pre-
dictable farmland tax expenses compared to those based on more volatile market values,
which Chicoine and Hendricks (1985) notes may lead to additional investment.
Empirical Model
We estimate the impact of UVA on two outcomes: (i) conversion of farmland to developed
use and (ii) conversion between dryland and irrigated agricultural production. Our empirical
approach combines generalized post-matching difference-in-differences (DD) estimation with
the contiguous border county-pair framework of Dube, Lester, and Reich (2010).
For each plot observation, ik, in the development model, we construct a binary variable
Di t that takes a value of 1 if the parcel is in developed use at time t and 0 otherwise. In our
k
notation, i indexes a physical plot of land we observe over time and k indexes the instances
of the plot’s time series appearing in the sample, which is explained further below. We treat
development decisions as irreversible and our model is therefore estimated using parcels
where Di 1 = 0. Parcels drop out of the estimation sample once they are converted, making
k
the panel unbalanced. The probability that an undeveloped parcel is developed between the
baseline period and time t is estimated using:
(5) Prob(Di t = 1) = β UB A D′ D
k 1t t KSi + α1tUt KSi +X i ρ1 + φ1tBPi + τ
k 1t + εikt.
The indicator variables UB
t and UA
t represent the respective periods before and after UVA
is implemented. Differential pre-treatment time effects associated with being located in the
UVA treatment area (KSi), Kansas, are thus measured by β1t. Common time fixed effects
are captured by τ1t. Of primary interest are the differential effects for Kansas in periods
13
after UVA was implemented, which are estimated by α1t. The model also accounts for
observable factors XD
i that may influence land development. The D superscript is included
on the vector of observable covariates to highlight that the explanatory factors included in
the development and irrigation models may differ.
Similar to Dube, Lester, and Reich (2010), to minimize potential bias stemming from
unobserved spatial heterogeneity, we include in (5) a complete set of cross-border contiguous
county pair indicator variables (BPD
i ). We construct this vector to cover all cross-border
k
contiguous county pairs in the study region. Since counties may be contiguous to more than
one cross-border county, each plot i entering (5) is indexed by k to account for the fact that a
plot may enter the dataset multiple times. For an individual parcel observation, k = 1, ..., Ki
indexes the number of times a plot’s county is contiguous to (or shares a border with) another
county on the opposite side of the Kansas border. We also allow the border-pair effects, φ1t,
to vary over time to account for unobserved temporal variation common to each border pair.
We construct a similar model to estimate the impact of UVA on irrigation adoption on
cropland along Kansas’s western border. For each parcel i, a binary indicator variable Iikt
takes the value of 0 if the parcel is in dryland production or the value of 1 if the parcel is
irrigated at time t. The probability that a parcel is irrigated at time t can be expressed:
(6) Prob( = 1) = B + ′
Ii t β2tUt KSi α2tU
A I
t KSi +X i ρ1 + φ I
k 2tBPi + τ
k 2t + εikt.
All terms in (6) are defined analogously to those in (5), although the study area (and
consequently the relevant border pairs in BP I
i ) and specific control variables included in X I
k i
differ. It is also important to note that we treat dryland-irrigated cropland conversions as
reversible, meaning that farmers may switch from dryland to irrigated or irrigated to dryland
production. In contrast to 5, estimation of 6 is thus conducted with a balanced panel and the
UVA effects for the irrigated conversion model β2t captures the net effect of UVA adoption
on irrigated conversion decisions. It is plausible that both types of transition would be
influenced by the UVA policy. Irrigation investment, entailing a switch from dryland to
irrigated production, could be brought about by UVA through reduced expenditures on
14
taxes. Likewise, a switch to UVA could also stimulate a greater resilience to increased
energy costs for irrigation amid declining aquifer levels, a major issue in the Ogallala region
(Scanlon et al., 2012).
Identification Strategy
Our identification strategy for estimating the effects of Kansas’s UVA policy consists of
five components. First, we exploit temporal discontinuities across states in the timing of
UVA adoption. Specifically, Kansas did not adopt UVA until 1989, well after neighboring
states had adopted similar policies. The change in Kansas’s UVA status serves as our source
of temporal variation to study the effects of farmland tax policy changes on land use and
irrigation investment decisions. Prior to 1989, the opportunity cost of developing farmland
would have been relatively lower in Kansas, since the net returns to farming would have
been lower due to the higher rate of farmland taxation. Likewise, we investigate whether the
tax savings from the switch to UVA resulted in more irrigated land in Kansas after 1989.
To utilize these differences in UVA adoption timing, we rely on a panel dataset of plot-
level observations of both land-use classification and irrigation status spanning a timeframe
covering years before and after Kansas implemented UVA (1982–2012).14
Second, to mitigate the potential for confounding factors to bias our estimates, we adopt
the contiguous-border county-pair strategy developed by Dube, Lester, and Reich (2010).
Specifically, we stack our data and include in the specifications a complete vector of border-
pair dummy variables for the counties located along the Kansas border. Stacking the data
is required because of the many-to-many relationship that characterizes contiguous cross-
border county pairs. Without stacking, equations (5) and (6) are not estimable due to the
fact that each plot’s county may share the border with more than one county, resulting
in perfect collinearity between the dummy variable regressors. In stacking the data, we
incorporate the observations for plot i for each of the Ki border pairs to which its county
belongs.15 In addition, we further control for time fixed effects common to each individual
border pair, which, for example, accounts for the possibility that the northern and southern
portions of the Kansas City metro area developed at different rates over time. The rationale
15
behind including the contiguous cross-border county-pair effects relies on the idea that plots
in adjacent counties will be more similar to one another in terms of local economic structure,
agricultural production, and other factors not captured explicitly in the model specification,
thus serving as a more valid control group to study the effects of changes in farmland tax
policy. Of course, the validity of the cross-border control group hinges on there being no
other contemporaneous changes to farmland tax policy or other state-specific factors that
affect irrigation and land development over the study timeframe.
Third, to further narrow the source of variation used to generate our estimates, the
samples are restricted to plots contained within a pre-specified bandwidth of the Kansas
state border. We generate estimates for bandwidths of 5–25 miles in five-mile increments.
The restriction is meant to further mitigate potential bias stemming from unobserved spatial
heterogeneity. This is particularly important given the size of some counties in the study
area. We also report estimates derived from a “no-bandwidth” sample containing all plots
in the relevant counties adjacent to the border.
Fourth, our preferred estimates are derived from a sample matched on propensity scores
measuring the likelihood of a plot being located in Kansas based on its observable character-
istics.16 As in Chen, Lewis, and Weber (2016), we use nearest-neighbor 1:1 propensity score
matching without replacement and a caliper equal to one-quarter of the standard deviation
of the estimated propensity scores.17 Note that the propensity score model is estimated
using the unstacked version of the estimation dataset. Otherwise, the matching coefficients
would be biased towards observations located in counties with a greater number of contigu-
ous cross-border neighbors. The matching attributes are measured at the plot level using
data from the USDA’s National Resources Inventory (NRI) and are described in greater
detail below. We estimate a single propensity score model for each study area (Kansas City
and western Kansas) and bandwidth (5–25 miles or no bandwidth). After estimating the
propensity scores, we apply the stacking procedure described above and only allow matches
to occur within cross-border contiguous county pairs. However, since a plot may not have
a match within all of its associated border pairs, the number of times each plot appears in
16
the matched sample will be less than or equal to the number of times the plot appears in
the unmatched sample. Matching is intended to produce a more comparable set of treat-
ment and control plots, which is especially important in our case given that we do not have
observation-level fixed effects in the model specifications.
Finally, we include a vector of observable covariates, XD
i or X I
i , to account for observ-
able factors that may influence the outcomes of interest, which increases the precision of
our estimates and further narrows the range of possible sources of unobserved confounding
variation. As described below, the controls include plot characteristics from the NRI and
various baseline county attributes.
In sum, the UVA estimates can be thought of as average treatment effects on the treated
(ATTs) identified through cross-border variation in plot-level development and irrigation
investment outcomes using a matched sample within a pre-specified bandwidth of the Kansas
border, conditional on common border pair time effects and observable plot characteristics
and county attributes. By stacking contiguous cross-border county pairs, the UVA estimates
are computed by averaging over all estimated local ATTs for each border pair. In this sense,
our adopted approach incorporates elements of DD, matching, and regression discontinuity
design estimation strategies. The models are estimated in linear probability form to avoid the
convergence issues associated with the incidental parameters problem in non-linear binary-
choice models.
A somewhat unique feature of our study is that we are studying the effect of UVA
in Kansas in relation to neighboring states, all of which had already adopted some form
of preferential tax assessment prior to the start of our study time period. This contrasts
with the typical two-by-two DD case comparing two groups without a treatment, which
is subsequently assigned to one of the two, with the other serving as a nonexperimental
control group. Our application can be thought of as a special two-group case of the more
generalized situation in which all groups eventually receive a treatment. As Goodman-Bacon
(2018) shows, if all observational units receive a treatment at some point, the eventual source
of variation used to identify the treatment effect is the same as that used in the present study.
17
This comes at the point when all but one unit has received the treatment, in which case the
remaining variation stems from control groups comprised of units that have already been
treated. An obvious implication of relying on this type of variation, however, is that we are
unable to use parallel trends as a means to determine the comparability of the treatment
and control groups prior to Kansas adopting UVA. The use of a matched sample should, in
part, alleviate concerns over the comparability of the treatment and control groups.
Data and Summary Statistics
In this section, we provide an overview of our data sources and present summary statistics
for the samples involved in our econometric application.
Data Sources
The USDA-Natural Resources Conservation Service’s (USDA-NRCS) NRI is the primary
data source for our analysis. The NRI is a panel dataset of plot-level land-use observations
covering 1982-2012 in five year increments. Derived from a combination of remote-sensing
analysis and physical site inspection, the NRI data consists of over 300,000 land segments
and 800,000 core sampling points covering the contiguous US (USDA-NRCS, 2012). Each
NRI observation is associated with an expansion factor (or survey weight) which represents
the acreage attributable to that plot.18 From the NRI data, we obtain information on
land use (cultivated/uncultivated cropland, pasture, urban) and irrigation status.19 The
NRI data also provide a number of plot-level attributes measuring land capability class,
slope, erodibility, flood potential, soil hydricity, and prime farmland.20 Prior to estimation,
we adjust the expansion factors to reflect the number of times each plot is stacked in the
estimation data set. For instance, if plot i is located in a county with Ki cross-border
adjacent counties, the expansion factor, xi, is rescaled to x(k)i = xi/Ki prior to estimation.
The expansion factors are readjusted after matching to reflect the number of times each plot
is stacked in the matched samples. Although the NRI plot locations are confidential, we
obtained supplemental data on several important spatial characteristics for the plots in our
sample, including distance to nearby urban areas and distance from each plot to the Kansas
border, which is critical to our identification strategy. Given its granular scale and consistent
18
panel structure, a number of prior studies have used the NRI data to estimate econometric
models of land-use change (e.g., Lubowski, Plantinga, and Stavins, 2006, 2008; Lewis et al.,
2011; Lawler et al., 2014; Claassen, Langpap, and Wu, 2017).
In addition to the NRI data, we obtain information on several county attributes. For
the land development model, we include measures of baseline population density, total pop-
ulation, median household income, and baseline farm real estate value ($/acre). In the
irrigation adoption model, in addition to the farm real estate value, we include baseline
shares of irrigated farmland, farmland rented, and farmland under a corporate ownership
structure. All of these county-level measures enter into our model specification using their
baseline, initial-year (1982) values, making them time-invariant. We do this to avoid the
bias that would result from confounding, for example, a change in population density with
the change in land development due to UVA.
Summary Statistics
Before presenting the estimation results, we present and discuss summary statistics for the
development and irrigation adoption samples.
Development model sample
The baseline sample for the land development model consists of all plots in the Kansas
City CSA that are not developed as of 1982 or in one of seven auxiliary land uses (other
rural land, rural transportation, small water, census water, federal land, and land enrolled
in the Conservation Reserve Program) in any year between 1982 and 2012. We also screen
the sample for plots that are at one point labeled as being developed, but then revert to
an undeveloped use, which reflect outliers that we presume are more likely reflective of
measurement error than a true land-use change. Overall, 5.8% of the development sample
is converted at some point over the 1982-2012 study period, with 5.3% and 6.3% of the
Kansas and Missouri samples being converted, respectively. The development pattern follows
the same general trajectory on both sides of the Kansas border, with conversions peaking
in the 1992-1997 and declining thereafter (table 1).21 The difference for 1982-87 is quite
stark, however, as development increased rapidly in Missouri and quite modestly in Kansas,
19
which highlights the importance of controlling for these pre-UVA trends in the econometric
specification.
We check for covariate balance between the Kansas and Missouri observations before and
after matching using normalized mean differences (NMDs) of the plot characteristics.22 In
the context of matching applications, Imbens and Wooldridge (2009) note that normalized
mean differences in excess of |0.25| can be problematic. Figure 2 displays the NMDs using
a 10-mile bandwidth and the full Kansas City CSA border sample for the matched and
unmatched samples.23 In the majority of cases, matching reduces the observable differences
between plots in Kansas and Missouri. One exception is for the distance to the second
nearest urban area, which is associated with the only NMD above |0.25| in each matched
sample. Other cases where the matched sample produces a larger NMD are few in number
and amount to modest increases with NMDs remaining below |0.25|.
Irrigation model sample
The irrigation adoption model sample consists of all NRI plots that are either solely used for
cropland or used for a mix of cropland and pasture/range for the entire 1982-2012 period.
With the exception of the Kansas-Colorado border, observations are included if they are
located in a county with at least 25% of its area overlying the Ogallala Aquifer. The Kansas-
Colorado border area is excluded from this latter screen because the 25% cutoff excludes some
larger counties that have sufficient aquifer coverage near the state border. Overall, 29% of
the land in the western Kansas sample is irrigated at some point over the 1982-2012 study
period. A total of 4.6% of land converts from dryland to irrigated use, 5.6% converts from
irrigated to dryland use, and 2.4% switches back and forth between the two uses, with the
remaining 87% undergoing no change in irrigated status.
Table 1 shows separate trends in irrigated acreage for each border segment using the total
samples in each county.24 Along the Colorado border, the trends differ early on in the sample
(1982-1997) before trending in similar directions from 1997 to 2012, with both states seeing
fairly substantial declines in irrigated acreage over the 1982-2012 period. For the Nebraska
border segment, the trends diverge substantially for the entire study period, which ends with
20
Nebraska seeing a much larger gain in irrigated acreage compared to Kansas. The aggregate
trends for the Oklahoma border are also opposite, such that the Kansas sample generally
experiences gains in irrigated land when losses occur in Oklahoma, and vice versa, resulting
in a decline in irrigated land in Oklahoma and a gain of equal magnitude in Kansas. Although
we emphasize that these figures convey simple trends with no controls, they indicate that
irrigated acreage has trended differently for the Kansas and neighboring-state plot samples,
particularly for the Nebraska and Oklahoma border segments.25 Similar to the development
samples, matching tends to improve covariate balance for the irrigation samples, particularly
in cases where unmatched NMDs are especially large (figure 3).26
Results
The econometric results are discussed below. Standard errors in each model are clustered at
the plot level to account for serial correlation in plot outcomes and the stacked nature of our
data which causes many plots to appear multiple times in a given year. We present only the
UVA effects in this section. Estimates for the other covariates in the models are provided in
Appendices D and E.
Development Model
The primary estimates of the UVA treatment effect on land development are presented in
table 3. Results indicate that Kansas saw large declines in developed acreage between 1982
and 1987. The 1987 point estimate is largest for the 5-mile buffer, amounting to a 4.2%
reduction in development, and more modest, ranging from -1.2% to -2%, in other years.
Kansas passed its UVA legislation in 1985 but did not formally adopt a UVA policy until
1989, so it is conceivable that some landowners were reluctant to develop their land in light
of the upcoming tax policy change. However, given the five-year gap between observations
in our data, it is not possible with our data to say with any degree of certainty whether or
not the 1987 effect can be attributed to the UVA policy. The point estimates are generally
negative, but noisy, for 1992 and 1997, the first years of our data following implementation
of UVA. From 2002 onward, we find consistently negative and significant effects across all
samples. The estimates indicate that development decreased in Kansas by 1.8-3.4% by 2002,
21
1.1-2% by 2007, and 1-2.7% by 2012. These estimates are always largest for the narrowest
bandwidth (5 miles), smallest for no-bandwidth sample, and largest in 2002. The lack of a
strong, consistent effect of the UVA policy on development until 2002 is potentially explained
by the irreversible nature of farmland conversion. Specifically, there are large option values
associated with the decision to develop farmland (Plantinga, Lubowski, and Stavins, 2002),
which may create a lag in how landowners update their expectations in response to the tax
policy change.
Overall, the matched sample produces results that are consistent with the intended goal
of the UVA policy, namely to provide Kansas farmers with tax relief that slows the con-
version of farmland. Appendix D shows results that illustrate the substantial effect that
matching can have on the UVA point estimates. With the exception of the 1987 effects,
which remain negative, the development model based on the unmatched sample produces
effects with opposite signs and/or significance levels in nearly all cases and broadly sug-
gest that development in Kansas increased in the near-term, followed by no further changes
throughout the study period aside from a small, marginally significant effect for 2012 with
the 5-mile bandwidth. The substantive effects matching can have on the estimation of pol-
icy treatment effects has been documented in previous studies (Ferraro and Miranda, 2014,
2017).
Although matching increases the covariate balance between treatment and control groups,
achieving perfect balance is not a practical goal for policy evaluations based on observational
data. As a robustness check, we assess the extent to which the UVA effects may result from
higher development likelihood for plots with certain characteristics after the UVA policy went
into affect. For example, distance to urban areas may have become more important after the
UVA policy went into effect given the change in incentives brought about by the program.
To account for this possibility, we follow Dempsey and Plantinga (2013) and reestimate the
matched version of equation 5 using a specification that includes interactions between the
plot attributes and a post-UVA dummy variable, which takes a value of 1 for all years after
1989 (i.e., 1992 and beyond). Results from this specification are unchanged, and exhibit
22
slightly improved precision over those derived from the baseline specification (Appendix D).
In Appendix D, we also show that the development results are insensitive to the use of a
sample matched with replacement.
Irrigation Model
Irrigation model estimates for each border segment are provided for the 10-mile, 20-mile,
and no-bandwidth samples in table 4. Estimates from alternative bandwidths are shown
in Appendix E. For Colorado, the UVA effect estimates indicate an increase in irrigation in
Kansas between 1982 and 1987 ranging from 1.8-5.2 but the effects are insignificant in half of
the samples. In 1992, the first year we observe after the UVA policy was in place, irrigation
increased in Kansas by 3.1-7.5%, with larger effects deriving from the smaller bandwidths.
Estimates for years after 1992 are always positive, but the precision is inconsistent across the
different bandwidth samples. The 5- and 10-mile bandwidth samples produce no significant
effects, while those for the 15- and 20-mile samples yield at least some precision in most years.
The larger bandwidths generally do not produce significant effects. UVA effects from the
unmatched samples are similar to the matched ones, but yield significant estimates for the
smaller bandwidths in later years (Appendix E). Estimates from the specification with post-
UVA attribute interactions are mostly positive but generally lack significance (Appendix E).
This suggests that the effects on irrigation outcomes, particularly in later years, may derive
from Kansas farmland being more amenable to irrigation. The 1992 effect for the 10-mile
buffer, however, remains significant at the 5% level.
Results for the Nebraska border segment, generally speaking, are opposite those found
along the Colorado border segment. With few exceptions, we find null effects for 1987 and
1992, as well as the 5- and 10-mile bandwidth samples. For bandwidths of at least 15 miles
and for later years, however, we find significant negative effects ranging from 3.1-10.8%, which
become larger over time but exhibit no clear pattern across the bandwidth sizes. A similar
pattern of negative effects is found with the unmatched samples and after allowing for post-
UVA attribute interactions (Appendix E). Effects for the Oklahoma sample are generally
insignificant, with signs flipping both over time and between different bandwidths. Without
23
matching (Appendix E), several of the earlier negative become significant, while several
additional positive effects emerge for later years. The post-uva interaction specification
produces similarly inconsistent estimates (Appendix E). Allowing for replacement in the
matching algorithm does not meaningfully alter the general conclusions from the irrigation
models (Appendix E).
Overall, the irrigation model estimates provide mixed support for the notion that the
tax break brought about by the UVA policy led to a higher rate of irrigation in Kansas.
The Colorado border segment provides evidence consistent with this potential outcome of
UVA, but we cannot rule out that some of the UVA effects may derive from differences in
plot characteristics which made Kansas farmland more suitable for irrigation. We noted
previously the similarity of farmland tax policies in Kansas and Colorado. In addition, the
institutions governing groundwater use are also quite similar between the two states. Both
states rely on the prior appropriation doctrine to administer groundwater pumping rights.
It is plausible that some minor policy changes in the later portion of our study period may
have influenced the irrigation model results, but, on the whole, Colorado is likely the best
control state for the irrigation model, particularly for 1997 and earlier.27 On balance, the
irrigation results for the Colorado border segment provide suggestive evidence, at best, that
farmers reinvested the tax savings brought about by the UVA program in the form of more
irrigation at the extensive margin.
Results for Nebraska provide no support for the notion that the UVA policy led to more
irrigation in Kansas. Nebraska, however, is perhaps the worst control group of the three
bordering states. In addition to having a far different program for assessing farmland taxes,
Nebraska relies on a correlative rights program to administer groundwater rights, which
essentially ties them to the ownership of land overlying the aquifer, as opposed to seniority.28
Oklahoma also relies on a correlative rights system, but has a farmland tax policy which is
similar in spirit to that of Kansas. For the Oklahoma border segment, the results are the
least consistent, and suggest that the UVA policy may have led to an immediate relative
decline in irrigation in Kansas, but led to a longer term increase in irrigated land throughout
24
the broader border-segment area.
Discussion and Conclusions
This study examines two possible outcomes of use-value assessment, arguably the most
widespread form of land-use regulation in the United States. Under UVA, farmland is taxed
according to its value in perpetual agricultural use, as opposed to its full market value
which reflects returns to converting the land to other uses, such as residential housing.
UVA is intended to reduce farmland owners’ incentives to develop land, but, despite its
ubiquity, there is little rigorous empirical evidence of its effect on land-use decisions. This
is largely due to the fact that most policies were implemented decades ago, when data and
methods necessary to study their effects were not well developed. Relying on a combination
of matching and difference-in-differences analysis using plot-level panel data, our application
exploits the spatial-temporal discontinuity in UVA adoption between Kansas and neighboring
states. The estimated UVA effects provide fairly strong and robust evidence that the UVA
policy reduced development likelihood on the Kansas side of the border of the greater Kansas
City area. The estimated effects dissipate over time, which is consistent with the notion
that UVA policies provide a temporary incentive to, at most, delay development (Hellerstein
et al., 2002). In other words, our results are consistent with the idea that property tax
incentives provide a meaningful lever to influence land-use decisions for a time, but returns
to development will eventually be sufficiently large such that the opportunity cost of not
developing is no longer justifiable on financial grounds. The fact that the development
effects do not emerge until 2002 is consistent with the notion that there are large option
values associated with development decisions, and that it took landowners time to update
their expectations in response to the new tax policy.
Some proponents of preferential taxation have argued that the smaller and more pre-
dictable tax burden brought about by UVA may encourage farmers to make additional
investments in their operation. To this end, we examined the degree to which UVA increases
farm investment by by looking at irrigation adoption in western Kansas. Overall, our results
to this end are mixed, with the most consistent results emerging for the Kansas-Colorado
25
border. While having the financial means to irrigate is likely viewed as a positive from
the perspective of farmers, there also may be external costs associated with increasing the
extensive margin of irrigation. The increased mining of the Ogallala Aquifer, along with po-
tential policy responses, is a topic that has received considerable attention from both natural
scientists and economists (see, e.g., Scanlon et al., 2012; Lin Lawell, 2016). The irrigation
adoption results, however, are not definitive in this regard. For the Colorado border segment,
we are unable to conclude that the bullk of the effects are not attributable to differences
in the attributes of Kansas and Colorado farmland. Of course, irrigation is not the only
investment possibility for landowners affected by the policy. Other possibilities include crop
switching, adoption of improved technology (e.g., drought-tolerant corn), land acquisition,
and alternative investments unrelated to farming.
Ultimately, our study indicates that tax incentives can have an impact on parcel-level
land-use decisions, and our findings carry a number of policy implications. Preferential
farmland taxation policies are not without criticism. To some extent, this controversy stems
from the distributional consequences of shifting a portion of the local tax base from farmland
owners to other landowners within a tax assessment jurisdiction (Anderson and Griffing,
2000). If farmers are investing a portion of their tax savings back into their operation
through irrigation or other land improvements, the additional revenue growth provided by
these investments could partly dampen the magnitude of the shift in the tax base. For
example, irrigated land generally yields higher tax revenues than nonirrigated land, thus
reducing the loss of total tax revenue from farmland after the adoption of UVA. However,
our results provide suggestive evidence, at best, that this may be occurring. While it does
appear that the UVA policy reduced the incentives of farmers to develop their land, it is
unlikely that accomplishing this through a large statewide property tax break is justifiable
from a social cost-benefit standpoint. From a broader research perspective, preferential
farmland tax policy is an understudied topic and there remain numerous aspects that could
be studied in greater detail, including these broader impacts on local public financing, public
goods provision, and investment.
26
Notes
1Most states also have similar provisions for forested land (Anderson, 2012).
2Anderson and England (2014) provide a comprehensive review of the history and char-
acteristics of preferential farmland taxation, including a description of how policies work in
each state.
3Apart from Wisconsin, which adopted its policy in 1995, Wyoming is the only other state
with an adoption year (1989) as recent as that of Kansas (Anderson, Giertz, and Shimul,
2015).
4Historical information on tax rates in Kansas was obtained through personal communi-
cation with the Kansas Department of Revenue.
5Another noteworthy feature of Kansas’s UVA policy is the absence of a development
penalty. In some states with UVA, penalties in the form of repayment of foregone tax
revenues are levied on land that is subject to preferential taxation and then subsequently
developed. Development penalties are also not present in the policies of the states neighbor-
ing Kansas.
6In Missouri, net operating income measures are taken as a ten-year rolling average, and
vary with land use, crop choice, and soil productivity. The net income measures are trans-
lated into a use value using the ten-year moving average of the farm real estate interest
rate reported by the Federal Reserve Bank of Kansas City. Although the Food and Agri-
cultural Policy Research Institute (2007) notes that farmland assessments were based on
a fixed market-value assessment ratio before 1985, farmland tax rates fell precipitously in
Missouri after 1975 at a time when farmland values in Missouri were rising faster than in
any other study area state (U.S. Department of Agriculture, 2014). This suggests that the
1975 legislation did lead to a divergence between farmland taxes and market conditions.
7According to the USDA National Resources Inventory data used in this paper, roughly
27
90% of irrigated land in our western Kansas study area is irrigated with groundwater.
8The use-value format used for agricultural land taxation in Colorado has changed little
over time, with the only major change taking place in 2012, when greater restrictions were
placed on how the provisions were applied to homesteads and other farm-adjacent residential
properties.
9For a detailed history of Nebraska’s farmland tax legislature, see Nebraska Department
of Revenue (2017).
10The Census of Agriculture did not begin asking about property tax expenses until 1987.
11As shown in Appendix A, similar results are obtained if we split the sample based on
the 50th percentile of the 1987 population density distribution.
12As a reviewer points out, however, it is possible that some of these characteristics could
change endogenously with the use of irrigation. Likewise, producers may also switch the
crops they grow when land is irrigated (e.g., corn over wheat), which implies a change to
the relevant components of P and Y that enter the farmer’s observed production decisions.
13Relaxing this constraint would entail a dynamic formulation, similar to that given above
for land development, which incorporates the investment in durable irrigation-related capital.
14Since Kansas passed the original UVA legislation in 1985, it is possible that our results
are influenced by expectations regarding the eventual form of the policy. Our estimates
can be considered conservative to the extent that any effects manifested prior to 1982. For
example, if development in Kansas was artificially low in 1982 due to policy expectations,
any further decline shown in our results would be smaller in magnitude than it would have
been if not for the anticipated policy outcomes.
15This is analogous to the stacking strategy based on the county-level observations in
Dube, Lester, and Reich (2010), where a county is included in the estimating dataset for as
many times as it can be paired with a neighbor on the opposite side of the state border.
28
16See Ferraro and Miranda (2014); Chen, Lewis, and Weber (2016); Ferraro and Miranda
(2017), among others, for recent applications of post-matching DD models.
17While matching without replacement has the advantage of yielding a balanced sample
of treated and untreated observations, a downside is that the matches are not optimal and
sensitive to the order in which the data are sorted (Dehejia and Wahba, 2002). In Appendices
D and E, we show that estimating our main models using a sample matched with replacement
produces similar results.
18NRI expansion factors are calibrated to match total land areas within each county,
given by the most recent estimates from the U.S. Census Bureau, as well as other known
control quantities (e.g., water areas and federal land). Importantly, each time the NRI
is adjusted, weights are recalibrated to match historical land use patterns using a process
known as iterative proportional scaling. For more information on how the NRI weights are
constructed, see Nusser and Goebel (1997); Fuller (1999); Goebel (2009).
19According to USDA-NRCS, 2012, irrigated land in the NRI are defined as "Land that
shows evidence of being irrigated during the year of the inventory or of having been irrigated
during 2 or more of the last 4 years. Water is supplied to crops by ditches, pipes, or other
conduits. For the purposes of the NRI, water spreading is not considered irrigation.”
20As a reviewer points out, one potentially important factor we do not observe is elevation,
which may be particularly relevant for the wide bandwidths along the Colorado border
segment. Since Colorado is, in general, at a higher elevation than Kansas, the aggregate
difference in elevation should be accounted for by the baseline Kansas effect, αS. To some
extent, the other soil-related variables (wind and water erosion and slope) should further
proxy for elevation patterns within each state.
21Development changes for the different bandwidths (5, 10, 15, 20, and 25 miles) are
provided in Appendix B. With few exceptions, the trends described here for the full samples
are similar for the subsamples contained within each border bandwidth.
29
22In addition to the plot characteristics included in the development and irrigation models,
the propensity score logit models also include the NRI expansion factor as an additional
covariate to control for plot size. The logit models themselves, however, are not weighted.
Results from the propensity score models are given in Appendix C. Mean values of the
development model county attributes for Kansas and Missouri are also presented in Appendix
C.
23NMD figures for the other bandwidths are presented in Appendix C.
24Irrigated acreage changes for the other bandwidths are presented in Appendix B.
25According to county-level data from the Census of Agriculture, the Colorado counties
in the study area saw a 13% net decline in irrigation over the 1982-2012 period, Nebraska
counties saw an increase of 23%, and the Oklahoma counties saw a 2% decline. The Kansas
counties along the Colorado, Nebraska, and Oklahoma border segments were subject to
net changes of -11%, 9%, and 31%, respectively. Thus, from an aggregate standpoint, the
Colorado border segment samples most closely match one another in terms of aggregate
irrigated acreage change over the study period.
26See Appendix C for the irrigation sample NMDs for alternative bandwidths. Note that
the dummy variable for flood potential drops out of the matched 5-mile Oklahoma sample
due to perfect collinearity. State-level mean values of the county attributes included in the
irrigation adoption models are also presented in Appendix C.
27The other policy changes include local incentives for high-efficiency drop nozzle sprin-
klers in Kansas brought about through federal and local state initiatives starting in 1998
(Pfeiffer and Lin, 2014), though the federal program was likely also available to Colorado
farmers. Moreover, our summary statistics indicate that irrigated area actually dropped
more in Kansas than Colorado after 1992 (table 1). There was, however, a small water
rights retirement program which went into place in Colorado in 2006 may have partly con-
tributed to this decline (Monger et al., 2018).
30
28Nebraska’s groundwater institutions have also undergone a number of relatively recent
changes related to groundwater conservation policy. For the Republican River Basin, the area
adjacent to western Kansas, regulation timing varied, with some natural resource districts
adopting volumetric irrigation as recently as 2004.
31
References
Anderson, J.E. 2012. “Agricultural use-value property tax assessment: Estimation and policy
issues.” Public Budgeting & Finance 32:71–94.
—. 1993. “Use-value property tax assessment: Effects on land development.” Land Economics
69:362–369.
Anderson, J.E., and R.W. England. 2014. Use-Value Assessment of Rural Land in the United
States. Lincoln Institute of Land Policy.
Anderson, J.E., S.H. Giertz, and S.N. Shimul. 2015. “Property taxes for agriculture: Use-
value assessment and urbanization across the United States.” Unpublished, Mercatus
Working Paper.
Anderson, J.E., and M.F. Griffing. 2000. “Use-value assessment tax expenditures in urban
areas.” Journal of Urban Economics 48:443–452.
Bigelow, D.P., A.J. Plantinga, D.J. Lewis, and C. Langpap. 2017. “How does urbanization
affect water withdrawals? Insights from an econometric-based landscape simulation.” Land
Economics 93:413–436.
Blase, M.G., and W.J. Staub. 1971. “Real property taxes in the rural-urban fringe.” Land
Economics 47:168–174.
Butsic, V., D.J. Lewis, and L. Ludwig. 2011. “An econometric analysis of land development
with endogenous zoning.” Land Economics 87:412–432.
Chambers, R.G., and R.E. Just. 1989. “Estimating multioutput technologies.” American
Journal of Agricultural Economics 71:980–995.
Chen, Y., D.J. Lewis, and B. Weber. 2016. “Conservation land amenities and regional
economies: A postmatching difference-in-differences analysis of the Northwest Forest
Plan.” Journal of Regional Science 56:373–394.
32
Chicoine, D.L., and A.D. Hendricks. 1985. “Evidence on farm use value assessment, tax
shifts, and state school aid.” American Journal of Agricultural Economics 67:266–270.
Claassen, R., C. Langpap, and J. Wu. 2017. “Impacts of Federal Crop Insurance on Land Use
and Environmental Quality.” American Journal of Agricultural Economics 99:592–613.
Colorado Assessors Association. 2019. “How Agricultural Property is Valued.”
Conklin, H.E., and W.G. Lesher. 1977. “Farm-value assessment as a means for reducing
premature and excessive agricultural disinvestment in urban fringes.” American Journal
of Agricultural Economics 59:755–759.
Coughlin, R.E., D. Berry, and T. Plaut. 1978. “Differential assessment of real property as
an incentive to open space preservation and farmland retention.” National Tax Journal
31:165–179.
Dehejia, R.H., and S. Wahba. 2002. “Propensity score-matching methods for nonexperimen-
tal causal studies.” The Review of Economics and Statistics 84:151–161.
Dempsey, J.A., and A.J. Plantinga. 2013. “How well do urban growth boundaries contain
development? Results for Oregon using a difference-in-difference estimator.” Regional Sci-
ence and Urban Economics 43:996–1007.
Dempsey, J.A., A.J. Plantinga, J.D. Kline, J.J. Lawler, S. Martinuzzi, V.C. Radeloff, and
D.P. Bigelow. 2017. “Effects of local land-use planning on development and disturbance
in riparian areas.” Land Use Policy 60:16–25.
Dinterman, R., and A.L. Katchova. 2019. “Property tax incidence on cropland cash rent.”
Applied Economics Perspectives and Policy in press:pp. 21.
Dube, A., T.W. Lester, and M. Reich. 2010. “Minimum wage effects across state boundaries:
Estimates using contiguous counties.” The Review of Economics and Statistics 92:945–964.
Dunford, R.W., and D.C. Marousek. 1981. “Sub-county property tax shifts attributable to
use-value assessments on farmland.” Land Economics 57:221–229.
33
Ferraro, P.J., and J.J. Miranda. 2017. “Panel data designs and estimators as substitutes for
randomized control trials in the evaluation of public programs.” Journal of the Association
of Environmental and Resource Economists 4:281–317.
—. 2014. “The performance of non-experimental designs in the evaluation of environmen-
tal policy: A design-replication study using a large-scale randomized experiment as a
benchmark.” Journal of Economic Behavior and Organization 107:344–365.
Food and Agricultural Policy Research Institute. 2007. “Calculating Agricultural Use Values
for Missouri Farmland.”
Fuller, W.A. 1999. “Estimation procedures for the United State National Resources Inven-
tory.” Annual conference proceedings, Statistical Society of Canada.
Geshkov, M.V., and J.S. DeSalvo. 2012. “The effect of land-use controls on the spatial size
of U.S. urbanized areas.” Journal of Regional Science 52:648–675.
Goebel, J.J. 2009. Working paper, USDA-Natural Resources Conservation Service.
Goodman-Bacon, A. 2018. “Difference-in-differences with variation in treatment timing.”
Working paper no. 25018, National Bureau of Economic Research.
Hapgood, J. 2010. “Agricultural Use Value (Presentation).”
Hellerstein, D., C. Nickerson, J. Cooper, P. Feather, D. Gadsby, D. Mullarkey, A. Tegene,
and C. Barnard. 2002. “Farmland Protection: The Role of Public Preferences for Ru-
ral Amenities.” Working paper No. AER-815, U.S. Department of Agriculture, Economic
Research Service.
Hildreth, W.B. 2018. “State-by-State Property Tax at a Glance, Kansas.” Working paper,
Lincoln Institute of Land Policy.
Hornbeck, R., and P. Keskin. 2014. “The historically evolving impact of the Ogallala Aquifer:
Agricultural adaptation to groundwater and drought.” American Economic Journal: Ap-
plied Economics 6:190–219.
34
Imbens, G.W., and J.M. Wooldridge. 2009. “Recent developments in the econometrics of
program evaluation.” Journal of Economic Literature 47:5–86.
Kansas Department of Revenue. 2018. “States Governing Ag Use 79-1476.”
Lapping, M.B., R.J. Bevins, and P.V. Herbers. 1977. “Differential assessment and other
techniques to preserve Missouri’s farmlands.” Missouri Law Review 42:369–408.
Lawler, J.J., D.J. Lewis, E. Nelson, A.J. Plantinga, S. Polasky, J.C. Withey, D.P. Helmers,
S. Martinuzzi, D. Pennington, and V.C. Radeloff. 2014. “Projected land-use change im-
pacts on ecosystem services in the United States.” Proceedings of the National Academy
of Sciences 111:7492–7497.
Lewis, D.J., A.J. Plantinga, E. Nelson, and S. Polasky. 2011. “The efficiency of voluntary
incentive policies for preventing biodiversity loss.” Resource and Energy Economics 33:192–
211.
Lin Lawell, C.Y.C. 2016. “The management of groundwater: Irrigation efficiency, policy,
institutions, and externalities.” Annual Review of Resource Economics 8:247–259.
Lubowski, R.N., A.J. Plantinga, and R.N. Stavins. 2006. “Land-use change and carbon
sinks: Econometric estimation of the carbon sequestration supply function.” Journal of
Environmental Economics and Management 51:135–152.
—. 2008. “What drives land-use change in the United States? A national analysis of
landowner decisions.” Land Economics 84:529–550.
Midwestern Office of the Council of State Governments. 2012. “The Hows and Whys of
Taxing Farmland.”
Monger, R.G., J.F. Suter, D.T. Manning, and J.P. Schneekloth. 2018. “Retiring land to save
water: Participation in Colorado’s Republican River Conservation Reserve Enhancement
Program.” Land Economics 94:36–51.
35
Morris, A.C. 1998. “Property tax treatment of farmland: Does tax relief delay land devel-
opment?” In W. E. Oates, ed. Local Government and Tax and Land Use Policies in the
United States: Understanding the Links. Lincoln Institute of Land Policy, pp. 144–167.
Nebraska Department of Revenue. 2017. “2017 Annual Report.”
Nechyba, T.J., and R.P. Walsh. 2004. “Urban sprawl.” Journal of Economic Perspectives
18(4):177–200.
Negri, D.H., N.R. Gollehon, and M.P. Aillery. 2005. “The effects of climatic variability on
US irrigation adoption.” Climatic Change 69:299–323.
Nusser, S., and J. Goebel. 1997. “The National Resources Inventory: a long-term multi-
resource monitoring programme.” Environmental and Ecological Statistics 4:181–204.
Pfeiffer, L., and C.C. Lin. 2014. “Does efficient irrigation technology lead to reduced ground-
water extraction? Empirical evidence.” Journal of Environmental Economics and Man-
agement 67:189–208.
Plantinga, A.J., R.N. Lubowski, and R.N. Stavins. 2002. “The effects of potential land
development on agricultural land prices.” Journal of urban economics 52:561–581.
Polyakov, M., and D. Zhang. 2008. “Property tax policy and land-use change.” Land Eco-
nomics 84:396–408.
Quigley, J.M., and L.A. Rosenthal. 2005. “The effects of land use regulation on the price
of housing: What do we know? What can we learn?” Cityscape: A Journal of Policy
Development and Research 8:69–137.
Scanlon, B.R., C.C. Faunt, L. Longuevergne, W.M. Reedy, Robert C. amd Alley, V.L.
McGuire, and P.B. McMahon. 2012. “Groundwater depletion and sustainability of irri-
gation in the US High Plains and Central Valley.” Proceedings of the National Academy of
Sciences 109:9320–9325.
36
Schwartz, S., D. Hansen, and T. Foin. 1975. “Preferential taxation and the control of urban
sprawl: An analysis of the California Land Conservation Act.” Journal of Environmental
Economics and Management 2:120–134.
Schwartz, S., D.E. Hansen, and T. Foin. 1976. “Landowner benefits from use-value assess-
ment under the California Land Conservation Act.” American Journal of Agricultural
Economics 58:170–178.
Stavins, R.N., and A.B. Jaffe. 1990. “Unintended impacts of public investments on private
decisions: The depletion of forested wetlands.” American Economic Review 80:337–352.
Towe, C.A., H.A. Klaiber, and D.H. Wrenn. 2017. “Not my problem: Growth spillovers from
uncoordinated land use policy.” Land Use Policy 67:679–689.
U.S. Department of Agriculture, E.R.S. 2014. “Farm Real Estate Historical Data Series.”
U.S. Department of Agriculture, Natural Resources Conservation Service (USDA-NRCS).
2015. “2012 National Resources Inventory Summary Report.”
Wrenn, D.H., and E.G. Irwin. 2015. “Time is money: An empirical examination of the effects
of regulatory delay on residential subdivision development.” Regional Science and Urban
Economics 51:25–36.
Zhang, W., and C.J. Nickerson. 2015. “Housing market bust and farmland values: Identifying
the changing influence of proximity to urban centers.” Land Economics 91:605–626.
Zhang, W., D.H. Wrenn, and E.G. Irwin. 2017. “Spatial heterogeneity, accessibility, and zon-
ing: An empirical investigation of leapfrog development.” Journal of Economic Geography
17:547–570.
37
Figures
Figure 1. Kansas City Combined Statistical Area and Ogallala Aquifer
Note: The polygons with the lattice overlay represent the counties included in our samples. In parentheses
under each state name is the year in which the state adopted its original use-value assessment policy.
38
Figure 2. Normalized mean differences in plot covariates for Kansas City CSA development
model sample, 10-mile bandwidth and full sample
39
Figure 3. Normalized mean differences in plot covariates for 10-mile bandwidth and full
county samples, irrigation model border segments
40
Tables
Table 1. Changes in Outcome Variables, by Treatment and Control Group, Full Samples
Five-year period ending: 1987 1992 1997 2002 2007 2012 Total
Development:
KS 2.2 15.8 20.9 13.3 8.0 2.6 62.8
MO 20.0 11.5 31.3 21.0 10.9 6.6 101.3
Irrigation, Colorado segment:
KS -7.2 5.7 -12.6 -49.1 -5.3 -6.8 -75.3
CO -20.5 -42.5 14.7 -14.1 4.2 -38.7 -96.9
Irrigation, Nebraska segment:
KS -0.8 13.8 12.3 -10.8 2.1 -0.8 15.8
NE 39.0 8.5 19.2 25.9 9.1 -3.4 98.3
Irrigation, Oklahoma segment:
KS -2.0 -7.2 11.7 15.9 4.4 2.7 25.5
OK 11.0 10.7 -9.6 -30.5 4.9 -13.0 -26.5
41
Table 2. Effect of Use-Value Assessment on Various Farmland Tax Variables, 1987-1992
Variable All counties Rural counties Urban counties
Log(Farmland property tax) -0.133 -0.145 -0.093
(0.025)*** (0.030)*** (0.030)***
[-12.49] [-13.52] [-8.94]
Log(Farmland property tax per acre) -0.146 -0.157 -0.106
(0.028)*** (0.032)*** (0.037)***
[-13.62] [-14.55] [-10.15]
Log(Farmland property tax per farm) -0.123 -0.131 -0.093
(0.029)*** (0.034)*** (0.040)**
[-11.57] [-12.34] [-8.96]
Log(Effective farmland property tax rate) -0.128 -0.140 -0.094
(0.042)*** (0.050)*** (0.060)
[-12.10] [-13.16] [-9.14]
Observations 886 664 222
Counties 443 332 111
Notes: These results were generated from a regression of the various variables listed in the
leftmost column on a set of county fixed effects, a 1992 dummy variable, and an interaction
between the 1992 dummy variable and a dummy variable denoting counties in Kansas. The
interaction term coefficients represent the effects of the use-value assessment policy and are
reported in the above table. Standard errors, reported in parentheses, are clustered by USDA
agricultural district. The bracketed terms represent the marginal effect of the UVA policy in
percentage terms. Urban and rural counties were determined based on the 75th percentile
of the 1987 distribution of population density.
42
Table 3. Land Development Model, UVA Effect Estimates by Year and Bandwidth,
Matched Sample
Bandwidth (mi.) 5 10 15 20 25 None
1987 -0.042 -0.014 -0.018 -0.020 -0.015 -0.012
(0.016)*** (0.005)*** (0.005)*** (0.005)*** (0.004)*** (0.003)***
1992 -0.013 <0.001 -0.005 -0.008 -0.007 -0.006
(0.016) (0.007) (0.007) (0.006) (0.005) (0.005)
1997 0.006 0.007 -0.007 -0.001 -0.010 -0.007
(0.009) (0.009) (0.009) (0.008) (0.007) (0.006)
2002 -0.034 -0.024 -0.024 -0.019 -0.020 -0.018
(0.012)*** (0.009)*** (0.007)*** (0.005)*** (0.005)*** (0.004)***
2007 -0.020 -0.013 -0.018 -0.013 -0.012 -0.011
(0.008)** (0.004)*** (0.004)*** (0.003)*** (0.003)*** (0.003)***
2012 -0.027 -0.018 -0.019 -0.013 -0.010 -0.009
(0.013)** (0.006)*** (0.005)*** (0.004)*** (0.003)*** (0.003)***
No. observations 2,420 5,994 8,872 10,782 12,500 12,758
No. plots (w/ repeats) 386 936 1,384 1,676 1,922 1,954
No. plots (no repeats) 263 631 908 1,120 1,280 1,307
Notes: The dependent variable in the above models is binary and takes a value of 1 once a plot is
developed, at which point it is removed from the sample. Asterisks denote significance at the 10% (*),
5% (**), and 1%(***) levels. Standard errors are clustered by National Resources Inventory (NRI)
plot. The baseline year is 1982 in all models. Estimates are weighted using NRI expansion factors
divided for the number of times each plot is stacked due to its county having multiple contiguous
cross-border counties. All models include a set of contiguous-county border-pair dummy variables,
time fixed effects, border-pair time fixed effects, a baseline Kansas dummy variable, a constant term,
and controls representing plot characteristics and county attributes.
43
44
Table 4. Irrigation Model, UVA Effect Estimates by Year, Border Segment, and Bandwidth, Matched Sample
Segment: Colorado Nebraska Oklahoma
Bandwidth (mi.): 10 20 None 10 20 None 10 20 None
1987 0.032 0.016 0.018 -0.017 -0.021 -0.018 -0.012 -0.020 -0.010
(0.021) (0.013) (0.010)* (0.011) (0.013) (0.009)** (0.011) (0.010)** (0.005)*
1992 0.075 0.039 0.031 -0.002 -0.040 -0.019 -0.027 -0.025 -0.015
(0.024)*** (0.016)** (0.013)** (0.015) (0.020)** (0.012) (0.020) (0.016) (0.013)
1997 0.041 0.039 0.034 -0.002 -0.071 -0.031 -0.009 -0.027 0.023
(0.032) (0.020)* (0.018)* (0.015) (0.023)*** (0.013)** (0.027) (0.019) (0.023)
2002 0.044 0.053 0.027 -0.017 -0.100 -0.056 0.027 -0.005 0.061
(0.029) (0.020)** (0.019) (0.017) (0.025)*** (0.016)*** (0.041) (0.032) (0.027)**
2007 0.048 0.040 0.022 -0.033 -0.108 -0.067 0.044 0.011 0.056
(0.030) (0.021)* (0.019) (0.027) (0.030)*** (0.020)*** (0.041) (0.033) (0.023)**
2012 0.036 0.042 0.029 -0.037 -0.108 -0.071 0.080 0.028 0.073
(0.031) (0.021)** (0.020) (0.027) (0.030)*** (0.021)*** (0.038)** (0.031) (0.027)***
No. observations 2,898 5,460 7,854 2,359 4,466 9,772 2,583 5,173 10,304
No. plots (w/ repeats) 414 780 1,122 337 638 1,396 369 739 1,472
No. plots (no repeats) 277 513 740 233 418 848 277 510 931
Notes: The dependent variable in the above models is binary and takes a value of 1 if a plot is irrigated. Asterisks denote significance at
the 10% (*), 5% (**), and 1%(***) levels. Standard errors are clustered by National Resources Inventory (NRI) plot. The baseline year
is 1982 in all models. Estimates are weighted using NRI expansion factors divided for the number of times each plot is stacked due to its
county having multiple contiguous cross-border counties. All models include a set of contiguous-county border-pair dummy variables,
time fixed effects, border-pair time fixed effects, a baseline Kansas dummy variable, a constant term, and controls representing plot
characteristics and county attributes.
Title: “AJAE appendix for A Tale of Two Borders: Use-Value Assessment, Land
Development, and Irrigation Investment”
Authors: Daniel P. Bigelow and Todd Kuethe
Date: February 11, 2020
Note: The material contained herein is supplementary to the article named in the title and
published in the American Journal of Agricultural Economics (AJAE).
1
Appendix A
(a) Farm real estate taxes per $100 of farm (b) Farm real estate tax expenses per farm
real estate value (effective tax rate)
(c) Farm real estate tax expenses per acre (d) Total farm real estate tax expenses
Figure A1. State-level trends in farm real estate value taxation, 1980-1993
2
3
Table A1. Effect of Use-Value Assessment on Various Farmland Tax Variables, 1987-1992, Alternative Estimates
Weighted by farmland acreage Alternative rural and urban split
Variable All counties Rural counties Urban counties Rural counties Urban counties
Log (Farmland property tax) -0.131 -0.136 -0.100 -0.142 -0.114
(0.028)*** (0.033)*** (0.030)*** (0.035)*** (0.028)***
[-12.27] [-12.79] [-9.57] [-13.33] [-10.84]
Log (Farmland property tax per acre) -0.140 -0.144 -0.114 -0.153 -0.131
(0.031)*** (0.036)*** (0.035)*** (0.037)*** (0.029)***
[-13.07] [-13.46] [-10.82] [-14.26] [-12.28]
Log (Farmland property tax per farm) -0.114 -0.119 -0.092 -0.135 -0.104
(0.034)*** (0.038)*** (0.045)* (0.039)*** (0.030)***
[-10.82] [-11.26] [-8.84] [-12.74] [-9.9]
Log (Effective farmland property tax rate) -0.157 -0.169 -0.107 -0.163 -0.087
(0.054)*** (0.062)*** (0.058)* (0.063)** (0.046)*
[-14.69] [-15.72] [-10.33] [-15.19] [-8.44]
Observations 886 664 222 442 444
Counties 443 332 111 221 222
Notes: These results were generated from a regression of the various variables listed in the leftmost column on a set of county
fixed effects, a 1992 dummy variable, and an interaction between the 1992 dummy variable and a dummy variable denoting
counties in Kansas. In the first three columns. all observations are weighted by the acreage of farmland in the county reported
for 1987. The interaction term coefficients represent the effects of the use-value assessment policy and are reported in the above
table. Standard errors, reported in parentheses, are clustered by USDA agricultural district. The bracketed terms represent the
marginal effect of the UVA policy in percentage terms. Urban and rural counties were determined based on the 75th percentile
of the 1987 population density distribution. In the final two columns, the urban and rural counties are designated by the 50th,
rather than the 75th percentile of the 1987 population density distribution.
Appendix B
Table B1. NRI Changes in Developed Acreage (1000s) in Kansas City CSA by Bandwidth
Five-year period ending: 1987 1992 1997 2002 2007 2012 Total
5-mile bandwidth
KS 0.5 5.0 4.2 1.6 1.7 0.7 13.7
MO 5.4 4.5 1.3 2.4 0.9 2.2 16.7
10-mile bandwidth
KS 1.9 10.4 14.0 8.8 4.8 1.5 41.4
MO 9.2 5.5 10.7 11.5 6.5 5.2 48.6
15-mile bandwidth
KS 2.1 13.0 18.2 10.7 6.1 1.9 52.0
MO 13.2 8.6 19.0 15.1 8.2 5.9 70.0
20-mile bandwidth
KS 2.1 15.5 20.2 12.7 7.7 2.4 60.6
MO 19.8 10.4 28.5 18.9 10.4 6.5 94.5
25-mile bandwidth
KS 2.2 15.8 20.9 13.3 8.0 2.6 62.8
MO 20.0 11.5 30.0 21.0 10.9 6.6 100.0
4
Table B2. NRI Changes in Irrigated Acreage (1000s) by Bandwidth, Colorado Border
Segment
Five-year period ending: 1987 1992 1997 2002 2007 2012 Total
5-mile bandwidth
KS 0.0 16.0 -4.4 -0.9 0.0 0.0 10.7
CO -14.0 -4.4 11.0 -7.9 -2.9 -2.9 -21.1
10-mile bandwidth
KS 1.3 17.3 -14.6 -7.0 -5.8 0.0 -8.8
CO -21.6 -9.4 22.7 -11.6 -9.6 -2.9 -32.4
15-mile bandwidth
KS 1.3 11.7 -12.1 -13.2 -5.6 0.0 -17.9
CO -17.1 -34.6 16.5 -19.4 -5.1 -24.4 -84.1
20-mile bandwidth
KS 1.3 10.7 -14.5 -50.4 -8.6 -1.0 -62.5
CO -15.3 -35.6 1.3 -33.4 -0.1 -14.4 -97.5
25-mile bandwidth
KS 1.3 10.4 -19.4 -54.5 -5.3 -6.8 -74.3
CO -15.3 -40.0 7.1 -34.8 -3.4 -14.4 -100.8
5
Table B3. NRI Changes in Irrigated Acreage (1000s) by Bandwidth, Nebraska Border
Segment
Five-year period ending: 1987 1992 1997 2002 2007 2012 Total
5-mile bandwidth
KS -0.8 4.1 3.4 -6.7 2.5 0.0 2.5
NE -0.3 -2.8 2.8 1.4 -0.7 2.7 3.1
10-mile bandwidth
KS -0.8 4.1 3.4 -6.7 -1.2 0.0 -1.2
NE 5.4 -4.0 -0.8 0.0 3.9 2.8 7.3
15-mile bandwidth
KS -0.8 4.1 3.4 -6.7 -1.2 0.0 -1.2
NE 16.0 -4.9 10.0 11.5 5.7 2.8 41.1
20-mile bandwidth
KS -0.8 4.1 3.4 -2.1 2.1 0.0 6.7
NE 30.5 7.1 10.8 22.1 11.7 -3.6 78.6
25-mile bandwidth
KS -0.8 7.5 7.3 -7.2 2.1 0.0 8.9
NE 39.0 8.5 19.2 25.9 9.1 -3.4 98.3
6
Table B4. NRI Changes in Irrigated Acreage (1000s) by Bandwidth, Oklahoma Border
Segment
Five-year period ending: 1987 1992 1997 2002 2007 2012 Total
5-mile bandwidth
KS 1.3 1.8 8.9 -0.5 -0.2 2.5 13.8
OK 2.0 8.3 4.3 -8.9 -2.0 -4.2 -0.5
10-mile bandwidth
KS 1.3 3.0 18.6 35.1 2.0 3.7 63.7
OK 3.7 15.8 4.3 -9.4 2.5 -4.5 12.4
15-mile bandwidth
KS 1.3 4.1 24.3 33.8 3.1 3.7 70.3
OK 9.0 11.4 4.3 -2.7 -4.2 -4.5 13.3
20-mile bandwidth
KS 1.3 2.1 20.4 33.5 1.4 2.7 61.4
OK 9.0 11.4 6.3 -3.8 -4.2 -4.5 14.2
25-mile bandwidth
KS -2.0 -5.6 17.0 21.5 0.5 2.7 34.1
OK 9.0 9.1 6.0 -14.0 -6.2 -4.5 -0.6
7
Appendix C
Additional balance plots and propensity score model results
Figure C1. Normalized mean differences in plot covariates for Kansas City CSA
development model sample, alternative bandwidths
8
Figure C2. Normalized mean differences in plot covariates for alternative bandwidths,
Colorado border segment
9
Figure C3. Normalized mean differences in plot covariates for alternative bandwidths,
Nebraska border segment
10
Figure C4. Normalized mean differences in plot covariates for alternative bandwidths,
Oklahoma border segment
11
Table C1. Means of Baseline (1982) County Attributes Used in Development Models
Kansas Missouri
Farm real estate value ($/acre) 1934 1953
Median household income ($1000s) 39.11 38.07
Population (1000s) 53.00 69.64
Population density 0.21 0.20
12
Table C2. Means of Baseline (1982) County Attributes Used in Irrigation Models, by
Border Segment
Colorado Nebraska Oklahoma
KS CO KS NE KS OK
Farm real estate value ($/acre) 993 682 995 1446 1018 811
Irrigated farmland share 0.12 0.09 0.02 0.14 0.16 0.07
Rented farmland share 0.49 0.37 0.47 0.45 0.57 0.46
Corporate-owned farmland share 0.06 0.04 0.00 0.06 0.09 0.10
13
Table C3. Propensity Score Logit Model Estimates, Development Sample
Bandwidth (mi.) 5 10 15 20 25 None
LCC 1 0.916 -0.015 -0.169 -0.645 -0.554 -0.436
(1.156) (0.722) (0.556) (0.489) (0.458) (0.445)
LCC 2 0.336 -0.309 -0.606 -1.189 -1.328 -1.371
(0.387) (0.277) (0.247)** (0.222)*** (0.207)*** (0.206)***
LCC 3 0.554 0.344 -0.240 -0.635 -0.675 -0.683
(0.340) (0.249) (0.220) (0.199)*** (0.188)*** (0.187)***
LCC 4 -0.151 0.433 -0.129 -0.396 -0.449 -0.449
(0.381) (0.270) (0.231) (0.210)* (0.200)** (0.199)**
Slope 0.004 -0.027 -0.074 -0.113 -0.121 -0.124
(0.023) (0.017) (0.016)*** (0.015)*** (0.014)*** (0.014)***
Hydric soil -3.779 -2.364 -2.392 -2.203 -2.196 -2.201
(1.063)*** (0.276)*** (0.224)*** (0.198)*** (0.183)*** (0.181)***
Flood potential -0.685 0.294 0.247 -0.018 0.049 0.074
(0.340)** (0.228) (0.197) (0.173) (0.162) (0.161)
Dist. nearest city 0.117 0.188 0.210 0.193 0.188 0.181
(0.034)*** (0.022)*** (0.019)*** (0.017)*** (0.016)*** (0.016)***
Dist. sec. nearest city 0.009 -0.018 -0.046 -0.045 -0.041 -0.042
(0.029) (0.016) (0.014)*** (0.013)*** (0.012)*** (0.012)***
Dist. KS border -0.138 -0.052 -0.053 -0.051 -0.049 -0.058
(0.079)* (0.025)** (0.014)*** (0.010)*** (0.008)*** (0.007)***
Expansion factor -0.179 -0.168 -0.173 -0.161 -0.161 -0.164
(0.023)*** (0.014)*** (0.011)*** (0.010)*** (0.009)*** (0.009)***
No. plots 595 1,330 1,912 2,393 2,709 2,770
Notes: The dependent variable in the above models is binary and takes a value of 1 if a plot is located
in Kansas. Results from these models are used to construct matched samples which form the basis
for the main development estimates shown and discussed in the paper (see Table 3). Asterisks denote
significance at the 10% (*), 5% (**), and 1%(***) levels
14
Table C4. Propensity Score Logit Model Estimates, Irrigation Sample
Bandwidth (mi.) 5 10 15 20 25 None
LCC 1 3.989 3.939 4.056 3.863 3.466 3.354
(0.815)*** (0.508)*** (0.400)*** (0.321)*** (0.268)*** (0.238)***
LCC 2 2.259 2.413 2.513 2.340 1.883 1.490
(0.642)*** (0.421)*** (0.335)*** (0.270)*** (0.230)*** (0.200)***
LCC 3 1.563 1.906 2.320 2.070 1.853 1.288
(0.598)*** (0.388)*** (0.313)*** (0.247)*** (0.210)*** (0.182)***
LCC 4 0.002 0.435 0.579 0.190 -0.116 -0.309
(0.603) (0.380) (0.306)* (0.246) (0.216) (0.191)
Prime farmland -1.592 -1.635 -1.328 -1.331 -1.297 -1.243
(0.301)*** (0.195)*** (0.151)*** (0.134)*** (0.116)*** (0.100)***
Slope 0.126 0.174 0.128 0.122 0.049 0.026
(0.093) (0.066)*** (0.052)** (0.048)** (0.043) (0.037)
Flood potential -0.424 -0.770 -0.472 -0.392 -0.313 -0.414
(0.414) (0.281)*** (0.225)** (0.196)** (0.171)* (0.159)***
Wind erosion 0.046 0.043 0.045 0.050 0.046 0.014
(0.016)*** (0.011)*** (0.009)*** (0.008)*** (0.007)*** (0.006)**
Water erosion -0.031 -0.076 -0.058 -0.056 -0.030 -0.015
(0.065) (0.043)* (0.034)* (0.031)* (0.028) (0.024)
Dist. KS border 0.028 0.066 0.021 0.008 0.018 -0.013
(0.067) (0.023)*** (0.012)* (0.008) (0.006)*** (0.003)***
Expansion factor 0.005 0.007 0.009 0.010 0.014 0.001
(0.006) (0.004)* (0.004)** (0.003)*** (0.003)*** (0.003)
No. plots 570 1,255 1,876 2,488 3,178 3,953
Notes: The dependent variable in the above models is binary and takes a value of 1 if a plot is
located in Kansas. Results from these models are used to construct matched samples which form
the basis for the main development estimates shown and discussed in the paper (see Tables 4 and
E1). Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels.
15
Appendix D
Table D1. Land Development Model, Control Variable Estimates by Year and Bandwidth,
Matched Sample
Bandwidth (mi.) 5 10 15 20 25 None
LCC 1 -0.023 -0.027 -0.019 -0.013 -0.003 -0.005
(0.010)** (0.008)*** (0.012) (0.018) (0.007) (0.006)
LCC 2 -0.001 -0.002 -0.005 0.002 0.001 -0.001
(0.009) (0.006) (0.006) (0.006) (0.004) (0.004)
LCC 3 -0.013 -0.009 -0.008 -0.001 -0.001 -0.004
(0.008)* (0.005)* (0.005) (0.005) (0.004) (0.003)
LCC 4 -0.004 -0.013 -0.010 -0.004 -0.004 -0.007
(0.013) (0.007)* (0.006)* (0.005) (0.004) (0.004)*
Slope <0.001 -0.001 <0.001 <0.001 <0.001 <0.001
(0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
Hydric soil -0.009 0.006 -0.003 0.007 0.006 0.006
(0.012) (0.007) (0.005) (0.004) (0.003)* (0.003)*
Flood potential 0.002 -0.004 >-0.001 -0.008 -0.006 -0.007
(0.011) (0.007) (0.005) (0.004)* (0.003)* (0.003)**
Dist. nearest city -0.003 -0.003 -0.002 -0.002 -0.002 -0.002
(0.001)*** (<0.001)*** (<0.001)*** (<0.001)*** (<0.001)*** (<0.001)***
Dist. sec. nearest city <0.001 0.001 0.001 0.001 <0.001 <0.001
(0.001) (<0.001)*** (<0.001) (<0.001)** (<0.001) (<0.001)
Dist. KS border <0.001 -0.001 -0.001 >-0.001 >-0.001 >-0.001
(0.002) (<0.001) (<0.001)* (<0.001) (<0.001)*** (<0.001)***
Pop. density 0.005 -0.041 -0.064 -0.049 -0.033 -0.035
(0.027) (0.021)* (0.017)*** (0.016)*** (0.015)** (0.013)***
Median income 0.002 0.001 0.001 0.001 0.001 0.001
(0.001)** (<0.001)* (<0.001)* (<0.001)*** (<0.001)*** (<0.001)**
Population <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
(<0.001)*** (<0.001)*** (<0.001)*** (<0.001)*** (<0.001)*** (<0.001)***
Farm real est. value >-0.001 <0.001 >-0.001 >-0.001 >-0.001 >-0.001
(<0.001)** (<0.001) (<0.001) (<0.001) (<0.001)** (<0.001)**
No. observations 2,420 5,994 8,872 10,782 12,500 12,758
No. plots (w/ repeats) 386 936 1,384 1,676 1,922 1,954
No. plots (no repeats) 263 631 908 1,120 1,280 1,307
Notes: Control estimates displayed above correspond to the models used to produce the UVA effects in Table 3. The
dependent variable in the above models is binary and takes a value of 1 once a plot is developed, at which point it
is removed from the sample. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels. Standard
errors are clustered by National Resources Inventory (NRI) plot. The baseline year is 1982 in all models. Estimates
are weighted using NRI expansion factors divided for the number of times each plot is stacked due to its county
having multiple contiguous cross-border counties. All models include a set of contiguous-county border-pair dummy
variables, time fixed effects, border-pair time fixed effects, a baseline Kansas dummy variable, a constant term, and
controls representing plot characteristics and county attributes.
16
Table D2. Land Development Model, UVA Effect Estimates by Year and Bandwidth,
Unmatched Sample
Bandwidth (mi.) 5 10 15 20 25 None
1987 -0.016 -0.008 -0.008 -0.010 -0.008 -0.007
(0.007)** (0.004)** (0.003)*** (0.003)*** (0.002)*** (0.002)***
1992 <0.001 0.009 0.008 0.009 0.009 0.009
(0.009) (0.005)* (0.004)* (0.004)** (0.003)*** (0.003)***
1997 0.012 0.009 0.006 0.004 0.005 0.005
(0.007) (0.008) (0.006) (0.005) (0.005) (0.005)
2002 -0.006 <0.001 0.001 0.001 0.002 0.002
(0.005) (0.005) (0.003) (0.003) (0.003) (0.002)
2007 0.003 -0.001 >-0.001 <0.001 0.001 0.001
(0.003) (0.003) (0.002) (0.002) (0.002) (0.001)
2012 -0.007 -0.003 -0.001 -0.001 >-0.001 >-0.001
(0.004)* (0.002) (0.002) (0.001) (0.001) (0.001)
No. observations 7,567 16,194 22,764 28,273 32,079 32,864
No. plots (w/ repeats) 1,166 2,501 3,515 4,353 4,920 5,033
No. plots (no repeats) 595 1,330 1,912 2,393 2,709 2,770
Notes: The dependent variable in the above models is binary and takes a value of 1 once a plot is
developed, at which point it is removed from the sample. Asterisks denote significance at the 10% (*),
5% (**), and 1%(***) levels. Standard errors are clustered by National Resources Inventory (NRI)
plot. The baseline year is 1982 in all models. Estimates are weighted using NRI expansion factors
divided for the number of times each plot is stacked due to its county having multiple contiguous
cross-border counties. All models include a set of contiguous-county border-pair dummy variables,
time fixed effects, border-pair time fixed effects, a baseline Kansas dummy variable, a constant term,
and controls representing plot characteristics and county attributes.
17
Table D3. Land Development Model, UVA Effect Estimates by Year and Bandwidth,
Matched Sample, Alternative Specification with Post-UVA Plot Attribute Interaction
Terms
Bandwidth (mi.) 5 10 15 20 25 None
1987 -0.042 -0.014 -0.018 -0.020 -0.015 -0.012
(0.016)** (0.005)*** (0.005)*** (0.005)*** (0.004)*** (0.003)***
1992 -0.013 -0.003 -0.007 -0.009 -0.009 -0.009
(0.017) (0.008) (0.007) (0.006) (0.005) (0.006)
1997 0.006 0.005 -0.009 -0.002 -0.012 -0.009
(0.010) (0.009) (0.009) (0.008) (0.007)* (0.006)
2002 -0.034 -0.027 -0.025 -0.020 -0.022 -0.021
(0.013)** (0.009)*** (0.007)*** (0.006)*** (0.005)*** (0.004)***
2007 -0.020 -0.016 -0.020 -0.014 -0.014 -0.014
(0.010)** (0.005)*** (0.005)*** (0.004)*** (0.003)*** (0.003)***
2012 -0.028 -0.021 -0.021 -0.014 -0.012 -0.012
(0.016)* (0.007)*** (0.005)*** (0.004)*** (0.003)*** (0.003)***
No. observations 2,420 5,994 8,872 10,782 12,500 12,758
No. plots (w/ repeats) 386 936 1,384 1,676 1,922 1,954
No. plots (no repeats) 263 631 908 1,120 1,280 1,307
Notes: The dependent variable in the above models is binary and takes a value of 1 once a plot is
developed, at which point it is removed from the sample. Asterisks denote significance at the 10% (*),
5% (**), and 1%(***) levels. Standard errors are clustered by National Resources Inventory (NRI)
plot. The baseline year is 1982 in all models. Estimates are weighted using NRI expansion factors
divided for the number of times each plot is stacked due to its county having multiple contiguous
cross-border counties. All models include a set of contiguous-county border-pair dummy variables,
time fixed effects, border-pair time fixed effects, a baseline Kansas dummy variable, a constant term,
controls representing plot characteristics and county attributes, and interaction variables between the
controls and a post-1987 dummy variable.
18
Table D4. Land Development Model, UVA Effect Estimates by Year and Bandwidth,
Matched Sample, Matching with Replacement
Bandwidth (mi.) 5 10 15 20 25 None
1987 -0.121 -0.015 -0.019 -0.038 -0.031 -0.040
(0.057)** (0.007)** (0.007)*** (0.012)*** (0.012)** (0.015)***
1992 -0.113 -0.032 -0.006 -0.023 -0.015 -0.011
(0.052)** (0.025) (0.011) (0.011)** (0.009)* (0.008)
1997 0.001 -0.014 -0.026 -0.013 -0.016 -0.018
(0.011) (0.023) (0.027) (0.012) (0.010) (0.015)
2002 -0.090 -0.092 -0.062 -0.057 -0.043 -0.050
(0.037)** (0.029)*** (0.025)** (0.022)*** (0.013)*** (0.017)***
2007 -0.030 -0.019 -0.025 -0.020 -0.029 -0.030
(0.015)** (0.008)** (0.010)** (0.008)** (0.011)*** (0.013)**
2012 -0.024 -0.011 -0.023 -0.013 -0.010 -0.011
(0.011)** (0.004)** (0.008)*** (0.004)*** (0.003)*** (0.003)***
No. observations 5,123 11,665 16,167 18,878 21,435 21,889
No. plots (w/ repeats) 793 1,800 2,503 2,923 3,297 3,356
No. plots (no repeats) 477 1,051 1,504 1,812 2,031 2,049
Notes: The dependent variable in the above models is binary and takes a value of 1 once a plot is
developed, at which point it is removed from the sample. Asterisks denote significance at the 10% (*),
5% (**), and 1%(***) levels. Standard errors are clustered by National Resources Inventory (NRI)
plot. The baseline year is 1982 in all models. Estimates are weighted using NRI expansion factors
divided for the number of times each plot is stacked due to its county having multiple contiguous
cross-border counties. All models include a set of contiguous-county border-pair dummy variables,
time fixed effects, border-pair time fixed effects, a baseline Kansas dummy variable, a constant term,
controls representing plot characteristics and county attributes, and interaction variables between
these terms and a post-1987 dummy variable.
19
Appendix E
20
21
Table E1. Irrigation Model, UVA Effect Estimates by Year and Border Segment, Other Bandwidths, Matched Sample
Segment: Colorado Nebraska Oklahoma
Bandwidth (mi.): 5 15 25 5 15 25 5 15 25
1987 0.052 0.028 0.015 -0.009 -0.009 -0.011 >-0.001 -0.012 -0.018
(0.026)** (0.012)** (0.011) (0.010) (0.008) (0.008) (<0.001) (0.008) (0.008)**
1992 0.075 0.051 0.032 -0.002 -0.013 -0.004 -0.034 -0.021 -0.021
(0.033)** (0.018)*** (0.014)** (0.012) (0.016) (0.014) (0.036) (0.016) (0.015)
1997 0.024 0.032 0.011 -0.002 -0.029 -0.022 -0.108 -0.021 -0.021
(0.053) (0.023) (0.021) (0.012) (0.019) (0.015) (0.061)* (0.020) (0.017)
2002 0.051 0.045 0.018 0.015 -0.050 -0.033 -0.110 0.009 -0.009
(0.045) (0.023)** (0.021) (0.016) (0.022)** (0.018)* (0.083) (0.029) (0.028)
2007 0.062 0.048 0.008 0.020 -0.073 -0.038 -0.091 0.021 0.003
(0.043) (0.023)** (0.022) (0.016) (0.031)** (0.024) (0.085) (0.030) (0.029)
2012 0.033 0.056 0.001 0.020 -0.076 -0.035 -0.045 0.044 0.016
(0.047) (0.023)** (0.023) (0.016) (0.031)** (0.024) (0.073) (0.027) (0.028)
No. observations 1,302 4,424 6,440 1,057 3,752 6,251 896 4,165 7,070
No. plots (w/ repeats) 186 632 920 151 536 893 128 595 1,010
No. plots (no repeats) 128 414 610 109 339 567 102 421 670
Notes: The dependent variable in the above models is binary and takes a value of 1 if a plot is irrigated. Asterisks denote
significance at the 10% (*), 5% (**), and 1%(***) levels. Standard errors are clustered by National Resources Inventory (NRI)
plot. The baseline year is 1982 in all models. Estimates are weighted using NRI expansion factors divided for the number
of times each plot is stacked due to its county having multiple contiguous cross-border counties. All models include a set of
contiguous-county border-pair dummy variables, time fixed effects, border-pair time fixed effects, a baseline Kansas dummy
variable, a constant term, and controls representing plot characteristics and county attributes.
Table E2. Irrigation Model, Colorado Border Segment, Control Variable Estimates by Year
and Bandwidth, Matched Sample
Bandwidth (mi.) 5 10 15 20 25 None
LCC 1 1.021 0.780 0.812 0.726 0.857 0.810
(0.226)*** (0.169)*** (0.115)*** (0.102)*** (0.098)*** (0.082)***
LCC 2 0.573 0.459 0.469 0.418 0.477 0.443
(0.158)*** (0.131)*** (0.090)*** (0.075)*** (0.077)*** (0.060)***
LCC 3 0.221 0.178 0.234 0.207 0.313 0.314
(0.154) (0.112) (0.072)*** (0.047)*** (0.064)*** (0.046)***
LCC 4 0.175 0.092 0.178 0.123 0.198 0.167
(0.149) (0.093) (0.066)*** (0.047)*** (0.059)*** (0.041)***
Prime farmland 0.281 0.425 0.416 0.474 0.426 0.471
(0.192) (0.139)*** (0.086)*** (0.089)*** (0.076)*** (0.064)***
Slope 0.168 0.055 0.068 0.054 0.049 0.069
(0.053)*** (0.039) (0.025)*** (0.020)*** (0.020)** (0.021)***
Flood potential 0.035 0.053 0.136 0.097 0.023 0.041
(0.244) (0.094) (0.080)* (0.063) (0.047) (0.048)
Wind erosion 0.017 0.021 0.019 0.023 0.021 0.022
(0.007)** (0.004)*** (0.005)*** (0.004)*** (0.004)*** (0.004)***
Water erosion -0.096 -0.015 -0.016 -0.011 -0.007 -0.026
(0.054)* (0.039) (0.012) (0.011) (0.011) (0.015)*
Dist. KS border -0.033 -0.014 -0.007 -0.005 -0.004 -0.002
(0.021) (0.006)** (0.003)** (0.002)** (0.002)** (0.001)**
Shr. farmland irrig. 0.645 -0.077 -0.221 -0.618 -0.356 -0.116
(0.621) (0.327) (0.346) (0.354)* (0.275) (0.344)
Shr. farmland rented -0.696 -0.567 -0.444 -0.329 -0.313 -0.213
(0.356)* (0.195)*** (0.194)** (0.170)* (0.156)** (0.129)
Shr. farmland corp. 0.537 0.826 0.702 0.696 1.265 0.932
(0.655) (0.436)* (0.350)** (0.274)** (0.325)*** (0.296)***
Farm real est. value >-0.001 >-0.001 >-0.001 <0.001 <0.001 <0.001
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
No. observations 1,302 2,898 4,424 5,460 6,440 7,854
No. plots (w/ repeats) 186 414 632 780 920 1,122
No. plots (no repeats) 128 277 414 513 610 740
Notes: The estimates displayed above are for the control variables in the models used to produce the
UVA effects in Tables 4 and E1. The dependent variable in the above models is binary and takes a value
of 1 if a plot is irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels.
Standard errors are clustered by National Resources Inventory (NRI) plot. The baseline year is 1982 in
all models. Estimates are weighted using NRI expansion factors divided for the number of times each
plot is stacked due to its county having multiple contiguous cross-border counties. All models include a
set of contiguous-county border-pair dummy variables, time fixed effects, border-pair time fixed effects,
a baseline Kansas dummy variable, a constant term, and controls representing plot characteristics and
county attributes.
22
Table E3. Irrigation Model, Colorado Border Segment, UVA Effect Estimates by Year and
Bandwidth, Unmatched Sample
Bandwidth (mi.) 5 10 15 20 25 None
1987 0.037 0.029 0.029 0.020 0.017 0.011
(0.016)** (0.013)** (0.012)** (0.010)** (0.008)** (0.006)*
1992 0.079 0.060 0.043 0.028 0.025 0.018
(0.025)*** (0.016)*** (0.015)*** (0.012)** (0.011)** (0.008)**
1997 0.028 0.016 0.022 0.024 0.014 0.011
(0.042) (0.021) (0.019) (0.016) (0.015) (0.011)
2002 0.071 0.040 0.040 0.034 0.022 0.005
(0.039)* (0.022)* (0.020)** (0.018)* (0.017) (0.013)
2007 0.087 0.052 0.042 0.032 0.021 0.004
(0.036)** (0.022)** (0.021)** (0.019)* (0.017) (0.013)
2012 0.090 0.053 0.053 0.035 0.023 0.013
(0.039)** (0.023)** (0.021)** (0.019)* (0.018) (0.013)
No. observations 2,863 6,132 9,303 12,068 14,980 19,404
No. plots (w/ repeats) 409 876 1,329 1,724 2,140 2,772
No. plots (no repeats) 212 459 690 899 1,139 1,489
Notes: The dependent variable in the above models is binary and takes a value of 1 if a plot is
irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels. Standard errors
are clustered by National Resources Inventory (NRI) plot. The baseline year is 1982 in all models.
Estimates are weighted using NRI expansion factors divided for the number of times each plot is
stacked due to its county having multiple contiguous cross-border counties. All models include a set
of contiguous-county border-pair dummy variables, time fixed effects, border-pair time fixed effects,
a baseline Kansas dummy variable, a constant term, and controls representing plot characteristics
and county attributes.
23
Table E4. Irrigation Model, Colorado Border Segment, UVA Effect Estimates by Year and
Bandwidth, Alternative Specification with Post-UVA Plot Attribute Interaction Terms
Bandwidth (mi.) 5 10 15 20 25 None
1987 0.053 0.031 0.028 0.015 0.015 0.017
(0.025)** (0.021) (0.012)** (0.013) (0.011) (0.010)*
1992 0.055 0.057 0.033 0.026 0.016 0.027
(0.033) (0.023)** (0.019)* (0.017) (0.016) (0.014)*
1997 0.003 0.023 0.015 0.026 -0.004 0.029
(0.051) (0.031) (0.024) (0.020) (0.020) (0.019)
2002 0.030 0.026 0.028 0.040 0.003 0.022
(0.041) (0.028) (0.023) (0.021)* (0.021) (0.020)
2007 0.041 0.030 0.032 0.027 -0.007 0.017
(0.040) (0.028) (0.023) (0.021) (0.022) (0.020)
2012 0.012 0.018 0.040 0.029 -0.014 0.025
(0.043) (0.029) (0.023)* (0.021) (0.023) (0.021)
No. observations 1,302 2,898 4,424 5,460 6,440 7,854
No. plots (w/ repeats) 186 414 632 780 920 1,122
No. plots (no repeats) 128 277 414 513 610 740
Notes: The dependent variable in the above models is binary and takes a value of 1 if
a plot is irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***)
levels. Standard errors are clustered by National Resources Inventory (NRI) plot. The
baseline year is 1982 in all models. Estimates are weighted using NRI expansion factors
divided for the number of times each plot is stacked due to its county having multiple
contiguous cross-border counties. All models include a set of contiguous-county border-
pair dummy variables, time fixed effects, border-pair time fixed effects, a baseline Kansas
dummy variable, a constant term, and controls representing plot characteristics and county
attributes, and interaction variables between the controls and a post-1987 dummy variable.
24
Table E5. Irrigation Model, Colorado Border Segment, UVA Effect Estimates by Year and
Bandwidth, Matching with Replacement
Bandwidth (mi.) 5 10 15 20 25 None
1987 0.030 0.069 0.040 0.071 0.026 0.027
(0.015)** (0.036)* (0.019)** (0.044) (0.022) (0.026)
1992 0.053 0.095 0.047 0.077 0.033 0.034
(0.021)** (0.036)*** (0.020)** (0.044)* (0.022) (0.027)
1997 -0.062 0.080 0.007 0.078 0.002 0.053
(0.107) (0.045)* (0.012) (0.046)* (0.030) (0.027)*
2002 0.018 0.084 0.008 0.065 0.001 0.039
(0.046) (0.044)* (0.014) (0.049) (0.030) (0.028)
2007 0.024 0.085 0.010 0.057 -0.006 0.036
(0.045) (0.044)* (0.015) (0.049) (0.030) (0.027)
2012 0.007 0.078 0.012 0.065 -0.009 0.037
(0.046) (0.044)* (0.015) (0.048) (0.030) (0.028)
No. observations 1,575 4,088 5,943 7,637 10,066 10,458
No. plots (w/ repeats) 225 584 849 1,091 1,438 1,494
No. plots (no repeats) 152 359 521 669 879 976
Notes: The dependent variable in the above models is binary and takes a value of 1 if a
plot is irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels.
Standard errors are clustered by National Resources Inventory (NRI) plot. The baseline year
is 1982 in all models. Estimates are weighted using NRI expansion factors divided for the
number of times each plot is stacked due to its county having multiple contiguous cross-border
counties. All models include a set of contiguous-county border-pair dummy variables, time
fixed effects, border-pair time fixed effects, a baseline Kansas dummy variable, a constant
term, and controls representing plot characteristics and county attributes.
25
Table E6. Irrigation Model, Nebraska Border Segment, Control Variable Estimates by Year
and Bandwidth, Matched Sample
Bandwidth (mi.) 5 10 15 20 25 None
LCC 1 0.772 0.714 0.751 0.654 0.637 0.670
(0.201)*** (0.143)*** (0.118)*** (0.110)*** (0.088)*** (0.087)***
LCC 2 0.032 0.096 0.104 0.092 0.061 0.027
(0.056) (0.055)* (0.055)* (0.061) (0.052) (0.051)
LCC 3 0.084 0.245 0.084 0.132 0.115 0.089
(0.104) (0.099)** (0.072) (0.067)** (0.064)* (0.062)
LCC 4 -0.059 -0.006 0.027 -0.011 0.017 0.019
(0.034)* (0.043) (0.043) (0.045) (0.042) (0.040)
Prime farmland -0.042 -0.044 -0.009 -0.069 -0.032 0.040
(0.029) (0.053) (0.043) (0.047) (0.039) (0.037)
Slope -0.004 -0.003 0.016 0.014 -0.003 <0.001
(0.012) (0.015) (0.009)* (0.010) (0.012) (0.009)
Flood potential -0.032 0.002 0.018 0.046 0.059 0.005
(0.044) (0.056) (0.044) (0.045) (0.039) (0.031)
Wind erosion 0.028 0.053 0.028 0.023 0.029 0.034
(0.011)** (0.010)*** (0.006)*** (0.006)*** (0.006)*** (0.006)***
Water erosion -0.005 -0.006 -0.009 -0.008 -0.001 >-0.001
(0.006) (0.007) (0.005)* (0.006) (0.006) (0.005)
Dist. KS border 0.019 0.011 0.009 0.004 0.003 0.003
(0.010)* (0.004)** (0.003)*** (0.002)** (0.001)** (0.001)**
Shr. farmland irrig. -0.575 -0.194 1.042 0.575 0.915 1.376
(0.623) (1.555) (1.387) (1.362) (1.145) (1.161)
Shr. farmland rented -1.197 -0.060 0.272 -1.843 -0.298 -0.858
(0.835) (0.975) (1.090) (1.022)* (0.798) (0.708)
Shr. farmland corp. -0.149 0.084 -0.625 -1.731 -0.702 -0.948
(0.548) (0.583) (0.627) (0.604)*** (0.504) (0.413)**
Farm real est. value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001)* (<0.001)*
No. observations 1,057 2,359 3,752 4,466 6,251 9,772
No. plots (w/ repeats) 151 337 536 638 893 1,396
No. plots (no repeats) 109 233 339 418 567 848
Notes: The estimates displayed above are for the control variables in the models used to produce the
UVA effects in Tables 4 and E1. The dependent variable in the above models is binary and takes a value
of 1 if a plot is irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels.
Standard errors are clustered by National Resources Inventory (NRI) plot. The baseline year is 1982 in
all models. Estimates are weighted using NRI expansion factors divided for the number of times each
plot is stacked due to its county having multiple contiguous cross-border counties. All models include a
set of contiguous-county border-pair dummy variables, time fixed effects, border-pair time fixed effects,
a baseline Kansas dummy variable, a constant term, and controls representing plot characteristics and
county attributes.
26
Table E7. Irrigation Model, Nebraska Border Segment, UVA Effect Estimates by Year and
Bandwidth, Unmatched Sample
Bandwidth (mi.) 5 10 15 20 25 None
1987 -0.003 -0.015 -0.026 -0.031 -0.028 -0.028
(0.005) (0.008)* (0.009)*** (0.009)*** (0.007)*** (0.007)***
1992 0.016 -0.005 -0.019 -0.037 -0.031 -0.028
(0.025) (0.013) (0.012) (0.012)*** (0.010)*** (0.010)***
1997 0.011 >-0.001 -0.028 -0.042 -0.038 -0.034
(0.027) (0.016) (0.014)** (0.013)*** (0.011)*** (0.011)***
2002 -0.002 -0.004 -0.044 -0.062 -0.059 -0.055
(0.025) (0.016) (0.015)*** (0.014)*** (0.013)*** (0.013)***
2007 0.013 -0.010 -0.055 -0.076 -0.069 -0.065
(0.029) (0.023) (0.019)*** (0.018)*** (0.015)*** (0.015)***
2012 -0.002 -0.017 -0.059 -0.074 -0.068 -0.065
(0.032) (0.024) (0.020)*** (0.018)*** (0.016)*** (0.016)***
No. observations 2,716 5,824 8,953 12,369 15,596 16,919
No. plots (w/ repeats) 388 832 1,279 1,767 2,228 2,417
No. plots (no repeats) 196 416 641 903 1,142 1,232
Notes: The dependent variable in the above models is binary and takes a value of 1 if a plot is
irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels. Standard
errors are clustered by National Resources Inventory (NRI) plot. The baseline year is 1982 in all
models. Estimates are weighted using NRI expansion factors divided for the number of times each
plot is stacked due to its county having multiple contiguous cross-border counties. All models
include a set of contiguous-county border-pair dummy variables, time fixed effects, border-pair
time fixed effects, a baseline Kansas dummy variable, a constant term, and controls representing
plot characteristics and county attributes.
27
Table E8. Irrigation Model, Nebraska Border Segment, UVA Effect Estimates by Year and
Bandwidth, Alternative Specification with Post-UVA Plot Attribute Interaction Terms
Bandwidth (mi.) 5 10 15 20 25 None
1987 -0.009 -0.017 -0.009 -0.021 -0.011 -0.019
(0.010) (0.011) (0.009) (0.013) (0.008) (0.009)**
1992 -0.005 0.008 >-0.001 -0.016 0.010 -0.018
(0.014) (0.016) (0.016) (0.018) (0.014) (0.013)
1997 -0.005 0.008 -0.016 -0.046 -0.008 -0.030
(0.014) (0.016) (0.018) (0.020)** (0.016) (0.014)**
2002 0.014 -0.007 -0.037 -0.075 -0.019 -0.055
(0.017) (0.017) (0.020)* (0.021)*** (0.017) (0.016)***
2007 0.019 -0.022 -0.060 -0.083 -0.024 -0.066
(0.020) (0.024) (0.029)** (0.027)*** (0.024) (0.020)***
2012 0.019 -0.027 -0.063 -0.083 -0.021 -0.070
(0.020) (0.024) (0.029)** (0.027)*** (0.024) (0.020)***
No. observations 1,057 2,359 3,752 4,466 6,251 9,772
No. plots (w/ repeats) 151 337 536 638 893 1,396
No. plots (no repeats) 109 233 339 418 567 848
Notes: The dependent variable in the above models is binary and takes a value of 1 if
a plot is irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***)
levels. Standard errors are clustered by National Resources Inventory (NRI) plot. The
baseline year is 1982 in all models. Estimates are weighted using NRI expansion factors
divided for the number of times each plot is stacked due to its county having multiple
contiguous cross-border counties. All models include a set of contiguous-county border-
pair dummy variables, time fixed effects, border-pair time fixed effects, a baseline Kansas
dummy variable, a constant term, and controls representing plot characteristics and county
attributes, and interaction variables between the controls and a post-1987 dummy variable.
28
Table E9. Irrigation Model, Nebraska Border Segment, UVA Effect Estimates by Year and
Bandwidth, Matching with Replacement
Bandwidth (mi.) 5 10 15 20 25 None
1987 -0.005 -0.020 0.011 -0.012 -0.001 -0.007
(0.005) (0.015) (0.011) (0.010) (0.007) (0.008)
1992 0.002 -0.011 -0.033 -0.099 0.001 -0.003
(0.009) (0.016) (0.037) (0.063) (0.011) (0.014)
1997 0.002 -0.011 -0.041 -0.120 -0.068 -0.015
(0.009) (0.016) (0.037) (0.061)** (0.051) (0.013)
2002 0.009 -0.029 -0.070 -0.095 -0.069 -0.041
(0.009) (0.025) (0.047) (0.067) (0.053) (0.014)***
2007 0.017 -0.065 -0.144 -0.110 -0.095 -0.043
(0.013) (0.043) (0.073)** (0.072) (0.058) (0.016)***
2012 0.017 -0.068 -0.145 -0.111 -0.094 -0.067
(0.013) (0.043) (0.073)** (0.072) (0.058) (0.026)**
No. observations 1,295 2,856 4,676 6,125 7,595 9,800
No. plots (w/ repeats) 185 408 668 875 1,085 1,400
No. plots (no repeats) 121 252 385 500 633 770
Notes: The dependent variable in the above models is binary and takes a value of 1 if
a plot is irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***)
levels. Standard errors are clustered by National Resources Inventory (NRI) plot. The
baseline year is 1982 in all models. Estimates are weighted using NRI expansion factors
divided for the number of times each plot is stacked due to its county having multiple
contiguous cross-border counties. All models include a set of contiguous-county border-
pair dummy variables, time fixed effects, border-pair time fixed effects, a baseline Kansas
dummy variable, a constant term, and controls representing plot characteristics and county
attributes.
29
Table E10. Irrigation Model, Oklahoma Border Segment, Control Variable Estimates by
Year and Bandwidth, Matched Sample
Bandwidth (mi.) 5 10 15 20 25 None
LCC 1 0.735 0.759 0.875 0.783 0.538 0.409
(0.290)** (0.222)*** (0.229)*** (0.127)*** (0.111)*** (0.149)***
LCC 2 0.627 0.119 0.205 0.151 -0.026 -0.038
(0.248)** (0.188) (0.220) (0.112) (0.097) (0.093)
LCC 3 0.072 -0.244 -0.131 -0.146 -0.318 -0.243
(0.193) (0.181) (0.217) (0.104) (0.092)*** (0.088)***
LCC 4 - -0.605 -0.397 -0.297 -0.351 -0.319
- (0.208)*** (0.244) (0.134)** (0.148)** (0.110)***
Prime farmland 0.441 0.191 0.041 0.052 0.073 0.203
(0.251)* (0.076)** (0.051) (0.047) (0.043)* (0.039)***
Slope 0.063 0.009 -0.020 -0.006 -0.021 0.022
(0.125) (0.022) (0.022) (0.020) (0.020) (0.014)
Flood potential - -0.046 -0.021 -0.031 -0.132 -0.124
- (0.044) (0.065) (0.070) (0.047)*** (0.047)***
Wind erosion 0.031 0.032 0.022 0.017 0.019 0.013
(0.011)*** (0.006)*** (0.006)*** (0.004)*** (0.004)*** (0.003)***
Water erosion 0.111 0.010 0.023 0.005 0.013 0.002
(0.165) (0.007) (0.015) (0.016) (0.012) (0.007)
Dist. KS border -0.062 0.006 -0.002 -0.006 -0.002 <0.001
(0.031)** (0.010) (0.005) (0.003)* (0.002) (0.002)
Shr. farmland irrig. -0.602 0.119 0.468 -1.020 1.293 1.361
(2.272) (0.990) (0.791) (0.753) (0.641)** (0.530)**
Shr. farmland rented -1.015 0.257 1.079 0.551 1.073 0.580
(1.229) (0.635) (0.508)** (0.398) (0.392)*** (0.330)*
Shr. farmland corp. -0.222 0.733 0.928 0.393 0.512 0.725
(1.377) (0.556) (0.419)** (0.370) (0.373) (0.317)**
Farm real est. value 0.001 <0.001 <0.001 0.001 <0.001 <0.001
(0.001) (<0.001) (<0.001) (<0.001)* (<0.001) (<0.001)
No. observations 896 2,583 4,165 5,173 7,070 10,304
No. plots (w/ repeats) 128 369 595 739 1,010 1,472
No. plots (no repeats) 102 277 421 510 670 931
Notes: The estimates displayed above are for the control variables in the models used to produce the
UVA effects in Tables 4 and E1. The dependent variable in the above models is binary and takes a value
of 1 if a plot is irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels.
Standard errors are clustered by National Resources Inventory (NRI) plot. The baseline year is 1982 in
all models. Estimates are weighted using NRI expansion factors divided for the number of times each
plot is stacked due to its county having multiple contiguous cross-border counties. All models include a
set of contiguous-county border-pair dummy variables, time fixed effects, border-pair time fixed effects,
a baseline Kansas dummy variable, a constant term, and controls representing plot characteristics and
county attributes.
30
Table E11. Irrigation Model, Oklahoma Border Segment, UVA Effect Estimates by Year
and Bandwidth, Unmatched Sample
Bandwidth (mi.) 5 10 15 20 25 None
1987 -0.010 -0.009 -0.017 -0.014 -0.012 -0.008
(0.011) (0.007) (0.007)** (0.006)** (0.005)** (0.005)*
1992 -0.043 -0.056 -0.041 -0.027 -0.021 -0.015
(0.027) (0.020)*** (0.017)** (0.013)** (0.012)* (0.009)
1997 -0.109 -0.050 -0.032 -0.019 -0.017 0.009
(0.043)** (0.023)** (0.019)* (0.015) (0.013) (0.017)
2002 -0.063 0.052 0.027 0.027 0.022 0.054
(0.052) (0.037) (0.027) (0.023) (0.021) (0.020)***
2007 -0.051 0.042 0.035 0.034 0.033 0.053
(0.052) (0.033) (0.025) (0.023) (0.020) (0.018)***
2012 -0.031 0.060 0.047 0.043 0.041 0.066
(0.051) (0.033)* (0.025)* (0.023)* (0.020)** (0.020)***
No. observations 3,262 6,503 9,121 11,473 14,511 19,278
No. plots (w/ repeats) 466 929 1,303 1,639 2,073 2,754
No. plots (no repeats) 219 461 671 860 1,122 1,513
Notes: The dependent variable in the above models is binary and takes a value of 1 if a plot is
irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels. Standard
errors are clustered by National Resources Inventory (NRI) plot. The baseline year is 1982 in all
models. Estimates are weighted using NRI expansion factors divided for the number of times each
plot is stacked due to its county having multiple contiguous cross-border counties. All models
include a set of contiguous-county border-pair dummy variables, time fixed effects, border-pair
time fixed effects, a baseline Kansas dummy variable, a constant term, and controls representing
plot characteristics and county attributes.
31
Table E12. Irrigation Model, Oklahoma Border Segment, UVA Effect Estimates by Year
and Bandwidth, Alternative Specification with Post-UVA Plot Attribute Interaction Terms
Bandwidth (mi.) 5 10 15 20 25 None
1987 >-0.001 -0.012 -0.012 -0.019 -0.018 -0.009
(0.001) (0.011) (0.008) (0.010)* (0.008)** (0.005)*
1992 -0.025 -0.087 -0.032 -0.041 -0.017 -0.020
(0.054) (0.027)*** (0.020) (0.022)* (0.019) (0.015)
1997 -0.102 -0.071 -0.032 -0.043 -0.016 0.017
(0.065) (0.033)** (0.024) (0.025)* (0.021) (0.024)
2002 -0.103 -0.037 <0.001 -0.019 -0.004 0.056
(0.081) (0.041) (0.029) (0.031) (0.029) (0.027)**
2007 -0.088 -0.021 0.012 -0.001 0.009 0.050
(0.082) (0.041) (0.030) (0.032) (0.029) (0.023)**
2012 -0.045 0.015 0.035 0.016 0.021 0.068
(0.082) (0.038) (0.028) (0.030) (0.029) (0.026)***
No. observations 896 2,583 4,165 5,173 7,070 10,304
No. plots (w/ repeats) 128 369 595 739 1,010 1,472
No. plots (no repeats) 102 277 421 510 670 931
Notes: The dependent variable in the above models is binary and takes a value of 1 if a
plot is irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***) levels.
Standard errors are clustered by National Resources Inventory (NRI) plot. The baseline year
is 1982 in all models. Estimates are weighted using NRI expansion factors divided for the
number of times each plot is stacked due to its county having multiple contiguous cross-border
counties. All models include a set of contiguous-county border-pair dummy variables, time
fixed effects, border-pair time fixed effects, a baseline Kansas dummy variable, a constant
term, and controls representing plot characteristics and county attributes, and interaction
variables between the controls and a post-1987 dummy variable.
32
Table E13. Irrigation Model, Oklahoma Border Segment, UVA Effect Estimates by Year
and Bandwidth, Matching with Replacement
Bandwidth (mi.) 5 10 15 20 25 None
1987 -0.001 -0.015 -0.004 -0.014 -0.007 -0.006
(0.001) (0.015) (0.003) (0.011) (0.004) (0.005)
1992 -0.005 -0.009 -0.019 -0.019 -0.015 -0.028
(0.037) (0.021) (0.015) (0.015) (0.011) (0.022)
1997 -0.062 0.015 -0.023 -0.014 0.001 0.014
(0.050) (0.025) (0.025) (0.021) (0.013) (0.026)
2002 -0.037 0.075 -0.003 -0.109 -0.010 0.042
(0.053) (0.043)* (0.041) (0.077) (0.048) (0.036)
2007 -0.026 0.084 0.003 -0.105 -0.013 0.027
(0.055) (0.043)* (0.041) (0.077) (0.047) (0.033)
2012 -0.026 0.111 0.039 -0.059 0.006 0.039
(0.056) (0.038)*** (0.030) (0.078) (0.046) (0.032)
No. observations 861 2,716 4,494 6,321 9,821 11,914
No. plots (w/ repeats) 123 388 642 903 1,403 1,702
No. plots (no repeats) 123 317 504 666 940 1,121
Notes: The dependent variable in the above models is binary and takes a value of 1 if a
plot is irrigated. Asterisks denote significance at the 10% (*), 5% (**), and 1%(***)
levels. Standard errors are clustered by National Resources Inventory (NRI) plot.
The baseline year is 1982 in all models. Estimates are weighted using NRI expansion
factors divided for the number of times each plot is stacked due to its county having
multiple contiguous cross-border counties. All models include a set of contiguous-
county border-pair dummy variables, time fixed effects, border-pair time fixed effects,
a baseline Kansas dummy variable, a constant term, and controls representing plot
characteristics and county attributes.
33