E STIMATING THE I MPACT OF D ROUGHT ON

A GRICULTURE U SING THE U.S.

D ROUGHT M ONITOR

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

Y USUKE K UWAYAMA , A LEXANDRA T HOMPSON , R ICHARD B ERNKNOPF ,
B ENJAMIN Z AITCHIK , AND P ETER V AIL

We estimate the impacts of drought, as defined by the U.S. Drought Monitor (USDM), on crop
yields and farm income in the United States during the 2001–2013 time period. Our empirical strat-
egy relies on panel data models with fixed effects that exploit spatial and temporal variability in
drought conditions and agricultural outcomes at the county level. We find negative and statistically
significant effects of drought on crop yields equal to reductions in the range of 0.1% to 1.2% for corn
and soybean yields for each additional week of drought in dryland counties, and 0.1% to 0.5% in irri-
gated counties. Region-specific results vary, with some regions experiencing no yield impacts from
drought, while yield reductions as high as 8.0% are observed in dryland counties in the Midwest for
every additional week of drought in the highest USDM severity category. Despite this impact on
crop yields, we find that additional weeks of drought have little to no effect on measures of farm in-
come. While precipitation and temperature explain most of the variability in crop yields, we find that
the USDM captures additional negative impacts of drought on yields.
JEL codes: Q1, Q15, Q25, Q54.

Key words: Drought, agriculture, irrigation.

Recent years have seen an increased interest (Climate Change Science Program 2008;
in the economic impacts of droughts in the Walthall et al. 2012). Droughts are associated
United States as a result of disasters such as with below-average precipitation, which can
the 2012 Midwest drought and the 2011–2017 affect crop yields, and reduce surface water
drought in California. This attention has been and groundwater supplies, which can affect ir-
reinforced by predictions that droughts will be- rigation and livestock watering. Furthermore,
come more frequent and more severe in the droughts are associated with periods of above-
future with the progression of climate change average temperatures that can exacerbate hy-
(Intergovernmental Panel on Climate Change; drological and biophysical stress. These
IPCC 2014). While droughts have the potential impacts can lead to a decrease in revenues
to affect an entire regional economy, the agri- from crop and livestock sales, changes in pro-
cultural sector is particularly vulnerable duction costs—or both—possibly reducing net
farm income and negatively affecting commu-
nities that are dependent on agriculture.
Yusuke Kuwayama is a fellow and Alexandra Thompson is a re- Despite this interest in the economic
search associate, both at Resources for the Future. Richard impacts of drought, few studies have quanti-
Bernknopf is a research professor in the Department of
Economics, University of New Mexico. Benjamin Zaitchik is an
fied the impact of droughts on physical meas-
associate professor in the Department of Earth and Planetary ures such as crop yield, or monetized
Sciences, Johns Hopkins University. Peter Vail is a research as- measures, such as farm income. One reason
sistant at Resources for the Future. The authors wish to thank
Sara Pesko for excellent research assistance, and David for this lack of research is that there is no uni-
Brookshire, Molly Macauley, Kevin Patrick, Matthew Rodell, versally accepted quantitative definition of
seminar participants at the University of Florida and the drought. Instead, drought is often defined
University of Nebraska-Lincoln, and three anonymous referees
for valuable comments and suggestions. This project was com- qualitatively as a deficit of water relative to
pleted with financial support from the National Aeronautics and normal conditions as referenced by water sup-
Space Administration, NNX09A01G. Correspondence to be sent
to: [email protected]
ply demand and management (Wilhite 2000;

Amer. J. Agr. Econ. 101(1): 193–210; doi: 10.1093/ajae/aay037

Published online July 10, 2018
C The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics
V
Association. All rights reserved. For permissions, please email: [email protected]
194 January 2019 Amer. J. Agr. Econ.

Lloyd-Hughes 2014). As a result, what consti- provide useful information for policymakers
tutes a drought varies across regions and time who are designing USDM-based eligibility
periods, and is also a function of the local so- requirements for drought assistance. Our sec-
cioeconomic context, including factors such as ond contribution is to add to existing evi-
the composition of the local economy, access dence regarding the effect of weather and
to infrastructure, and household income levels climate on agricultural outcomes. While pre-

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

and distribution. At the same time, govern- cipitation and temperature have received sig-
ments at the federal, state, and municipal lev- nificant attention as determinants of current
els often use some form of composite drought and future agricultural output, quantitative
index instead of individual thresholds for hy- evidence on the effect of drought as an ex-
drologic measures (e.g., precipitation or reser- treme event is scarce. Drought, as defined by
voir levels) for informing drought-related the USDM and other composite drought in-
policies such as State of Emergency proclama- dices, is driven not only by precipitation and
tions and eligibility for drought disaster assis- temperature measures; it is also determined
tance. Furthermore, public sector drought by indicators such as soil moisture, stream-
policies in the United States act as triggers for flow, vegetation indices, reservoir and lake
the distribution of a significant amount of levels, groundwater levels, and snowpack.
public resources to agricultural communities. These other dimensions of drought can gen-
For example, in fiscal year 2014, the federal erate impacts on farm income beyond those
government designated $873 million across that are identified for below-average precipi-
11 western states for drought-related crop in- tation or above-average temperatures.
surance and programs that were a result of Our analysis is based on a panel dataset
drought emergency declarations (Mount et al. that was constructed by matching crop yield
2016). data from the USDA’s National Agricultural
In the United States, the composite Statistics Service (NASS), and farm income
drought index most often used by policy- data from the Bureau of Economic Analysis
makers is the U.S. Drought Monitor (BEA) to drought intensity categorizations in
(USDM). The USDM consists of a weekly the USDM at the county level. Our empirical
map that indicates which regions of the coun- strategy relies on panel data models with
try are currently in drought, as well as the in- fixed effects that exploit spatial and temporal
tensity of those droughts. It is used by a variability in drought conditions and crop
variety of federal agencies including the yield and farm income, allowing us to esti-
USDA and the Internal Revenue Service mate an average nationwide impact as well as
(IRS), as well as state government agencies, impacts in specific regions. These estimates
primarily for programs related to the agricul- reflect the impact of an additional week of
tural sector (National Drought Mitigation drought on crop yield and annual farm in-
Center; NDMC 2016). Given the key role come, inclusive of all the on-farm activities
that the USDM plays in U.S. drought policy that are available to agricultural producers
in the agricultural sector, evidence regarding that help mitigate the biophysical impact of
the relationship between USDM drought cat- drought on crops and livestock. We run sepa-
egorizations and realized agricultural out- rate regressions for irrigated and dryland
comes is desirable. counties because we expect the effect of
In this paper, we estimate the impacts of drought on agricultural outcomes to vary be-
drought, as defined by the USDM, on crop tween these two types of counties.
yields and farm income in the United States For the average dryland county, we find
during the 2001–2013 time period. Our analy- negative and statistically significant impacts
sis makes two main contributions to the liter- of drought on crop yields equal to reductions
ature. First, to our knowledge, this is the first in the range of 0.1% to 1.2% for corn and
study to confirm that drought information soybeans for each additional week of
from the USDM is correlated with observed drought. Impacts in irrigated counties are
agricultural outcomes. This evidence is valu- smaller in magnitude, ranging from 0.1% to
able because the USDM plays a critical role 0.5% for each additional week of drought.
in decisions regarding the allocation of finan- Region-specific results are mixed, with some
cial resources in response to drought. regions experiencing no yield impacts from
Furthermore, by quantifying the impacts on drought, while yield reductions in dryland
crop yields and farm income that are associ- counties in the midwest are as high as 8.0%
ated with USDM drought categorizations, we for corn and 3.1% for soybeans for every
Kuwayama et al. Estimating the Impact of Drought on Agriculture Using the U.S. Drought Monitor 195

additional week of exceptional drought, the sorghum, oats, soybeans, and wheat. However,
highest level of drought severity in the USDM the authors note that because of high prices
categorization. We find that additional weeks resulting from the drought, non-drought area
of drought have little to no effect on the value production, inventory sell-offs, and irrigation,
of cash receipts and production expenses, per- net agricultural income in 1988 rose slightly
haps due to farmers receiving higher prices as from the previous non-drought year despite

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

a result of drought-induced local scarcity in ag- the crop losses. The authors also find that neg-
ricultural commodities. Finally, we find that ative impacts for specific individuals and
precipitation and temperature explain most of regions were substantial. Wheaton et al.
the variability in crop yields during the time (2008) evaluate drought impacts on agriculture
period of our analysis, but the USDM does by comparing production values during the
capture additional negative impacts of drought 2001–2002 Canadian drought to that in bench-
on yields. mark years. Estimated crop production value
The paper proceeds as follows. The next losses were in the range of $1.7 to $2.4 billion
section contains a brief review of existing depending on the region, while impacts on
studies on the relationship between droughts cash receipts were smaller due to inventory
and agricultural outcomes and a description adjustments.
of the USDM and its current use in federal Horridge, Madden, and Wittwer (2005)
drought policy. This is followed by sections take an agricultural production function ap-
describing our data and empirical strategy, proach to estimate the impacts of the 2002–
respectively. In the next section, we present 2003 Australian drought, aggregating values
our estimates for the effect of drought on from 38 sectors and 45 regions. The authors
crop yields and farm income, while the final find significant aggregate effects from agricul-
section concludes. ture on the national economy, despite the rel-
atively small role of the sector in Australia,
with income losses of up to 20% and a 1.6%
Background reduction in Gross Domestic Product. More
recently, Howitt et al. (2014, 2015) and
Droughts and Agricultural Outcomes Medellın-Azuara et al. (2016) estimated the
economic impact of drought in California on
From a biophysical standpoint, the effects of the state’s agricultural sector. Using an eco-
drought on crops are well studied, particu- nomic optimization model of crop choice that
larly in the context of potential future includes regional water availability con-
impacts due to climate change. In addition to straints, the authors calculated the net water
retrospective analyses of past droughts on shortage (5 million acre feet) to result in sig-
crop productivity (e.g., Ciais et al. 2005; Zhao nificant losses in crops ($2 billion) and dairy
and Running 2010), biophysical research on and livestock ($553 million), as well as addi-
droughts and crops has addressed the benefits tional groundwater pumping costs ($1.3 bil-
of drought tolerance practices (e.g., lion) and lost jobs (43,000) in 2014, 2015, and
Cattivellia et al. 2008; Craine et al. 2013) and 2016. Unlike these previous ex post studies,
the role of extreme heat (e.g., Lobell et al. which estimate costs to the agricultural sector
2013). The effects of drought on livestock for entire drought events, our econometric
have also been studied, with a focus on devel- approach yields estimates of economic im-
oping countries and the role of livestock act- pact for marginal increases in the duration
ing as a buffer against income shocks from and intensity of droughts. This information
drought (Fafchamps, Udry, and Czukas 1998; may be useful for policymakers who need to
Kinsey, Burger, and Gunning 1998; Kazianga allocate scarce financial resources for drought
and Udry 2006). assistance and emergency programs.
A surprisingly small number of studies con- A different strand of research addresses
duct ex post estimations of the economic cost specific types of farmer adaptation to drought.
of droughts on the agricultural sector. For ex- For example, Cavatassi et al. (2011) investi-
ample, Riebsame, Changnon, and Karl (1991) gate whether farmers in Ethiopia adopt mod-
present a detailed narrative of the 1988–1989 ern, drought-tolerant varieties of sorghum as a
drought in the United States, including a quan- risk reduction strategy in the face of drought.
titative impact analysis that yields an esti- Moreover, Ding, Schoengold, and Tadesse
mated crop loss of $15 billion due to output (2009) investigate the relationship between
and price changes in corn, barley, grain drought and flood events and the likelihood
196 January 2019 Amer. J. Agr. Econ.

that farmers will adopt conservation tillage Table 1. Categories of Drought Intensity
practices, using county-level data from Iowa, Used by the U.S. Drought Monitor
Nebraska, and South Dakota. While our data
Category Drought Intensity Level Percentile
do not allow us to observe the evolution of
specific on-farm practices in response to D0 Abnormally dry 20 to  30
changes in drought conditions, our estimates D1 Drought, moderate 10 to  20

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

can be interpreted as including the mitigating D2 Drought, severe 5 to  10
effects of some of these practices, namely, D3 Drought, extreme 2 to  5
those that can be adopted in response to the D4 Drought, exceptional  2
onset of a drought.
Our paper is perhaps most similar to a
group of studies that addresses the relation-
ship between agricultural outcomes and spe- by the USDA, the National Oceanic and
cific weather components, such as Atmospheric Administration (NOAA), and
precipitation and temperature. The main goal the National Drought Mitigation Center
of these studies is to use these estimated rela- (NDMC) on a weekly basis.1 The maps desig-
tionships to predict the effects of future cli- nate general drought areas as being in one of
mate change on U.S. agriculture. Mendelsohn, five intensity classes, ranging from
Nordhaus, and Shaw (1994) adopt a hedonic “abnormally dry” (abbreviated as D0),
approach by modeling farmland values and “moderate drought” (D1), “severe drought”
rents as a function of temperature, precipita- (D2), “extreme drought” (D3), and
tion, soil type, and other physical and socio- “exceptional drought” (D4). Each intensity
economic variables. The authors then use class is associated with its probability of oc-
uniform precipitation and temperature currence, expressed as a percentile, on the
increases to simulate how agricultural farm- basis of a 1932–2001 data record of drought
land values would vary under climate change, indicators (Svoboda et al. 2002; Houborg
finding effects that are highly nonlinear and et al. 2012). Table 1 lists the percentiles asso-
that vary by geography and season. Schlenker, ciated with each drought intensity class; the
Hanemann, and Fisher (2005) refined this ap- classification scheme indicates, for example,
proach by differentiating irrigated from that droughts of intensity D3 or worse have a
rainfed agriculture, finding statistically signifi- 5% chance of occurring in any given location.
cant differences in coefficient estimates. The USDM is not strictly a drought index,
Desch^ enes and Greenstone (2007) take a dif- but rather a composite product developed
ferent approach by exploiting year-to-year from a suite of climate indices and numerical
weather fluctuations rather than climate differ- models, as well as from input from regional
ences across counties, and by using yields and and local experts. The six key physical indica-
profits as outcome variables instead of farm- tors used by the USDM authors are a drought
land values, concluding that climate change index (specifically, the Palmer Drought
impacts will be neutral or possibly positive due Severity Index), percentiles from a soil mois-
to adaptation. However, Fisher et al. (2012) ture model, daily streamflow percentiles, the
find different results when replicating the percent of normal precipitation, a standard-
Desch^ enes and Greenstone (2007) analysis ized precipitation index, and a remotely-
while including inventory adjustments and cor- sensed vegetation health index (Svoboda
rection for data irregularities. The replication et al. 2002). The authors also rely on supple-
finds significant negative economic impacts of mentary indicators such as humidity and tem-
climate change on agricultural production. As perature departure from normal, reservoir,
mentioned earlier, our analysis differs from and lake levels, surface water supply indices,
these studies because our explanatory varia- snowpack, and groundwater levels. The
bles consist of drought categorizations that are authors then seek input and verification from
actively used in federal drought policy, and re- regional and state climatologists, agricultural
flect a larger universe of weather and hydro- and water resource managers, National
logic conditions. Weather Service field employees, and others
to help ground-truth the maps based on local
The U.S. Drought Monitor
The USDM is an expert-based national map 1
The USDM map for any given week can be accessed at
of drought conditions that is produced jointly http://droughtmonitor.unl.edu/.
Kuwayama et al. Estimating the Impact of Drought on Agriculture Using the U.S. Drought Monitor 197

knowledge of drought conditions and sales, and emergency loans through Emergency
impacts. All of this information is subjectively Disaster Designations and Declarations.
incorporated into a final map of drought in- The LFP, which was made a permanent
tensity categorizations that are considered program by the U.S. Agricultural Act of
with reference to their historical frequency of 2014, provides eligible livestock producers
occurrence for a location and time of year, so payments that are equal to 60% of their

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

that categorizations reflect local impacts and monthly feed costs. Producers are eligible for
vulnerability (Svoboda et al. 2002). one monthly payment if they own or lease
The fact that the USDM is a composite grazing land or pastureland located in a
product developed by a panel of experts may county rated by the USDM as having, during
make it less “objective” than specific weather the normal grazing period, a D2 drought for
indicators such as precipitation and tempera- eight consecutive weeks or more, and addi-
ture. However, drought is driven by multiple tional monthly payments for weeks in D3 or
environmental stresses that may combine in D4 drought (USDA 2017a). Between 2011
ways that are not entirely predictable and and 2016, more than $6 billion in LFP funds
that may cause cascading impacts (National were awarded (USDA 2017b). The LAGP,
Research Council; NRC 2007). The USDM the predecessor to the LFP, awarded $50 mil-
has the benefit of incorporating many more lion in 2007 alone (USDA 2006). Persistent
types of information that can improve under- drought conditions between 2001 and 2003 in
standing of drought conditions. In addition, the central United States prompted a
the USDM is a timely and easily interpreted Commodity Credit Corporation (CCC) sell-
data product, making it readily usable by reg- off of surplus dry milk to supplement feed sup-
ulators, producers, and the general public in plies to impacted livestock producers.
drought-related decisions. Eligibility for these sales hinged on USDM
Another drawback of using the USDM to drought designations; if on March 11 2003, any
determine the impacts of drought on agricul- part of a county was included in the D4 cate-
ture is that it includes a measure of vegeta- gory, or if it was in the D3 category and expe-
tion health. In the context of our empirical rienced D4 sometime between September 3
analysis, the inclusion of vegetation health 2002 and March 11 2003, surplus dry milk sales
may make our regressors endogenous. were available to producers in the county
Because of the way the USDM is developed, (USDA 2003). Automatic disaster designa-
it is impossible to determine how much em- tions triggered by the USDM authorize emer-
phasis is placed on vegetation, and as a result, gency loans to producers in affected counties
it is impossible to determine the degree to as well as in adjacent counties; more than $56
which the endogeneity is problematic for esti- million in emergency farm loans were pro-
mation. Nonetheless, to the extent that the vided in the 2015 and 2016 fiscal years (al-
USDM is used to inform drought policy, our though not all loans are related directly to
empirical results can help decision makers drought; USDA 2017c). The IRS also uses the
understand how USDM categorizations re- USDM to determine the time frame for waiv-
late to agricultural outcomes. ing gains from livestock replacement pur-
Despite its drawbacks, the USDM is for- chases due to drought (U.S. Department of
mally used to inform several major drought Treasury; USDT 2016). Finally, the private
management decisions. As of 2012, the sector uses the USDM in order to make deci-
USDA’s Secretarial disaster designation pro- sions about resource allocation that may be
cess provides for nearly automatic designation affected by the regional and temporal distribu-
for a county when, during the growing season, tion of drought (Bernknopf et al. 2018). In
any portion of the county is classified by the many states, USDM categorizations are
USDM as being affected by a drought with an combined with location-specific levels of its
intensity level of D2 for eight consecutive component metrics and serve as triggers for
weeks, or a higher drought intensity level for action in drought preparedness plans.
any length of time (USDA 2015). In addition,
the USDA has utilized the USDM to deter-
mine drought disaster assistance program eligi-
bility in programs such as the Livestock Forage Data
Disaster Program (LFP), the Livestock
Assistance Grant Program (LAGP), The key explanatory variables in our data set
Commodity Credit Corporation surplus stock are based on observations of USDM drought
198 January 2019 Amer. J. Agr. Econ.

categorizations, precipitation, and tempera- additional week of D1 drought covering all of

ture for 3,080 counties for every year between the agricultural area in the county.
2001 and 2013. Our dependent variables are Summary statistics for the aggregated
county-level yields of corn and soybeans, and USDM data are presented in the upper por-
measures of farm income per agricultural tion of table 2. As would be expected, the av-
acre. Yield and farm income data are not erage number of weeks that counties are

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

available for all counties for all years, so the assigned to a particular drought intensity
number of county-year observations varies class is smaller for more intense drought clas-
across our econometric specifications. ses. In addition, some regions of the United
States have experienced greater variability in
drought conditions relative to other regions.
Figure 1 is a map that illustrates regional dif-
USDM Data
ferences in the variability of drought condi-
Weekly USDM drought intensity categoriza- tions observed over the time period covered
tions do not necessarily follow county bound- by our USDM data. The map shows that
aries. Given our interest in drought’s effects counties with the most variable drought con-
on agricultural outcomes, we develop annual ditions are located along a band starting in
county-level USDM measures that reflect Texas and extending northward through the
drought occurrence in agricultural areas High Plains. A portion of the Southeastern
within counties during the twelve months United States also exhibits high variability in
preceding the end of the growing season. drought. Because our econometric specifica-
Specifically, for each week, we quantify the tions rely on the use of county fixed effects,
percentage of county agricultural area our coefficient estimates will be driven by the
experiencing each USDM drought category relationship between drought and agricul-
using a geographic information system (GIS). tural outcomes in these high-variability
Values for each drought level are summed counties.
from October of the previous year to
September of the current year, representing Crop Yield Data
the general time window during which crops
are affected by drought. County agricultural In order to evaluate the effect of drought (as
areas were determined by aggregating agri- designated by the USDM) on agricultural
cultural land cover categories observed in the production, we utilize corn and soybean yield
2008 Cropland Data Layer.2 Thus, our data, available annually from 2001 through
drought variables are defined as the number 2013 from NASS.3 Yield variables represent
of weeks that a county experiences a drought crop-specific ratios of total county production
of a given severity level, where each week is to total county acres harvested.4 Yield data,
weighted by the percentage of the county’s while surveyed annually, are not available for
agricultural area affected by that level of every county-year. Missing observations rep-
drought during that week. For example, our resent county-years with no production,
drought variable D1ist , representing exposure county-years with a sufficiently small number
to drought intensity level D1 for county i in of producers such that information is not dis-
state s year t, can be expressed as closed for privacy reasons, or county-years
that simply were not surveyed.
X
ð1ÞD1ist ¼ ðproportion of county i
w Farm Income Data
agricultural area in D1 drought
during week wÞ Farm income data were obtained from the
Bureau of Economic Analysis (BEA) for the
where w is the index for weeks falling be-
tween October of year t – 1 and September of
3
year t. Used as an explanatory variable in a NASS crop yield data can be accessed at: https://quickstats.
nass.usda.gov/.
regression, the coefficient associated with this 4
Existing studies on the impacts of weather on agricultural
variable can be interpreted as the effect of an outcomes use either harvested acres (e.g., Annan and Schlenker
2015) or planted acres (e.g., Desch^ enes and Greenstone 2007) to
calculate crop yields. There is little guidance in the literature on
which acreage measure is preferable. We use harvested acres,
2
The 2008 Cropland Data Layer can be accessed at: https:// which is the default choice in reporting by NASS and the Food
nassgeodata.gmu.edu/CropScape/. and Agriculture Organization of the United Nations (FAO).
Kuwayama et al. Estimating the Impact of Drought on Agriculture Using the U.S. Drought Monitor 199

Table 2. Summary Statistics

Variable Mean Std. dev. Minimum Maximum Observations
# weeks in D0 (weighted by % 8.47 7.34 0 50.98 40,040
agricultural area affected)
# weeks in D1 (weighted by % 5.66 7.11 0 51.05 40,040

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

agricultural area affected)
# weeks in D2 (weighted by % 3.87 6.8 0 52 40,040
agricultural area affected)
# weeks in D3 (weighted by % 2.26 5.84 0 52 40,040
agricultural area affected)
# weeks in D4 (weighted by % 0.8 3.55 0 51.15 40,040
agricultural area affected)
Cash receipts from marketings plus 100,858 201,018 0 5,033,282 9,240
value of inventory change (in
thousands of U.S. dollars)
Production Expenses (in thousands of 94,441 161,973 0 3,661,699 9,240
U.S. dollars)
Cash receipts from marketings plus 0.5 0.97 0 44.47 9,126
value of inventory change / farm
acres (in thousands of U.S. dollars)
Production Expenses / farm acres (in 0.48 0.81 0 37.04 9,126
thousands of U.S. dollars per acre)
Soy yield (bushels per acre) 37.59 10.44 0.7 68.5 21,247
Corn yield (bushels per acre) 126.04 37.12 4.5 244 25,188
Maximum harvested irrigated cropland 0.19 0.3 0.000015 1 2,909
acres / Harvested cropland acres
Soybean moderate heat degree days 1.9 0.54 0.22 3.22 40,027
(Celsius and days, thousands)
Corn moderate heat degree days 1.87 0.52 0.22 3.15 40,027
(Celsius and days, thousands)
Soybean extreme heat degree days 0.49 0.59 0 6.1 40,027
(Celsius and days, hundreds)
Corn extreme heat degree days (Celsius 0.71 0.74 0 7.06 40,027
and days, hundreds)
Precipitation (meters) 0.56 0.23 0.01 1.51 40,027
Note: Monetary values expressed in 2009 U.S. dollars.

three USDA Census of Agriculture years has been shown to be an important compo-
that overlap our USDM data timeframe: nent of farmer decision making in the pres-
2002, 2007, and 2012. Although the BEA pro- ence of weather fluctuations (Riebsame,
vides farm income estimates for every year, Changnon, and Karl 1991; Wheaton et al.
only census-year data were used because they 2008; Fisher et al., 2012). The major field
are the only years for which farm income esti- crops in the United States are often stored by
mates are consistently based on census data farmers in years with high yields or low pri-
and not interpolated or imputed (Bureau of ces, and these stocks are depleted in years of
Economic Analysis; BEA 2015). We choose high prices or low yields. Similarly, livestock
two outcomes of interest that reflect farm producers may alter their herd sizes in reac-
revenues and costs, both of which can be af- tion to market or weather conditions. Total
fected by drought through a variety of chan- cash receipts from marketings combined with
nels: (a) total cash receipts from marketings the value of inventory change represents rev-
plus the value of inventory change; and (b) enues generated by commodities produced
production expenses. under the current year’s conditions. Since
We rely on BEA farm income data instead these data constitute total county values, they
of census data because BEA provides esti- are normalized by the total county agricul-
mates of the value of inventory change, which tural acres for that year; data on agricultural
200 January 2019 Amer. J. Agr. Econ.

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

Figure 1. Variance of weighted drought index based on USDM drought classifications 2001
through 2013
Note: Weighted drought index is calculated by summing annual drought variables after they have been multiplied by a factor corresponding to severity (e.g.,
weeks in D0 is the identity and weeks in D4 is multiplied by five). The figure illustrates the regional variance of drought 2001 through 2013. The greatest vari-
ation occurs in the Western and Southeastern United States (darker shades).

acres are also available from NASS. Thus, we the four census years and use it as a time-
explore the effect of drought on revenues invariant variable that describes the potential
from current production and production for farmland in the county to be irrigated.6
expenses per agricultural acre.5 Although our We define a county to be irrigated if the pro-
farm income measures are only available for portion of harvested cropland that is irrigated
three years, county coverage is better than is greater than 15%. Counties with dryland
for our crop yield data, covering most coun- farming are those in which the proportion of
ties in the lower 48 states. harvested cropland that is irrigated is less
than 15%.7
Irrigated Acreage Data
In order to explore the potential mitigating Weather Data
effects of irrigation on drought impacts, we
We rely on modeled precipitation and tem-
collected additional data from NASS that
perature data, developed by Schlenker and
allows us to assign counties to one of two
Roberts (2009), to determine annual growing
groups: irrigated counties and dryland coun-
season weather conditions in agricultural
ties. In order to designate a county as irri-
areas. Specifically, we calculate cumulative
gated or dryland, we first take the area of
precipitation and both moderate and extreme
harvested cropland in the county that is irri-
heat degree days (Snyder 1985) between
gated and divide it by the total area of har-
vested cropland in that county as reported in
the 1997, 2002, 2007, and 2012 censuses. This 6
For some counties, data on irrigated harvested cropland area
value represents the proportion of harvested is not available in all Census years, and as a result, the proportion
cropland in the county that is irrigated in of harvested cropland that is irrigated cannot be calculated for
every Census year. For these counties, we take the maximum
each census year. We then take the largest value of the proportion of harvested cropland that is irrigated
proportion observed for each county across across years in which data are available.
7
The choice of a 15 percent cutoff for the proportion of har-
vested cropland that is irrigated is consistent with definitions of
irrigated and non-irrigated counties in previous studies, which
5
All monetary figures are expressed in terms of 2009 US use cutoffs that range from 5 percent to 20 percent [Schlenker
dollars. et al., 2005; Desch^enes and Greenstone, 2007].
Kuwayama et al. Estimating the Impact of Drought on Agriculture Using the U.S. Drought Monitor 201

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

Figure 2. County observations included in regression models employing different dependent
variables
Note: Light shades represent dryland counties and dark shades represent irrigated counties

April and September. County averages of are dropped does not lead to significantly dif-
daily precipitation and degree days observed ferent results.
in agricultural areas are summed throughout
the growing season. Following Annan and
Schlenker (2015), we implement different
growing degree day temperature thresholds Empirical Strategy
depending on the crop in question. Bounding
temperatures for moderate heat degree days The Impacts of Drought on Crop Yields
for corn and soybeans are 10-29 C and 10-
Our empirical strategy identifies the net
30 C, respectively; extreme heat degree days
impacts of additional weeks of drought in a
are those above 29 C and 30 C for corn and
given year, at various levels of intensity, on
soybeans, respectively.
crop yields and measures of farm income at
Summary statistics for all agricultural and
the county-year level, exploiting spatial and
weather data are presented in table 2.
inter-temporal variation in agricultural out-
Figure 2 presents four maps that indicate the
comes, drought intensity, and drought dura-
irrigated and dryland counties that are repre-
tion. First, we consider a parsimonious fixed-
sented in our corn yield, soybean yield, and
effects specification for log crop yield:
farm income regressions. Our panel is unbal-
anced due to inconsistent availability of crop
yield data across years. The empirical analy- ð2Þ yist ¼ a þ Dist 0  C þ kt þ ui þ gs ðtÞ þ eist
sis presented below is based on this unbal-
anced panel, although using a balanced panel where Dist represents a vector of five varia-
in which counties with missing years of data bles indicating the area-weighted number of
202 January 2019 Amer. J. Agr. Econ.

weeks in year t that agricultural areas of We estimate models (2), (3), and (4) using
county i in state s were categorized as observed yields of corn as well as soybeans.
experiencing each of the five USDM drought We run separate regressions for irrigated and
intensity levels. County fixed effects (ui ) help dryland counties. Irrigation is known to help
obtain unbiased estimates in the presence of counteract the deleterious effects of drought,
unobserved time-invariant characteristics of at least in the short run (Madariaga and

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

counties that affect their agricultural outcomes McConnell 1984; O’Brien et al. 2001;
in the face of drought, while year dummies (kt) Hansen, Libecap, and Lowe 2009; Hornbeck
help control for common trends that may be and Keskin 2014), and studies have shown
correlated with explanatory variables such as that accounting for irrigation in an analysis of
drought occurrence. We also control for the the impacts of climate change on U.S. agri-
fact that yields are trending upward over time culture leads to qualitatively different results
by using a state-specific linear time trend gs ðtÞ. (Schlenker, Hanemann, and Fisher 2005,
The vector C contains our coefficients of inter- 2006).
est, which capture the effect of an additional It is important to note that while time-
week of drought at each USDM intensity level invariant characteristics of counties that af-
on crop yields. We omit a variable representing fect crop yields are controlled for by our fixed
the number of weeks that a county is not indi- effects specifications, the impacts of drought
cated as being in any drought status, so the and weather on crop yields estimated in C
coefficients in C can be interpreted as the im- and D may include some mitigating factors as-
pact on crop yields when all of a county’s agri- sociated with farmer behavior and relevant
cultural area is affected by an additional week institutional, market, and policy environ-
of drought of a given severity (D0 through D4) ments that change over time differently
relative to not being in any drought at all. across counties. For example, farmers may be
To provide a comparison to models from able to curtail the impacts of drought and
previous studies that have estimated the rela- weather fluctuations on crop yields within the
tionship between crop yields and weather vari- growing season by the following: increasing
ables such as temperature and precipitation, we irrigation; changing harvest patterns and tim-
also estimate a version of equation (2) in which ing; and augmenting labor, fertilizer, and
we replace the vector of drought variables Dist other production inputs. Other factors that
with a vector of weather variables, Wist : may benefit farmers differentially in the pres-
ence of drought include crop insurance pay-
ð3Þ yist ¼ a þ Wist  D þ kt þ ui þ gs ðtÞ þ eit : ments and government drought disaster
assistance.
Following Annan and Schlenker (2015), We forgo specifications of Equations (2),
for our crop yield regressions, our weather (3), and (4) that implement state-by-year
variables include (a) the number of degree fixed effects in the place of year dummies and
days of moderate heat, (b) the number of de- state-specific time trends. State-by-year fixed
gree days of extreme heat, (c) annual precipi- effects may seem appealing because they will
tation, and (d) annual precipitation squared. help control for state-level inter-temporal
These regressions will help illustrate whether shocks such as changes in state-level agricul-
or not the USDM drought variables are supe- tural policy programs, agricultural markets,
rior to the weather variables traditionally or technological change. However, in our em-
used to estimate crop yield outcomes— pirical context, state-by-year fixed effects
namely, temperature and precipitation. have the potential to absorb a significant
A third set of crop yield regressions amount of variance in drought conditions,
includes both drought variables Dist and leading to large standard errors. This concern
weather variables Wist : was highlighted by Fisher et al. (2012), who
criticized the use of state-by-year fixed effects
by Desch^enes and Greenstone (2007) in
yist ¼ a þ Dist  C þ Wist  D þ kt þ ui
ð4Þ econometric specifications for identifying the
þ gs ðtÞ þ eist : effect of temperature on agricultural yield
and profits. As a result, the source of varia-
The purpose of these regressions is to ex- tion in our econometric specification is funda-
amine whether the two types of variables— mentally different from that which is used to
drought and weather—can explain different identify the impacts of weather on agricul-
aspects of variability in crop yields. tural outcomes in existing studies. Studies
Kuwayama et al. Estimating the Impact of Drought on Agriculture Using the U.S. Drought Monitor 203

that have employed the hedonic approach corn yields for an additional week of D4
(Mendelsohn, Nordhaus, and Shaw 1994; drought.
Schlenker, Hanemann, and Fisher 2005) rely Column 2 of the same table reports results
on cross-sectional variation across counties, from the estimation of equation (3) for corn
while Desch^ enes and Greenstone (2007), who yield outcomes in dryland counties. This re-
pioneered the panel approach, use both gression serves as a check for consistency

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

county fixed effects and state-by-year fixed with the existing literature, which has already
effects such that parameters are identified addressed the relationship between crop
from the county-specific deviations in yields and weather variables such as precipi-
weather from county averages after adjusting tation and temperature. Consistent with these
for shocks common to all counties in a state. existing studies, we find that degree days of
We account for potential spatial correlation moderate heat are beneficial for crop yields,
in the error terms by using an OLS estimator but degree days of extreme heat are harmful
with nonparametric estimation of the for yields. However, in our data, only the ex-
variance-covariance matrix for eist, developed treme heat variable is statistically significant.
by Hsiang (2010). Like in previous studies, we find that precipi-
tation has a positive and statistically signifi-
The Impacts of Drought on Farm Income cant relationship with yields, and the
quadratic term for precipitation is negative
We also estimate equations (2), (3), and (4) and statistically significant. The magnitudes
using two measures of farm income—cash of these estimated effects are similar to those
receipts from marketings (net of the value of presented in studies that use the same tem-
inventory changes) and production perature and precipitation measures (Annan
expenses—as our dependent variables. The and Schlenker 2015).
counties and years for which farm income When drought and weather variables are
data are available are not the same as those simultaneously included as explanatory varia-
for crop yield data, so results from the farm bles for corn yield (equation (4)), the coeffi-
income regressions are not directly compara- cient estimates for the number of weeks of
ble to the results from the crop yield regres- D2 and D3 drought remain negative and sig-
sions. Other than the fact that they are nificant (table 3, column 3). However, the
monetized outcomes, our farm income meas- magnitudes of these coefficients are less than
ures differ from our crop yield measures in half of the magnitudes estimated in the speci-
that the coefficient estimates may also in- fication that only includes drought variables.
clude the effect of changes in local prices of At the same time, the magnitude and signifi-
agricultural commodities due to scarcity that cance of the coefficients for the weather vari-
results from droughts, which may increase ables are similar to those estimated in the
net farm income if yields and production weather-only regressions. We also note that
expenses are not significantly affected. As the R-squared statistic for the drought-plus-
with the crop yield regressions, our farm in- weather specification is much closer to the
come regressions are performed separately R-squared statistic in the weather-only speci-
for irrigated and dryland counties. fication than to the R-squared statistic in the
drought-only specification. Collectively, these
regression outcomes suggest that precipita-
Results tion and temperature explain most of the
observed variability in crop yields. While the
Column 1 in table 3 reports results from the drought variables add little explanatory
estimation of equation (2) for dryland coun- power in the estimation of equation (4), the
ties when the dependent variable is log corn statistically significant coefficients for D2 and
yields. The coefficient estimates are all nega- D3 drought do represent meaningful impacts;
tive and statistically significant, indicating that is, even after controlling for precipitation
that additional weeks of drought in any sever- and temperature, an additional week of D2
ity category are associated with reduced corn or D3 drought is associated with reductions
yields. The magnitude of this negative impact in corn yields of 0.3% to 0.5%. Furthermore,
is larger for additional weeks of drought in a Wald test indicates that the five drought
higher severity categories and ranges from a variables are jointly significant at the 1%
0.1% decrease in corn yields for an additional level. These additional impacts captured by
week of D0 drought up to a 1.2% decrease in the drought variables suggest that there are
204 January 2019 Amer. J. Agr. Econ.

Table 3. Impact of Additional Weeks of Drought and Weather in Agricultural Areas on Log
Corn Yield
Dryland Counties Irrigated Counties
(1) (2) (3) (4) (5) (6)
Drought Weather Drought Drought Weather Drought

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

monitor only monitor & monitor only monitor &
only weather only weather
# weeks in D0 0.00142c 8.37e-05 0.000939b 0.000258
(0.000425) (0.000365) (0.000374) (0.000349)
# weeks in D1 0.00318c 0.000441 0.000394 0.000413
(0.000519) (0.000430) (0.000439) (0.000393)
# weeks in D2 0.00622c 0.00284c 0.00135c 0.000229
(0.000735) (0.000572) (0.000395) (0.000381)
# weeks in D3 0.0109c 0.00484c 0.00125b 0.000311
(0.00107) (0.000851) (0.000512) (0.000450)
# weeks in D4 0.0119c 0.00166 0.00485c 0.000953
(0.00205) (0.00140) (0.000940) (0.000822)
Moderate degree 0.0724 0.0926 0.109a 0.110a
days (thousands) (0.0570) (0.0566) (0.0569) (0.0574)
Extreme degree 0.571c 0.534c 0.205c 0.203c
days (hundreds) (0.0227) (0.0230) (0.0169) (0.0170)
Precipitation (meters) 0.635c 0.572c 0.0867 0.0917
(0.124) (0.122) (0.110) (0.110)
Precipitation squared 0.408c 0.381c 0.0613 0.0672
(0.0872) (0.0862) (0.0831) (0.0828)
Constant 4.365c 4.420c 4.378c 4.481c 4.939c 4.931c
(0.100) (0.174) (0.172) (0.0751) (0.161) (0.162)
Observations 17,182 17,182 17,182 5,061 5,061 5,061
R2 within 0.3043 0.4525 0.4618 0.1936 0.2801 0.2815
R2 between 0.0525 0.2937 0.3074 0.0043 0.0000 0.0001
R2 overall 0.1459 0.3273 0.3471 0.0086 0.0268 0.0281
Note: Independent variables included in each specification are labeled in the second row. Standard errors account for potential spatial correlation in the error
terms and appear in parentheses. All models include county and year fixed effects and state-specific linear time trends. All drought variables are weighted by
the percentage of county agricultural area affected. Significance is denoted as follows: a ¼ 10% level, b ¼ 5% level, and c ¼ 1% level.

hydrologic and climatic factors other than significant. In the drought-plus-weather specifi-
precipitation and temperature that may influ- cation (column 6), the coefficients on the
ence crop yields. weather variables are very similar to those esti-
Columns 4, 5, and 6 in table 3 report results mated in the weather-only specification, and
from the estimation of equations (2), (3), and the coefficients on the drought variables are
(4) for corn yield in irrigated counties. As not statistically significant. These results suggest
would be expected, yields are less sensitive to that, once controlling for temperature and pre-
drought and weather in these irrigated counties cipitation, the USDM contributes little to
relative to the dryland counties. In the drought- explaining crop yields in irrigated counties.
only specification (column 4), additional weeks Our regression results for soybean yields
of drought are associated with negative impacts are similar to those for corn yields in dryland
on corn yield at every drought severity level ex- counties (columns 1, 2, and 3 in table 4).
cept for D1, but the magnitude of these impacts Drought is associated with lower soybean
is smaller than those estimated for dryland yields at every severity level, with negative
counties, with reductions ranging from 0.1% to impacts for an additional week of drought
0.5%. The weather-only specification (column ranging from 0.2% to 0.8%. In the weather-
5) yields coefficient estimates for degree days only regression, degree days of moderate
of moderate heat and extreme heat that are heat are associated with higher soybean
negative and statistically significant, while the yields, and unlike the case for corn, this rela-
relationships between the precipitation varia- tionship is statistically significant. Degree
bles and corn yields are not statistically days of extreme heat are harmful for soybean
Kuwayama et al. Estimating the Impact of Drought on Agriculture Using the U.S. Drought Monitor 205

Table 4. Impact of Additional Weeks of Drought and Weather in Agricultural Areas on Log
Soy Yield
Dryland Counties Irrigated Counties
(1) (2) (3) (4) (5) (6)
Drought Weather Drought Drought Weather Drought

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

monitor only monitor & monitor only monitor &
only weather only weather
# weeks in D0 0.00184c 0.000119 0.000127 0.00143b
(0.000446) (0.000375) (0.000628) (0.000574)
# weeks in D1 0.00602c 0.00266c 0.00216c 0.000498
(0.000561) (0.000449) (0.000713) (0.000657)
# weeks in D2 0.00591c 0.00183c 0.00292c 0.00117a
(0.000683) (0.000526) (0.000725) (0.000702)
# weeks in D3 0.00862c 0.00183b 0.00410c 0.00200c
(0.00109) (0.000774) (0.000771) (0.000696)
# weeks in D4 0.00657b 0.00357b 0.000768 0.00349c
(0.00288) (0.00182) (0.00129) (0.00112)
Moderate degree 0.409c 0.418c 0.290c 0.241b
days (thousands) (0.0576) (0.0577) (0.0943) (0.0939)
Extreme degree 0.696c 0.670c 0.300c 0.288c
days (hundreds) (0.0235) (0.0234) (0.0308) (0.0307)
Precipitation (meters) 1.490c 1.444c 0.645c 0.682c
(0.114) (0.115) (0.178) (0.173)
Precipitation squared 0.884c 0.872c 0.345c 0.382c
(0.0793) (0.0789) (0.130) (0.127)
Constant 3.108c 1.925c 1.928c 3.625c 2.785c 2.876c
(0.157) (0.170) (0.173) (0.215) (0.311) (0.309)
Observations 15,366 15,366 15,366 3,391 3,391 3,391
R2 within 0.2408 0.4732 0.4816 0.1851 0.2863 0.2974
R2 between 0.0998 0.202 0.2049 0.0044 0.1078 0.0887
R2 overall 0.1683 0.2896 0.2972 0.0122 0.0205 0.0107
Note: Independent variables included in each specification are labeled in the second row. Standard errors account for potential spatial correlation in the error
terms and appear in parentheses. All models include county and year fixed effects and state-specific linear time trends. All drought variables are weighted by
the percentage of county agricultural area affected. Significance is denoted as follows: a ¼ 10% level, b ¼ 5% level, and c ¼ 1% level.

yields, and this negative effect is an order of drought and weather variables relative to
magnitude larger than the beneficial effect of dryland counties (as was the case with corn
moderate heat degree days, which is consis- yields). In the drought-only specification,
tent with previous literature. In addition, co- negative impacts for an additional week of
efficient estimates for both the linear and drought range from 0.2% to 0.4%, while in
quadratic precipitation terms are statistically the weather-only specification, the coefficients
significant and indicate an inverted U-shape for the precipitation and temperature variables
relationship with soybean yields. As with are half the magnitudes of those estimated for
corn yields, when drought variables are dryland counties. We struggle to find an intui-
added to the weather variables as explana- tive explanation for the positive and significant
tory variables in the soybean regression, the effect of additional weeks of D4 drought in the
magnitudes of the coefficient estimates for drought-plus-weather specifications for both ir-
the weather variables are similar to those es- rigated and dryland counties.
timated in the weather-only regression, and Table 5 reports results from our three econo-
coefficients for some of the drought variables metric specifications, for dryland and irrigated
remain statistically significant. A Wald test counties, using the log of cash receipts per acre
indicates that the five drought variables are from marketings plus the value of inventory
jointly significant at the 1% level for soybean change per acre as the dependent variable.
yields in dryland counties. Table 6 reports results from these same specifi-
In irrigated counties (columns 4, 5, and 6 in cations using the log of production expenses
table 4), soy yields are less sensitive to per acre as the dependent variable. We find
206 January 2019 Amer. J. Agr. Econ.

Table 5. Impact of Additional Weeks of Drought and Weather in Agricultural Areas on

Farm Income (Log Cash Receipts from Marketings Plus Value of Inventory Change / Farm
Acres)
Dryland Counties Irrigated Counties
(1) (2) (3) (4) (5) (6)

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

Drought Weather Drought Drought Weather Drought
monitor only monitor & monitor only monitor &
only weather only weather
# weeks in D0 0.000253 0.000292 0.000419 0.000179
(0.000215) (0.000228) (0.000393) (0.000379)
# weeks in D1 6.21e-05 0.000179 6.76e-05 8.98e-05
(0.000216) (0.000228) (0.000292) (0.000330)
# weeks in D2 0.000318 0.000322 0.000142 0.000246
(0.000280) (0.000281) (0.000323) (0.000350)
# weeks in D3 0.00104c 0.000916c 0.000221 0.000372
(0.000313) (0.000318) (0.000343) (0.000337)
# weeks in D4 0.000288 0.000313 0.000575 6.89e-05
(0.000480) (0.000527) (0.000558) (0.000616)
Moderate degree 0.0564 0.0485 0.134c 0.150b
days (thousands) (0.0360) (0.0348) (0.0512) (0.0587)
Extreme degree 0.0283c 0.0258c 0.00746 0.00657
days (hundreds) (0.00702) (0.00691) (0.0103) (0.0101)
Precipitation (meters) 0.0132 0.00405 0.00832 0.00733
(0.0430) (0.0437) (0.0773) (0.0803)
Precipitation squared 0.0189 0.00951 0.0400 0.0386
(0.0328) (0.0328) (0.0656) (0.0686)
Constant 0.105c 0.00434 0.0123 0.268c 0.616c 0.662c
(0.0298) (0.0932) (0.0902) (0.0222) (0.125) (0.141)
Observations 6,078 6,078 6,078 2,598 2,598 2,598
R2 within 0.5079 0.5078 0.5101 0.2918 0.2964 0.2972
R2 between 0.0757 0.0758 0.0804 0.0108 0.0248 0.0242
R2 overall 0.1145 0.1148 0.1192 0.0001 0.0059 0.0064
Note: Independent variables included in each specification are labeled in the second row. Standard errors account for potential spatial correlation in the error
terms and appear in parentheses. All models include county and year fixed effects and state-specific linear time trends. All drought variables are weighted by
the percentage of county agricultural area affected. Significance is denoted as follows: a ¼ 10% level, b ¼ 5% level, and c ¼ 1% level.

that the relationships between drought, cash receipts from marketings and production
weather, and these farm income measures are expenses—exclude payments from crop insur-
much less clear than the relationship between ance and drought disaster assistance programs
drought, weather, and crop yields, with a gen- that may become available to farmers during
eral lack of statistical significance across coeffi- drought episodes. However, depending on the
cient estimates, especially in irrigated counties. timing of these payments, farmers may be able
In dryland counties, additional weeks of D3 to use income from these payments to adjust
drought are associated with a decrease in cash their on-farm practices in order to offset the
receipts and additional weeks of D0 and D2 negative impacts of drought on crop yields.
drought are associated with increases in pro- We also note that our farm income results rely
duction expenses, but the magnitudes of these
effects are very small.
The lack of statistically significant responses 8
We are unable to conduct a sub-analysis of the relationship
of farm income measures to drought and between drought and the price of agricultural commodities at the
weather may be due to the fact that farmers same level of detail as the rest of our analysis because price data
are more able to curtail the impacts of drought are only available to us at the state level. Regressions using state-
level prices of corn and soybeans as dependent variables and
on farm income than on yield. This is likely to drought variables aggregated to the state level as independent
be the case if local scarcity of agricultural out- variables result in positive and significant effects of drought on
corn and soybean prices for D4 drought only, and these coeffi-
puts caused by drought raises prices received cient estimates become insignificant once year dummies are in-
by farmers.8 Our measures of farm income— cluded in the specifications.
Kuwayama et al. Estimating the Impact of Drought on Agriculture Using the U.S. Drought Monitor 207

Table 6. Impact of Additional Weeks of Drought and Weather in Agricultural Areas on

Production Expenses (Log Production Expenses/Farm Acres)
Dryland Counties Irrigated Counties
(1) (2) (3) (4) (5) (6)
Drought Weather Drought Drought Weather Drought

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

monitor only monitor & monitor only monitor &
only weather only weather
# weeks in D0 0.000534c 0.000636c 2.41e-06 9.46e-05
(0.000189) (0.000198) (0.000367) (0.000348)
# weeks in D1 0.000244 8.91e-05 0.000210 2.22e-06
(0.000191) (0.000199) (0.000269) (0.000282)
# weeks in D2 0.000429a 0.000565b 0.000177 0.000436
(0.000240) (0.000240) (0.000299) (0.000313)
# weeks in D3 0.000339 0.000238 0.000224 0.000283
(0.000282) (0.000283) (0.000322) (0.000312)
# weeks in D4 6.06e-05 0.000188 0.000233 0.000132
(0.000499) (0.000526) (0.000513) (0.000590)
Moderate degree 0.0980c 0.113c 0.103b 0.124b
days (thousands) (0.0311) (0.0302) (0.0480) (0.0552)
Extreme degree 0.00408 0.00703 0.00204 0.00172
days (hundreds) (0.00603) (0.00606) (0.00951) (0.00927)
Precipitation (meters) 0.0187 0.0202 0.0332 0.0337
(0.0414) (0.0413) (0.0617) (0.0632)
Precipitation squared 0.0152 0.0227 0.0926a 0.0912a
(0.0308) (0.0305) (0.0505) (0.0521)
Constant 0.0846b 0.342c 0.363c 0.240c 0.517c 0.566c
(0.0843) (0.0333) (0.0811) (0.0149) (0.112) (0.129)
Observations 6,078 6,078 6,078 2,598 2,598 2,598
R2 within 0.4959 0.4956 0.498 0.2295 0.2388 0.2396
R2 between 0.0771 0.0538 0.0533 0.0111 0.0249 0.0219
R2 overall 0.1116 0.0884 0.0874 0.0000 0.0054 0.0056
Note: Independent variables included in each specification are labeled in the second row. Standard errors account for potential spatial correlation in the error
terms and appear in parentheses. All models include county and year fixed effects and state-specific linear time trends. All drought variables are weighted by
the percentage of county agricultural area affected. Significance is denoted as follows: a ¼ 10% level, b ¼ 5% level, and c ¼ 1% level.

on only 3 years of data (from 2002, 2007, and Impacts are particularly large in dryland coun-
2012), in contrast to the 13 years of data that ties in the Midwest, where an additional week
are available for crop yields. It is possible that of D4 drought is associated with an 8.0% de-
the small number of years used in our farm in- crease in corn yields and a 3.1% decrease in
come regressions are not representative of soybean yields. These regional results suggest
year-to-year fluctuations in farm income due that the USDM may be more useful for identi-
to coincident changes in agricultural policy, fying communities affected by drought in the
the ethanol boom and corresponding increases High Plains, Midwest, and South. Our results
in crop prices, and the 2012 Midwest drought. also suggest that drought disaster assistance
Finally, we estimated equation (2) for corn programs that apply the same eligibility criteria
and soybean yields in dryland and irrigated for all regions using USDM categorizations
counties in six different regions of the United (e.g., a county must have experienced D3 or
States: High Plains, Midwest, Northeast, South, D4 drought for at least eight consecutive
and Southeast.9 We find that the expected neg- weeks) may not target the farmers who are
ative relationship between drought and crop most vulnerable to drought.
yields is consistently observed in dryland coun-
ties in the High Plains, Midwest, and South
regions but less so in the other two regions. Conclusion

Despite growing interest within the research

9
We omit the western region of the United States from our anal- and policy communities, few studies quantify
ysis due to a lack of crop yield data from counties in this region. the effect of drought on agricultural outcomes
208 January 2019 Amer. J. Agr. Econ.

in the United States. In this paper, we estimated our framework can help policymakers form di-
the impacts of drought, as defined by the rect links between the provision of drought di-
USDM, on crop yields and farm income. We saster assistance and the economic impacts that
find negative and statistically significant effects these programs seek to mitigate.
of drought on crop yields in the average county, Building on our results, there are several
equal to yield reductions in the range of 0.1% to ways in which future work could delve deeper

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

1.2% for corn and soybean yields for each addi- into the impacts of drought on agriculture. For
tional week of drought in dryland counties, and example, there is evidence that crop production
0.1% to 0.5% in irrigated counties. However, is affected not only by the intensity and dura-
region-specific outcomes vary, with some tion of a drought, but also its timing with re-
regions experiencing no yield impacts from spect to the growing season (Walthall et al.
drought, while we observe yield reductions as 2012). Future work could examine whether
high as 8.0% in dryland counties in the Midwest drought categorizations are particularly harm-
for every week of drought in the highest severity ful during certain parts of the calendar year, al-
category. Despite this impact on crop yields, we though this would require careful consideration
find that additional weeks of drought have little of the econometric approach as effects are
to no effect on measures of farm income. likely to be heterogeneous across regions and
Using regression specifications that include crop types. In addition, future work could quan-
weather variables in addition to drought vari- tify the degree to which on-farm practices, crop
ables, we find that precipitation and tempera- insurance programs, and disaster assistance pol-
ture explain most of the observed variability icies affect observed agricultural outcomes in
in crop yields. However, additional weeks of the presence of drought.
drought in certain severity categories are as-
sociated with negative and statistically signifi-
cant impacts on crop yields even after
controlling for precipitation and temperature. Supplementary Material
This finding suggests that the USDM contains
information regarding impacts on crop Supplementary material are available at
yields above and beyond what can be drawn American Journal of Agricultural Economics
from observing only temperature and precipi- online.
tation, and serves as evidence to support the
official use of the USDM as a complement to
weather information for making resource al- References
location decisions within government drought
disaster assistance programs. At the same Annan, F., and W. Schlenker. 2015. Federal
time, our analysis suggests that the USDM Crop Insurance and the Disincentive to
may be more helpful for identifying drought Adapt to Extreme Heat. American
impacts in some regions of the United States Economic Review 105 (5): 262–6.
and less so in others. In particular, we find Bureau of Economic Analysis. 2015. Local
that the USDM is more correlated with crop Area Personal Income Methodology.
yield outcomes in dryland counties than in ir- Washington DC: BEA.
rigated counties, and is more correlated with Bernknopf, R., D. Brookshire, Y.
crop yield in the High Plains, Midwest, and Kuwayama, M. Macauley, M. Rodell, A.
South than in the Northeast and Southeast. Thompson, P. Vail, and B. Zaitchik.
Our analysis focused specifically on the ef- 2018. The Value of Remotely Sensed
fect of additional weeks of drought on crop Information: The Case of a GRACE-
yields and components of farm income, but our Enhanced Drought Severity Index.
USDM-based framework can be used to esti- Weather, Climate, and Society 10 (1):
mate other types of relationships between 187–203.
droughts and agriculture. For example, our Cattivellia, L., F. Rizza, F.W. Badeck, E.
econometric approach could be used to esti- Mazzucotelli, A.M. Mastrangelo, E.
mate the effects of droughts with a longer dura- Francia, C. Marè, A. Tondelli, and A.M.
tion; such an analysis may help policymakers Stanca. 2008. Drought Tolerance
choose the appropriate number of weeks that Improvement in Crop Plants: An
counties need to be in drought, as categorized Integrated View from Breeding to
by the USDM, before being eligible for Genomics. Field Crops Research 105 (1–
drought disaster assistance. More generally, 2): 1–14.
Kuwayama et al. Estimating the Impact of Drought on Agriculture Using the U.S. Drought Monitor 209

Cavatassi, R., L. Lipper, and U. Narloch. Adaptation to Groundwater and

2011. Modern Variety Adoption and Drought. American Economic Journal:
Risk Management in Drought Prone Applied Economics 6 (1): 190–219.
Areas: Insights from the Sorghum Horridge, M., J. Madden, and G. Wittwer.
Farmers of Eastern Ethiopia. 2005. The Impact of the 2002-2003
Agricultural Economics 42 (3): 279–92. Drought on Australia. Journal of Policy

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

Ciais, P., M. Reichstein, N. Viovy, A. Modeling 27 (3): 285–308.
Granier, J. Ogee, V. Allard, M. Aubinet, Houborg, R., M. Rodell, B. Li, R. Reichle,
et al. 2005. Europe-Wide Reduction in and B. F. Zaitchik. 2012. Drought
Primary Productivity Caused by the Heat Indicators Based on Model-Assimilated
and Drought in 2003. Nature 437 (22): Gravity Recovery and Climate
529–33. Experiment (GRACE) Terrestrial Water
Climate Change Science Program. 2008. The Storage Observations. Water Resources
Effects of Climate Change on Agriculture, Research 48 (7): doi:10.1029/
Land Resources, Water Resources, and 2011WR011291.
Biodiversity in the United States. Howitt, R.E., J. Medellın-Azuara, D.
Washington DC: U.S. Department of MacEwan, J.R. Lund, and D.A. Sumner.
Agriculture. 2014. Economic Analysis of the 2014
Craine, J.M., T.W. Ocheltree, J.B. Nippert, Drought for California Agriculture.
E.G. Towne, A.M. Skibbe, S.W. Kembel, Davis, CA: Center for Watershed
and J.E. Fargione. 2013. Global Diversity Sciences, University of California.
of Drought Tolerance and Grassland ———. 2015. Economic Analysis of the 2015
Climate-Change Resilience. Nature Drought for California Agriculture.
Climate Change 3 (1): 63–7. Davis, CA: Center for Watershed
Desch^enes, O., and M. Greenstone. 2007. Sciences, University of California.
The Economic Impacts of Climate Hsiang, S.M. 2010. Temperatures and
Change: Evidence from Agricultural Cyclones Strongly Associated with
Output and Random Fluctuations in Economic Production in the Caribbean
Weather. American Economic Review 97 and Central America. Proceedings of the
(1): 354–85. National Academy of Sciences 107 (35):
Ding, Y., K. Schoengold, and T. Tadesse. 15367–72.
2009. The Impact of Weather Extremes Intergovernmental Panel on Climate Change.
on Agricultural Production Methods: 2014. Climate Change 2014: Synthesis
Does Drought Increase Adoption of Report. Contribution of Working Groups
Conservation Tillage Practices? Journal I, II and III to the Fifth Assessment
of Agricultural and Resource Economics Report of the Intergovernmental Panel on
34 (3): 395–411. Climate Change. Geneva.
Fafchamps, M., C. Udry, and K. Czukas. Kazianga, H., and C. Udry. 2006.
1998. Drought and Saving in West Consumption Smoothing? Livestock,
Africa: Are Livestock a Buffer Stock? Insurance and Drought in Rural Burkina
Journal of Development Economics 55 Faso. Journal of Development Economics
(2): 273–305. 79 (2): 413–46.
Fisher, A.C., W.M. Hanemann, M.J. Roberts, Kinsey, B., K. Burger, and J.W. Gunning.
and W. Schlenker. 2012. The Economic 1998. Coping with Drought in
Impacts of Climate Change: Evidence Zimbabwe: Survey Evidence on
from Agricultural Output and Random Responses of Rural Households to Risk.
Fluctuations in Weather: Comment. World Development 26 (1): 89–110.
American Economic Review 102 (7): Lloyd-Hughes, B. 2014. The Impracticality of
3749–60. a Universal Drought Definition.
Hansen, Z.K., G.D. Libecap, and S.E. Lowe. Theoretical and Applied Climatology 117
2009. Climate Variability and Water (3–4): 607–11.
Infrastructure: Historical Experience in the Lobell, D.B., G.L. Hammer, G. McLean, C.
Western United States. NBER Working Messina, M.J. Roberts, and W.
Paper No. 15558. Cambridge, MA. Schlenker. 2013. The Critical Role of
Hornbeck, R., and P. Keskin. 2014. The Extreme Heat for Maize Production in
Historically Evolving Impact of the the United States. Nature Climate
Ogallala Aquifer: Agricultural Change 3 (5): 497–501.
210 January 2019 Amer. J. Agr. Econ.

Madariaga, B., and K.E. McConnell. 1984. Analysis of Optimal Growing

Value of Irrigation Water in the Middle Conditions. Review of Economics and
Atlantic States: An Econometric Statistics 88 (1): 113–25.
Approach. Southern Journal of Snyder, R.L. 1985. Hand Calculating Degree
Agricultural Economics 16 (2): 91–8. Days. Agricultural and Forest
Medellın-Azuara, J., D. MacEwan, R.E. Meteorology 35 (1–4): 353–8.

Downloaded from https://academic.oup.com/ajae/article-abstract/101/1/193/5051676 by St Petersburg State University user on 19 July 2020

Howitt, D.A. Sumner, and J.R. Lund. Svoboda, M., D. LeComte, M. Hayes, R.
2016. Economic Analysis of the 2016 Heim, K. Gleason, J. Angel, B. Rippey,
Drought for California Agriculture. et al. 2002. The Drought Monitor.
Center for Watershed Sciences, Bulletin of the American Meteorological
University of California, Davis. Society 83 (8): 1181–90.
Mendelsohn, R., W.D. Nordhaus, and D. U.S. Department of Agriculture. 2003. Fact
Shaw. 1994. The Impact of Global Sheet: Sales of Surplus Non-Fat Dry
Warming on Agriculture: A Ricardian Milk. Washington DC: USDA.
Analysis. American Economic Review 84 ———. 2006. Fact Sheet: Livestock Assistance
(4): 753–71. Grant Program. Washington DC: USDA.
Mount, J., E. Hanak, C. Chappelle, B. Colby, ———. 2015. Fact Sheet: Emergency Disaster
R. Frank, G. Gartrell, B. Gray, et al. Designation and Declaration Process.
2016. Improving the Federal Response to Washington DC: USDA.
Western Drought. San Francisco, CA: ———. 2017a. Fact Sheet: Livestock Forage
Public Policy Institute of California. Disaster Program. Washington DC:
National Drought Mitigation Center. 2016. USDA.
U.S. Drought Monitor Background. ———. 2017b. Livestock Forage Payments as
Lincoln, NE. of March 2, 2017. Washington DC:
National Research Council. 2007. USDA.
Understanding Multiple Environmental ———. 2017c. Farm Loan Programs
Stresses: Report of a Workshop. Program Data. Washington DC: USDA.
Washington DC. U.S. Department of Treasury. 2016. Internal
O’Brien, D.M., F.R. Lamm, L.R. Stone, and Revenue Bulletin: 2016-42 Notice 2016-60
D.H. Rogers. 2001. Corn Yields and Extension of Replacement Period for
Profitability for Low-Capacity Irrigation Livestock Sold on Account of Drought.
Systems. Applied Engineering in Washington DC: USDT.
Agriculture 17 (3): 315–21. Walthall, C.L., J. Hatfield, P. Backlund, L.
Riebsame, W.E., S.A. Changnon Jr., and Lengnick, E. Marshall, M. Walsh, S.
T.R. Karl. 1991. Drought and Natural Adkins, et al. 2012. Climate Change and
Resources Management in the United Agriculture in the United States: Effects
States: Impacts and Implications of the and Adaptation. Washington DC: U.S.
1987-89 Drought. Boulder, CO: Department of Agriculture, Technical
Westview Press. Bulletin No. 1935.
Schlenker, W., and M.J. Roberts. 2009. Wheaton, E., S. Kulshreshtha, V. Wittrock,
Nonlinear Temperature Effects Indicate and G. Koshida. 2008. Dry Times: Hard
Severe Damages to U.S. Crop Yields un- Lessons from the Canadian Drought of
der Climate Change. Proceedings of the 2001 and 2002. Canadian Geographer/Le
National Academy of Sciences 106 (37): Geographe Canadien 52 (2): 241–62.
15594–8. Wilhite, D.A. 2000. Drought as a Natural
Schlenker, W., W.M. Hanemann, and A.C. Hazard: Concepts and Definitions. In
Fisher. 2005. Will U.S. Agriculture Drought: A Global Assessment, Volume
Really Benefit from Global Warming? 1, ed. D.A. Wilhite, 3–18. London:
Accounting for Irrigation in the Hedonic Routledge.
Approach. American Economic Review Zhao, M., and S.W. Running. 2010. Drought-
95 (1): 395–406. Induced Reduction in Global Terrestrial
———. 2006. The Impact of Global Warming Net Primary Production from 2000
on U.S. Agriculture: An Econometric through 2009. Science 329 (5994): 940–3.