Regional Crop Diversity and Weather

Shocks in India
Maximilian Auffhammer and Tamma A. Carleton∗

Agriculture in both the developing and developed country context is highly

sensitive to weather shocks. The intensity of these shocks is likely to increase
under climate change, leading to an ongoing debate regarding the ability
of farmers to insulate yields and income against accelerating environmental
extremes. We study crop diversity as an avenue for increased resilience.
Diversity in agricultural systems has been suggested in the agroecology and
environmental economics literatures as a powerful means of on-farm insurance,
both through physical and market-based channels. However, large-scale
empirical evidence of its effectiveness is lacking, and crop diversity is largely
absent from the empirical climate impacts literature. We examine the insurance
benefits of crop diversity in the context of India at the height of the Green
Revolution, a period of rapid change in agricultural diversification due to the
increased penetration of a small set of high-yielding variety crops. Building
on a basic empirical model from the climate impacts literature, we show
that areas with higher crop diversity of planted area display measurably more
drought resilience, both in terms of gross and net revenues. We decompose this
aggregate result to show that diversification has implications for farmer welfare
both through physical (yield) and market (price) channels.

Keywords: agriculture, climate change, crop diversity, weather shocks

JEL codes: Q10, Q15

I. Introduction

Food production, and hence farmer welfare, is highly vulnerable to weather

shocks, particularly in the developing world, where few insurance mechanisms
exist to buffer farmers against extreme weather events. The link between
climatic conditions and agricultural yields has been extensively empirically
documented for major crops in the North American context (Burke and Emerick
2016, Schlenker and Roberts 2009), and these findings have been replicated
across the globe (Carleton and Hsiang 2016, Auffhammer and Schlenker 2014,

∗
Maximilian Auffhammer (corresponding author): Professor and Associate Dean of Social Sciences, University
of California, Berkeley. E-mail: auffhammer@berkeley.edu; Tamma A. Carleton: PhD Candidate, University of
California, Berkeley. E-mail: tcarleton@berkeley.edu. We thank the participants at the Asian Development Review
Conference in Seoul in August 2017, the managing editor, and two anonymous referees for helpful comments and
suggestions. The usual disclaimer applies.

Asian Development Review, vol. 35, no. 2, pp. 113–130 © 2018 Asian Development Bank and
https://doi.org/10.1162/adev_a_00116 Asian Development Bank Institute.
Published under a Creative Commons
Attribution 3.0 International (CC BY 3.0) license.
114 Asian Development Review

IPCC 2014). Indian agriculture is particularly susceptible to yield damage

under adverse climate events, as production is heavily dependent on uncertain
monsoon rainfall (Fishman 2016; Auffhammer, Ramanathan, and Vincent 2012),
crop-damaging temperatures regularly depress yields (Carleton 2017; Welch et al.
2010; Auffhammer, Ramanathan, and Vincent 2006), and formal crop insurance
is rare. By increasing the intensity of extreme climate conditions, global climate
change is likely to exacerbate this vulnerability, raising important questions
regarding whether and how the agricultural sector will adapt.
There are many adaptive behaviors and technologies that enable farmers
to cope with weather-based uncertainty. These include investment in irrigation;
development and use of new drought- and/or submergence-tolerant crop varieties
(Dar et al. 2013); transfer of land and labor to nonagricultural sectors (Colmer 2016;
Mendelsohn, Nordhaus, and Shaw 1994); shifting timing of planting or harvesting
(Wright and Gardner 1995); and investment in formal crop insurance (Annan
and Schlenker 2015). However, recent empirical work focusing on crop-specific
avenues of adaptation has uncovered little to no evidence that farmers have
successfully reduced yield sensitivity to accelerating climate change (Burke and
Emerick 2016; Lobell et al. 2014; Schlenker, Roberts, and Lobell 2013; Schlenker
and Roberts 2009). In contrast, a parallel body of work in agroecology has turned
attention to across-crop solutions, suggesting that crop diversity has the potential
to buffer both productivity (Bellora et al. 2017; Zimmerer 2010; Di Falco and
Chavas 2008, 2009) and income (Baumgärtner and Quaas 2009, Di Falco and
Perrings 2003) from adverse climate conditions. While researchers have proposed
many mechanisms through which diversification of crop portfolios can enable
adaptation, there is a dearth of large-scale empirical evidence of its effectiveness.
Here, we use rich panel data from 270 districts across India during a period of
rapid agricultural change to test whether shifts in macroscale (i.e., district-level)
crop diversification have observable impacts on farmer income in times of
extreme weather. Consistent with prior farm-level analyses (see, for example, Di
Falco, Bezabih, and Yesuf 2010), we detect benefits of macroscale diversification
for farmer revenues. In a novel test that isolates physical and market-based
mechanisms, we show that diversification mitigates aggregate yield losses in times
of drought, while simultaneously weakening drought-induced price shocks.
Our findings contribute to two distinct bodies of literature, which to date have
remained independent. First, our results suggest that crop diversity is an effective,
yet understudied, form of adaptation in the rapidly growing empirical literature on
the agricultural impacts of climate change. This literature has demonstrated that
while high temperatures and low rainfall are generally damaging to staple crops,
there is substantial heterogeneity across crops in the weather–yield relationship.
For example, rice in India is sensitive to rising nighttime temperatures (Welch et al.
2010; Auffhammer, Ramanathan, and Vincent 2006), while other major staple crops
like maize, soy, and cotton suffer disproportionately under extreme daytime heat in
 Regional Crop Diversity and Weather Shocks in India 115

India and the United States (Burgess et al. 2014, Schlenker and Roberts 2009). Rice
production throughout South and Southeast Asia has been found to be sensitive to
the availability of rainfall during the monsoon season for nonirrigated systems, and
to the availability of economically viable groundwater and surface water resources
for irrigated systems (Auffhammer, Ramanathan, and Vincent 2012; 2006). Rainfall
appears, however, much less important for crop production relative to temperature
when researchers consider other crops and other regions of the world (Schlenker
and Lobell 2010).
This heterogeneity suggests that diversification of crop planting, either at
the farm level or at a more aggregate scale, may facilitate the resilience of total
production and farm income to adverse climate conditions. However, this literature
has focused on other means of adaptation. For example, the breeding of new
crop varieties, which can more easily withstand climatic extremes, has received
substantial attention. Dar et al. (2013) use a randomized field experiment in
Orissa (now known as Odisha), India to show that a novel submergence-tolerant
rice variety has substantial positive impacts on the mean and variance of rice
yield. Maybe the most obvious avenue of adaptation to more uncertain rainfall
regimes available to farmers is the installation of irrigation infrastructure, which
has been observed throughout the major agricultural production areas globally.
While demonstrably beneficial, irrigation as adaptation ceases to be feasible in
the presence of insufficient groundwater or surface water (Fishman 2018). Other
hypothesized forms of adaptation include shifting the planting calendar, switching
crops, or exiting agriculture altogether. In this paper, we focus on the possibility that
crop diversification may provide a means of adaptation that has thus far remained
absent from climate change impacts research.
The second body of literature we contribute to is the active area of research
on the benefits of biodiversity. Diversification is often cited as a means of reducing
vulnerability to shocks, both within agriculture and in ecological systems more
broadly (Bellora et al. 2017, Tilman et al. 2001, and Tilman and Downing 1994).
Similarly, diversification as insurance is a well-accepted principle in finance and
underlies modern portfolio theory. In the agricultural and agroecology literatures,
crop diversification has been argued to enhance the ability of farmers and food
production systems to respond to climatic variability by increasing the efficiency
of water and nutrient use, enhancing species complementarities, and ensuring that
species with heterogeneous climatic sensitivities exist jointly (Di Falco, Bezabih,
and Yesuf 2010). However, much of the empirical agroecology literature has
focused on a relatively narrow spatial scale and spectrum of crops (Gil et al.
2017), with findings often diverging (Di Falco, Bezabih, and Yesuf 2010). This
is an active area of research with empirical studies covering much of the globe
(Zimmerer 2010).
A number of economic studies have examined the benefits of crop
diversification at various temporal and spatial scales. One literature builds on the
116 Asian Development Review

seminal paper by Rosenzweig and Binswanger (1993) looking at crop choice, farm
level profitability, and weather risk. The majority of these papers have engaged in
case studies of single farms or local production systems, which often specialize in
a small number of crops. Gaudin et al. (2015) find significant yield benefits from
cropping sequence diversification over a 31-year period of field trials in Ontario,
Canada. They found that diversification led to yield increases of 7% for corn and
22% for soybeans. Di Falco and Chavas (2008) show that for cereals in Southern
Italy, higher biodiversity is consistent with better production outcomes during
negative rainfall events. Di Falco, Bezabih, and Yesuf (2010) demonstrate similar
findings for a panel of farms in Ethiopia. They show that higher crop diversity leads
to better production outcomes during bad rainfall events, suggesting that diversity
is consistent with higher resilience. These papers and the references cited therein
comprise a flourishing literature, which appears to suggest empirical evidence of
the benefits to resilience from higher crop diversification. The advent of massive
new data sets of high resolution land cover imagery will certainly revolutionize this
literature as one will be able to look at both temporal and spatial changes in crop
diversity locally and globally.
In this paper, we seek to make a modest contribution to the literature with two
innovations. First, we test whether we can detect evidence of increased resilience at
a macroscale of diversification, in contrast to the field- and farm-level analyses that
dominate previous work.1 Second, we separately identify two key pathways linking
diversity and climate resilience. Nearly all studies focus on the production benefits
of diversification, citing or testing for particular mechanisms like the species
complementarities discussed above. However, the manner in which these production
effects manifest as changes in farmer welfare depend critically on endogenous
price responses. Baumgärtner and Quaas (2009) describe a model in which formal
market insurance and “natural insurance” realized through diversification of a crop
portfolio are substitutes. In this conceptual framework, farmers adopt diversification
in the presence of uncertain climate conditions when formal market insurance is
absent or costly, as is the case in much of India. However, the advantages of such
natural insurance depend on the extent to which local prices move in tandem with
diversity-moderated yield shocks. To our knowledge, we provide the first empirical
evidence that diversification influences both physical production and local price
levels.
Our study is set in the context of India during the onset of the Green
Revolution, which dramatically transformed Indian agriculture. Following decades
of frequent famines, the early 1960s saw the rollout of high-yielding varieties
(HYVs) of cereals (particularly wheat), increased rollout of irrigation infrastructure,

1
See Zimmerer (2010) for a discussion of the role of geographic scale in agroecological studies of crop
diversity.
 Regional Crop Diversity and Weather Shocks in India 117

and more frequent use of pesticides and fertilizers as well as machinery. On the
institutional side, this period saw land reforms and the related consolidation of
landholdings as well as the broad availability of credit access for farmers. As
a consequence, diversity generally fell through increased adoption of HYVs and
other agricultural inputs that facilitated a transition toward more monoculture-based
production systems. This transition was particularly stark in North India. We use
these spatial and temporal differences as a source of variation in crop diversity at
the district level, where the average Indian district is slightly larger than the average
county in the United States.
We use annual weather, agricultural price, production, and cost data
measured in each district and each year to test whether district-level crop diversity
is correlated with higher gross or net farm revenues per hectare in years of drought.
Our estimation results suggest that in drought years, districts with higher levels of
agricultural diversity experience higher gross and net revenues per hectare. These
aggregate results can arise not only through physical benefits of diversification on
crop yields, but also through changes in local prices to the extent that markets are
not fully integrated across space. In what we believe is a novel decomposition of the
income effect, we test whether the impacts of diversity during drought periods affect
crop yields and agricultural prices simultaneously. We find evidence that revenues
are affected both through physical impacts on yields and through a farm gate price
mechanism. Our estimation results show that these impacts are robust across space
and to the inclusion of controls for unobservables in the form of district and year
fixed effects as well as regional trends. The remainder of the paper is organized as
follows. Section II briefly describes the data employed in our estimation. Section
III describes the empirical model. Section IV presents and discusses the results.
Section V concludes with some suggestions for further research.

II. Data

The agricultural data at the district level used in this paper come from the
India Agriculture and Climate Data Set. We use the original data set created by
Sanghi, Kumar, and McKinsey (1998), which covers the period 1956–1987. These
data contain area planted, output, and farm gate prices for major crops (pearl millet,
sorghum, maize, rice, and wheat) as well as minor crops (barley, cotton, groundnut,
chickpea, jute, potato, finger millet, pigeon pea, rapeseed and mustard, sesame,
soy, sugarcane, sunflower, tobacco, and a collective category called “other pulses”).
Area planted is reported in thousands of hectares, output is reported in thousands of
tons.
As Zimmerer (2010) discusses, the study of crop diversity and how it is
measured depends greatly on the spatial scale of analysis. Studies of single crops
often extend down to the level of different types of strawberry plants. We do not
118 Asian Development Review

have data at this scale. Also, indicators of diversity often take into account different
spatial arrangements of planting patterns for individual crops—the “surrounding
landscape,” which is thought to play an important ecological role. We do not have
a field-level database, which would allow us to generate an indicator to take these
important dimensions into account. Finally, there is the diversity in crop rotation
studied by Gaudin et al. (2015), which looks at temporal measures of diversity.
We only observe annual data and hence cannot investigate the order of planting
and harvest. Given these limitations, we create a simple yet widely used indicator
of concentration, the Herfindahl–Hirschman Index (HHI), based on area planted
to different crops in a given year and district. While this measure is relatively
coarse when compared with field-level analyses, it enables us to investigate whether
the microscale findings from the existing literature scale to an aggregate level.
Moreover, this macroscale indicator enables us to investigate the market response
to diversification, as measured by local prices.
Our HHI for district i and year t is defined as follows:

J
HHIit = s2i jt (1)
j=1

ai jt
where si jt = J
ai jt
is the share of total planted area in district i dedicated to crop
j=1
j in year t. J is the total number of crops, which in our data set comprises the 20
major and minor crops listed above.2
Our weather data are monthly averages and were aggregated up from weather
stations to districts and weighted by the inverse squared distance from district
centroid to each station. We define the kharif (monsoon crop) growing season
as June through September and rabi (winter crop) as November and December
(Carleton 2017, Guiteras 2009). We use cumulative growing season rainfall and
average growing season temperature as explanatory climatic variables. We create
an indicator for extremely low levels of rainfall, using the lowest tercile of
kharif season rainfall for each district as the cutoff value, following Auffhammer,
Ramanathan, and Vincent (2012), and Burgess et al. (2014).
Our agricultural price index is based on Burgess et al. (2014), which is a
simple index with time-invariant crop weights. Each crop j in each district i is
weighted by the average percentage of total crop revenue derived from crop j. That
is, the weight for crop j in district i is calculated by

1
T
Pjit Q jit
w ji = J (2)
T t=1 j=1 Pjit Q jit

2
We used the HHI over the Shannon Index since our data contain many crop and year combinations with zero
area planted.
 Regional Crop Diversity and Weather Shocks in India 119

The price index for each district i in year t is calculated as


J
Pit = Pjit w ji (3)
j=1

where Pjit is the crop-specific price (in rupees/quintal) as reported in the raw data
(Sanghi, Kumar, and McKinsey 1998).
We first conduct our analysis for all districts with available data across India.
To test for spatial heterogeneity based on the location of emergence of the Green
Revolution, we split India into North (Gujarat, Haryana, Punjab, Rajasthan, and
Uttar Pradesh) and South (Andhra Pradesh, Bihar, Karnataka, Madhya Pradesh,
Maharashtra, Odisha, Tamil Nadu, and West Bengal). We run versions of our
models allowing the effect of crop diversity during drought years to vary between
the North and the South. Table 1 provides the complete summary statistics for our
sample of districts and also breaks them out by geographic region. The summary
statistics indicate that the northern districts are significantly drier during the kharif
season on average and in terms of extremes. This is not surprising since the southern
states are part of the monsoon belt, where much of the rain-fed rice agriculture
is based. The northern districts are also warmer during the kharif season with an
average monthly temperature that is almost 1.5 degrees Celsius higher than the
southern states.
To create indicators of farmer income, we construct measures of both gross
and net revenues per hectare of planted area. Gross revenues are simply the sum
of all production returns for each crop, where prices and quantities are taken
directly from the India Agriculture and Climate Data Set. Net revenues subtract
estimated labor and fertilizer costs from gross revenues. However, these input costs
are heavily interpolated and results for net revenues should therefore be interpreted
with caution. The statistics on net revenue per hectare in Table 1 indicate similar net
and gross revenues per hectare in the North and the South. Yields in terms of tons
per hectare are not statistically different in the North than in the South. There is no
statistical difference in area planted per district between the two groupings of states.
If we consider the share of irrigated crops across the two groupings, 24% of area
planted was irrigated in the North across the entire sample, while the same average
is only 18% for the South. Further, the farm gate price index is slightly higher in
the South, although the difference is not statistically significant.
Figure 1 displays our main variable of interest, the HHI of crop diversification
across time, separately for the North and South. We see a trend consistent with
the onset of the Green Revolution in the mid-1960s. The North experienced a
steady increase in HHI over time. Two things stand out. First, there is the drastic
increase in the index over time in the North, which is indicative of a move toward
more monoculture. This relationship is not clear in the South, where the HHI stays
somewhat constant on average. Figure 2 displays the HHI plotted against the share
 Table 1. Summary Statistics—All India, North India, and South India
All India North South
Variable Obs. Mean Std. Dev. Min Max Mean Std. Dev. Mean Std. Dev.
Precip JJAS 8,401 0.86 0.51 0.01 5.43 0.71 0.36 0.96 0.57
Precip ND 8,401 0.04 0.07 0.00 0.76 0.01 0.02 0.05 0.09
120 Asian Development Review

Temp JJAS 8,401 28.53 2.07 21.10 34.52 29.92 1.50 27.62 1.88
Temp ND 8,401 20.64 2.41 10.57 27.66 19.28 2.22 21.55 2.07
Total Revenue/ha (thousands) 8,401 1.07 0.95 0.00 29.24 1.02 0.88 1.09 1.00
Net Revenue/ha (thousands) 8,401 0.19 1.13 −44.07 19.56 0.20 0.68 0.19 1.35
Yield (tons/ha) 8,401 0.86 0.46 0.00 15.08 0.89 0.45 0.84 0.46
Area (thousand ha) 8,401 535.35 293.89 3.10 2,177.41 540.69 275.03 531.82 305.72
Share irrigated 8,401 0.20 0.16 0.00 0.92 0.24 0.15 0.18 0.16
Price Index 8,401 134.75 85.74 6.36 832.35 129.47 84.31 138.25 86.51
HYVs Share of Area Planted 8,401 0.11 0.12 0.00 0.58 0.12 0.13 0.10 0.12
HHI Diversification Index 8,401 0.37 0.21 0.11 1.00 0.30 0.16 0.42 0.23
ha = hectare; HHI = Herfindahl–Hirschman Index (calculated over area planted to 20 different crops); HYVs = high-yielding varieties; JJAS = June,
July, August, and September; ND = November and December.
Notes: “Precip” refers to cumulative seasonal rainfall measured in meters. “Temp” refers to average seasonal temperature measured in degrees Celsius.
Source: Authors’ own calculation from data constructed by Sanghi, Apurva, K. S. Kavi Kumar, and James W. McKinsey Jr. 1998. India Agriculture
and Climate Data Set. Washington, DC: World Bank.
 Regional Crop Diversity and Weather Shocks in India 121

Figure 1. Time Series of Crop Diversity Index

HHI = Herfindahl–Hirschman Index (calculated over area planted to 20 different crops).

Notes: North India includes Gujarat, Haryana, Punjab, Rajasthan, and Uttar Pradesh. South India includes Andhra
Pradesh, Bihar, Karnataka, Madhya Pradesh, Maharashtra, Odisha, Tamil Nadu, and West Bengal.
Source: Authors’ own calculation from data constructed by Sanghi, Apurva, K. S. Kavi Kumar, and James W.
McKinsey Jr. 1998. India Agriculture and Climate Data Set. Washington, DC: World Bank.

of HYVs in terms of area planted. The HYVs rollout in the North correlates strongly
with decreased diversity, as is evidenced by the increase in the HHI from 0.20 to
0.34 for the districts in Punjab, Uttar Pradesh, and Haryana. This means that while
Green Revolution technologies were taken up at relatively similar rates across India,
only in the North did this imply lower crop diversity. Second, there is significantly
more variability in the South in the HHI than in the North.
We now use these data to examine whether more diverse cropping systems
are more resilient to rainfall shocks in the form of a drought.

III. Empirical Model

Auffhammer, Ramanathan, and Vincent (2012) previously examined the role

of rainfall extremes on rice yields throughout India using aggregate state-level data.
They show that higher incidence of drought and an increase in more intense rainfall
events, conditional on total monsoon rains, lead to measurable depressed yields for
122 Asian Development Review

Figure 2. Crop Diversity Index and Share of Area Planted to HYVs

HHI = Herfindahl–Hirschman Index (calculated over area planted to 20 different crops), HYVs = high-yielding
varieties.
Notes: North India includes Gujarat, Haryana, Punjab, Rajasthan, and Uttar Pradesh. South India includes Andhra
Pradesh, Bihar, Karnataka, Madhya Pradesh, Maharashtra, Odisha, Tamil Nadu, and West Bengal.
Source: Authors’ own calculation from data constructed by Sanghi, Apurva, K. S. Kavi Kumar, and James W.
McKinsey Jr. 1998. India Agriculture and Climate Data Set. Washington, DC: World Bank.

rain-fed rice. We adopt a similar econometric specification as a starting point, which

is given by

yit = αi + θt + β1 HHIit−1 + β2 HHIit−1 ∗ droughtit + β3 droughtit

+ β4 Tempit + β5 Precipit + εit (4)

where yit is one of our four outcome variables (gross revenue per hectare, net
revenue per hectare, yield per hectare, and price index) in district i and year t;
αi , and θt are district and year fixed effects, respectively;3 droughtit is a dummy
variable indicating a low rainfall year; Tempit is a vector of seasonal (kharif, rabi)
average temperature; and Precipit is a vector of seasonal (kharif, rabi) cumulative
rainfall. The additive error term εit is assumed to be serially correlated across time,
yet independent across districts.

3
For robustness, we show additional results including region-specific time trends in place of year fixed effects.
 Regional Crop Diversity and Weather Shocks in India 123

The variable yit will be one of four different variables. In a first set of
regressions we will use gross revenues (in rupees) per area planted, which is simply
the sum of revenues from all crops grown in a district-year. In a second set of
regressions, we will use net revenues, which account for costs of major inputs to
production. In a third set of regressions we will use yield, which is simply total
output (in tons) aggregated across crops and measured per unit of area planted. This
measure of course does not directly link to farmer income, as weight in tons and the
economic value of crops are not necessarily positively correlated. In a final set of
regressions, we will put our price index based on equation (3) as the dependent
variable. We will run all regressions for all of India and then examine whether there
is evidence of spatial heterogeneity between the North and South, and between more
and less heavily irrigated areas.
The variable drought is a dummy equal to 1 when cumulative kharif season
rainfall is in the first tercile of rainfall for district i in year t, based on the distribution
of that district’s rainfall over the entire sample time period. Tempit comprises two
variables, kharif and rabi average monthly temperature, and is an important control,
given the negative impacts of heat on yields in India (Burgess et al. 2014; Guiteras
2009; Auffhammer, Ramanathan, Vincent 2006).
The variables of interest in these regressions are HHIit−1 and HHIit−1 ∗
droughtit . The coefficient β1 will measure the relationship between lower crop
diversity and each dependent variable directly and hence is of interest in itself. The
coefficient β2 will measure the relationship between lower crop diversity during
drought years and each dependent variable, and is the main coefficient of interest.
If higher diversity makes districts more resilient to droughts, we would expect a
negative coefficient on this interaction term for the net and gross revenue and yield
regressions.4
In our regressions we include HHI and its interaction with the drought
variable as lagged by one period, which makes them econometrically predetermined
in the time series sense. We estimate all equations with district and year fixed
effects to control for temporal effects within each district and time-invariant
state characteristics, respectively. Identification hence comes from within-district
variation, conditional on the year fixed effects and other observable controls.
The coefficients on all weather variables—Temp, Precip, and drought—can be
interpreted causally as these within-location realizations are plausibly exogenous
(Carleton and Hsiang 2016). However, given significant temporal dependence in
the diversity measure, our use of a lagged independent variable is imperfect and
β2 may be affected by unobserved heterogeneity. In the absence of a readily
available instrument, we see this approach as the best one can do to assuage
endogeneity concerns. We interpret all our findings regarding the HHI interaction

4
Recall that a higher HHI indicates higher concentration in area planted, and hence lower diversity.
124 Asian Development Review

Table 2. Ordinary Least Squares Estimation Results—All India

Gross Gross Net Net
Revenue Revenue Revenue Revenue Yield Yield
HHI −0.059 −0.255 −1.212 **
−2.371 ***
0.107 −0.417
(0.831) (1.057) (0.516) (0.698) (0.452) (0.546)
HHI × Drought −0.211*** −0.160** −0.565* −0.565* −0.077*** −0.074***
(0.072) (0.070) (0.323) (0.306) (0.027) (0.025)
Drought 0.027 0.003 0.155 0.126 0.003 −0.010
(0.026) (0.027) (0.105) (0.095) (0.012) (0.011)
Temp JJAS −0.006 0.170*** 0.055* −0.019 −0.030** 0.003
(0.024) (0.015) (0.034) (0.021) (0.012) (0.006)
Precip JJAS 0.011 0.092** 0.021 −0.034 0.016 0.054***
(0.050) (0.042) (0.055) (0.061) (0.024) (0.017)
Temp ND −0.018 −0.062*** −0.072*** −0.066*** −0.007 −0.019***
(0.012) (0.006) (0.026) (0.008) (0.005) (0.003)
Precip ND 0.413 0.019 −0.714 −0.602 −0.014 −0.041
(0.349) (0.318) (1.140) (0.889) (0.168) (0.141)
District FEs Yes Yes Yes Yes Yes Yes
Year FEs Yes No Yes No Yes No
Region Trends No Yes No Yes No Yes
Observations 8,401 8,401 8,401 8,401 8,401 8,401
R-squared 0.711 0.683 0.388 0.373 0.697 0.699
FEs = fixed effects; HHI = Herfindahl–Hirschman Index (calculated over area planted to 20 different crops);
JJAS = June, July, August, and September; ND = November and December.
Notes: “Precip” refers to cumulative seasonal rainfall. “Temp” refers to average seasonal temperature. “Drought”
refers to a binary indicator defined as equal to 1 when cumulative JJAS season rainfall is in the first tercile of
the district-specific rainfall distribution. Gross and net revenues are measured in thousands of rupees per hectare;
yield is measured in tons per hectare. Standard errors are clustered by administrative district and are reported in
parentheses. *** p < 0.01, ** p < 0.05, and * p < 0.1.
Source: Authors’ calculations.

as correlational. All of our estimates cluster standard errors by district. The next
section discusses the results from estimating different versions of equation (4).

IV. Estimation Results

Table 2 displays the estimation results for our main specification. The first
two columns show the estimation results from estimating equation (4) using gross
revenues per hectare as the dependent variable for all of India, controlling for
year fixed effects (column 1) versus region-specific trends (column 2). All models
control for district fixed effects. We find a negative yet statistically insignificant
parameter estimate on the diversity index, suggesting that there is no correlation
between crop diversity and gross revenues per hectare, conditional on the included
observables and fixed effects or trends. The interaction of the HHI indicator and the
drought dummy is significantly different from zero and negative, which is consistent
with more diverse districts having higher gross revenues per hectare during drought
 Regional Crop Diversity and Weather Shocks in India 125

years. Its point estimate suggests that a one sample standard deviation higher HHI
during a drought year would result in 4.4% less gross revenue per hectare for that
district. If we use net revenues, the sign of the interaction term is similar, yet
the point estimate and associated uncertainty are larger. A one standard deviation
increase in the HHI has a roughly two and a half times larger negative effect during
drought years, which of course works through gross revenues as well as labor and
input expenditures. The latter are not well measured in the data set, which leads
us to suggest that the gross revenue regressions provide a more reliable estimate of
impacts. The last two columns in the table conduct a simple yield regression and the
results carry the same sign and statistical significance. In average years, diversity
has no distinguishable impact on crop yields, but in times of drought, districts with
lower diversity suffer substantially lower yields. The standardized magnitude of
the effect on yields is smaller than that for gross revenues, suggesting that a one
standard deviation increase in the HHI is consistent with a 1.9% decrease in yields.
Table 3 examines whether there is heterogeneity in the estimated effects
in terms of location of the district or how heavily irrigated the district is. To test
whether the northern districts are more or less resilient than the southern districts,
we include an interaction term of our variable of interest with a dummy for the
northern states. The first three columns of the table show these results. The point
estimates on the interaction term of interest between the drought dummy and the
HHI are almost identical to those in Table 2. The interaction term with the North
India dummy is not statistically different from zero, suggesting that there is little
evidence of heterogeneity in terms of a North–South divide, despite substantial
differences in the rollout of Green Revolution technologies.
We then generate a dummy variable called “Irrig,” which equals 1 if more
than 30% of a district’s area is irrigated. This represents the top quartile of the
districts in terms of the share of area irrigated. The interaction is included in the
last three columns of Table 3. In the gross revenue and yield regression, the point
estimate on the interaction term is opposite in sign to the now slightly larger point
estimates on the regular interaction, which suggests that the resilience benefit from
higher diversification appears to be concentrated in districts with lower shares of
irrigated area.
The effect of crop diversification on revenue resilience in light of drought
depends both on physical responses of crops to the diversity of the local portfolio,
as well as on the degree to which diversity can moderate the influence of
weather shocks on local prices. For example, if markets throughout India were
perfectly integrated, local prices would fail to reflect local climatic conditions and
diversification would matter for farming households only through its effects on crop
yields. However, if local prices were fully determined by local production, negative
supply shocks in drought years would be accompanied by price increases, helping
to buffer income effects through a natural market mechanism. To test whether and
126 Asian Development Review

Table 3. Ordinary Least Squares Estimation Results—All India Heterogeneity Effects

Gross Net Gross Net
Revenue Revenue Yield Revenue Revenue Yield
HHI −0.069 −1.206 **
0.098 −0.097 −1.176 **
0.090
(0.831) (0.533) (0.452) (0.832) (0.492) (0.452)
HHI × Drought −0.213*** −0.564* −0.079*** −0.293*** −0.486** −0.114***
(0.073) (0.327) (0.028) (0.069) (0.224) (0.025)
HHI × Drought × North 0.061 −0.040 0.055
(0.116) (0.274) (0.041)
HHI × Drought × Irrig 0.303*** −0.293 0.136***
(0.100) (0.418) (0.041)
Drought 0.021 0.159* −0.003 0.023 0.159 0.001
(0.032) (0.093) (0.013) (0.027) (0.109) (0.012)
Temp JJAS −0.007 0.056 −0.032*** −0.007 0.056* −0.031***
(0.024) (0.036) (0.012) (0.024) (0.033) (0.012)
Precip JJAS 0.010 0.022 0.014 0.003 0.029 0.012
(0.050) (0.056) (0.024) (0.050) (0.055) (0.024)
Temp ND −0.018 −0.072*** −0.007 −0.015 −0.075*** −0.006
(0.012) (0.026) (0.005) (0.012) (0.029) (0.005)
Precip ND 0.415 −0.716 −0.011 0.424 −0.725 −0.009
(0.349) (1.132) (0.168) (0.348) (1.144) (0.170)
District FEs Yes Yes Yes Yes Yes Yes
Year FEs Yes Yes Yes Yes Yes Yes
Observations 8,401 8,401 8,401 8,401 8,401 8,401
R-squared 0.711 0.388 0.698 0.712 0.389 0.698
FEs = fixed effects; HHI = Herfindahl–Hirschman Index (calculated over area planted to 20 different crops); JJAS
= June, July, August, and September; ND = November and December.
Notes: “Precip” refers to cumulative seasonal rainfall. “Temp” refers to average seasonal temperature. “Drought”
refers to a binary indicator defined as equal to 1 when cumulative JJAS season rainfall is in the first tercile of the
district-specific rainfall distribution. “North” refers to a binary indicator defined as equal to 1 when the district is
within Gujarat, Haryana, Punjab, Rajasthan, or Uttar Pradesh. “Irrig” refers to a binary indicator defined as equal to
1 when more than 30% of a district’s planted area is irrigated. Gross and net revenues are measured in thousands of
rupees per hectare; yield is measured in tons per hectare. Standard errors are clustered by administrative district and
are reported in parentheses. *** p < 0.01, ** p < 0.05, and * p < 0.1.
Source: Authors’ calculations.

how local prices are influenced by crop diversity in addition to the yield effects
documented above, we regress our price index on the same set of covariates we use
in the previous two tables. The first two columns in Table 4 use the pooled all-India
sample. Columns 3 and 4 look at whether the price response varies across space.
These regressions suggest a statistically significant correlation between the price
index and the interaction of the diversification index with the drought indicator.
However, this effect is quantitatively small: a one standard deviation increase in the
HHI during a drought year is consistent with a roughly 1.5% higher price index. If
we look at the interaction in the last two columns, this effect is much larger in North
India, where the effect is 4%. The interaction between the drought variable and the
diversification index suggests that in drought years, less diverse districts see slightly
higher farm gate prices, alongside lower yields.
 Regional Crop Diversity and Weather Shocks in India 127

Table 4. Ordinary Least Squares Estimation Results—Price Index

Price Index Price Index Price Index Price Index
HHI −53.558 ***
16.505 −55.131 ***
15.207
(12.050) (15.445) (11.951) (15.355)
HHI × Drought 10.101** 12.103*** 9.801** 11.637***
(4.235) (4.046) (4.141) (3.998)
HHI × Drought × North 9.940* 14.588**
(5.598) (5.945)
Drought −1.135 −1.496 −2.176 −3.003
(2.002) (2.007) (2.127) (2.144)
Temp JJAS 6.788*** 19.910*** 6.583*** 19.511***
(1.729) (1.300) (1.730) (1.306)
Precip JJAS 1.959 7.306** 1.701 6.936*
(3.232) (3.592) (3.235) (3.590)
Temp ND −0.779 −2.801*** −0.807 −2.811***
(0.983) (0.591) (0.987) (0.591)
Precip ND −5.210 −33.327*** −4.779 −32.741***
(12.539) (10.959) (12.525) (10.928)
District FEs Yes Yes Yes Yes
Year FEs Yes No Yes No
Region Trends No Yes No Yes
Observations 8,401 8,401 8,401 8,401
R-squared 0.816 0.782 0.816 0.782
FEs = fixed effects; HHI = Herfindahl–Hirschman Index (calculated over area planted to 20 different
crops); JJAS = June, July, August, and September; ND = November and December.
Notes: “Precip” refers to cumulative seasonal rainfall. “Temp” refers to average seasonal
temperature. “Drought” refers to a binary indicator defined as equal to 1 when cumulative JJAS
season rainfall is in the first tercile of the district-specific rainfall distribution. “North” refers to a
binary indicator defined as equal to 1 when the district is within Gujarat, Haryana, Punjab, Rajasthan,
or Uttar Pradesh. Total and net revenues are measured in thousands of rupees per hectare; yield is
measured in tons per hectare. Standard errors are clustered by administrative district and are reported
in parentheses. *** p < 0.01, ** p < 0.05, and * p < 0.1.
Source: Authors’ calculations.

V. Conclusions

Our empirical results suggest that Indian districts with a more diverse crop
mix are indeed more resilient in the presence of droughts. We show economically
small, yet statistically significant, effects on gross and net revenues. We document
that these impacts materialize both through a physical effect on yields, as well
as through changes in local prices. We see no heterogeneity between North India
and South India in the revenue and yield regressions, but do detect a higher price
response in the North compared to the South. We show some suggestive evidence
that these benefits seem to be negated in heavily irrigated districts, where irrigation
serves as a backstop for bad rainfall outcomes, as long as irrigation water is
available from groundwater or surface sources. These findings contribute both to
the literature seeking to understand possible adaptation pathways for agriculture
under anthropogenic climate change, as well as to the body of research quantifying
the benefits of increased diversification in agricultural and agroecological systems.
128 Asian Development Review

There are several caveats to our analysis. This observational study suggests
correlations only, as we have to rely on the fact that our lagged diversification index
is predetermined. A proper, and hugely costly, experimental design would randomly
encourage farmers in different districts to increase the diversity of crops grown. This
is clearly not feasible in the short run at this scale. A meta-analysis of the studies
listed in Zimmerer (2010) might serve as an interesting next step for future research.
Many outstanding questions regarding the resiliency benefits of crop
diversity go beyond the resources available for this study. They include a more
careful examination of which type of heterogeneity is responsible for the largest
gains in resilience. For example, are the aggregate gains from diversification due
to heterogeneity in minor crops? Or, are the observed benefits only realized when
minor and major crops together provide a diversified portfolio? We hope that
increased availability of satellite imagery over time will allow us to more closely
examine these questions for large-scale panel data. Closer collaboration between
ecologists, agricultural economists, and data scientists might provide a fruitful path
forward to disentangle the mechanisms behind these effects and better understand
the associated economic benefits and costs.

References

Annan, Francis, and Wolfram Schlenker. 2015. “Federal Crop Insurance and the Disincentive to
Adapt to Extreme Heat.” American Economic Review 105 (5): 262–66.
Auffhammer, Maximilian, V. Ramanathan, and Jeffrey R. Vincent. 2006. “Integrated Model
Shows that Atmospheric Brown Clouds and Greenhouse Gases Have Reduced Rice
Harvests in India.” Proceedings of the National Academy of Sciences 103 (52): 1966–72.
_____. 2012. “Climate Change, the Monsoon, and Rice Yield in India.” Climatic Change 111 (2):
411–24.
Auffhammer, Maximilian, and Wolfram Schlenker. 2014. “Empirical Studies on Agricultural
Impacts and Adaptation.” Energy Economics 46: 555–61.
Baumgärtner, Stefan, and Martin F. Quaas. 2009. “Agro-Biodiversity as Natural Insurance and
the Development of Financial Insurance Markets.” In Agrobiodiversity, Conservation, and
Economic Development, edited by Andreas Kontoleon, Unai Pascual, and Melinda Smale,
293–317. London: Routledge.
Bellora, Cecilia, Élodie Blanc, Jean-Marc Bourgeon, and Eric Strobl. 2017. “Estimating the
Impact of Crop Diversity on Agricultural Productivity in South Africa.” National Bureau
of Economic Research Working Paper No. w23496.
Burgess, Robin, Olivier Deschenes, Dave Donaldson, and Michael Greenstone. 2014. The
Unequal Effects of Weather and Climate Change: Evidence from Mortality in India.
Cambridge, US: Massachusetts Institute of Technology, Department of Economics.
Burke, Marshall, and Kyle Emerick. 2016. “Adaptation to Climate Change: Evidence from US
Agriculture.” American Economic Journal: Economic Policy 8 (3): 106–40.
Carleton, Tamma A. 2017. “Crop-Damaging Temperatures Increase Suicide Rates in India.”
Proceedings of the National Academy of Sciences 114 (33): 8746–51. https://doi.org/10
.1073/pnas.1701354114.
 Regional Crop Diversity and Weather Shocks in India 129

Carleton, Tamma A., and Solomon M. Hsiang. 2016. “Social and Economic Impacts of Climate.”
Science 353 (6304): aad9837.
Colmer, Jonathan. 2016. “Weather, Labour Reallocation, and Industrial Production: Evidence
from India.” London School of Economics Working Paper.
Dar, Manzoor, Alain de Janvry, Kyle Emerick, David Raitzer, and Elisabeth Sadoulet. 2013.
“Flood-Tolerant Rice Reduces Yield Variability and Raises Expected Yield, Differentially
Benefitting Socially Disadvantaged Groups.” Scientific Reports 3 (3315). doi: 10.1038
/srep03315.
Di Falco, Salvatore, Mintewab Bezabih, and Mahmud Yesuf. 2010. “Seeds for Livelihood:
Crop Biodiversity and Food Production in Ethiopia.” Ecological Economics 69 (8): 1695–
702.
Di Falco, Salvatore, and Jean–Paul Chavas. 2008. “Rainfall Shocks, Resilience, and the Effects
of Crop Biodiversity on Agroecosystem Productivity.” Land Economics 84 (1): 83–
96.
_____. 2009. “On Crop Biodiversity, Risk Exposure, and Food Security in the Highlands of
Ethiopia.” American Journal of Agricultural Economics 91 (3): 599–611.
Di Falco, Salvatore, and Charles Perrings. 2003. “Crop Genetic Diversity, Productivity, and
Stability of Agroecosystems. A Theoretical and Empirical Investigation.” Scottish Journal
of Political Economy 50 (2): 207–16.
Fishman, Ram. 2016. “More Uneven Distributions Overturn Benefits of Higher Precipitation for
Crop Yields.” Environmental Research Letters 11 (2): 024004.
_____. 2018. “Groundwater Depletion Limits the Scope for Adaptation to Increased Rainfall
Variability in India.” Climatic Change 147 (1–2): 195–209.
Gaudin, Amélie C. M., Tor N. Tolhurst, Alan P. Ker, Ken Janovicek, Cristina Tortora, Ralph C.
Martin, and William Deen. 2015. “Increasing Crop Diversity Mitigates Weather Variations
and Improves Yield Stability.” PLoS ONE 10 (2): e0113261. https://doi.org/10.1371
/journal.pone.0113261.
Gil, Juliana D. B., Avery S. Cohn, John Duncan, Peter Newton, and Sonja Vermeulen. 2017. “The
Resilience of Integrated Agricultural Systems to Climate Change.” WIREs Climate Change
8. doi: 10.1002/wcc.461.
Guiteras, Raymond. 2009. “The Impact of Climate Change on Indian Agriculture.” Manuscript,
Department of Economics, University of Maryland.
Intergovernmental Panel on Climate Change (IPCC). 2014. Climate Change 2014 Impacts,
Adaptation, and Vulnerability: Regional Aspects. Cambridge, UK: Cambridge University
Press.
Lobell, David B., Michael J. Roberts, Wolfram Schlenker, Noah Braun, Bertis B. Little, Roderick
M. Rejesus, and Graeme L. Hammer. 2014. “Greater Sensitivity to Drought Accompanies
Maize Yield Increase in the US Midwest.” Science 344 (6183): 516–19.
Mendelsohn, Robert, William D. Nordhaus, and Daigee Shaw. 1994. “The Impact of Global
Warming on Agriculture: A Ricardian Analysis.” American Economic Review 84 (4): 753–
71.
Rosenzweig, Mark R., and Hans P. Binswanger. 1993. “Wealth, Weather Risk and the Composition
and Profitability of Agricultural Investments.” The Economic Journal 103 (416): 56–78.
Sanghi, Apurva, K. S. Kavi Kumar, and James W. McKinsey Jr. 1998. India Agriculture and
Climate Data Set. Washington, DC: World Bank.
Schlenker, Wolfram, and David B. Lobell. 2010. “Robust Negative Impacts of Climate Change
on African Agriculture.” Environmental Research Letters 5 (1): 014010.
130 Asian Development Review

Schlenker, Wolfram, and Michael J. Roberts. 2009. “Nonlinear Temperature Effects Indicate
Severe Damages to US Crop Yields under Climate Change.” Proceedings of the National
Academy of Sciences 106 (37): 15594–98.
Schlenker, Wolfram, Michael J. Roberts, and David B. Lobell. 2013. “US Maize Adaptability.”
Nature Climate Change 3 (8): 690–91.
Tilman, David, and John A. Downing. 1994. “Biodiversity and Stability in Grasslands.” Nature
367 (6461): 363–65.
Tilman, David, Peter B. Reich, Johannes Knops, David Wedin, Troy Mielke, and Clarence
Lehman. 2001. “Diversity and Productivity in a Long-Term Grassland Experiment.”
Science 294 (5543): 843–45.
Welch, Jarrod R., Jeffrey R. Vincent, Maximilian Auffhammer, Piedad F. Moya, Achim
Dobermann, and David Dawe. 2010. “Rice Yields in Tropical/Subtropical Asia Exhibit
Large but Opposing Sensitivities to Minimum and Maximum Temperatures.” Proceedings
of the National Academy of Sciences 107 (33): 14562–67.
Wright, Brian D., and Bruce L. Gardner. 1995. Reforming Agricultural Commodity Policy.
Washington, DC: American Enterprise Institute.
Zimmerer, Karl S. 2010. “Biological Diversity in Agriculture and Global Change.” Annual Review
of Environment and Resources 35: 137–66.