Content uploaded by Navin Ramankutty
Author content
All content in this area was uploaded by Navin Ramankutty on Nov 27, 2018
Content may be subject to copyright.
Contents lists available at ScienceDirect
Global Food Security
journal homepage: www.elsevier.com/locate/gfs
How much of the world's food do smallholders produce?
Vincent Ricciardi
a,b,⁎
, Navin Ramankutty
a,b
, Zia Mehrabi
a,b
, Larissa Jarvis
a,b
,
Brenton Chookolingo
a,b
a
The Institute for Resources, Environment, and Sustainability, University of British Columbia, Canada
b
School of Public Policy and Global Affairs, University of British Columbia, Canada
ABSTRACT
The widely reported claim that smallholders produce 70–80% of the world’s food has been a linchpin of agri-
cultural development policy despite limited empirical evidence. Recent empirical attempts to reinvestigate this
number have lacked raw data on how much food smallholders produce, and have relied on model assumptions
with unknown biases and with limited spatial and commodity coverage. We examine variations in crop pro-
duction by farm size using a newly-compiled global sample of subnational level microdata and agricultural
censuses covering more countries (n=55) and crop types (n=154) than assessed to date. We estimate that farms
under 2ha globally produce 28–31% of total crop production and 30–34% of food supply on 24% of gross
agricultural area. Farms under 2ha devote a greater proportion of their production to food, and account for
greater crop diversity, while farms over 1000ha have the greatest proportion of post-harvest loss.
1. Introduction
It has been widely reported that smallholder farmers (defined gen-
erally as being less than 2 ha) produce 70–80% of the world's food ETC,
2009;Maass Wolfenson, 2013;FAO, 2014), are central to conserving
crop diversity (Altieri, 2008; Badstue et al., 2005; Conway, 2011),
produce more food crops than larger farms (Horrigan et al., 2002;
Naylor et al., 2005), and yet are largely food insecure (IFAD and UNEP,
2013). These arguments have been a linchpin in recent agricultural
development policy. For example, in 2014, the ‘International Year of
the Family Farm’, the United Nations (UN) and other food security
agencies reiterated these arguments to garner increased support for
family farmers, who are predominantly smallholders (FAO, 2014). The
COP21 agreement (the 2015 UN Conference of Parties on Climate
Change) includes mitigation and adaptation commitments pertaining to
agriculture from 179 countries that include the need to bolster small-
holder adaptive capacity to climate change. Goal 2 of the UN Sustain-
able Development Goals (SDGs) aims to end hunger and achieve food
security through sustainable agriculture; a key target (SDG 2.3) is by
‘2030, [to] double the agricultural productivity and incomes of small-
scale food producers, in particular women, indigenous peoples, family
farmers, pastoralists, and fishers’(UN, 2015). Yet, despite progress in
steering development policy towards smallholder farmers, there is scant
empirical data on smallholder farms, and their role in the food system.
Key to enacting and monitoring progress on these international
agreements and policies is a global baseline on the contribution of
smallholders to global food production and security. However, the data
underlying three widely reported claims on smallholder crop produc-
tion remain non-transparent or contradictory. First, the source of var-
ious UN reports citing smallholder production is a communiqué from
the ETC group (ETC, 2009), which suggests that ‘peasants’grow at least
70% of the world's food; yet, the derivation of the estimate is obscure in
this report. Second, the claim that smaller farms produce more food
directly consumed by people, with larger industrialized farms produ-
cing more non-food crops, such as biofuels and animal feed (Horrigan
et al., 2002; Naylor et al., 2005), has been brought into question by the
observation that smaller farms have larger amounts of post-harvest loss
due to lack of market and cold storage access (Hodges et al., 2011;
Tefera, 2012). Thirdly, while some authors argue that economies of
scale are needed for farms to produce a diversity of crops (Rahman and
Kazal, 2015), others suggest that larger farms face labor constraints that
hamper mixed-cropping systems (Van den Berg et al., 2007), so it is
unknown if smaller farms produce a greater diversity of crop species
than larger farms. In sum, our current understanding how much food
smallholders produce, what kinds of food they produce, where their
food is destined in the food system, and how much nutrition it contains,
are all key knowledge gaps in global agricultural research.
The need to fill these knowledge gaps has been recently recognized
by scientists (Graeub et al. (2016);Herrero et al. (2017);Lowder et al.
(2016);Samberg et al., 2016 (referred to as Graeub, Lowder, Herrero,
https://doi.org/10.1016/j.gfs.2018.05.002
Received 8 January 2018; Received in revised form 8 May 2018; Accepted 9 May 2018
⁎
Corresponding author at: The Institute for Resources, Environment, and Sustainability, University of British Columbia, Canada.
E-mail address: vinnyricciardi@gmail.com (V. Ricciardi).
Global Food Security 17 (2018) 64–72
2211-9124/ © 2018 Elsevier B.V. All rights reserved.
T
and Samberg respectively hereafter). In 2016, a pair of studies eval-
uated the contribution of smallholders and family farms to global crop
and food production. Lowder was the first to report on global farm size
trends from 1960 to 2010 derived from 167 countries in the World
Census of Agriculture (WCA). They found that small-farms (defined as
being < 2 ha) constituted only 12% of the global available farmland,
but represented 84% of all farms. Their study did not report on crop
production, but their results implied that smallholders do not produce
70% global crops; it is unlikely they could produce this much food on
12% of available farmland, even if we assumed that small farms had
higher yields and produced more food crops than larger farms. The
second of these studies (Graeub) quantified the number and extent of
family farms in the world and their production contributions. By using
national family farm definitions, defining family farms based on farm
size, or a combination thereof to represent regionally appropriate fa-
mily farm definitions they estimated that ~98% of all farms globally
are family farms, collectively managing 53% of all cropland, and
meeting an estimated 36–114% of domestic caloric requirements for
different countries. While Graeub's study highlighted the contribution
of family farms, they also challenge the idea that all family farms are
small farms. For example, farms in Brazil may be family owned but are
large in size (while ~ 85% of farms in Brazil are family owned and
cover ~ 25% of agricultural land, only 21% of farms are less than 2 ha
in size and cover only 0.25% of the agricultural area). Together these
two studies, quantified the global number of smallholders or family
farmers, their cropping area, and detailed the differences between
smallholders and family farms.
Two additional studies were recently published that tried to better
estimate the proportion of food coming from smallholder farmers glob-
ally. Samberg estimated the contributions of smallholders in an analysis
of 41 crops and 83 countries in smallholder dominant regions (Latin
America, sub-Saharan Africa, and South and East Asia) that represent
35% of global cropland. They estimated that smallholders (which they
defined as all administrative units with a “mean agricultural area”<5
ha) produced 52.5% of food calories in their cross-regional sample.
While, this study was a valuable step in mapping the geographic dis-
tribution of smallholders, using mean agricultural area within an ad-
ministrative unit as an index of smallholder production is problematic
because farm size distributions are highly skewed (e.g. Lowder).
Following this, Herrero presented an analysis which modeled crop and
livestock production, micro-nutrition production, and agricultural land-
scape diversity. Crop and animal data were related to farm size classes by
combining crowd-sourced data on field sizes (Fritz et al., 2015) with
national farm size distributions (Lowder) as a proxy for per pixel pro-
duction by farm size. They reported that farms < 50 ha produce 56% of
commodities and nutrients in their sample of 41 crops, 7 livestock, 14
aquaculture and fish products, across 161 countries. They also estimated
that ~18% of food calories globally come from farms < 2 ha, and
highlighted the valuable micronutrient contribution of smallholders,
with farms < 20 ha producing ~ 70% of the world's vitamin A. While
both Samberg and Herrero provided clear steps forward in understanding
the role of smallholders in the food system, and in particular Herrero
covering both animal and crop products, they did not use direct mea-
surements of crop production and/or area by farm size, compute di-
versity calculations based on these direct calculations of production and/
or area, or report on the broader role of smallholders in the food system
(e.g. how much of their food is wasted and destined to non-food crops).
To fill these gaps, we compiled the first open source dataset to es-
timate crop production by farm size derived from actual farmer surveys
containing crop-specific measurements of production or area that are
cross-tabulated against each farm size class. Our dataset includes 154
crop types and covers 55 countries, which represents 51.1% of global
agricultural area. We compare these direct estimates to those from the
previous modeling studies (e.g., Herrero et al 2017; Samberg et al.,
2016).In addition, we provide global estimates of the type of produc-
tion (i.e., food, feed, processing, seed, waste, and other) across farm
sizes and within each farm size class, to understand if more production
from small farms is wasted from storage and transportation, and if this
cancels the larger losses to biofuels and animal feed grown on large
farms. Finally, we evaluate how the type of crops grown, crop species
diversity, and macro-nutrient production varies by farm size. Our study
is the first to directly evaluate the relationship between farm size, crop
types, and crop diversity across a large range of farm sizes and geo-
graphic regions, and to assess how this diversity influences the amount
of macro-nutrients available from crops. Together, these results provide
the most comprehensive empirically grounded estimates of crop pro-
duction by farm size currently available.
2. Methods
2.1. Data compilation
We compiled a global convenience sample of datasets that directly
measured crop production and/or area by farm size for 55 countries at
either the national, or subnational level (for a total of 3410 national or
subnational units; see Fig. 1). These datasets were either agricultural
census data or nationally (or sub-nationally) representative sample
surveys, aggregated by administrative unit (n= 34 countries) or
available at the micro-level (e.g., anonymized individual household
level records) (n= 21 countries; of which 18 were household surveys
and 3 were censuses that captured both family and non-family farms).
The median year of the data was from 2013, with the oldest datasets
from 2001 and the newest from 2015. The database has 154 crops
which we matched with commodity names outlined in the Food and
Agricultural Organization's (FAO) statistical database (2017) [FAO-
STAT hereafter]. Where farm size and production were not cross-ta-
bulated in the survey instrument (i.e. for 33 countries), we calculated
production by farm size by first extracting either harvest area, culti-
vated area, crop area, or planted area to calculate farm size, and then
converted area to production using FAOSTAT's national yield data. We
tested the validity of this method, and found it to slightly underestimate
production (full details of bias tests, inclusion criteria, variable de-
scriptions, summary statistics, and per country statistics are given in the
accompanying Data in Brief article). When farm size data was not
available for a country, but we had micro-level data, we used the sum of
farm plot areas for a given household as a proxy for farm size. Internal
validation of the use of micro-data to fill in data gaps was not possible
with our data, because we did not have both micro-data and farm size
metrics for any of our countries, but we think the impact of using ag-
gregate plot area is likely to be negligible for our results, as this was
only used on 4.8% of administrative units in our dataset. Finally, all
crop production data was tallied per country and validated against
available national level reports, and to the FAOSTAT crop production
database, both of which are computed from aggregated crop area es-
timates. In total, our dataset captures 51.1% of global crop production
and 52.9% of global cropland area. We harmonized the datasets to
match the WCA farm size categories: 0–1 ha, 1–2 ha, 2–5 ha, 5–10 ha,
10–20 ha, 20–50 ha, 50–100 ha, 100–200 ha, 200–500 ha,
500–1000 ha, and above 1000 ha. While we recognize that per country
definitions of smallholders may not fall within these farm size bins, the
majority of the datasets included reported these farm size breaks. We
report our estimates by each WCA farm size class and cumulatively to
allow flexible definitions of smallholders that are consistent with past
attempts to quantify the relationship between farm size and crop pro-
duction. Future researchers may use the accompanying, open-access
dataset to redefine smallholders based on country specificdefinitions.
Where European data included a > 100 ha category, we included this
in the 100–200 ha range, making our classification less precise in >
100 ha groupings, in comparison to < 100 ha. Future researchers may
wish to aggregate all ‘large’farms into a > 100 ha bin for their specific
needs, but here we present the results maintaining the disaggregation
for surveys that reported it.
V. Ricciardi et al. Global Food Security 17 (2018) 64–72
65
2.2. Crop allocation
Following data compilation, we converted all tonnes of production
to their kilocalorie (kcal/capita/day) equivalents using FAOSTAT con-
version values per crop per country per year. We then applied the
percent of feed, food, processing, seed, waste, or ‘other’based on
FAOSTAT's food balance sheets per crop per country per year. For ex-
ample, in many countries maize can be used for human consumption,
animal feed, a processed biofuel commodity, and seed, while some
maize may be lost due to storage and transportation. FAOSTAT contains
national totals for each of these types of crop allocation categories. We
used these totals to calculate percentages per crop per country per year
to allocate a certain portion of each crop's production towards food,
feed, and the other crop allocation categories. While this approach does
not account for the actual distribution of crop allocation by farm size, it
is the most detailed information available and represents a proxy in-
dicator based on what type and quantities of crops each farm size
produces.
While certain FAOSTAT categories were straightforward to interpret
and contained detailed definitions (e.g., ‘feed’towards livestock and
poultry and ‘seed’set aside for sowing or planting), the processing ca-
tegory was ambiguous and required us to make assumptions. We fol-
lowed Cassidy et al. (2013) and assumed that the reported oil crop
processing category already separated out oil crop production for
human consumption from that for industrial use, as well as protein
dense cakes for animal feed. The waste category encompassed any loss
of a given commodity during storage and transportation; losses incurred
before and during harvest were excluded, as were losses due to
household consumption. The ‘other’category encompassed any uses not
already accounted for.
After allocating all crop production to type of production (e.g., feed,
food, other, etc.) in kcal/capita/day we evaluated how the global
quantity of each varied across farm size classes. We also provide cu-
mulative distributions of our estimates to encompass a sliding scale of
definitions of small-farms (e.g., farms under 2 ha, farms under 50 ha,
etc.), as may be required by different researchers and regional policy
makers who might define ‘small’using different thresholds. In addition
to comparing how the type of production varies across farm sizes, we
also analyzed how the types of production are distributed within each
farm size class.
To obtain global estimates for the proportions presented in this
manuscript, we computed 95% confidence intervals using the ac-
celerated bias-corrected percentile limits bootstrap method (BCa), with
1000 iterations. BCa is useful extension of the basic percentile
bootstrap, that decreases coverage error by accounting for bias in
sample parameters (i.e., when the sample parameter —computed from
the 55 countries —does not equal that of the average bootstrapped
parameter, which in our case is our best estimate of global trends), and
allowing the standard deviation of the bootstrap parameter to vary with
the sample parameter (Manly, 2006). We chose to bootstrap all of the
parameters of interest at the level of the country (n = 55), not the
administrative unit (n = 3410), in an attempt to account for de-
pendencies amongst administrative units in the same country and
sampling campaign. Accuracy in uncertainty estimation for the global
trends could be improved in future by adding to the number of coun-
tries in the dataset. While the BCa does not make any assumptions
about the distribution of underlying random variable we use the natural
log transform of production in our analysis for data visualization.
2.3. Crop species diversity and crop types
To estimate the relationship between crop diversity and farm size,
we counted the proportion of unique number of species each farm size
category produced within each administrative unit, and estimated the
95% CI's for each category using BCa. We note that different survey
instruments have different crops included, and that farmer responses
may not include the full diversity of crops that farmers actually pro-
duce. Thus, our estimates represent our current state of knowledge
given empirical data, and are likely to be conservative. We present BCa
estimates of crop diversity for each farm size using the administrative
unit level to compare crop diversity distributions across farms within
similar biogeographical landscapes (e.g., climate, soil, etc.). We also
present BCa estimates of crop diversity while controlling for cumulative
farm area, to give an indication of how diversity scales across the world
in each farm size class. To do this we plotted cumulative numbers of
unique species against cumulative area of administrative units for each
farm size class, and estimated uncertainty for these curves by resam-
pling the distribution of administrative units for each size class at
random, a 1000 times (taking the 2.5th and 97.5th percentiles as lower
and upper bounds, respectively).
To examine the variation in crop groups by farm size, we aggregated
our crop species data into major commodity groups according to
FAOSTAT definitions of cereals, fruit, oil crops, pulses, roots and tubers,
tree nuts, vegetables, and other, and we estimated 95% CI's using BCa.
Relying on the FAOSTAT classification has its limitations. For example,
soy was classified as an oil crop, but it is also a pulse; therefore, this
classification should be used as a guideline (see accompanying Data in
Brief for crop grouping details). In order to examine whether different
Fig. 1. Spatial coverage and resolution of our data on crop production by farm size. Countries shaded purple had directly measured data on crop production or
harvested area.
V. Ricciardi et al. Global Food Security 17 (2018) 64–72
66
farm sizes grew a different portfolio of crop groups, we used Sorensen's
similarity index (Chao et al., 2006):
=+
C
SS
C
C2
ij
ij
ij
+
S
S
i
j
where C
i,j
is the number of species two farm size classes have in
common, Sis the total number of species found in the given farm size
class, and iand jare the two farm size classes being compared; a score
of 1.0 would represent perfect overlap in the crop groups grown be-
tween the two farm size classes.
2.4. Macro-nutrient production
We converted production of each crop in our dataset to its macro-
nutrient (i.e., carbohydrate, protein, or fats in grams/capita) equivalent
using FAOSTAT food balance sheets, and conversion factors per crop
per country for the year matching the farm size data survey year. Any
temporal data gaps in FAOSTAT were linearly interpolated per crop and
country. As with production, we analyzed how macro-nutrient pro-
duction varied both across farm-size classes and within farm-size classes
and computed 95% CI's using BCa at the country level to estimate
global figures.
Fig. 2. A-F) Distribution of total global crop production (in kcal equivalents) across farm size groups different uses (e.g., food, feed, other, etc.). Grey shows
bootstrapped 95% confidence intervals and red indicates the average. G) Allocation of use of production within each farm size class. H) Cumulative percent of global
food production by farm size group with 95% confidence intervals. See Table S1 for underlying data. (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article).
V. Ricciardi et al. Global Food Security 17 (2018) 64–72
67
3. Results
3.1. Crop allocation
The smallest two farm size classes (0–1 ha and 1–2 ha) are the
greatest contributors to global food production compared to all other
classes. Farms less than 2 ha produce 28–31% of total crop production
and 30–34% of the global food supply (by calories; Fig. 2A-H) as ex-
trapolated from the 55 countries in our dataset. Their contribution is
slightly higher than their areal coverage of 24% of gross harvested area,
suggesting small farmers have greater cropping intensity or higher
yields than larger farms.
We found smallholders (farms < 2 ha) also allocate the largest
percentage (55–59%) of their crop production to food compared to all
other farm size classes (Fig. 2G). Generally, larger farms devote more of
their production towards feed and processing. Farms between 200 and
500 ha have the largest allocation of their production to feed (16–29%)
compared to farms < 2 ha who allocate 12–16% to feed. Farms >
1000 ha allocated 12–32% of their production to processing.
Farms < 2 ha contribute the most 28.1% (26–30%) to total food
waste (on-farm and post-harvest loss) (Fig. 2E); however, this is mainly
driven by this farm size group's large contribution to the total crop
production. In our dataset, only 4% (2.3–6.1%) of smallholder pro-
duction is wasted, compared to farms > 1000 ha that have the greatest
amount of within farm size class waste at 7.5% (0.0–18.5%). However,
the large uncertainty indicates both that there is substantial variation
within large farms, and low confidence in the trend between farm size
and waste holds at the global level. All farm sizes have fairly consistent
allocations towards seed (means ranged from 2% to 5% with over-
lapping 95% CI's), while there is a trend that smaller farms allocate
more to the ‘other’category.
3.2. Crop species diversity and crop types
We found that species richness declined with increasing farm size
(Fig. 3A). Diversity also scaled differently with area within different
farm size classes, with greater turnover in unique species in small farms
than in land allocated to larger farms (Fig. 3B).
Between farm size dissimilarity in species shows that larger farms,
while harboring less diversity, and lower turnover in crop diversity
across space, show greater specialization in certain crop groups than
other farm sizes. Farms < 5 ha grow similar crops as each other
(Sorensen's coefficient of 0.94), and farms > 100 ha have a perfect
overlap in crops grown (Sorensen's coefficient of 1.0; Fig. 4). But farms
greater than 20 ha grow a different array of crops compared to farms
smaller than 20 ha (Sorensen's coefficient of 0.4–0.67) and farms
greater than 100 ha have the lowest overlap with other farm size
classes.
The crop portfolio of each farm size class shows that smaller farms
(< 2 ha) produce a greater share of the world's fruits, pulses, and roots
and tubers, while medium sized farms produce more vegetables and
nuts, and large farms produce more oil crops and 'other' (Fig. 5). While
all farm sizes contribute a large proportion to cereals, smaller farms
devote a greater percentage of their overall production to cereals
compared to other farm size classes.
3.3. Macro-nutrient production
The trends in macro-nutrient (carbohydrates, proteins, and fats)
production across farm sizes follows that of the food production. Yet, of
their own production, smaller farms produce a slightly higher percen-
tage of carbohydrates (~ 0.08% more than the largest farm size class)
while larger farms grow a slightly higher percentage of proteins (~
0.05% more than the smallest farm size class). But these differences are
minute, and considering the uncertainty estimates, there are no sig-
nificant differences in the percentage of macro-nutrients produced
within each farm size class (Fig. 6).
4. Discussion
4.1. Comparison to previous studies
Our dataset is the first global sample of direct crop-specific mea-
surements of production or area by farm size. We found that farms <
2 ha produce 28–31% of total crop production and 30–34% of the food
supply on 24% of gross agricultural land when using our directly
measured farm size dataset. While our dataset covers 55 countries, with
distinct datagaps in smallholder dominant Southeast and East Asia, our
findings are in line with Samberg and Herrero's global estimates. This
suggests that these three studies, using different methodologies, agree
Fig. 3. A) Distribution of total species richness across farm size classes. Grey represents bootstrapped 95% confidence intervals; red is the bootstrapped average. The
area of each administrative unit polygon weighted the data. See Table S2 for underlying data. B) Cumulative area to cumulative species richness curves. 1000
iterations generated the cumulative distributions between species richness and farm size. The starting point for cumulative distributions were randomly chosen each
iteration. The lighter the colors, the larger the farm size classes. (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article).
V. Ricciardi et al. Global Food Security 17 (2018) 64–72
68
that the previous estimate of smallholders producing 70–80% of global
food production needs to be revised.
While our results are similar to the previous two modeling studies
that estimated global smallholder production, there are several key
differences. Our results offer more refined estimates using direct mea-
surements of production by farm size instead of relying on modeling,
includes a larger range of crop species than previously assessed, and our
accompanying open-access dataset allows individual countries to have
a reliable SDG baseline for how much of their food production is grown
by smallholders (according to their own regional definitions of farm
size). Samberg reported that farms < 5 ha produced 55% of global food
calories, which is slightly larger than our equivalent estimate of
44–48% (Table 1, Samberg A.). To arrive at this estimate, they divided
the total calories produced in each farm size category in their 83-
country sample by total global calories produced by all countries. Their
estimate could be considered a global estimate if one assumes that their
sample of smallholder dominant regions account for most of the world's
small farms (and their purposeful sample might suggest that inter-
pretation). An alternative interpretation (which is similar to ours) is
that the 83 countries in their sample are globally representative; in that
case one would divide the calories produced by each farm size class by
the total calories produced in those 83 countries. By this estimate,
farms < 5 ha produced 76% of global food calories (Table 1, Samberg
B.). The differences between Samberg and our dataset may be due to
the countries and crops sampled and that Samberg relying on modeled
results instead of direct measurements. Samberg used 41 crop species
(while we included 154), and they use mean agricultural area instead of
farm size distributions to understand crop production in smallholder
dominant areas rather than crop production by farm size. Both Samberg
and our study relied on household sample surveys to varying degrees
(Samberg relied primarily on household surveys, while our dataset re-
lied on them for 22.5% of total crop production). Household sample
surveys systematically do not sample non-family farms and hence may
be presumed to over-represent smaller farms when compared to agri-
cultural censuses that survey all farm types. However, in our accom-
panying Data in Brief article we show that using household surveys to
estimate national production is not significantly different than using
FAOSTAT's national production estimates.
Our estimates were also close to Herrero's global estimates. < 50
ha. We found that farms < 2 ha produce 30-34% of the worlds food
and < 50 ha produce 62–66% of the world's food, which is near
Herrero's estimate of 18% and 56%, respectively (as in Herrero et al.
(2017) Table 3). Our two studies capture different aspects of the global
food system. Herrero incorporated livestock and fisheries, which are
important source of nutrients and income for smallholders, while we
only focus on crop production; our focus was due to data constraints
and definitional mismatches between using farm size versus herd size,
fishing area, or common pasture land. There are also crop species dif-
ferences between the datasets, where Herrero used 41 crop species
while we used 154. One key analytical difference is that Herrero's
modeled results used field size as a proxy for farm size instead of actual
reported farm size; we used field size as a proxy for farm size for only
4.8% of our data and direct measurements of production by farm size
instead of modeled estimates. We found that using field size as a farm
size proxy measure may slightly over estimate small-farms’production
since it does not account for non-field elements of a farm (see the ac-
companying Data in Brief article). Additionally, Herrero disaggregated
production to the pixel level based on field size, while Samberg dis-
aggregated pixel-level production based on mean agricultural areas.
Essentially, both methods assume a constant yield for each farm size
class since they cannot directly link crop production with farm size.
There is a widely observed inverse relationship between yield and farm
size (IR), where smaller farms have higher yields (Bevis and Barrett,
2016; Henderson, 2015; Sen, 1962). For 66.7% of our dataset we also
needed to use constant yields since direct data on production by farm
size was not always available (we did have harvest area per crop by
farm size, and minimally used data on planted area, cropped area,
plotted area). Our dataset allowed us to test for the bias introduced by
constant yield methods, and provides the relationship which re-
searchers may use to correct for it. In the accompanying Data in Brief
article, we found a small effect size that using constant yields slightly
underestimates small-farms’production. Hence, our numbers, Herrero's
and Samberg's may all slightly underestimate smallholders’crop pro-
duction owing to this assumption.
Fig. 4. Heat-map of Sorensen's coefficient between each farm size class pair. Purple indicates a greater similarity of crops grown between pairs of farm size classes,
while brown indicates greater dissimilarity between crops grown.
V. Ricciardi et al. Global Food Security 17 (2018) 64–72
69
4.2. Crop allocation
Our new findings on crop allocation across different farm sizes has
important implications for food access and availability, as well as
farmer livelihoods, since food, feed, processing, and seed market prices
may differ from one another. We found nearly 60% of smallholder
production is allocated to food. A smaller percentage is allocated to-
wards feed (12–16%), which was surprising since smallholders often
engage in mixed crop-animal farming systems (Smith et al., 2012); this
finding may be explained by the fact that smallholders likely rely more
on rearing animals that graze on pasture compared to largeholders.
Our results counter common thinking about smallholders’post-
harvest loss, where improving cold storage and road infrastructure is a
common development intervention to improve smallholder income by
reducing wastage. Our dataset suggests that only a small percentage of
smallholders’production is wasted. However, one reason for the low
amount of smallholder waste in our results may be due to food allocated
to the ‘other’category. From our data, 19–23% of smallholder pro-
duction went towards ‘other’uses. This may be indicative of the need
for smaller, farms to make use of all grown material in integrated
farming systems (e.g., using rice stocks as a cover crop to promote soil
health). Smallholders’large allocation towards ‘other’may be in-
dicative that waste reduction practices are common since smallholders
are often resource poor and would achieve higher relative benefit
Fig. 5. A-H) Distribution of global production by crop type across farm size classes. Grey shows bootstrapped 95% confidence intervals and red is the average. I) Crop
type portfolio within each farm size class. See Table S3 for underlying data.
V. Ricciardi et al. Global Food Security 17 (2018) 64–72
70
compared to largeholders to find a use for wasted crops.
While, interventions aimed to reduce smallholder post-harvest loss
are still needed in many locales, there is also a need for agricultural
researchers to identify why larger farms are wasting crops, because this
group showed the greatest proportion of waste in any category (al-
though this was country dependent, as shown by the wide bootstrapped
confidence intervals). An estimated 1/4 of global food production from
croplands is wasted from farm to market (Kummu et al., 2012). The
waste data we used takes into consideration the quantities lost in the
transformation of crop to processed goods (FAOSTAT, 2017). Hence,
one possible reason for the increased wastage of larger farms’is that
large farms on a whole engage in more crop production allocated for
processing. Since FAOSTAT's definition of waste also encompassed
waste incurred from poor distribution and storage, it was surprising
that smaller farms did not have larger proportions of their crop wasted
when the majority of smallholders are in countries with hot and humid
climates and poorer storage infrastructure (Cohn et al., 2017). Future
studies should disaggregate the types of crop production waste each
farm size contributes to and local dependencies for these relationships.
4.3. Crop species diversity, crop types, and macro-nutrient production
Our data suggests a negative relationship between farm size and
crop species richness. This adds significantly to the evidence on the
literature's mixed finding on this relationship (Assunção and Braido,
2007; Rahman and Kazal, 2015; Van den Berg et al., 2007), as our study
contains a wider range of farm sizes and more crop species than ever
compared in previous studies. Due to the heterogeneity in data sources
used to construct our dataset, there were not always a wide list of crop
species included in each national survey, which may indicate a larger
portion of primary crops were documented compared to local species.
This limitation indicates that our findings are conservative and suggest
that smaller farms, which are associated with producing many non-
primary crops (Fifanou et al., 2011; Keleman et al., 2013), may have
even higher degrees of crop diversity than we found.
There are several food access, nutrition, and climate resilience im-
plications of higher crop diversity in smallholder systems. Since
smallholders may be tied to subsistence-surplus production models and
constrained to highly localized rural markets, their food access is often
more reliant on their local communities’crop production compared to
large farms (Baiphethi and Jacobs, 2009). The differences in types of
crops produced by different farm sizes, and macro-nutrient contents
that follow overall food production trends, supports the differences in
micro-nutrient and macronutrient production across farm sizes, as
found in Herrero. However, there are discrepancies. While smallholders
produce a large amount of the world's protein rich pulses, we did not
find that they produced a greater relative percentage of proteins than
larger farms (i.e., all farm sizes allocated a similar percentage of their
production to proteins). This suggests potential benefits from the pro-
motion of mixed animal-crop systems for smallholders to access protein
of which they are often deficient (Smith et al., 2012).
Our results suggest a nuanced view of the benefits of landscapes
harboring different farm sizes, beyond the basic relationship between
farm size and crop species richness. More diversified farming land-
scapes may need to include smaller farms because they collectively
grow a higher diversity of crops than large farms, but also include
larger farms because of their unique crop composition. Each farm size
produces a greater quantity of certain types of crops than other farm
sizes: smaller farms produce more fruits, pulses, and roots and tubers,
Fig. 6. A) Percentage of macro-nutrient production across farm size classes with 95% confidence intervals. B) Percentage macro-nutrient production within each
farm size class with 95% confidence intervals. See Table S4 for underlying data.
Table 1
Comparison between global estimates for the percentage of food smallholders
produce. For Samberg, A. uses estimates compared to global total food pro-
duction, while B. compares estimates to total food production within their 83
sampled countries.
< 2 ha < 5 ha < 50 ha Methodological Distinctions
Our study 30–34% 44–48% 62–66% Direct measurements
154 Crops
55 Countries
Herrero 18% –56% Modeled estimates
41 Crops; 7 Livestock; 14 Aquatic
Species
Near global coverage
Samberg A.
a
37% 55% –Modeled estimates
41 Crops
Samberg B. 52% 76% 92% 83 Countries
Mean Agricultural Area as farm
size proxy
a
Note that we do not provide < 50 ha estimates for Samberg A because we
cannot support the assumption that there are no farms < 50 ha outside of the
83 countries sampled by Samberg.
V. Ricciardi et al. Global Food Security 17 (2018) 64–72
71
while medium sized farms produce more treenuts and vegetables, and
larger farms produce more oil crops. Promoting a diversity of farm sizes
may encourage a greater diversity of crop types at the landscape level
that can better provide more balanced diets and non-food needs, while
potentially mitigating climate risks to the food system as a whole.
5. Conclusion
This study attempted to provide a global baseline for international
policy measures aimed to support smallholder agriculture. These in-
clude a need for improved monitoring of SDG Goal 2.3, which aims to
double food production of smallholders and increase nutrient avail-
ability; yet, Goal 2.3's monitoring framework does not use crop pro-
duction by farm size as a national indicator (Sustainable Development
Solutions Network (SDSN), 2015). Our findings suggest that previous
estimates of the percentage of food produced by smallholders were
either overinflated by public-sector opinions (ETC, 2009;Maass
Wolfenson, 2013;FAO, 2014) or still needed directly measured data to
assess quality (Herrero et al., 2017; Samberg et al., 2016), and that a
nutrient diverse farming landscape would include a diversity of farm
sizes, since each farm size produces a unique crop portfolio.
Critically, while our dataset is the first to use directly measured crop
specific data on production or area by farm size, we were only able to
find 55 countries with the necessary data to do this analysis. To monitor
SDG Goal 2.3, there needs to be increased effort to build on datasets like
ours through leveraging stakeholder networks. Ongoing efforts to use
and add to our dataset will enable continuous food system monitoring
over time with more geographic precision. We urge researchers and
food system advocates towards data-driven policy monitoring to accu-
rately assess the scale and progress of policy interventions.
Declarations of interest
None.
Role of funding source
This research was funded through the University of British
Columbia 4 Year Doctoral Fellowship & Social Sciences and Humanities
Research Council (SSHRC) Insight Grant (#435-2016-0154). These
funding sources had no role in data collection or analysis.
Appendix A. Supplementary material
Supplementary data associated with this article can be found in the
online version at http://dx.doi.org/10.1016/j.gfs.2018.05.002.
References
Altieri, M.A., 2008. Small Farms as a Planetary Ecological Asset: Five Key Reasons Why
we Should Support the Revitalisation of Small Farms in the Global South. Third
World Network, Penang, Malaysia.
Assunção, J.J., Braido, L.H.B., 2007. Testing household-specific explanations for the in-
verse productivity relationship. Am. J. Agric. Econ. 89, 980–990. http://dx.doi.org/
10.1111/j.1467-8276.2007.01032.x.
Badstue, L.B., Bellon, M.R., Berthaud, J., Ramírez, A., 2005. Collective Action for the
Conservation of On-farm Genetic Diversity in A Center of Crop Diversity: an
Assessment of the Role of Traditional Farmers' Networks.
Baiphethi, M.N., Jacobs, P.T., 2009. The contribution of subsistence farming to food se-
curity in South Africa. Agrekon 48, 459–482. http://dx.doi.org/10.1080/03031853.
2009.9523836.
Bevis, L.E.M., Barrett, C.B., 2016. Close to the Edge: Do Behavioral Explanations Account
for the Inverse Productivity Relationship? Cornell University working paper, Mimeo
(accessed May. 10, 2018). http://barrett.dyson.cornell.edu/files/papers/Close%20to
%20the%20Edge%20July%202017%20Bevis%20&%20Barrett.pdf.
Cassidy, E.S., West, P.C., Gerber, J.S., Foley, J.A., 2013. Redefining agricultural yields:
from tonnes to people nourished per hectare. Environ. Res. Lett. 8, 34015. http://dx.
doi.org/10.1088/1748-9326/8/3/034015.
Chao, A., Chazdon, R.L., Colwell, R.K., Shen, T.J., 2006. Abundance-based similarity
indices and their estimation when there are unseen species in samples. Biometrics 62,
361–371. http://dx.doi.org/10.1111/j.1541-0420.2005.00489.x.
Cohn, A.S., Newton, P., Gil, J.D., Kuhl, L., Samberg, L., Ricciardi, V., Manly, J.R.,
Northrop, S., 2017. Smallholder agriculture and climate change. Ann. Rev. Envir.
Res. 42, 347–375.
Conway, G., 2011. On Being a Smallholder. Conference new Dir. Smallhold. Agric. IFAD
Rome 1–17.
ETC -group, 2009. Who Will Feed Us? Questions for the Food and Climate Crises.
FAO, 2014. The State of Food and Agriculture 2014: Innovation in Family Farming Food
and Agriculture Organization of the United Nations.
FAOSTAT, 2017. Food and Agriculture Organization of the United Nations FAOSTAT
database. 〈http://www.fao.org/faostat/en/〉. Last (Accessed 18 January 2018).
Fifanou, V.G., Ousmane, C., Gauthier, B., Brice, S., 2011. Traditional agroforestry systems
and biodiversity conservation in Benin (West Africa). Agrofor. Syst. 82, 1–13. http://
dx.doi.org/10.1007/s10457-011-9377-4.
Fritz, S., See, L., McCallum, I., You, L., Bun, A., Moltchanova, E., Duerauer, M., Albrecht,
F., Schill, C., Perger, C., Havlik, P., 2015. Mapping global cropland and field size.
Global change biology 21 (5), 1980–1992.
Graeub, B.E., Chappell, M.J., Wittman, H., Ledermann, S., Kerr, R.B., Gemmill-Herren, B.,
2016. The state of family farms in the world. World Dev. 87, 1–15. http://dx.doi.org/
10.1016/j.worlddev.2015.05.012.
Henderson, H., 2015. Considering technical and allocative efficiency in the inverse farm
size-productivity relationship. J. Agric. Econ. 66, 442–469. http://dx.doi.org/10.
1111/1477-9552.12086.
Herrero, M., Thornton, P., Power, B., Bogard, J., Remans, R., Fritz, S., Gerber, J., Nelson,
G., See, L., Waha, K., Watson, R., West, P., Samberg, L., van de Steeg, J., Stephenson,
E., van Wijk, M., Havlík, P., 2017. Farming and the geography of nutrient production
for human use: a transdisciplinary analysis. Lancet Planet. Health 1 (1), e33–e42.
Hodges, R.J., Buzby, J.C., Bennett, B., 2011. Postharvest losses and waste in developed
and less developed countries: opportunities to improve resource use. J. Agric. Sci.
149, 37–45. http://dx.doi.org/10.1017/S0021859610000936.
Horrigan, L., Lawrence, R.S., Walker, P., 2002. How sustainable agriculture can address
the environmental and human health harms of industrial agriculture. Environ. Health
Perspect. 110, 445–456. http://dx.doi.org/10.1289/ehp.02110445.
IFAD, UNEP, 2013. Smallholders, food security and the environment. Rome Int. Fund.
Agric. Dev. 54.
Keleman, A., Hellin, J., Flores, D., 2013. Diverse varieties and diverse markets: scale-
related maize “profitability crossover”in the central Mexican highlands. Hum. Ecol.
41, 683–705. http://dx.doi.org/10.1007/s10745-013-9566-z.
Kummu, M., de Moel, H., Porkka, M., Siebert, S., Varis, O., Ward, P.J., 2012. Lost food,
wasted resources: global food supply chain losses and their impacts on freshwater,
cropland, and fertiliser use. Sci. Total Environ. 438, 477–489. http://dx.doi.org/10.
1016/j.scitotenv.2012.08.092.
Lowder, S.K., Skoet, J., Raney, T., 2016. The number, size, and distribution of farms,
smallholder farms, and family farms worldwide. World Dev. 87, 16–29.
Maass Wolfenson, K.D., 2013. Coping with the food and agriculture challenge: small-
holders' agenda. In: Proceedings of the 2012 United Nations Conference on
Sustainable Development. 47.
Manly, B.F.J., 2006. Randomization Bootstrap and Monte Carlo Methods in Biology, 3rd
ed. Chapman and Hall/CRC.
Naylor, R., Steinfeld, H., Falcon, W., Galloway, J., Smil, V., Bradford, E., Alder, J.,
Mooney, H., 2005. Losing the links between. Science 80, 1621–1622. http://dx.doi.
org/10.1126/science.1117856.
Rahman, S., Kazal, M.M.H., 2015. Determinants of crop diversity in the regions of
Bangladesh (1990–2008). Singap. J. Trop. Geogr. 36, 83–97. http://dx.doi.org/10.
1111/sjtg.12086.
Samberg, L.H., Gerber, J.S., Ramankutty, N., Herrero, M., West, P.C., 2016. Subnational
distribution of average farm size and smallholder contributions to global food pro-
duction. Environ. Res. Lett. 11, 124010. http://dx.doi.org/10.1088/1748-9326/11/
12/124010.
Sen, A.K., 1962. An aspect of Indian agriculture. Econ. Wkly. 14, 243–246.
Smith, J., Sones, K., Grace, D., MacMillan, S., Tarawali, S., Herrero, M., 2012. Beyond
milk, meat, and eggs: role of livestock in food and nutrition security. Anim. Front. 3,
6–13. http://dx.doi.org/10.2527/af.2013-0002.
Sustainable Development Solutions Network (SDSN), 2015. Indicators and a Monitoring
Framework for Sustainable Development Goals - Launching a Data Revolution for the
SDGs.
Tefera, T., 2012. Post-harvest losses in African maize in the face of increasing food
shortage. Food Secur. 4, 267–277. http://dx.doi.org/10.1007/s12571-012-0182-3.
UN, 2015. Sustainable Development Goals (SDGs) [WWW Document]. URL 〈https://
sustainabledevelopment.un.org/topics/sustainabledevelopmentgoals#〉(Accessed 1
January 2017).
Van den Berg, M.M., Hengsdijk, H., Wolf, J., Van Ittersum, M.K., Guanghuo, W., Roetter,
R.P., 2007. The impact of increasing farm size and mechanization on rural income
and rice production in Zhejiang province, China. Agric. Syst. 94, 841–850. http://dx.
doi.org/10.1016/j.agsy.2006.11.010.
V. Ricciardi et al. Global Food Security 17 (2018) 64–72
72