Over the last twenty years, agricultural production has become
increasingly concentrated on larger farms. According to Census of
Agriculture data, farms from 1,000 to 10,000 acres increased in number
by 14.3% between 1982 and 2002 and total farmland controlled by these
large operations increased by 20.6%. (1) In contrast, over this period
farms with fewer than 1,000 acres declined in number and amount of
farmland controlled. Increasing concentration of agricultural production
has coincided with an increasing share of government payments going to
large farms: between 1982 and 2002 the share of all payments going to
farms from 1,000 to 10,000 acres increased from 41.1% to 49.5%.
In recent years interest groups, politicians, and newspaper
editorials have expressed concern that payments unfairly advantage large
operations and have argued that government payments are a key factor
contributing to the growth in concentration and farm sizes (e.g.,
Williams-Derry and Cook 2000; Becker 2001; Nelson 2002). Concerns about
the link between agricultural payments and farm size have helped
motivate congressional efforts to tighten payment caps on large-scale
producers (U.S. Department of Agriculture 2003). (2)
Claims that government payments unfairly advantage large farms are
usually supported with statistics showing a steady growth in farm size
and the strong association between farm size and payment levels.
However, while government payments and production have both become
increasingly concentrated, this concurrence of trends does not
necessarily imply a causal link between payments and farm size. The
design of government agricultural programs is such that payment levels
are tied to the amount of land farmed and the land's production
history. Thus, regardless of what caused farms to increase in size,
payments would have become more concentrated on larger farms (MacDonald,
Hoppe, and Banker 2005).
To what extent are government agricultural programs and their
associated
payments contributing to the concentration of production? Most
studies that have attempted to explain changes in the size and survival
of individual farms have focused on characteristics of the farm operator
or farm, not on the role of government payments (Sumner and Leiby 1987;
Hallam 1993; Zepeda 1995; Kimhi and Bollman 1999; Weiss 1999).
Exceptions are two recent studies by Key and Roberts (2006; 2007) that
found government payments positively and significantly associated with
the survival rate and duration of individual farm businesses, as well as
farm size conditional upon survival. These studies, however, consider
effects of payments on the growth or survival of individual farms, which
cannot predict the effects of an increase in payments on aggregate farm
structure. This is because studies of individual farms cannot account
for how induced changes on one farm affect other, neighboring farms, or
how payments influence the size and number of entering farms. For
example, consider a hypothetical case where payments cause consolidation
of two neighboring farms of equal size and equal payment levels. At the
level of an individual farm, the average payment effect on size may
appear to equal zero: one farm increased in size by 100% while the other
declined in size by 100%. Nevertheless, the aggregate effect on land
concentration is substantial.
Some past studies have estimated the effect of agricultural
payments on aggregate measures of farm structure, including the national
agricultural bankruptcy rate (Shepard and Collins 1982), the total
number of farms (Tweeten 1993), and average farm size (Huffman and
Evenson 2001). While taking very different approaches, these studies
treat government payments as exogenous and have used current payments to
explain current indicators of farm structure. A problem with this
approach is that it is difficult to attribute a causal mechanism to an
observed cross-sectional association between payments and farm size. To
do so requires confidence that determinants of farm structure, other
than agricultural payments, are adequately accounted for. A particular
concern is the heterogeneity of land and farms across regions in the
United States. A finding that farms are larger in areas with higher
payments might be explained by the fact that government programs target
field crops (e.g., corn, soybeans, and wheat), which require more land
to be profitably farmed. Another concern is the endogeneity of payments:
farm acreage decisions influence payment levels, so causation may go in
the opposite direction.
This study compares zip codes with different per-acre payment
levels to subsequent percentage changes in land concentration. That is,
it examines whether concentration growth rates are higher in areas with
higher historical payments per acre relative to areas with lower
payments per acre. Even if programs happen to target regions where farms
are larger because of the crops grown, we see no obvious reason to
expect programs to target regions inclined to subsequently experience
faster growth in concentration over time--by examining growth rates we
control for time-invariant factors associated with concentration. In
other words, a correlation between payments and the subsequent change in
land concentration is less likely to result from reverse causality.
The study supplements the simple comparisons described above with a
semiparametric generalized additive model (GAM) to control for location,
historical concentration, historical sales per acre, and the share of
land in agriculture in a flexible way. These variables control for
time-varying factors that may create a spurious relationship between
payments and concentration growth.
Concentration is measured at the zip code level. Farmland
concentration is defined as the acre-weighted median farm size: the farm
size such that half the farmland within each zip code resides on larger
farms and half resides on smaller farms (cropland concentration is
analogously defined). We focus on this measure, rather than the more
commonly used mean or median farm size, because these measures are
extremely sensitive to the definition of a farm, which has changed
implicitly or explicitly over time, and are heavily influenced by a
growing number of small "hobby" farms. In table 1, notice the
growth in the number of smallest farms (0-50 acres) and larger farms
(1,000-10,000 and 10,000 + acres) and the marked decline in the number
of middle-sized farms (150-500 and 500-1000 acres).
The analysis uses microdata from the 1987, 1992, 1997, and 2002
agricultural censuses and includes all U.S. zip codes with at least
three farms in all four censuses. The zip code analysis improves upon
national, state, or county-level analyses by providing more observations
and more variation across observations in both concentration and payment
levels. Sufficient variation at a local level is important when using an
empirical technique that controls for factors that vary geographically.
The census data are the only data available for such a fine-scale
analysis: the census attempts to collect information on every U.S. farm
business with expected sales of at least $1,000.
Determinants of Concentration: Farm Size and Survival
As the amount of U.S. farmland and cropland has remained relatively
stable, changes in concentration from one period to the next depend on
the size distribution and growth rate of surviving farms, and on the
sizes of entering farms (Vesterby et al. 2006). The literature on firm
size and survival therefore provides some insight into the determinants
of farm structure. In this literature, the relationship between firm
size and survival is often modeled as a dynamic process wherein firms
(or entrepreneurs) are uncertain about their own competitiveness at
startup (Jovanovic 1982; Ericson and Pakes 1992; and Pakes and Ericson
1998). In these models, firms gradually learn about their abilities over
time and the longer they operate, the more they learn about their
competitiveness. As managers revise their perceptions of their
firm's ability upward, they tend to expand, while those revising
downward tend to contract or exit. Thus, the longer a firm has existed,
the bigger it will become and the less likely it will be to fail.
Empirical studies generally confirm these theoretical predictions
(Dunne, Roberts, and Samuelson 1988; Baldwin and Gorecki 1991; Audretsch
1991; Audretsch and Mahmood 1995; among others).
Theory does not provide unambiguous predictions as to how a change
in government payments would influence farm growth and survival.
Consider, for example, a model of a representative farm where the
quantity of agricultural land is fixed, but labor and capital are mobile
between agricultural and nonagricultural sectors (Kislev and Peterson
1982). In this model, farm size is a function of the ratio of wages to
the cost of capital. An increase in government payments increases
returns to farming, but these additional profits are capitalized into
the price of land. Hence, a change in government payments has no clear
direct effect on the cost of labor relative to capital, and therefore
has no effect on farm size.
In more complex economic models that allow for transaction costs
and a range of farm sizes, there arise a variety of mechanisms through
which payments could influence farm structure. For example, if per-acre
payments are unequally distributed across farms of different sizes then
an increase in payments could alter farm structure. Such a pattern may
arise if there are fixed transactions costs associated with program
participation, so that larger farms have a stronger incentive to
participate than smaller farms. Higher payments per acre for a
particular farm size group would allow this group to expand and bid up
the prices of fixed resources--especially land--and cause other size
farms to shrink or exit.
An unequal distribution of per acre and/or total payments might
also influence farm size and survival through capital or labor market
mechanisms. Borrowing constraints could cause a farm's cost of
capital to depend on its net worth: farms with greater net worth face
lower borrowing costs because they have more resources with which to
secure a loan (e.g., Hubbard 1998). If this were the case, an increase
in income from government payments would raise the net worth of a farm,
making it less costly for a farmer to obtain financing to increase farm
size. Similarly, anticipated payments may give farm operators more
leverage with agricultural lenders. Because larger farms presumably
require more capital, both per-acre and/or total payments may influence
borrowing costs. If large farms are credit constrained and small farms
are not--a counterintuitive but distinct possibility given increasing
returns to scale and the fact that larger farms tend to be more
leveraged--then an increase in payments causes large farms to expand and
increase in number, which bids up land prices and causes small farms to
shrink and decline in number (Key and Roberts 2005). If both large and
small farms are credit con strained, then the effect of an increase in
government payments on farm size and survival is ambiguous.
Total payments may also influence farm size and survival by
altering farm operator labor-leisure decisions via a wealth effect
combined with transactions costs. Payments could encourage farmers
receiving them to work less; and if there are transaction costs
associated with hiring labor or finding employment, higher payments may
cause a reduction in the supply of farm labor (Lopez 1984; Strauss
1986). Less farm labor could mean less production and a smaller farm.
However, under certain conditions, a higher shadow wage for farm labor
could mean greater capital utilization and thus an increase in farm
size, as in Kislev and Peterson. (3)
Trends in Concentration and Government Payments
Census of Agriculture data illustrates the increasing concentration
of production in U.S. agriculture. Table 1 shows a marked increase in
the prevalence of farms with between 1,000 and 10,000 acres. (4) Between
1982 and 2002, these large farms increased from 6.7 to 8.0% of all farms
and increased their share of total farmland from 34.0 to 41.8%. Growth
in the number of these large farms came mainly at the expense of farms
between 150 and 500 acres, which declined as a share of all farms from
29.7 to 23.5%. The number of farms with fewer than 50 acres increased
markedly, as did their share of farmland. Although these farms comprised
37.8% of all farms by 2002, they made up only 1.7% of all farmland. Very
large farms--those with more than 10,000 acres of farmland--increased
slightly in number and declined slightly in their share of all farmland.
Table 2 presents four measures of representative farm size from
1982 to 2002 for all farms, and for farms with fewer than 10,000 acres.
For all farms, mean farm size increased by 2.4%, from 431 acres in 1982
to 441 acres in 2002. However, median farm size actually declined by
22.1% over this period: falling from 122 to 95 acres. The decline in
median farm size reflects the growing proportion of very small farms
mentioned above. (5)
The acre-weighted mean and the acre-weighted median are alternative
indicators of land concentration. The weighted mean farm size averages
farm sizes over acres rather than over farms. It can be thought of as
the expected farm size associated with a randomly chosen acre of
farmland. (6) The acre-weighted median is the size of a farm such that
half of all farmland is operated by larger farms and half by smaller
farms. Both of these measures emphasize the production unit (acre),
rather than the farm, as the unit of interest. This emphasis makes sense
when focusing on land concentration, because we want a statistic that
measures the farm size associated with a typical production unit. The
weighted median and particularly the weighted mean are much larger than
the unweighted measures, reflecting the fact that large farms control
most farmland. Table 2 shows that for all farms, the weighted mean
increased by 96.0% between 1982 and 2002, while the weighted median
increased by 35.2%. The weighted median row indicates that in 1982 half
of all farmland was operated by farms larger than 1,620 acres; by 2002,
half of all farmland was controlled by farms larger than 2,190 acres.
Comparing all farms to farms with fewer than 10,000 acres (bottom
of table 2), we find similar patterns over time for the mean, median,
and weighted median (the levels are smaller but the changes over time
are similar). However, the weighted mean increased by a smaller amount.
The analysis of changing land concentration in the ne