Content uploaded by Lori Lynch
Author content
All content in this area was uploaded by Lori Lynch
Content may be subject to copyright.
Is There Evidence of a Critical Mass
in the Mid-Atlantic Agriculture Sector
Between 1949 and 1997?
Lori Lynch and Janet Carpenter
Ongoing farmland loss has led county planners to ask “is there a critical mass of farmland needed?”
to retain a viable agricultural sector. This study examines whether counties lost farmland at a faster
rate if the number of agricultural acres fell below a critical threshold. Results from six Mid-Atlantic
states over the period 1949 to 1997 indicate that counties with fewer agricultural acres lost farmland
at a faster rate. However, after splitting the study period into two time segments (1949S1978 and
1978S1997) and modeling separately, this result was not found for the later time period, suggesting
a uniform critical mass level may not exist. Population growth in a county accelerated farmland loss
over all time periods.
Key Words: critical mass, development pressure, farmland loss, panel data, use-value taxation
As farmland has decreased over the last 50 years,
some have questioned whether there is a critical
mass of agricultural land needed to sustain a viable
agricultural sector.1 The total amount of land in
farms decreased by 20% in the United States be-
tween 1949 and 1997. In the Mid-Atlantic region,
the rate of decrease was 50%. Agricultural land has
been converted to alternative uses at an even higher
rate in metropolitan areas (Lockeretz, 1989; Gard-
ner, 1994). Metropolitan residents have expressed
concern about the loss of the amenities the farmland
provides. Rural economies that are highly dependent
on agricultural industries suffer negative conse-
quences when agricultural land is converted. Society
Lori Lynch is an associate professor, Department of Agricultural and Re-
source Economics, University of Maryland; Janet Carpenter is a biotech-
nology advisor, U.S. Agency for International Development. Support for
this project was provided by the Maryland Center for Agro-Ecology, Inc.
This paper was presented at the Land Use Policy Workshop of the
Northeastern Agricultural and Resource Economics Association annual
meetings, Harrisburg (Camp Hill), PA, June 9S11, 2002.
1 This is a slightly different question than that posed in the 1970s
when citizens advocated farmland preservation for food security reasons.
Several research studies, including those in the early 1980s by Fischel
(1982) and Dunford (1983), analyzed whether the rate of farmland con-
version would affect the national agricultural production capacity. They
found that while farmland was disappearing from certain regions, there
were sufficient national land resources still available to ensure the nation’s
food security.
may want to retain an agricultural sector to maxi-
mize its welfare. Many counties are trying to deter-
mine how much agricultural land must be retained
to ensure a viable agricultural economy and the pre-
servation of these amenities. Government officials
need to determine “how much agricultural land is
enough” to ensure this retention. Therefore, know-
ing whether the rate of farmland loss is affected by
the level of agricultural activity within an area is
important.
Most land use models assume agricultural land-
owners will farm the land until the value in an
alternative use exceeds the agricultural value. These
models have focused on the effect of changes in the
nonagricultural use values on the decision to convert
rather than the effect of changes in the agricultural
use value (e.g., Bockstael, 1996; Chicoine, 1981;
Clonts, 1970; Dunford, Marti, and Mittelhammer,
1985; Hardie, Narayan, and Gardner, 2001; Bell and
Bockstael, 2000; Muth, 1961; Nickerson and Lynch,
2001; Shi, Phipps, and Colyer, 1997).
The net market value in many developed uses is
greater than in agricultural use. Thus, the market
will allocate the land to this privately optimal devel-
oped use. However, the conversion of farmland to
a developed use may not maximize society’s wel-
fare. First, the amenity benefits of farmland are lost.
Agricultural and Resource Economics Review 32/1 (April 2003): 116S128
Copyright 2003 Northeastern Agricultural and Resource Economics Association
Lynch and Carpenter Is There a Critical Mass in the Mid-Atlantic Agriculture Sector? 117
Second, the developed use may impact the surround-
ing farmland’s use value. Some researchers have
considered the external benefits of nearby land
uses—such as agriculture, parks, or forest lands—
on residential use values (Kitchen and Hendon,
1967; Weicher and Zerbst, 1973; Hammer, Cough-
lin, and Horn, 1974; McMillan, 1974; Peiser and
Schwann, 1993; Irwin and Bockstael, 2001; Geog-
hegan, Wainger, and Bockstael, 1997; Geoghegan,
2002: Geoghegan, Lynch, and Bucholtz, 2003;
Cheshire and Sheppard, 1995). However, few
models have considered the external benefits or
costs of adjacent land use on agricultural use value.
If the landscape surrounding the existing farmland
changes, the agricultural use value may also change.
The agricultural use value could be affected by both
the new uses on adjacent land and the resulting ad-
justment in the agricultural support sector to fewer
farmland acres.
Adjacent land use can affect agricultural land use
in several ways. Population growth or suburbani-
zation near farming areas can create problems for
farmers. Nonfarm neighbors may object to nuisances
related to traditional farming practices, such as in-
sects, noise, odor, dust, and slow-moving equipment
on the roads, and may advocate limitations to these
practices. Even in rural areas, incompatible activ-
ities in the surrounding landscape may affect the
profitability of farms. Farmers may earn more profit
operating within a thriving agricultural community
than in an area dominated by other land uses—be
they city, forest, or recreational. Alternatively, the
proximity of nearby suburban cities may generate
new marketing opportunities.
The loss of support industries may also affect the
agricultural use value. The exodus of the support
industries in an area may indicate that a critical
mass of agricultural land is important to long-term
viability. A critical mass threshold implies econo-
mies of scale exist in both input and output busi-
nesses which are essential to agriculture. Support
businesses will close or relocate as farm production
levels decrease below a certain threshold.
For example, growers in a region may specialize
in peas. They sell the peas to a local processing
plant that freezes and packages their product. The
plant will remain in business as long as local pro-
duction is sufficient to sustain the firm’s production
at competitive operating costs. If local pea produc-
tion decreases, the processing plant may not achieve
economies of scale and might close at this location.
Remaining pea producers would face decreased
returns, as the shipping costs to more distant pro-
cessing plants are higher. Consequently, they might
switch to producing another, less profitable crop.
As the agricultural use value decreases, the relative
return to converting the farmland increases. Thus
the rate of farmland loss could increase.
Few studies have explored the existence of a crit-
ical mass for the agricultural sector. In an early
study, Dhillon and Derr (1974) estimated the
critical level of production necessary to operate at
or close to the minimum per unit production cost
level for various agricultural commodities grown in
the Philadelphia–New York–Boston corridor. More
recently, Daniels and Lapping (2001) proposed a
critical mass threshold definition of (a) at least
100,000 acres, (b) $50 million in agricultural sales,
and/or (c) 20,000 acres of preserved farmland in a
county.
We conduct an econometric analysis to deter-
mine if we can find statistical evidence of, and the
level of, a critical mass threshold for six Mid-
Atlantic states. The research seeks to determine if
a critical mass threshold exists. We analyze whether
counties lost farmland at a faster rate if the number
of acres fell below a “threshold” level of acres. The
study area includes six Mid-Atlantic states over the
period 1949 through 1997. Farmland loss can be
affected by a variety of factors besides a critical
mass. Thus, the effects of factors such as agricul-
tural net returns and development pressure on the
rate of farmland loss are also examined. We esti-
mate the impacts and conduct a sensitivity analysis
of these factors, and then explore whether the
effects change over time.
The remainder of the paper proceeds as follows.
The next section outlines the economic model and
the econometric model and analysis. Next, a descrip-
tion of the data and the study area is provided. The
results are then presented, including the sensitivity
analysis. We close with a discussion of the results
and offer some concluding remarks to guide future
research efforts.
The Economic and Econometric Models
Researchers and policy makers hypothesize that a
critical mass threshold exists in the agricultural
sector. The basis for this notion is the existence of
economies of scale in both input and output busi-
nesses which are essential to agriculture. As pro-
duction levels decrease below a certain threshold,
support businesses will close local facilities as their
revenues decrease or their costs increase. After an
input or output firm closes, the alternative input
118 April 2003 Agricultural and Resource Economics Review
supplier or output buyer may be farther away. In
addition, the new supplier or buyer might charge
higher prices for inputs or pay lower prices for out-
put due to less competition or higher operating
costs.2 The change in the local agricultural econ-
omy alters the agricultural use value. This, in turn,
changes the relative returns of converting or of idling
the remaining agricultural land.
To investigate whether a critical mass threshold
exists in the Mid-Atlantic agricultural sector, we
analyze the difference in the rate of farmland loss
for counties with varying levels of farmland acre-
age over time, holding constant all other variables.3
The county’s rate of farmland loss is modeled as a
function of the number of productive agricultural
acres, the net return in an agricultural use, the net
value in a residential use, the existence of agricul-
tural preservation policies, and the possible off-
farm income opportunities. When a county has a
high number of productive acres, the farm sector
can sustain a viable support sector and local agri-
culture remains competitive. Thus farmland loss
rates are hypothesized to be a function of the num-
ber of productive farmland acres, holding all other
variables constant.
Farmland will be converted to nonfarm uses when
the net value of an alternative use is higher than an
agricultural use. This would increase the rate of
farmland loss. The farmland loss rate may also be
high if the area has low or negative agricultural net
return even if residential, commercial, and indust-
rial uses have low net values. The farmland would
be idled.
Agricultural preservation programs purchase de-
velopment rights on agricultural parcels and prefer-
ential taxation programs decrease a farmer’s property
tax. Thus, these programs can increase the relative
return of remaining in an agricultural use, thereby
contributing to the retention of farmland acres and
slowing the rate of farmland loss.
When off-farm income opportunities are high,
farmers might choose to leave farming and enter
other professions. This would increase the rate of
farmland loss unless this land is sold to another
farmer. Alternatively, off-farm employment oppor-
tunities may decrease the farmland loss rate if
farmers supplement their farm income with off-
farm income. Off-farm income can be an important
diversification strategy for farmers. Therefore, off-
farm employment may increase or decrease the rate
of farmland loss.
The Econometric Model
Several models were estimated to determine which
of the following was the most appropriate econo-
metric technique to use for the panel data: pooling
the data, pooling the data with fixed effects repre-
senting each five-year time period and/or each crop
reporting district, or estimating a random-effects
model. A random-effects estimation procedure is
found to be the most efficient, using Lagrange
multiplier (LM) and Hausman tests (HT) (LM[2] =
1,581.33; HT[14] = 17.53). Thus, the unexplained
variation in the rate of farmland loss, or the residual
for the estimated model, is comprised of three parts:
git, µi, and wt. The means of the three disturbances
are assumed to be zero, and each has a variance
equal to σ2
g, σ2
µ, σ2
w, respectively. The covariances
between the error terms are also assumed to be zero.
The model incorporates both the within and the
between random components.
The random-effects model to be estimated is de-
fined by the following equation (Greene, 1995):
(1) yit 'α%βNxit %git %µi%wt,
where yit is the vector of the county-level rate of
farmland loss for counties in crop reporting district
i in the five-year time period t, α is the vector of
constants, β is the vector of estimated coefficients,
and xit is the matrix of county-level characteristics
that explain farmland loss for crop reporting district
i in the five-year time period t, such as sales per
acre, percentage change in housing units, and the
unemployment rate. The error terms git, µi, and wt
are the effects of unobserved variables that vary
over both crop reporting district i and five-year time
period t, and within each crop reporting district and
within each time period.
The critical mass threshold may evolve over
time. Improved communications and transportation
infrastructure may reduce the costs of purchasing
inputs and facilitate marketing. Growers may adapt
when suppliers and processors exit the area. Adap-
tations could include switching to crops or animal
products which are less reliant on these support
industries. Farmers could shift to direct marketing
rather than wholesaling.
2 Alternatively, if smaller input and output firms are consolidated, leav-
ing fewer, larger firms, these remaining firms may achieve greater econ-
omies of scale. The major effect on farmers’ costs would then be the
increase in transportation costs.
3 Another possible method for studying this issue would be to examine
input suppliers’ and processing firms’ cost structure. However, even if
businesses would permit us to do so, many of the businesses we would
want to study have exited the region over the past 50 years.
Lynch and Carpenter Is There a Critical Mass in the Mid-Atlantic Agriculture Sector? 119
Figure 1. Distribution of counties among crop reporting districts in
the six Mid-Atlantic states
In addition, farming has experienced technolog-
ical and structural changes over the last 50 years.
The United States lost almost half of its farms be-
tween 1950 and 1970, due in part to mechanization
and the consolidation of farms (Gardner, 2002).
U.S. average farm size and output per farm grew
rapidly through the 1970s, and have grown more
slowly since then. Labor costs have decreased as a
portion of total input costs beginning in the 1980s.
Because of these changes, we estimate a general
model (model 1) for the entire time period (1949S
1997), and then split the sample into two sub-
periods and estimate two additional models. The
two subperiods are designated as 1949S1978 (model
2a) and 1978S1997 (model 2b). This approach
allows us to determine if the critical mass threshold
and importance of the other factors varied by time
period.
Description of Data and Study Area
Data were compiled from the Census of Agriculture
and the Census of Population and Housing at the
county level for the years 1949 through 1997 [U.S.
Department of Commerce, Bureau of the Census;
U.S. Department of Agriculture, National Agricul-
tural Statistics Service (USDA/NASS)]. The Mid-
Atlantic states of Delaware, Maryland, New Jersey,
New York, Pennsylvania, and Virginia are included
in the data set. In 1997, these six Mid-Atlantic states
accounted for over 26 million acres of farmland
(3%) of total U.S. farmland and for $12 billion (6%)
of total U.S. sales.
The analysis uses data on 269 counties4 and 10
time periods of 4S5 years each. These time periods
correspond to the years the agricultural censuses
were taken. The data set was constructed as a panel
by crop reporting district and by time periods. A
county’s data were included in the crop reporting
district to which it belonged. The USDA National
Agricultural Statistics Service defines these crop
reporting districts to reflect similar geography, soil
types, and cropping patterns (figure 1).
Because the two censuses were conducted on
different schedules, population and housing census
data were adjusted to coincide with the years of the
agricultural census data. The population and housing
census is collected every 10 years, while the agri-
cultural census is collected every four to five years.
The population and housing census data were inter-
polated by calculating a constant change in the
variables between the census years. This constant
change was used to adjust the population and hous-
ing census data to the year the agricultural census
was collected. Thus, if the population change was
4 Independent cities of Virginia are also included in the analysis. In
several cases, due to either aggregation in data or actual boundary
changes during the study period, counties and/or independent cities have
been combined for this analysis.
120 April 2003 Agricultural and Resource Economics Review
Table 1. Definitions of Variables and Descriptive Statistics for the Entire Sample (269 counties),
1949S
SS
S1997
Entire Sample
Variable Name Definition Mean Std. Dev.
PCFLAND Percentage reduction in farmland (%) 7.58 0.1256
HCLAND Harvested cropland (1,000 acres) 54.372 47.097
HCLAND2Harvested cropland squared (1,000,000,000 acres) 5.1724 9.710
PAGFFM Percentage of adults employed in agriculture, forestry, fisheries, and
mining (%) 9.99 0.1056
SALESPERA Sales per acre (1997 $/acre) 549.07 2,394.11
EXPPERA Expenses per acre (1997 $/acre) 331.51 2,227.93
POPPERA Population per acre 0.5773 1.8430
MADUMMY = 1 if county is in metropolitan area (%) 33.72 0.4728
PCTOTHU Percentage change in total housing units (%) 8.09 0.0689
PCMFINC Percentage change in median family income (%) 11.92 0.0838
PCMHVAL Percentage change in median housing value (%) 11.66 0.1017
PHIGHSCH Percentage of adults with at least a high school education (%) 48.41 0.0185
PUNEMP Unemployment rate (%) 5.49 0.0223
STAX =1 if state has preferential taxation program for agricultural land (%) 56.63 0.4957
PRESPROG =1 if state and/or county has purchase or transfer of agricultural
conservation easement program (%) 8.47 0.2785
Note: The percentage change variables use the initial year of the time period as the ending year of the percentage change calculation. Thus the
percentage change in housing units for time period t was calculated as (HUt ! HUt!1)/HUt!1, where HUt is the total housing units at time t.
25% for the 10-year period, it was assumed the pop-
ulation grew 2.5% each year. Data from the 2000
census were not yet available. Therefore, extrapo-
lations of the 1990 population and housing census
data were conducted for 1992 and 1997. These
values were calculated based on the change in
the variables between 1980 and 1990. The rates
of change were assumed to remain constant during
the 1990s.
Counties with fewer than five farms in 1949
were excluded f rom the entire analysis. Six counties
were excluded due to limited agricultural activity in
1949: Bronx, Queens, Richmond, Kings, and New
York counties of New York State, and Arlington
County of Virginia. If the sales per acre data were
not available for a county for confidentiality reasons,
the county was deleted for that particular time per-
iod but not for the entire analysis.
Because farmland loss is affected by changes in
agricultural returns per acre, demand for land for
nonagricultural purposes, farmers’ alternative em-
ployment opportunities, and preservation policies,
variables are included to control for these factors.
Table 1 gives the names, definitions, and descrip-
tive statistics for the variables included in the
analysis.
The dependent variable was the rate of farmland
loss for time period t. It was calculated as
At%1&At
At
,
where At is the number of acres in the initial period.
The rate of farmland loss averaged 7.58% over the
study period. Some counties lost 100% of their
farmland in a time period. One county gained
77.65% more farmland in t (many counties gained
during the 1974S1978 period). Farmland is defined
by the U.S. Census of Agriculture to consist of land
used for crops, pasture, or grazing. Woodland and
wasteland acres are included if they were part of the
farm operator’s total operation. Conservation Re-
serve Program and Wetlands Reserve Program acre-
age is also included in this count.
The percentage change variables use the initial
year of the time period as the ending year of the
percent change calculation. Thus the percentage
change in housing units for time period t was cal-
culated as
HUt&HUt&1
HUt&1
,
where HUt is the total housing units at time t.
Lynch and Carpenter Is There a Critical Mass in the Mid-Atlantic Agriculture Sector? 121
County-level harvested cropland acres (HCLAND)
in t proxy the critical mass threshold acres. Harvest-
ed cropland includes land from which crops were
harvested or hay was cut, and land in orchards,
citrus groves, Christmas trees, vineyards, nurseries,
and greenhouses. These acres are better indicators
of the level of agricultural activity. Idled farmland
or acreage enrolled in the Conservation Reserve
Program, for example, requires the purchase of few
inputs, produces no output, and may not contribute
in the same manner to maintaining a viable agricul-
tural support sector. It is hypothesized that the
county’s rate of farmland loss will increase if the
level of harvested cropland falls below the threshold
needed to sustain a viable agricultural support
sector. Harvested cropland acreage is also included
as a squared term. By including harvested cropland
in this manner, we can compute an acreage threshold
required to ensure a critical mass. Harvested crop-
land acres averaged 54,372 acres per county. The
highest number of harvested cropland acres in any
one county was 334,294 acres.
The percentage of the county population em-
ployed in agricultural, forestry, fishing, or mining
activities (PAGFFM) in t is also included to indicate
the dominance of these resource-based activities in
the county. Counties varied from almost no employ-
ment in this type of work to 70% of the adult popu-
lation employed in these activities, with an average
percentage of 9.99%.
The agricultural net returns are proxied by coun-
ty-level agricultural sales per acre (SALESPERA) and
expenses per acre (EXPPERA) in t. Farmers are more
likely to remain in agriculture if sales increase more
than expenses. Sales per acre averaged $549.07 in
1997 dollars, and expenses per acre averaged
$331.51. Despite the almost 50% decrease in land
devoted to agriculture in the six Mid-Atlantic states,
total revenue has decreased by only 1% in real terms
between 1949 and 1997. Per acre sales have nearly
doubled during this period. Price and technology
changes are reflected in these expense and sales
numbers. In addition, these numbers reflect shifts
to alternative crops. By 1997, 42% of the study’s
counties derived their largest share of income from
a different commodity or animal source than in
1949 (figure 2). Decreases in agricultural net
returns may explain the farmland loss that occurred
in areas where the population decreased. Figure 3
depicts the areas where farmland decreased when
the population decreased and vice versa for one
decade of the study period—between 1987 and
1997.
Several variables are included to represent
demand for land for nonfarm uses: the population
level scaled by the number of acres in the county
(POPPERA), whether the county is in a metropolitan
area (MADUMMY), the percentage change in total
housing units (PCTOTHU), the percentage change in
median family income (PCMFINC), and the percent-
age change in median housing value (PCMHVAL). As
population increases, demand for land in residential
and commercial uses will also increase. Thus popu-
lation growth is hypothesized to increase the rate of
farmland loss. Total population in the six states has
increased by 43% since 1950, climbing from 35
million to 50 million people.
Given that the number of individuals per housing
unit has decreased, we include a direct indicator of
the rate of growth in the housing stock. As the
growth rate of housing units increases, the rate of
farmland loss is expected to increase. The percent-
age change in total housing units averaged 8.09%,
with some counties losing housing units at a rate of
19% while others had a growth rate of 60%. As
family income increases, people may demand larger
homes. Larger homes usually sit on larger parcels.
Therefore, an increase in income is expected to
increase the demand for farmland and accelerate the
farmland loss rate. Similarly, an increase in the med-
ian housing value may indicate an increase in the
demand for land (Hardie, Narayan, and Gardner,
2001) and accelerate the rate of farmland loss.
An increasing proportion of farmers supplement
their farm income with off-farm employment. Only
33% of Mid-Atlantic farmers reported working
over 100 days off the farm in 1949. By 1997, 44%
did so. Their off-farm income opportunities will be
greater if they are better educated and the unemploy-
ment rate in the county is low. However, an increase
in off-farm opportunities will increase the relative
benefit of selling the land and shifting full-time to
alternative employment. Off-farm opportunities are
proxied by both the percentage of the county
population with at least a high school education
(PHIGHSCH) and the percentage of unemployment
(PUNEMP). These opportunities could have either a
positive or negative effect on the rate of farmland
loss. Education attainment increased over the 1949S
1997 time period, with an average of 48.41% of
adult residents having a high school education. The
unemployment rate averaged 5.49%, with a range
across counties of 0.07% to 14.5%. Increases in
median family income might also signal a strong
local economy and the possibility of more off-farm
employment opportunities.
122 April 2003 Agricultural and Resource Economics Review
Figure 2. Counties in the six Mid-Atlantic states that changed crop or livestock commodity
from which they received their largest share of gross income between 1949 and 1997
Figure 3. Changes in farmland and population in the six Mid-Atlantic states
between 1987 and 1997
Lynch and Carpenter Is There a Critical Mass in the Mid-Atlantic Agriculture Sector? 123
Policy variables are included to indicate whether
the county has a preferential property tax program
for agricultural land and/or some type of farmland
preservation program. Preservation and taxation
programs are hypothesized to slow the rate of farm-
land loss. We consider four different program types:
state preferential property tax programs, state pur-
chase of agricultural conservation easement pro-
grams, local purchase of agricultural conservation
easement programs, and local transfer of develop-
ment rights programs. Information was collected on
the existence of these programs by county (Amer-
ican Farmland Trust, 1997, 2001a,b, c). The binary
variable STAX indicates whether the state had estab-
lished a preferential property tax program by t, and
the binary variable PRESPROG indicates if the county
had one or more local- or state-level preservation
programs in place by t. Counties were credited with
having a program if any locality within the county
had a program that had preserved at least one acre.
By 1982, all the states had established preferential
property tax programs. By 1997, 44% of the coun-
ties in the study area had a local or state preservation
program in place.
Results
Results of Model 1
The estimated coefficient on harvested cropland
acres indicates a negative relationship between crop-
land acres and the rate of farmland loss over the full
study period 1949 to 1997 (table 2). Counties with
fewer acres of harvested cropland had higher rates
of farmland loss. The estimated coefficient on the
squared term of harvested cropland is significant
and positive. When a county has 189,240 harvested
cropland acres, the slope of the rate of farmland
loss as a function of harvested cropland acres equals
zero. Thus counties below 189,240 harvested crop-
land acres have a higher rate of farmland loss. The
identified threshold, however, is nearly out of the
data range. Only 2S7 out of 269 counties exceed
189,240 acres of harvested cropland in any time
period. Therefore, the interpretation of this number
as a threshold should be made cautiously.
The estimated coefficients in model 1 also sug-
gest the rate of farmland loss is explained by sales
per acre, expenses per acre, population per acre,
unemployment rate, percentage change in median
family income, and percentage change in housing
units (table 2). The rate of farmland loss decreases
as harvested cropland acres, sales, and percentage
change in income increase. As expenses, popula-
tion, percentage change in total housing units, and
percentage unemployment increase, the rate of farm-
land loss increases. Counties with preferential taxa-
tion programs experienced a lower rate of farmland
loss than counties without a program. All else the
same, metropolitan counties lost farmland at a
higher rate than nonmetropolitan counties.
The predicted rate of farmland loss for the 1949S
1997 period is computed at the average value of the
continuous variables and at zero for the binary vari-
ables using the estimated coefficients. The predicted
rate of farmland loss is 7.9%. We then estimate
how much the predicted rate will change for a 10%
increase in each variable with a statistically signif-
icant parameter estimate. For the binary variables,
the rate of farmland loss is computed if they were
equal to one. Table 3 reports the predicted rate and
the new rate given the 10% increase.
The farmland loss rates increase in counties with
lower harvested cropland acres. The model predict-
ed an average rate of farmland loss of 7.9%. The rate
of farmland loss decreases to 7.67% if the harvested
cropland acres are increased 10% (table 3).
The sales and expenses per acre affect the farm-
land loss rate. A 10% change in sales per acre has
a greater effect than an equal percentage change in
expenses. A 10% increase in sales per acre would
decrease the rate of farmland loss to 7.83% (a
change of !0.07), while a 10% increase in expenses
per acre would increase the rate of farmland loss to
7.92% (a change of 0.02).
Changes in the proxies for development pressure
also impact the rate of farmland loss in model 1. A
10% increase in the population per acre increases
the rate of farmland loss to 8.01%. Similarly, if the
growth of housing stock increases 10%, the rate of
farmland loss increases to 8.02%. Metropolitan
counties lost farmland at a rate of 8.94% compared
to 7.9% for the nonmetropolitan counties.
Higher income growth levels and employment
opportunities decrease the rate of farmland loss. A
10% increase in median family income growth
lowers the rate of farmland loss to 7.73%. If the un-
employment rate increases by 10%, the rate of farm-
land loss increases to 8.09%. Educational attainment
has no impact on the rate of farmland loss.5
5 There was correlation among the three variables PCMFINC, PHIGHSCH,
and PUNEMP. The percentage of adults with high school education and the
percentage change in family income variables have a correlation co-
efficient of !0.52; and the percentage of unemployment and percentage
change in family income variables have a correlation coefficient of !0.32.
This may explain in part the insignificant parameter estimates on
PHIGHSCH in this and the two remaining models.
124 April 2003 Agricultural and Resource Economics Review
Table 2. Results of Models 1, 2a, and 2b for Farmland Loss, Including All Observations Using
Harvested Cropland as the Critical Mass Indicator
Model 1 (1949S1997) Model 2a (1949S1978) Model 2b (1978S1997)
Variable
Coefficient
(Std. Error)
Coefficient
(Std. Error)
Coefficient
(Std. Error)
HCLAND (harvested cropland) !0.00058994
(0.0001)
*** !0.00969508
(0.0002)
*** !0.00004426
(0.0002)
HCLAND2 (harvested cropland squared) 0.0015587
(0.0006)
*** 0.00268124
(0.0007)
*** !0.00015372
(0.0010)
PAGFFM (% resource employment) 0.0415
(0.0334)
0.0477
(0.0381)
0.0500
(0.0972)
SALESPERA (sales per acre) !0.00001
(0.0000)
*** !0.00002
(0.000002)
*** 0.00002
(0.000005)
***
EXPPERA (expenses per acre) 0.000005
(0.0000)
*** 0.00001
(0.000002)
*** !0.00003
(0.00001)
***
POPPERA (population per acre) 0.0187
(0.0017)
*** 0.0175
(0.0019)
*** 0.0148
(0.0053)
***
MADUMMY (metropolitan area) 0.0103
(0.0059)
* 0.0193
(0.0082)
** 0.0078
(0.0088)
PCTOTHU (% change in housing units) 0.1587
(0.0401)
*** 0.1780
(0.0498)
*** 0.0823
(0.0861)
PCMFINC (% change in family income) !0.1321
(0.0540)
** !0.1416
(0.0608)
** !0.0588
(0.1311)
PCMHVAL (% change in housing value) 0.0236
(0.0307)
!0.0025
(0.0448)
0.0462
(0.0651)
PHIGHSCH (% high school education) 0.0141
(0.0318)
0.0251
(0.0462)
!0.0032
(0.0584)
PUNEMP (% unemployment) 0.3207
(0.1255)
* 0.3313
(0.1643)
** 0.2931
(0.2201)
STAX (preferential tax) !0.0404
(0.0105)
*** !0.0358
(0.0112)
***
PRESPROG (preservation program) !0.0047
(0.0095)
!0.0082
(0.0111)
Constant 0.0875
(0.0387)
** 0.1162
(0.0415)
*** 0.0172
(0.0514)
R2
N
0.1647
2,604
0.2344
1,574
0.0623
1,030
Note: Asterisks (*, **, and ***) indicate, based on an asymptotic t-test, the H0: B = 0 is rejected using a 0.10, 0.05, and 0.01 criterion, respectively.
The preferential taxation programs were found to
have a significant effect on the rate of farmland loss.
Counties with preferential taxation programs had a
farmland loss rate of 4.06% compared to counties
without such a program at 7.9%. The presence of
other programs (purchase of development rights,
transfer of development rights, or purchase of agri-
cultural conservation easements) did not impact the
rate of farmland loss.
Results of Models 2a and 2b
Results of model 2a (covering subperiod 1949S
1978) and model 2b (1978S1997) demonstrate that
the effects of the variables changed over time. The
rate of farmland loss slowed about halfway through
the study period. The average five-year rate of farm-
land loss in 1949S1978 was 9.2%, and for 1978S
1997, 5.1%. Both agriculture and the pattern of city
and housing development changed during this time.
Our findings reveal the variables’ impacts were not
consistent over the two time periods. A likelihood-
ratio test indicated that estimating the two models
separately for these time periods is statistically dif-
ferent than pooling the data (χ2
[13] = 77.78). Model
2a’s results were similar to those reported above for
model 1. Different results are found for model 2b.
Estimated coefficients are reported in table 2.
Lynch and Carpenter Is There a Critical Mass in the Mid-Atlantic Agriculture Sector? 125
Table 3. Effects of a 10% Increase in Significant Continuous Variables and Binary Variables
Equaling 1 on Rate of Farmland Loss for Each of the Estimated Models (%)
Description
Model 1
1949S1997
Model 2a
1949S1978
Model 2b
1978S1997
Predicted Probability: 7.90 10.12 5.01
Probability After 10% Increase
Continuous Variables:
HCLAND (harvested cropland) 7.67 9.82
PCMFINC (% change in family income) 7.73 9.99
SALESPERA (sales per acre) 7.83 10.08 5.10
EXPPERA (expenses per acre) 7.92 10.25 4.91
POPPERA (population per acre) 8.01 10.30 5.08
PCTOTHU (% change in housing units) 8.02 10.36
PUNEMP (% unemployment) 8.09 10.37
Binary Variables:
STAX (preferential tax) 4.06 6.62
MADUMMY (metropolitan area) 8.94 12.13
The negative relationship between harvested
cropland and the rate of farmland loss is statistically
significant in the early period model (2a), but is not
observed in the later period model (2b). The critical
mass in model 2a was estimated to be 180,795
harvested cropland acres—similar to the threshold
of 189,240 harvested acres in model 1. However, as
before, few counties actually had more than 180,000
acres of harvested cropland acres. And given the
insignificant coefficient on cropland acres in model
2b, the early period appears to drive the threshold
result of model 1.
As observed from table 3, the predicted rate of
farmland loss in model 2a was 10.12% and in model
2b was 5.01%. In the early model, a 10% increase
in harvested cropland acreage resulted in a lower
farmland loss rate of 9.82% (a change of 0.30),
while in model 2b, the estimated coefficient for har-
vested cropland was not significant.
Similar to the results of model 1, a higher net rev-
enue decreased the rate of farmland loss in model
2a. However, expenses per acre had a bigger impact
than sales per acre. A 10% increase in sales per acre
decreased the rate of farmland loss to 10.08% (a
change of !0.04). If expenses increased by 10%,
the rate of farmland loss increased to 10.25%
(a change of 0.13). In the later period model, sur-
prisingly, the opposite relationship is observed. In
model 2b, a 10% increase in sales per acre increases
the rate of farmland loss from the predicted 5.01%
to 5.10%. A 10% increase in per acre expenses
decreases the rate of farmland loss to 4.91%.
The effects of other variables in model 2a were
similar to those reported above for model 1. The
resulting farmland loss rate due to a 10% increase
in these variables for model 2a can be found in
table 3. In model 2b, except for population per acre,
the other estimated parameters were insignificant.
A 10% increase in population per acre increased the
rate of farmland loss from 5.01% to 5.08%. The
overall explanatory power of model 2a, while not
high, was greater than that of model 2b. For the
1949S1978 period, model 2a yielded an R2 of 0.23;
the corresponding R2 for model 2b for the 1978S
1997 period was only 0.06. Thus, model 2b did not
explain 94% of the variation in farmland loss rates
for these counties between 1978 and 1997.
Discussion
Some evidence of a critical mass existed for the six-
state Mid-Atlantic study area during the early period
(1949S1978). However, the scale of agricultural
activity in the latter part of the study period did not
impact the rate of farmland loss. This finding raises
some interesting questions. First, to what extent
have farmers adapted to the difficulties associated
with shrinking input and output markets by shifting
to alternative crops or alternative marketing mech-
anisms? Further research is needed to determine the
extent to which this has occurred, and to assess the
implications of different changes. For 42% of the
269 counties examined here, the data reveal the agri-
cultural activity receiving the highest gross income
126 April 2003 Agricultural and Resource Economics Review
in 1997 was different from the activity that had gen-
erated the most income in 1949 (figure 2). Yet the
implications of a county-level change from dairy to
vegetables or from row crops to livestock need
further investigation.
Second, had the major technological changes in
agriculture, in terms of improved mechanization
and yield improvements, occurred by the mid-
1970s? Third, how did land development patterns
change, due to either changes in housing consumers’
preferences or changes in policies? How did these
changes impact the rate of farmland loss? Fourth,
how have counties responded to the high rate of
farmland loss between 1949 and the early 1970s?
Counties implemented preferential taxation pro-
grams and agricultural preservation programs,
which we have considered in this analysis.
However, other responses might be equally or more
important.
The impact of sales and expenses per acre
changed for the latter part of the study period. In
the early period, the expected result was con-
firmed—i.e., increased sales or decreased expenses
resulted in a lower rate of farmland loss. However,
from 1978 to 1997, the opposite result was found.
Farmers with the most marginal agricultural land
may have been the first to exit agriculture, leaving
only the most productive land under cultivation.
County average per acre sales therefore would
increase. Also, as mentioned above, farmers could
have switched crops. If farmers shifted to higher
value, smaller acreage crops such as berries or
vegetables, one would observe farmland loss
simultaneously with higher per acre sales. Of
course, this begs the question of why farmers had
not shifted to higher value crops at an earlier time
period.
The health of the local economy was also found
to impact the rate of farmland conversion. Counties
with higher median family incomes and lower
unemployment experienced lower rates of farmland
loss. These findings could be a function of better
off-farm employment opportunities or people with
higher incomes choosing to purchase a farm and
keeping it in production. Farmers in counties with
high unemployment may have had fewer off-farm
opportunities for themselves or their family
members, and may have chosen to sell the farm and
relocate. Policies that focus attention on local or
regional economic performance could promote
farmland retention. Examining farmland prices,
Hardie, Narayan, and Gardner (2001, p. 131)
conclude: “Policies developed for broader purposes
may have as much or more effect on farmland
prices as policies targeted directly at improving agri-
cultural returns.”
Population growth resulted in higher rates of
farmland loss in all three models. In the early
period, the growth rate of the total number of
housing units was also positively related to the rate
of farmland loss. Population growth and housing
development leads to the conversion of agricultural
land. Local communities can exercise control over
the extent and pattern of new development through
thoughtful planning. Given the Chesapeake Bay
Foundation’s (2002) finding that the rate at which
land is being consumed exceeds the population
growth rate by almost 2.5 times, policies could focus
on reducing land consumption per house or per
person to limit the impacts of both population growth
and housing development on agriculture.
Metropolitan counties had higher rates of farm-
land loss over and above losses related to their pop-
ulation and changes in housing stock. Metropolitan
counties may need to be even more active in
implementing policies and programs to encourage
farmland retention and to strengthen the agricultural
economy. Alternatively, states might decide to target
regions far from metropolitan areas for preservation
and retention programs. This approach could retain
agriculture on a statewide basis and allow states to
use their limited resources efficiently.
Preferential property taxation programs were
found to slow the rate of farmland loss. All six
states in this study had enacted such programs by
1982. Additional evaluation of these programs may
be warranted, as well as an examination of who
participates and who does not. A further property
tax reduction might slow the rate of farmland loss
even more. Such a reduction could potentially be
financed through a higher conversion tax rate. The
state or counties could thus recapture some of the
benefits the farmers accrue from the preferential tax
program. This conversion tax could be collected
when landowners choose to convert the land from
agriculture to a nonfarm use.
Other agricultural preservation programs (pur-
chase of development rights, transfer of develop-
ment rights, or purchase of agricultural conserva-
tion easements) did not impact the rate of farmland
loss. Few of these programs existed before the
1980s. Some of them have not had the resources to
preserve many acres. Farmland preservation pro-
grams may have more impact in the future if they
have greater resources and can enroll a higher
number of acres.
Lynch and Carpenter Is There a Critical Mass in the Mid-Atlantic Agriculture Sector? 127
Conclusions
The results do not provide clear evidence of a
critical mass in agricultural economies in the Mid-
Atlantic region. The results suggest that counties
with fewer farmland acres lost farmland at higher
rates. An acreage threshold level was computed.
Yet, these calculated critical mass levels (189,240
acres per county in model 1, and 180,795 per county
in model 2a) exceed the harvested cropland acres of
most counties (97%) in the Mid-Atlantic region. In
addition, a critical mass of agricultural acreage was
not found in the analysis of the 1978 to 1997 period.
Counties with fewer harvested cropland acres did
not have a higher rate of farmland loss in the later
period. Thus, even if the computed levels were
convincingly strong, the critical mass may have
altered in the latter part of the study period given
the many changes over the time period.
Additional research is needed to more fully
identify factors affecting the rate of farmland loss
and whether a critical mass exists. Farmers may
have adapted to a more limited support sector in
their regions by shifting to alternative crops or
products that are less reliant on nearby suppliers or
buyers. Specific sectors or commodities might be
doomed once an area loses a certain number of
acres, i.e., the farmers cannot continue to produce
these commodities profitably. However, adaptation
could ensure the viability of the farm sector as a
whole. A model incorporating all U.S. agricultural
economies might further demonstrate whether and
how much the level of a critical mass depends on
cropping patterns and geography. A more micro-
level analysis could reveal information obscured by
the county-level analysis. A case study approach
could provide insights about specific industries or
agricultural sectors in specific regions during spe-
cific time periods. Or further analysis could consider
the individual farming sectors in different areas and
how they have evolved over time.
While this analysis does not provide a clear prog-
nosis for the economic viability of the Mid-Atlantic’s
agricultural sector, the analysis suggests that the farm
community has been resilient to large losses of farm-
land over time, that the health of the local economy
in the county matters, and that controlling population
growth and housing development is very important
to slowing farmland loss. It also suggests that the
recent emphasis on preserving a critical mass of agri-
cultural land may be insufficient to ensure the long-
term viability of an agricultural sector. Decision
makers need to examine other policy objectives to
sustain a viable agricultural sector.
References
American Farmland Trust. (1997). Saving American Farmland:
What Works. Northampton, MA: American Farmland Trust
Publications Division.
———. (2001a). “Status of Selected Local PACE Programs.”
Fact Sheet. American Farmland Trust Technical Assistance,
Northampton, MA. Online website: http://www.farmland-
info.org.
———. (2001b). “Transfer of Development Rights.” Fact
Sheet. American Farmland Trust Technical Assistance,
Northampton, MA. Online website: http://www.farmland-
info.org.
———. (2001c). “Status of State PACE Programs.” Fact
Sheet. American Farmland Trust Technical Assistance,
Northampton, MA. Online website: http://www.farmland-
info.org.
Bell, K., and N. Bockstael. (2000). “Applying the Generalized-
Moments Estimation Approach to Spatial Problems Involv-
ing Micro-Level Data.” Review of Economics and Statistics
82(1), 72S82.
Bockstael, N. E. (1996, December). “Modeling Economics and
Ecology: The Importance of a Spatial Perspective.” Amer-
ican Journal of Agricultural Economics 78, 1168S1180.
Chesapeake Bay Foundation. (2000). “Land and the Chesa-
peake Bay.” Annapolis, MD.
———. (2002). “Future Growth in the Washington, DC,
Metropolitan Area.” Annapolis, MD. Online. Available at
http://www.savethebay.org/land/landuse/maps/future_
growth.html.
Cheshire, P., and S. Sheppard. (1995, May). “On the Prices of
Land and the Value of Amenities.” Economica 62, 247S267.
Chicoine, D. L. (1981, August). “Farmland Values at the Urban
Fringe: An Analysis of Sale Prices.” Land Economics 57,
353S362.
Clonts, H. A., Jr. (1970, November). “Influence of Urbanization
on Land Values at the Urban Periphery.” Land Economics
46, 489S497.
Daniels, T., and M. Lapping. (2001, November). “Farmland
Preservation in America and the Issue of Critical Mass.”
Paper presented at the American Farmland Trust National
Conference, November 13, 2001, Chicago, IL.
Dhillon, P. S., and D. A. Derr. (1974). “Critical Mass of Agri-
culture and the Maintenance of Productive Open Space.”
Journal of the Northeastern Agricultural Economics Council
3(1), 23S34.
Dunford, R. W. (1983). “Further Evidence on the Conversion
of U.S. Farmland to Urban or Transportation Uses.” Con-
gressional Research Service, Washington, DC.
Dunford, R. W., C. E. Marti, and R. C. Mittelhammer. (1985,
February). “A Case Study of Rural Land Values at the
Urban Fringe Including Subjective Buyer Expectations.”
Land Economics 61, 10S16.
Fischel, W. A. (1982). “The Urbanization of Agricultural Land:
A Review of the National Agricultural Lands Study.” Land
Economics 58, 236S259.
Gardner, B. L. (1994). “Commercial Agriculture in Metropol-
itan Areas: Economics and Regulatory Issues.” Agricultural
and Resource Economics Review 23(1), 100S109.
———. (2002). American Agriculture in the 20th Century:
How It Flourished and What It Cost. Cambridge, MA: Har-
vard University Press.
128 April 2003 Agricultural and Resource Economics Review
Geoghegan, J. (2002). “The Value of Open Spaces in Resi-
dential Land Use.” Land Use Policy 19(1), 91S98.
Geoghegan, J., L. Lynch, and S. Bucholtz. (2003, April).
“Capitalization of Open Spaces into Housing Values and the
Residential Property Tax Revenue Impacts of Agricultural
Easement Programs.” Agricultural and Resource Economics
Review 32(1), 33S45 [this journal issue].
Geoghegan, J., L. Wainger, and N. E. Bockstael. (1997). “Spa-
tial Landscape Indices in a Hedonic Framework: An Ecolog-
ical Economics Analysis Using GIS.” Ecological Economics
23(3), 251S264.
Greene, W. (1995). LIMDEP, Version 7.0: A User’s Guide.
Plainview, NY: Econometric Software, Inc.
Hammer, T. R., R. E. Coughlin, and E. T. Horn, IV. (1974,
July). “The Effect of a Large Urban Park on Real Estate Val-
ues.” American Institute of Planners Journal, pp. 274S277.
Hardie, I. W., T. A. Narayan, and B. L. Gardner. (2001). “The
Joint Influence of Agricultural and Nonfarm Factors on Real
Estate Values: An Application to the Mid-Atlantic Region.”
American Journal of Agricultural Economics 83(1),
120S132.
Irwin, E. G., and N. E. Bockstael. (2001). “The Problem of
Identifying Land Use Spillovers: Measuring the Effects of
Open Space on Residential Property Values.” American
Journal of Agricultural Economics 83(3), 698S704.
Kitchen, J. W., and W. S. Hendon. (1967, August). “Land
Values Adjacent to An Urban Neighborhood Park.” Land
Economics 43, 357S361.
Lockeretz, W. (1989). “Secondary Effects on Midwestern Agri-
culture of Metropolitan Development and Decreases in
Farmland.” Land Economics 65(3), 205S216.
McMillan, M. (1974, November). “Open Space Preservation in
Developing Areas: An Alternative Policy.” Land Economics
50, 410S417.
Muth, R. F. (1961). “Economic Change and Rural-Urban Land
Conversion.” Econometrica 29(1), 1S23.
Nickerson, C. J., and L. Lynch. (2001). “The Effect of Farm-
land Preservation Programs on Farmland Prices.” American
Journal of Agricultural Economics 83(2), 341S351.
Peiser, R. B., and G. M. Schwann. (1993, Summer). “The
Private Value of Public Open Space Within Subdivisions.”
Journal of Architectural and Planning Research 10,
91S104.
Shi, Y. J., T. T. Phipps, and D. Colyer. (1997). “Agricultural
Land Values Under Urbanizing Influences.” Land Eco-
nomics 73(1), 90S100.
U.S. Department of Agriculture, National Agricultural Statistics
Service. (1999). 1997 Census of Agriculture, 1A, 1B, 1C
(CD-ROM set). USDA/NASS, Washington, DC.
———. (2001). Agricultural Statistics. USDA/NASS, Wash-
ington, DC.
U.S. Department of Commerce, Bureau of the Census. Census
of Agriculture. Various years, 1950 to 1982. USDC, Wash-
ington, DC.
———. Census of Population and Housing. Various years,
1950 to 1990. USDC, Washington, DC.
Weicher, J. C., and R. H. Zerbst. (1973, February). “The Exter-
nalities of Neighborhood Parks: An Empirical Investiga-
tion.” Land Economics 49, 99S105.
