Agriculture Reference
In-Depth Information
C
i
where
is an unobserved farmland contract response variable;
C
i
is the observed dichoto-
mous choice of land contract for plot
i
, which is equal to 1 for cropshare contracts and equal
X
i
to 0 for cash rent contracts;
is a row vector of exogenous variables including the constant;
β
i
i
is a plot-specific error term. We use
a logit model to generate maximum likelihood estimates of the model given by equations
(7.8) and (7.9) for a various contract samples.
In general, the model predicts that cash rent contracts should be more common and output
shares should be higher in the second period. For this estimation we also use the same
variables (ACRES, AGE, FAMILY, HAY, INPUTS CHANGED, IRRIGATED, DENSITY,
ROW CROP, and YEARS DURATION) in our previous estimates to control for farm size,
age, whether or not contracts are between family members, whether the number of inputs
changed, the presence of irrigation, local population density, the type of crop, and the
number of years the landowner and farmer have contracted with each other.
In particular, including AGE allows us to control for the effects of farming experience
and assures that our NEW FARMER variable is representing farmers who are new to the
contracted plot of land. We test these predictions using the entire contract sample, as well
as several subsamples.
is a column vector of unknown coefficients; and
Cropshare versus Cash Rent
To test predictions 7.1 and 7.2 we use a sample that contains both cropshare and cash rent
contracts.
19
Table 7.3 presents the coefficient estimates for several logit regressions on the
choice of contract. The dependent variable in all cases is one if the contract is cropshare and
zero if the contract is cash rent. The first column presents the results using the entire contract
sample. The other estimates come from three additional subsamples: contracts for which
the farmer and landowner are unrelated (column 2), only oral contracts (column 3), and
annual or short-term contracts (column 4). Each of these subsamples allows us to control
for various factors that might influence the magnitude of the potential ratchet effects. In
each case the ratchet effect is expected to be larger than the full sample case. For example,
it might be expected that the ratchet effect is stronger with unrelated individuals because
more information is known about family members to begin with and ratcheting up incentives
on relatives may be frowned on within a family. On the other hand, oral contracts involve
less commitment, and as a result, ratchet effects should be more common. Likewise, shorter
annual contracts may imply less commitment and a higher likelihood of ratchet effects.
We directly test for the presence of ratchet effects by including the variables NEW
FARMER and NEW LANDOWNER in all the logit regressions. Prediction 7.1 implies
a negative coefficient for the NEW FARMER variable, but the estimated coefficients for
NEW FARMER are all positive and none are statistically significant, so we cannot reject
