A Translog Cost Function Analysis of U.8.

Agriculture, 1939-77
Subhash C. Ray

The translog cost function provides a convenient framework for analyzing U.S.
agricultural production in a multioutput context. Treating crops and livestock as two
distinct outputs, this study utilizes standard results of neoclassical duality theory to
obtain measures of pairwise elasticities of substitution between inputs, price elasticities

Downloaded from http://ajae.oxfordjournals.org/ at University of Ulster on February 16, 2015

of factor demands, and the rate of Hicks-neutral technical change. Results obtained from
joint GLS estimation of parameters of cost and share equations indicate a declining trend
in the degree of substitutability between capital and labor. Price elasticity of demand for
all inputs increased over time. The measured rate of technical change was 1.8% per year.

Key words: elasticities of substitution, Hicks-neutral technical change, neoclassical

duality, translog.

m
Over the past four decades major changes
have occurred in U.S. agriculture. A new era
(1) In C = In k + L a, In qi
i=l
in agricultural production in the United States
began in 1939. World War II gave a boost to m m n
agriculture still bogged in the depression. Pro- + ILL di j In qi • In qj + L b; In Wr
i=lj=l r=l
ductivity increases over the period 1939-77
have been quite impressive. Measured crudely n n
by the ratio of the indices of output and input, + ILL irs In Wr • In W s
the productivity index (1967 = 1(0) increased r=ls=1
from 59 in 1939 to 119 in 1977. To study this m n
phenomenon, a multi-input translog cost func-
tion incorporating Hicks-neutral technical
+ LL gir In a. . In W r + hT.
i=1 r=1
change (HNTC) was specified. Annual time-
series data for the United States was used to Here C is cost, qi is output of product i, w, is
estimate the model. Neoclassical duality the- the price of input r, and T is an annual index of
ory provides an approach for computing the time. This translog (dual) cost function can be
pairwise elasticities of substitution between regarded as a quadratic approximation to the
inputs for each year as well as the annual unspecified "true" cost function. In this con-
(own- and cross-) price elasticities of demand text, di j = dji and j., = Isr for all i.j,s, and r.
for inputs. Any sensible cost function must be
homogenous of degree 1 in input prices. In the
translog function (1), this requires that
The Translog Cost Function for Multiple Out- n n
puts L b; = 1, L Isr = 0 (r = 1,2, ... , n) and
r=1 s=1
m
The m-output-n-input translog cost function Lgir =0 (r = 1,2, ... n). Essentially, the
incorporating Hicks-neutral technical change i=1
(HNTC) is same set of restrictions on parameters follows
from the "adding up" requirement of the fac-
Subhash C. Ray is a visiting lecturer, Department of Economics, tor shares.
University of California, Santa Barbara. The translog form is flexible because
The author is grateful to an anonymous referee for valuable specific features of technology (like returns to
comments on an earlier version of this paper and is indebted to
Thomas Cooley for suggestions for improvement. All remaining scale or homotheticity) may be tested by ex-
errors are the author's responsibility. amining the estimated model parameters.

Copyright 1982 American Agricultural Economics Association

Ray Translog Cost Analysis of u.s. Agriculture 491

Specifically, if the technology is homothetic, estimated is large. For an m-output, n-in-

the dual cost function is multiplicatively put model with matrices (dij) and (frs) sym-
separable in output quantities and input metrical, one needs to estimate Hm + n)
prices. The cost function C = C(q,w) is of the (3 + m + n) parameters. This does not include
form h(q) . t(w), where q and ware the vectors the intercept and the rate of Hicks-neutral
of output quantities and input prices. In the technical progress. For example, in this case
translog case, this requires that gir = O(for all i with two outputs and five inputs, we must
and r) so that the quadratic interaction terms estimate thirty-seven parameters. In general,
between output levels and input prices should it is difficult to obtain a sample large enough to
disappear. estimate the full cost function. Moveover,
Cobb-Douglas, or some other exact spec- with time-series observations needed on so
ification, implies constant elasticities of sub- many variables, we are likely to encounter
stitution between inputs and homotheticity of severe multicollinearity. Thus, estimating the
technology. Generalized cost functions like cost function as a single-equation model even

Downloaded from http://ajae.oxfordjournals.org/ at University of Ulster on February 16, 2015

the translog allow us to test for specific with restrictions imposed for linear homoge-
characteristics of technology. They are not re- neity in the input prices may be either impos-
quired as prior assumptions. This largely ex- sible or inappropriate.
plains the increasing popularity of generalized Estimating the full dual system (i.e., cost
cost (and production) functions, of which the and share equations together) leads to much
translog is an example. Berndt and Christen- higher efficiency. Such joint estimation can
sen; Berndt and Wood; Christensen, Jorgen- compensate for the information inadequacy in
son, Lau; and Binswinger used translog mod- the equation (1) alone. From the translog cost
els. Other forms are analyzed in Diewert, function and Shephard's lemma, we derive the
Hanoch, and Phillips. Multi-output translog input share equations:
applications are found in Darrough and
Heineke, Fuss and Waverman, and Cooley (2) s, = bi + hlln WI + ... + hn ln W n
and Bothwell. Hasenkamp estimated the multi- + gil ln qi + ... + gimln qm'
output production technology of U.S. rail- where (i = 1, ... ,n), s, = WiX)C, and x, is the
roads using several functional forms. quantity of the ith input. These shares must
In view of the widespread use of the trans- add up to 1. This is true for all prices and
n n
log specification in recent years, a note of
warning is in order. As a quadratic approxima-
outputs. It requires I b, = 1, L Isr = 0,
m pI
tion to the unspecified "true" cost function, and L. gir
~l

= 0 (for r = 1,2, ... ,n). Note that

the translog has some limitations. In a recent i=1
Monte Carlo study by Guilkey and Lovell, a these are the same conditions implied by linear
translog approximation to a CES production homogeneity of the cost function in input
function performed rather poorly in estimating prices. If we additionally assume marginal
the elasticity of substitution between inputs, cost pricing for the outputs, we obtain the
when the true parameter was as high as 3.0. relations
Therefore, the results reported here and else- Yk = aln Cjaln qk = (aCjaqk) .
where should be used with caution. It is (qkj C) = Pkqk!C
difficult to define returns to scale in the mul-
tioutput case. However, in the special case of for each output k. This leads to the "revenue
linear homogeneity of the cost function in the share" equations, as follows:
output levels, we may assert constant returns
to scale (CRS). The conditions for CRS in
(3) Y« = ak + dkiin qi + + dkmln qm
m m + gkl In WI + + gkn ln W n,
equation (1) are I a, = 1, I dij = 0, and where (k = 1, ... , m).
n , i=1 i=1
Lgir
r=1
=
0 for allj and r. (See Darrough and While few people will disagree with the
cost-share equations, the revenue-share equa-
Heineke or Berndt and Christenson for restric- tions in (3) may not be readily accepted. Some
tions implying homotheticity, CRS, and other might argue that agricultural products are in-
features of the technology.) elastically supplied and government support
One problem with multi-output, multi-input programs tend to repudiate the marginal cost-
cost functions is that, even for a few outputs pricing assumption. This could be a valid ob-
and inputs, the number of parameters to be jection, but the way out dependent variables
492 August 1982 Amer. J. Agr. Econ.

are measured in equation (3) will reduce or price

eliminate this problem. Instead of using price

indices of agricultural products to obtain
PkqklC, we directly measured the numerator
(Pkqk) as market sales revenue plus govern-
ment payments. It is assumed that government
payments are designed to help maintain farm
income by equating effective average revenue
to marginal cost. This is illustrated by figure 1.
Assume that farmers expect a price equal to
BD and produce output equal to OD. At this
level price would equal marginal costs. How-
ever, from the demand curve, the market- d

clearing price for this supply is only CD. We

Downloaded from http://ajae.oxfordjournals.org/ at University of Ulster on February 16, 2015

assume in our model that the government sub-
o D quantity
sidy, however paid, equals BC. The effective
average revenue equals marginal cost at the Figure 1. Dlustration of price support pro-
actual production level. gram
Note that the revenue shares (Yk = PkqklC),
unlike the cost shares (s, = XiWi/C), need not
add up to 1. Even under competitive condi- generalized least squares procedure. Although
tions, total revenue may exceed or fall short of anyone parameter appearing in several equa-
costs in the short run. Moreover, economic tions has the same estimated value, the asso-
profits should be calculated after the opportu- ciated asymptotic t statistics may differ. We
nity cost of entrepreneurial efforts and im- found that the t-values were higher in the
puted wages for family labor are considered. share equations than in the cost equation it-
Here we assume that changes in these two self. In all such cases, we accepted the higher
inputs (family labor and entrepreneurship) are value.
not related to output changes. This assump- A Taylor's series approximation to any
tion permits us to calculate marginal costs of function is defined relative to a specific point
the outputs from the translog cost function. of approximation. In this case better results
Hence, we are really approximating a variable were obtained using 1967 values for scaling
cost function rather than a total cost function. (rather than the sample means). By using the
Fuss and Waverman in their study of Cana- 1967values, we approximate the cost function
dian telecommunications used the revenue at that specific point. All time-series data were
shares only for products whose price elasticity converted into index numbers with base year
of demand was absolutely greater than 1. For 1967 before estimation.
this analysis, we found including the revenue-
share equations led to more acceptable pa-
rameter estimates, including correct signs of Variable Definitions and Data Description
the price elasticities of input demands com-
puted from the estimated model. Since con- In this model, we have two outputs: qt is live-
cavity of the cost function at all data points stock output and q2 is crop output. Both are
requires that these elasticities be uniformly measurbd by index numbers of production
negative, we retained the revenue shares in (1967 = 100). Five inputs are included: hired
our model. labor; farm capital, consisting of farm real es-
Some additional econometric consid- tate, motor vehicles, and machinery; fertiliz-
erations are the following: (a) Because cost ers and lime; (purchased) feed, seed, and live-
shares must add to 1, one of the share equa- stock; and miscellaneous inputs (pesticides,
tions from (2) is redundant. (b) The individual ginning, electricity, telephone).
equations of the system are seemingly unre- In the dual system, input prices are used
lated in the sense of Zellner. (c) Restrictions rather than physical quantities. They are wage
must be imposed across equations to ensure of hired labor (w t ) ; user cost of farm capital
uniqueness of estimated parameters which (W2); fertilizer price (w a ); a divisia index of
occur in more than one equation. feed, seed, and livestock prices (W4); miscel-
The system was estimated using the joint laneous input prices measured by the price
Ray Translog Cost Analysis of u.s. Agriculture 493

index of all commodities purchased for farm try to test for such linear homogeneity in the
production (w s ) ' The trend variable, T, in- cost function taking the equation (1) sepa-
cluded as an explanatory variable to capture rately as a single-equation model. This, al-
HNTC starts with T = 39 for 1939 and in- though theoretically possible, proved to be
creases by 1 annually. impractical in our case due to severe mul-
The dependent variables are as follows: ticollinearity.
SI is (total wages paid to hired labor) A test for profit-maximizing behavior was
(farm production expenses); illustrated by Appelbaum. He imposed linear
S2' (repairs and operation expenses on capi- homogeneity on the cost-function parameters,
tal items + depreciation and other consump- adding up restrictions on share-equation pa-
tion of farm capital + taxes on farm property rameters separately and then tested for
+ interest on farm mortgages + net rent paid uniqueness of parameters across equations. It
to nonfarm landlords) -:- (farm production ex- was assumed that if firms really maximized
penses); profits (and were efficient in the neoclassical

Downloaded from http://ajae.oxfordjournals.org/ at University of Ulster on February 16, 2015

S3 , (expenses on fertilizers and lime) -:- (farm sense) parameters appearing in cost and share
production expenses); equations simultaneously would have unique
S4' (expenses on feed, seed, and livestock estimated values. This test of internal consis-
purchased) -:- (farm production expenses); tency could not be carried out in our case
ss, (miscellaneous expenses) -;- (farm pro- because the problem of singularity for the
duction expenses); cost-function estimation remained even after
Yl' (livestock marketing revenue) -:- (farm the parameter restrictions were imposed on
production expenses); equation (1). Equations (2) and (3) provide the
Y2' (crop marketing revenue + government additional information necessary to resolve
payments + inventory adjustments) -:- (farm this multicollinearity problem. It should be
production expenses); and theoretically possible to obtain estimates of
C, index of farm production expenses (1967 the parameters of the cost function (1) except
= 100). for the intercept and the time trend factor from
The data have been taken from various U.S. the share equations (2) and (3). However, our
government publications cited in references attempt to so obtain these parameters failed to
(1957, 1972, 1976 and 1978).1 yield acceptable estimates. In particular when
equation (1) is excluded and parameters are
estimated from (2) and (3) only, the price elas-
Empirical Results ticities of input demand computed from such
parameters are not of the right sign at every
Alternative versions of the model were esti- data point. We need to include all the
mated to test for constant returns to scale and equations-(1), (2), and (3)-to estimate the
homotheticity. The final choice of model was parameters of the cost function.
based on various F tests. (Note that the stan- The model was first estimated to test for
dard F test proposed by Fisher is to be mod- constant returns in the special sense of con-
ified in the context of a j oint generalized least stant average variable costs. Linear homoge-
squares estimation. See Theil for an exact de- neity of the cost function in factor prices and
scription of the test). In view of the adding up internal consistency were assumed. The com-
requirement of the input shares, one equation puted F-value (with 8 and 224 degrees of free-
(for miscellaneous inputs) was excluded from dom) was 62.025. This leads to a strong rejec-
the estimated system. tion of CRS. Next we tested for separability of
Ideally, we should first test for linear homo- the cost function in outputs and input prices.
geneity in the cost function with input prices. Such functional separability indicates homo-
This, however, cannot be tested in our com- theticity of the technology. The computed
plete model [equations (1), (2), and (3) all F-value (with 10 and 244 degrees of
included]. The reason is that the adding up freedom) was 4.98 and significant. Conse-
restrictions on the input-share equations (2) quently, we rejected homotheticity as well.
imply linear homogeneity of the cost function The model finally retained was the profit
in the input prices. As an alternative one might maximizing, linear homogenous version (in
input prices). The estimated parameters of the
1 For detailed references to tables from which variables were
taken and an explanation of the construction of farm capital price system and the associated asymptotic
index, see Ray, pp 10-11. t-values are reported in table 1.
494 August 1982 Amer. J. Agr. Econ.

Table 1. Estimated Coefficients of the Trans- sical duality theory to compute these elas-
log Cost Function ticities of substitution from the data using es-
timated parameters. Uzawa showed that in a
(Asymptotic) profit-maximizing, competitive model, the
Parameter Value t Ratio
Allen partial elasticities of substitution be-
k (intercept) -7.150465 -1.1201 tween inputs i andj can be derived from the
r (trend) -0.018004 -9.9244 (dual) cost function as
al 1.152584 3.0331
a2 1.170444 2.2545 Sij = (iJ 2 c / aw/JWj . C)/(aC/aWi . ac/aWj).
d ll 0.038830 0.5493
d 12 -0.139765 -1.9625 In the translog model, we obtain
d 22 0.022331 0.1869
b1 0.595642 4.3981 (4) Sij = (lij + Si . Sj)/(Si . Sj).
b2 0.780157 4.0054
b3 0.104822 1.6336 In equation (4) the hats indicate estimated

Downloaded from http://ajae.oxfordjournals.org/ at University of Ulster on February 16, 2015

b4 -0.679995 -4.0591 values. When i = j, we obtain
b5 0.199374 0.5233
III 0.004591 0.4106 (5) Sii = rAi + Si(Si - 1)]/Si2 •
112 -0.011978 -1.3456
113 0.029050 6.5084 The pairwise elasticities of factor substitution
114 0.015430 1.2087 computed from our model and the data set are
115 -0.037093 -4.6166 reported in table 2.
122 0.021669 1.4739 It has been proved (see, for example,
123 0.004641 1.4011
Binswanger) that the (own-) price elasticities
124 -0.054936 -4.9146
125 0.040604 7.1280 of demand for individual inputs can be ob-
133 0.029263 6.4161 tained from equation (5) as
134 -0.010268 -1.4728
las -0.052687 -5.8779 (6)
144 0.101900 4.7802
145 -0.052126 -3.4954 Similarly, the cross-partial elasticities of de-
155 0.101302 1.1485 mand for inputs are
gll -0.079501 -3.6859
g12 -0.067489 -2.9179 (7) eij = (axtlaw j) . wf x, = Sij . Sj
g13 -0.023726 -2.3451
g14 0.204859 7.7450 for pairs of inputs i and j. These cross elas-
g15 -0.034142 -2.0256 ticities of demand are not symmetrical as are
g21 -0.028521 -1.1805 the substitution elasticities. Price elasticities
g22 -0.012118 -0.3740
0.014151 1.1624 of input demands are reported in tables 3 and
g23
g24 0.009516 0.2784 4.
g25 0.016971 0.8961 The estimated parameters reported in table
1 may be used to compute partial and overall
A primary objective was to measure the scale economies at each data point. These
elasticities of substitution between pairs of in- scale economies measure the relative changes
puts during the four decades covered by the in outputs when expenses change but input
data. We used standard results from neoclas- prices are held constant. These are reciprocals

Table 2. Elasticities of Substitution between Pairs of Inputs (Selected Years)

Year S12 S13 S14 S15 S23 S24 S25 S34 S35 S45

1939 0.8546 4.6947 1.5196 -0.4423 1.2192 0.3137 1.5857 -0.2677 -6.5114 -0.9661
1949 0.8035 5.2288 1.4198 -0.9056 1.2334 0.4834 1.7205 0.1438 -7.2964 -0.5339
1959 0.7542 5.5655 1.4974 -1.1354 1.2057 0.5006 1.6592 0.2851 -5.5520 -0.3295
1969 0.6722 6.2039 1.6595 -1.4686 1.1776 0.4984 1.5773 0.3898 -3.5189 -0.1885
1977 0.6196 6.9961 1.8846 -1.5140 1.1667 0.4522 1.4787 0.3344 -3.0374 -0.1094
Sample
Mean 0.7482 5.7250 1.5272 -1.1473 1.2117 0.4865 1.6607 0.2249 -5.9461 -1.0278
Note: $12 is elasticity of substitution between labor and capital; $13 is elasticity of substitution between labor and fertilizers; S14 is elasticity
of substitution between labor and feed, seed, and livestock; SUI is elasticity of substitution between labor and miscellaneous inputs; S23 is
elasticity of substitution between capital and fertilizers; $24 is elasticity of substitution between capital and feed, seed, and livestock; Sn is
elasticity of substitution between capital and miscellaneous inputs; $34 is elasticity of substitution between fertilizers and feed, seed, and
livestock; $35 is elasticity of substitution between fertilizers and feed, seed, and livestock; $35 is elasticity of substitution between
fertilizers and miscellaneous inputs; $45 is elasticity of substitution between feed, seed, and livestock and miscellaneous inputs.
Ray Translog Cost Analysis of u.s. Agriculture 495

Table 3. Own-Price Elasticities of Demand for Inputs (Selected Years)

Feed, Seed,
Year Labor Capital Fertilizers & Livestock Miscellaneou s

1939 -0.8032 -0.4952 -0.3216 -0.2426 -0.1799

1949 -0.8202 -0.5283 -0.3345 -0.3444 -0.1107
1959 -0.8436 -0.5322 -0.4065 -0.3503 -0.1681
1969 -0.8597 -0.5340 -0.4737 -0.3504 -0.2343
1977 -0.8640 -0.5236 -0.4875 -0.3320 -0.2925

Table 4. Cross-Partial Elasticities of Demand for Inputs (Mean Values)

Price of

Downloaded from http://ajae.oxfordjournals.org/ at University of Ulster on February 16, 2015

Feed, Seed,
Demand for Labor Capital Fertilizers & Livestock Miscellaneous

Labor -0.8389 0.3134 0.3084 0.3928 -0.1757

Capital 0.0922 -0.5297 0.0646 0.1261 0.2469
Fertilizers 0.6751 0.5071 -0.1275 0.4534 -0.8553
Feed, Seed,
& Livestock 0.1813 0.2032 0.0937 -0.3432 -0.2233
Miscellaneous -0.1316 0.6984 -0.3050 -0.3905 -0.1608

of the cost elasticities with respect to output. be used to obtain the marginal costs of each
If only one output (qi) is varied, while other output, using the relation
outputs and input prices are held constant, the (8) Me.• = (I/SCE.) . (C/qi) (. 1 2)
partial scale economy may be measured as . . I = , .
Next, consider nonjointness in the outputs.
In the absence of jointness in production, the
This is the inverse of the revenue share of marginal cost of one output is independent of
output i defined in equation (3). The overall the production of the other. Stated rigorously
scale economy is obtained as SCE = in the translog model, the condition for non-
[I i (1/ S CE i )]- I . It may be shown that for jointness is d l 2 = - a la2' This is a nonlinear
any given proportionate increase in each out- restriction and not directly tested in this linear
put simultaneously, the computed value of model. However, from the estimated param-
SCE is greater than 1, when total costs in- eters, dl 2 = -0.139765 whereas al • a2 =
crease by a lower proportion than the outputs. 1.349035. This is not a conclusive test, but it
Similarly, SCE is less than 1 when costs in- does suggest that non-jointness of the two
crease by a greater proportion than outputs. outputs does not hold.)"
The overall scale economy may thus be inter- Finally, consider Hicks-neutral technical
preted as measuring the returns to scale in a change. The rate of technical change in U.S.
multiproduct situation. In this particular agriculture captured by the HNTC coefficient
study, the scale economies relate to variable on the time variable measured up to 1.8% per
costs only. The partial and overall scale econ- year. It is statistically very significant, and
omies are reported for selected years in table
5. The computed values of SCEi (i = 1,2) may 2 Noniointness of the outputs would require that the marginal
cost of one should be independent of the level of production of the
Table S. Partial and Overall Scale Economies other, i.e.,
(Selected Years) iJ(MCj)/iJqJ = tflC/iJqjiJqJ = 0
(for all i andj). It may be shown that when the above condition
Year SCE! (Livestock) SC& (Crop) SCE (Overall) holds, we obtain the relation,
1939 1.2003 1.3517 0.6357 iJ2lnC/ iJlnql . iJlnq2 = -(iJInC/iJlnql) . (iJlnC/iJlnq2)'
1949 1.2050 1.4278 0.6534
1959 1.3273 1.5237 0.7094 This is a necessary condition for nonjointness and must hold for all
levels ofq and w. If we setqj = W r = 1 for alli andr, the condition
1969 1.5205 1.6313 0.7870 reduces to d l2 = -ala 2. Note that this is a necessary but not a
1977 1.7044 1.6340 0.8342 sufficient condition for nonjointness.
496 August 1982 Amer. J. Agr. Econ.

may be compared with other estimates of pro- (e) The overall scale economy is less than 1,
ductivity increase. Schultz found that during implying that the average variable costs rise
the period 1910-50 (at 1946-48 prices) the when both outputs increase.
index of productivity measured by the ratio of (j) Nonhomotheticity of technology (evi-
output to input indexes in agriculture in- denced by the F-tests) indicate that the cost
creased at a 1.35% annual rate. For the subpe- function is not multiplicatively separable in
riod 1924-50, the annual growth rate was 2%. outputs and input prices. This violates Sol-
Such simple measures incorporate both tech- ow's condition for the existence of an aggre-
nological changes and factor substitution due gate production function for U.S. agriculture.
to price changes. We also computed a similar We may compare these findings with those
exponential trend using data from U. S. De- of Binswanger, who also used a translog ap-
partment of Agriculture (USDA) Statistical proximation for the cost function for U.S. ag-
Bulletin No. 612, obtaining an annual HNTC riculture. There are some major differences
rate of 1.77%. The measure of technological between his model and ours. He estimated a

Downloaded from http://ajae.oxfordjournals.org/ at University of Ulster on February 16, 2015

change obtained from our translog model is single output cost function using pooled
net of factor substitution and, therefore, cross-section and time-series data for forty-
should be more precise. Still, all these mea- eight states for the years 1949; 1954, 1959, and
sures of productivity change are of compara- 1964. Moreover, he estimated only the cost-
ble size. Recall that technological change neu- share equations and inferred the cost function
trality was assumed specifically in order to parameters from them. Further, he assumed
obtain a single measure of productivity biased (rather than neutral) technical change
growth. Hence, the possibility that technical and specified his share equations accordingly.
change in U.S. agriculture was biased in favor Finally, his definition of inputs is somewhat
of specific inputs was ruled out rather than different than here. For instance, he treated
tested in this analysis. land and machinery as separate inputs, while
From the results of this work the following in our model they are pooled into farm capital.
broad conclusions may be drawn about U.S. Binswanger found significant substitutability
agriculture: between labor and land as well as between
(a) Farm capital; fertilizers; and feed, seed, labor and machinery. His average elasticities
and livestock all substitute for hired labor in of substitution were 0.204 between land and
varying measures. However, the degree of labor and 0.851 between labor and machinery.
substitution between labor and capital is much Our average elasticity of substitutionbetween
smaller than between labor and fertilizers. The labor and capital was 0.7482. However,
high degree of substitutability between labor Binswanger found complementarity between
and fertilizers is consistent with the steady labor and fertilizers as opposed to substitution
decline in labor use and the steep increase in in this case. Even recognizing the fact that his
fertilizers use. data are for an earlier subperiod, this is some-
(b) Farm capital is a substitute for all other what surprising. The strong substitutability re-
inputs. The possibility of substituting fertiliz- lation between labor and fertilizers estimated
ers for farm capital may be of interest for here is more consistent with observed histori-
developing economies. Intensive use of bio- cal trends.
chemical inputs (principally inorganic fertiliz- Considering the price elasticities of factor
ers) may be made in those countries instead of demands, our measures for labor (-0.8389 at
massi ve agricultural mechanization in order to the mean) is not too different from his average
increase production. value of -0.9109. But our elasticity of demand
(c) Substitutability between labor and capi- for fertilizers was much less than his. For ma-
tal has declined, while that between labor and chinery, his average elasticity was -1.0886
fertilizers or labor and feed, seed, and live- and for land it was -0.3356. Our average elas-
stock has increased. ticity of demand for capital was -0.5297,
(d) The (own-) price elasticities of demand which is similar to an appropriately weighted
for all the inputs are less than 1 in absolute average of Binswanger's estimates.
value. Farm labor has the highest price elastic- This model was used to simulate the effects
ity of demand. Each input experienced in- of some hypothetical (though not unrealistic)
crease in its price elasticity of demand over changes in the exogenous variables. The index
time. This probably reflects greater use of pur- of the user's cost of farm capital was actually
chased rather than farm-supplied inputs. 522.4 in 1979 (1967 = 100). A 10% increase in
Ray TransLog Cost Analysis of u.s. AgricuLture 497

real estate and machinery prices along with an growth rate of productivity sustained over de-
interest rate of 16% in 1982 would cause this cades is quite impressive.
price index of capital to increase to 999.8. We The overall scale economies computed from
further assume that farm wages would rise by the cost function indicate that while U.S. ag-
12% and fertilizers price by 10% during 1979- riculture operated under diminishing returns,
82. In that case it would cost in 1982 about the returns to scale factor increased over time.
72% more to produce the same amounts of The cost function was tested for homothe-
crops and livestock as in 1977. ticity and rejection of this property indicates
Using these projected prices we find that the that aggregation of different outputs into a
marginal cost of livestocks and crops [com- single index of agricultural production is in-
puted using equation (8)] would be $830.1 mil- valid. Further, it was found that there is joint-
lion and $690 million, respectively, in 1982. If ness in the production of crops and livestock
prices are to correspond to marginal costs, the in the sense that the marginal cost of one is
livestock price index should rise by 66% and influenced by the output of the other. In this

Downloaded from http://ajae.oxfordjournals.org/ at University of Ulster on February 16, 2015

crop price index by 60% during 1977-82. Ac- respect, we need to realize that our model with
tuallevels of production and prices would, of two outputs (although a step in the right direc-
course, be determined by demand and costs tion) is not disaggregative enough. This is an
together. unfortunate limitation imposed by consid-
The impact of this steep increase in the erations of computational manageability with-
user's cost of farm capital on the demand for in the scope of this study.
labor may also be calculated from the esti- An important limitation is noninclusion of
mated elasticities. Using the 1977 value of the family labor as an input. This is primarily due
cross-partial elasticity the 148% increase in the to the facts that no data were available on the
price of farm capital relative to wages causes family labor component of the production ex-
the demand for farm labor to be 38% higher in penses and no figures were found for the wage
1982 than in 1977. This would probably mod- Tate of family labor on farm. Exclusion offam-
erate the out-migration of labor from farming. ily labor assumes functional separability of the
cost function in this and the other inputs. This
assumption could not be tested in our model.
Conclusion However, the usual practice of evaluating fam-
ily labor at the market wage rate is unsatisfac-
This study attempted to analyse the structure tory on two counts. First, this approach ig-
of agricultural production in the United States nores the market demand conditions for farm
using the translog approximation to the cost labor. Second, to the extent that family labor
function. Neoclassical duality results were ex- consists principally of managerial decision
tensively used for this purpose. In view of the making, it should be treated as a distinct input
relatively competitive structure of the farming and not to be added to hired labor.
sector in the United States, this is quite ap- Ideally, this study should have used farm
propriate. Crops and livestocks were treated level behavioral data. Use of aggregate data
as two distinct outputs instead of being here (as in all similar models using economy
lumped together into an aggregate product. Of wide observations) introduces a measure of
primary interest for this study was estimation aggregation bias. Besides, as already noted the
of the elasticities of substitution between in- translog approximation used in this paper has
puts and the price elasticities of factor de- .some limitations. Our estimates therefore
mands. Labor was found to be a substitute for should be regarded as broad indicators.
all other inputs. So was farm capital. The re-
[Received April 1981,. revision accepted
sults showed a decline in the substitutability
January 1982.]
between labor and capital along with an in-
crease in the substitutability between labor
and fertilizers. This information is potentially
useful for policy. If this trend is to continue, References
we may expect slowing down in the rate of
mechanization and intensified use of chemi- Appelbaum, E. "Testing Neoclassical Production
cals in U.S. agriculture. Another important Theory." J. Econometrics 7(1978):87-102.
finding is the rate of Hicks-neutral technical Berndt, E. R., and L. R. Christensen. "The Translog
progress in U.S. agriculture. A 1.8% annual Function and the Substitution of Equipment, Struc-
498 August 1982 Amer. J. Agr. Econ.

tures, and Labor in U.S. Manufacturing 1929-68." J. and Indirect Implicit Additivity. " Econometrica
Econometrics 1(1973):81-113. 43( 1975): 395-420.
Berndt, E. R., and D. O. Wood. "Technology, Prices and Hasenkamp, G. Specification and Estimation of Multiple
Derived Demand for Energy." Rev. Econ. and Output Production Functions. New York: Springer-
Statist. 52(1975):259--68. Verlag, 1976.
Binswanger, H. P. "A Cost Function Approach to the Phillips, L. .• Factor Demand in the Provision of Police
Measurement of Factor Demand and Elasticities of Protection." Economic Models ofCriminal Behavior,
Substitution." Amer. J. Agr. Econ. 56(1974):377-86. ed. J. M. Heineke, pp. 211-57. Amsterdam: North-
Christensen, R., D. W. Jorgenson, and L. R. Lau. Holland Publishing Co., 1978.
"Transcendental Logarithmic Utility Functions." Ray, S. C. "Factor Substitution and Technological
Amer. Econ. Rev. 65(1975):367-83. Change in U.S. Agriculture: A Multi-output Translog
Cooley, T. F., and J. L. Bothwell. "Efficiency in the Cost Function 1939-77." Work. Pap. in Econ. No.
Provision of Health Care." Mimeographed. Univer- 161, University of California, Santa Barbara, 1980.
sity of California, Santa Barbara, 1979. Schultz, T. W. Economic Organization of Agriculture.
Darrough, M. N., and J. M. Heineke. "The Multi-output New York: McGraw-Hill Book Co., 1953.

Downloaded from http://ajae.oxfordjournals.org/ at University of Ulster on February 16, 2015

Translog Cost Function: The Case of Law Enforce- Solow, R. M.••Production Function and the Theory of
ment Agencies." Economic Models of Criminal Be- Capital." Rev. Econ. Stud. 23(1956): 101-08.
havior, ed. J. M. Heineke, pp. 259--302. Amsterdam: Theil, H. Principles of Econometrics. New York: John
North-Holland Publishing Co., 1978. Wiley & Sons, 1971.
Diewert, W. E. "An Application of the Shephard Duality U. S. Department of Agriculture. Agricultural Statistics.
Theorem: A Generalized Leontief Production Func- Washington DC, 1957.
tion." J. Polit. Econ. 79(1971):481-507.
- - . Agricultural Statistics. Washington DC, 1972,
Fisher, F. M. "Test of Equality between Sets of
1978.
Coefficients in Two Linear Regressions: An Exposi-
tory Note." Econometrica 38(1970):361-66. - - . Balance Sheet of the Farming Sector, Supp. No.
Fuss, M., and L. Waverman. "Multi-product Multi-input 1. Agr. Info. Bull. No. 416, Washington DC, 1978.
Cost Function for a Regulated Utility: The Case of - - . Changes in Farm Production and Efficiency 1977.
Telecommunications in Canada." Paper presented at Statist. Bull. No. 612, Washington DC, 1978.
National Bureau of Economic Research Conference U.S. Department of Commerce. Historical Statistics of
on Public Regulations, Washington DC, Dec. 1977. the United States, bicentennial ed. Washington DCl
Guilkey, D. K., and C. A. K. Lovell. "On the Flexibility 1975.
of the Translog Approximation." Int. Econ. Rev. Uzawa, H. "Production Functions with Constant Elas-
21(1980): 137-47. ticities of Substitution." Rev. Econ. Stud. 29(1962):
Hanoch, G. "Production and Demand Models with Direct 291-99.