Download as pdf or txt
Download as pdf or txt
You are on page 1of 23

Journal of Development Economics 37 (1992) 265-287.

North-Holland

Hanan G. Jacoby
Universily of Rochesrer, Rochester. NY 14627. USA

Received July 1989, final version received December 1990

This paper estimates the productivity of men and women in the peasant agriculture of the
Peruvian Sierra, using recent household survey data. A sexual division of labor on the farm
implies that male and female labor are not perfectly substitutable. Evidence is found for female
specialization in livestock production. A translog production function reveals that the use of
animal traction and land affect the marginal productivity of male and female labor differently,
suggesting that the two types of labor cannot be aggregated. Overall, adult male labor is found
to contribute more to farm output at the margin than adult female labor, though the extent of
the difference is sensitive to how farm output and the labor inputs are measured.

1. Intrduction
Productivity of agri&tural labor is central to any discussion of rural
development. In order to understand the labor supply, schooling or fertility
decisions of peasant households, not only must the returns to farm work be
considered, but also how these returns differ according to the sex and age of
the worker. Econometric studies of consumer-producer agricultural house-
holds [such as Lau, Lin and Yotopoulos (1978) and Barnum and Squire
(1979)J estimate production or profit functions, but generally assume the
labor of men and women to be perfectly substitutable ic production. This
assumption allows such models to be estimated “recursivelly’;that isi by f.‘rst
estimating the technology, followed by a consumer demand system taking
farm profits as exogenous income.
household time allocation studies of
Rosenzweig ( 1980), men and women (an
types of workers, each with their own s

*This rePear&
____ _ _ was conducted while the author wgs z consn!tar?t ?O thp WP!~FE ZFC! HUAT~P
Resources Division of The World Bank.
author and should not be attributed to
comments of Hailu Mekonnen, Marty
acknow!edgcd.

0304-3878/91/$03.50 t!‘: 1991~Eisevier Science Publishers .V. All rights ITserved


266 H.G. Jacoby, Prdrctisitv c$ men and women

productivity. By equating market wages to these shadow values of time, even


for non-participants in the labor market, these studies can ignore the
underlying agricultural technology.
In countries or regions where labor market participation is limited, district
or village level wage rates, if available, may be poor proxies for the marginal
productivity of men and women in most households. To obtain adequate
cs;imates of these productivities, the ~001s of production theory must be
employed. The aim of this paper is to estimate the labor productivity of men
and women in the peasant agriculture of the Peruvian highlands or Sierra
region, using household level data from the Peruvian Living Standards
Survey (PLSS) of 198519%.
This paper asks two basic questions. First, in a traditional agricultural
setting, does male and female farm productivity differ much on average?
Alternative ways of measuring the inputs and outputs of subsistence farms
complicates this question. Before examining these issues in the context of a
production function, however, the question of which production function to
estimate must be addressed. Functional form of the agricultural technology,
particularly the substitutability of labor, is central to modeling the
consumer-producer farm household. But perfect substitutability of male and
female farm labor, postulated in much of the literature, is hard to reconcile
with the fact that in traditional agriculture there tends to be a marked sexual
division of labor; i.e., a distinction between predominantly maie and female
---L
tasks, which may IX inherently compleiiicucary.
After describing the data in section 2, section 3 explores one type of male-
female task specialization in the Peruvian highlands. n;imeiy that between
crop and livestock production. Section 4 then estimates a translog produc-
tion function, which accommodates x more ge?W!r sexual division of !abor
by allowing male and female productivities to be differentially affected at the
margin by the levels of other physical inputs employed on the farm-r With
a suitably restricted version of the translog in hand, section 5 examines its
robustness to various alterations in the empirical specification. The final
section of the paper summarizes the findings and discusses their implications.

2. The data akld construction of the variables


The Peruvian Living Standards Survey, conducted by The World Bank’s
Living Standards Unit and the Peruvian Instituto National de Estadistica
(INE), provides data on the agricultural activities of a large number of
households throughout the country. Peru can be divided into three starkly
different agro-climatic regions. The eastern jungle or Selva is a vast, sparsely
“‘iheoniy !mvlous sludy to rigorously e:<amir.e the i;;l$icdi;tiils of a sexual division of labor
on the structure of the agricultural technology is Laufer (1985) using Indian data and a flexible
form for the agricultural production function.
populated, tropical ra in forest wirh relatively little c~~tjvat~d Ias;d. To the
west, the Pacific coastal strip is a desert, dotted with oases that support
nearly al1 of Peru”s industriahzea plantation agriculture.
Sixty percent of Peru’s arable land is located in the Sierra region, as is
about half the country’s population. Here, the unit of agricultural production
is primarily the peasant farm household, the majority of whom are Indian.
The large population of subsistence farmers grows mainly potatoes and
grains, and raises livestock. Modern inputs and high yielding varieties have
been slow to arrive to the Sierra, and mechanization is often infeasible on the
steep mountain slopes. Households diversify their production activities,
sometimes by cultivating different crops in different ecological zones, which
are determined by the extremely qrariable terrain.
From the original 5,000 households in the survey, 1,549 agricultural
households from the Sierra region are chosen based on whether they had
worked any land during the survey year and had reported harvesting some
crops. About 70 percent of the households surveyed in tile Sierra fall into
this group. Since the focus is on the productivity of men and women, the
sample is reduced further by selecting only the 1,034 households in which at
least one adult male and female worked on the family farm. This latter
selection may lead to biased estimates of the technological parameters
because unobserved agricultural inputs (soil quality, for example) could well
be correlated with the off-farm work decisions of farm househo!ds. Section 5
explores this po)ssibf?ity of se!ertion b:bas.
Table 1 describes the variables used in the production function estimation
and reports their sample statistics. Because most farm households in the
Sierra grow several crops ar,d raise livestock as well, and since data on input
use ar not available by crop or activity, the different outputs are aggregated
using prices. Estimating value of output or gross revenue functions for multi-
output farms is not uncommon in the agriculturai economics literature [e-g.,
Huffman (1976)] although such linear aggregation can be quite restrictive.2
With regards to crop qmdvctinn, ea,ch hmsehdd in the srcmpk repsrts the
quantities of ihe different crops it harvested during the survey year and the
prices they would currently fetch on the market. Huge variation in these
prices for a given crop within a village, undoubtedly due to measurement
errors,3 necessitate using village level median prices to comp!atc the vzlve of
crop output. This level of price disaggregation is essential to de
Bardhan’s (1973) criticism of the practice of calculating the values of output

‘Strauss (1984) for example, estimates a more general constant elasticity of trans~o~a~~o~
production possibhty frontier for Sierra Leone agricuhre.
‘Quantilies and nominal prices are reported in he local cro
latter are converted into price per kilogram and then de IMy
index (Peru’s inflation was rampant at the time of the survey).
268 H.G. Jacoby, Productivity of men and women

Table II i

Variables for the production function.


--_-___---___ - _-_-- - -. -. ..-. -
Standard
Varia.bie Definition Mean deviation
- -___
(A) Value of livestock 4,809 7,854
(B) Sales of animal by-products 207 1,307
Value of output Value of all crops harvested + 0.2 x (A) + (B) 8,885 55,309
Land Land area in hectares 4.6 20
Equipment Value of farm equipment 580 8,880
Farm animals Value of oxen, horses and mules 1,541 3,414
Insecticide Expenditures on insecticide 71 215
Fertilizer Expenditures on fertilizer 142 567
Transport Expenditures on transportation 66 342
Livestock Expenditures on livestock inputs 173 568
Hired labor (Days of labor hired +days received in labor 292 1,235
exchange) x 10
Adult male labor Hours farm work, males age > 19 2,354 1491
Adult female labor Hours farm work, females age > 19 1,940 1,311
Teenager labor Hours farm work, ages 12-19 1,050 1,609
Child labor Hours farm work, ages 6-l 1 541 963

Farm and household characteristics


Irrigation Proportion of land irrigated 0.30 0.40
Head’s age Age of household head 48.1 13.5
Head’s schooiing Years schooling of head 2.9 2.9
Permanent crops? Dummy: 1 if grew permanent crops 0.41 0.49
(0 otherwise)
Harvest season? Dummy: 1 if interviewed during harvest 0.23 0.42
season
Planting season? Dummy: 1 if interviewed during planting 0.28 0.45
season
Off-season? Dummy: 1 if interviewed between harvest 0.25 0.43
and planting season
North? Dummy: 1 if live in northern Sierra 0.21 0.41
Central? Dummy: 1 if live in central Sierra 0.33 0.47
~~---_. _ _ - --_
“All variables refer to the twelve-month period prior to the date of interview. Monetary values
are in June 1985 Intis.

from the same set of prices for all farms4 However, these village prices
might also contain an undesirable seasonal component, due to different dates
of interview (seasonal corrections are discussed below).
Measuring the value of output from livestock production (e.g., cattle or
sheep herding) poses greater diffkulty. Part of this output consists of sales of
dairy, wool and other animal products, which are available from the survey.
The survey also provides the information necessary to compute the value of
4Bardhan’s argument is that if farmers face the same prices and the true production possibility
frontier is concave, rather than linear, then crop composition cannot be allowed to vary across
farms, since farmers are assumed to have the same technology. If crop composition is indeed
variable in the sample, movements along a given production possibility frontier will be
construed as shifts in the value of output, i.e., as scale effects.
H.G. Jacoby, Productit&y of men and women 269

the hous:hold’s stock of animals. Presumably, every year this stock appre-
ciates m value (through growth and births net of deaths) by an amount that
depends on the level of inputs such as hours of herding, food and medicine.
Unfortunately, this appreciation is unobservabie,s so in what follows the
return is simply assumed proportional to the value of the stock. Further-
more, since oxen, horses and mules are usually inputs into the producaion of
crops, they arc excluded from the calculation of the value of livestock (their
value appears as a right-hand side variable in the production function),
leaving only cows, sheep, pigs, goats and llamas to be considered. Thus, the
total value of output is obtained as a sum of the value of ail crops harvested,
the sales of animal products and some fraction of the value of the
household’s herd as just defined. This fraction is set, rather arbitrarily, at 0.2
throughout the analysis, though other values will be tried below to see how
robust the estimates are to this choice.
The input of land is measured as the amount at the household’s disposal
(owned, rented or sharecropped) in the year of the survey, but it is not
known whether this land was actually cultivated. Some of the land may have
been left fallow, though in Peru such land is usually used to graze livestock.
FWhermore, the proportion of ‘rand irrigated is known, which is more likely
to have been cultivated. The irrigation variable captures the quality dimen-
sion of land in two other ways. First, irrigaiion extends the growing season,
which is otherwise tied to the climate in the Sierra (normally between
November and March). Secondly, irrigation is most common on valley
floors, where land is most versatile and fertile. Other relevant farm character-
istics included in the production function are a dummy variable for whether
or not permanent or perennial crops are grown, and a sef of regional
dummies for the northern and central parts of the Sierra.6
Only expenditure data are available for thz ‘variable’ physical inputs:
insecticides, fertilizer, transportation, and livestock inputs (mainly food and
medicine). The use of such data in place of quantities in the production
function can lead to biased estimates if input price variation is substantial.
Yet, taking this route seems preferable to ignoring these inputs altogether
and suffering an omitted variables problem. The input of farm equipment
(mainly animal plows) is also expressed in value terms.
Farm labor is divided into five categories: adult (older than 19) male an
female, teeilagr: { :2 3 Is), child (6 to 12), and no&anally labor. Phe !atter
Eype of labor, which is not differentiated in the survey by age or sex, is
%ales, purchases and own consumption of livestock are observed. but these bear no I retic
relationship to returns. Births and deaths of animals are also reported in the survey, are
found to be rid&d with inconsis?encies.
Q the Sierra the crops !hat can be grown, the fertility of the soil arid even the
animals that can be raised, depend crucially on the altitude.
variation in altitude over a very s
dummies cannot adequately capture
270 H.G. Jacoby, Productioity of men and women

almost exclusively supplied by adult males, according to Figueroa (1984).


Deere (1982) finds in her 1976 Peasant Family Survey (conducted in the
scrra province of Cajamarca) that only about 127; of the days of’non-family
labor employed oin peasant farms are contributed by women. This category
includes both the days of hired labor and the days of labor received from
other households under either the traditional ‘Ayni’ or ‘Minka’ payment
systems.’
To ascertain each member’s farm labor input the questionnaire uses a
seven-day recall period, asking for both the actual or ‘effective’ and the
‘usual’ hours (per week and weeks per year) spent working in any productive
activity. If the individual did not work in the peek in which his or her
household was interviewed then a IZmonth reference period is used, and the
usual hours worked per week and weeks per year are asked. If the person
identifies himself as a farmer’ who is a ‘self-employed or unpaid family
worker’ in that particular job, his usua: hours, calculated on an annual basis,
are taken as the farm labor input. There is an extremely high correlation
{about 0.95 for both adult males and females) between the actual hours
reported worked on the farm during the last seven days and usual hours
worked. Although this correlation is slightly lower than that between non-
farm actual and usual hours, it indicates that either seasonality of hours is
not significant in the Sierra or that respondents simply base their estimate of
‘usual’ hours on their most recent work experience; i.e., on their actual hours.
If there is indeed marked seasonality in hours so that this ‘survey bias’
interpretation of the high correiation is the correct one, then it should be the
case that usual hours reported will vary according to the season of interview.
Figueroa (1984, pp. 61-63) describes an approximate agricultural calendar
for the Sierra consisting of a planting season (October-December), a harvest
season (April-June), and a pair of relatively slack seasons in betweenOgFigs.
1 and 2 show how the avetage farm iabor input (usuai hours) of adult males
and females vary by month of interview, and associated season.‘0 Although
reported hours of males display slight pro-seasonal patterns, these figures do
not show significant seasonality (relative to the standard deviation in yearly
hours) and hence survey bias does not appear to be a major problem. The
figures suggest that during slack seasons Peruvian peasants spend about the

‘The Ayni system is a straight exchange of labor, while Minka icvoivtrs a pureiy non-
monetary paymeni. In fact, mosi hired agricult- tira! workers in the Sierra Teceive a substantial
fraction of their daily remuneration in goods or services. Thus, some of the observed variation in
monetary wages across communities can be attributed to differences in the average fraction of
non-monetary payments. See Figueroa { i 984) for details.
*The survey asks, ‘have you worked on the farm or raised livestock belonging to your
household? [Grootaert and Arriagada (1986, p. 38)].
‘Animal husbandry, on ihe other hand, is not a seasonal activity.
“Interviews were conducted ill the Sierra thrcxghout the year 19851986, and were not
heavily concentrated in a particu!ar area during a given month.
Month of interview
Harvest ....._..April
Msy
June

Off ...._........July
August
Septxber

Planting.....October
November
December

Off . .. .......January
February
Mars2 _ _
0 500 ?Q@O 1690 3009 2590 3000 35nn
Mean yearly hours

Fig. 1. Mean hours of farm work for adult males, by month c~finterview.

Month of interview
Harvest . .....__April
May
June

Off ..... ........ July


AI!?& _’
Septemb -

Planting.....October
November
December

Off . ......._.January
February
March - -_.. ..-- - -_
0 500 1000 1500 2000 2500 3000
Mean yearly hours

Fig. 2. Mean hours of farm work for adult females. by month of interview.

same amount of time on the farm as in peak seasons. a


engaged in quite different tasks. In order to address a
albeit in a crude way, a set of three seasonal dummies
production function.
Finally, two variables representing characte~stics
age and years of SC E-c~~c~~
female, is also the primary
decision
to follow are not appreciably af’feet
educated farm work :r in the ho
variables are ass

affects the marginal products of 1


marginail rate of technical su!Mtution
una&cted.

3, of la tir R

The existenceof mmparati d imp&es a


complementarity between the 1
prod&an functions that impose t s~bst~tutabi~ity.~* One dr
the PLSS is that it does not disag
task, so the degee of sptxiahzatiion is inn
However, mdirect evidcxz is available on the di;fsi~z labijr bCiw6xn cro
and Sivebtock production since separate measures o
activities arc available, All but 96 of the I,034 s~~e~t~d1
&stock in ad&non to growing crops.
Consider two separate Cobb-Douglas (C-D)
for crop and livestock output.“’ In the first, the ~~~rithm of t
crop output is regressed on the logarithms of all th,: inputs, except livestock
inputs. This equation w d be a ‘pure’ crop pr~d~~t~~n munition if the
actual amount of each ty of labor and the quantity of land an
equipment going into crop production alone were Kiowa.
sexmd pseudo-production fu ion takes the value of livestock p
measured in table 1 as the d dent variable and drops physical i~~~t~ used
exeiusively for crop production (insecticide, fertilizer, transportation, farm

“A special case of Sato’s (1957) nested CES production junction provides a direct test of the
perkct substitutability of male and female Labor. h&ding an a~~~e~ate of male and female
labor inputs of the form QLR,+( I -y’, )Lf)I’@ in an otherwise CobbDouglas production function
for total farm output nests the hypothesis of perfect substitutability (p= 1) and the Cobb
Douglas form [p=O). The Non-linear Least Squares estimate of p is 0.
error of 0.870 neither of the above hypotheses can be rejected. The in
estimate of the elasticity of substitution between the le and female labor inputs suggests that
perhaps the elasticity is far from constant across s e points and that a flexible functional
form ought to be considered.
production function for total out Stock) is rejected in the
next section, the C form is a convenient rna~~tai~e loratcry exercise.
dult mde and female Eabor, which are always positive by cunst~ct~on 0
the sample. The choice of the additive constant is arbitrary, except for the
presumption that it be ‘near’ zero, or at Ieast small relative to the avera
input value.
Table 3 reports Ordinary Least
ions on the reduced samp

their respective output measure as well.


the co,lumns of table 3 is the relative

13The sample reduction is not ly selective. so that the


hm those in table 1.
244

Insecticide

Fertilizer -

Transport

Livestock

Hired labor 8.119


(0.014)
Adult male labor 0.153
(0.039)
Adult female labor 0.018
(0.043)
Teenage labor 0.015 0.014
(0.009) (0.011)
Child labor 0.010 O.OiR
(O.QIO) 10.013)
Irrigation 0.274 -
(0.082)
OfVIl
Head’s age -“.WI 0.003
(0.003)
Head’s schooling 0.033 ii.027
(0.013) (0.017)
Permanent crops? 0.345 -
(0.067)
Harvest season? 0.150 (3.218
(0.095) (0.122)
Planting season? 0.262 o.QO2
(0.090) (ii.1 ‘4)
Off-season? 0.177 -0.203
(0.094) (0.119)
North? -0.185 - 0.747
(0.093) (0.116)
Central? 0.042 - 0.426
(0.077) (0.099)
R2 0.47
“Standard errors in parentheses.
the familiar e~~a~~ty
of marginal rates of technical substitution (

Douglas case is if the ratio of the respective output elasticities is equal (i.e.,
%4.crsp h. crop = %4. E%s!k/~~.,~~~k).‘~ The estimates of table 3 show the output
elasticity ratio in crop production to be 8.5, compared v+ith only 0.57 in
hvestock production. Thus. if peasant households are optimally allocating
their time across farm tasks, the evi ence suggests a sexual division of labor;
with women spending relatively m re of their time than men in livestock
production, and men specializing in field work.
Indeed, in the Peruvian highlands, yeomen and young children do the
greatest share of animal care, which is a very time-intensive activity,” and
labor is usually not hired for this purpose. On the basis of her aforemen-
tioned Cajamarcan survey, Deere ( 1983) makes the folIowing observations:
Animal care is the mother’s responsibility in 62 percent of all housebof
. . . [TJhe participation of chiidren in this activity is vi
of the households children take full responsibility for
another 36.9 percent, children actively par&i
Colttrast to agricultural [crop] ~rodu~t~o~, where i
households women participate actively in tasks that are
men’s responsibihty. the partici ation of men i animsli care is

‘%ecall lhat in the C-D case MRT,j=(r,;:rj) (LJL,). where the 2s are the coefkients on the
logs of the labor inputs, I_..
’ 5Deere (1982) finds that lthe average peasant household in Caiamarca devczs 30 hours per
week to animal-raising activities.
276 H.G. Jacoby, Productivity of men and women

Fathers participate in animal care in only 9.8 percent of the households


and are the principal adults responsible for the activity in only 4.4
percent of the households - all of which are dairy farms. (p. 115.)
It is still somewhat surprising, given that women are not entirely specialized
on average, thai the estimated contribution of female labor in crop produc-
tion is essentially zero.

4. Separability and the translog production function


With a sexual division of labor, of which the crop-livestock allocation is
merely one manifestation, the labor productivity of men and women will be
differentially affected by the presence of other farm inputs. In other words,
the MRT between the labor of men and women will vary with the levels of
other farm inputs. Perfect substitutability of labor implies that the MRT is
constant, while the C-D production function imposes separability, since the
MRT depends only on (the ratio of) the male and female labor inputs, not
on the level of any other input. The purpose of this section is to find out
which inputs affect the male-female MRT, an exercise which will shed further
light on the sexual division of labor in the Peruvian Sierra.
Ccilsider the arbitrary logarithmic production function,

ln I’= F(G(ln L,, In L,, In X,), In X,) + H(ln X,), (1)

where the Li, i = M, F, represent the labor inputs of adult males and females,
respectively, and the Xj are some other inputs. This function embodies three
forms of separability: (I) the Li are strongly separable (ZS) with respect to
X3, (2) the Li are weakly separable (WS) with respect to AZ, and (3) the Li
are non-separable (NS) with respect to X,. Based on these definitions, the
C-D production function is SS in all the inputs. SS is, of course, more
restrictive than WS, and is a sufficient, but not a necessary, condition for the
MRT to be independent of the other inputs.
In the present context, the concern is not to ‘test down’ to a more
parsimonious specification [as in Denny and Fuss (1977)], but rather to
search for a potentially nun-separable technology. Hence, a sequential testing
procedure is adopted, where first SS is tested against the joint alternatives of
WS or KS and, if this hypothesis is rejected, WS is tested against NS alone.
If this secop,d r,uii ica *nip toA fr\r
rbpcwA ,“I tka
- particu!ar se: of inputs in question, the
production function is a possible non-separable technology.
Denny and Fuss (1977) formulate tests for separability based on the
interpretation of a translog production function as a second-order approxi-
mation to some unknown arbitrary productio function. Take, for example,
H.G. Jacoby, Prodwriir) of mm and women 277

the partition PI0 of the input sei into it’, = (LM,L, ‘, and AT2= (
1,-‘*, lo>, where k indexes the ten other inputs in the modeLr6 The ‘scaled’
translog function in Z = (N,, IV, ) is given by

lnY=a,+Ca,dInZi+1/2CC~ij~lnZ,dlnZj,
i i j

where d In Zi=ln Zi-In Zf. Expression (2) is a quadratic approximation


around the point Z * to a production function SS with res
partition PI0 if

yMk=yFk=O for all k in N,. (3)

That is, under condition (3) the MRT between L, and L, is jnde~~dent of
all the X,. Expression (2) is a quadratic approximation to a production
function WS with respect to P,, if

for all
h&F = YWkh-Fk k in N,. (4)

A very convenient feature of this scaled translog is that the ‘xiare just equal
to the production elasticities at the point of expansion, i.e., at Z* all the
second-order terms vanish. In what follows, the translog is scaled at the
geometric means of the data.
With the imposition of symmetry, namely *Jij=^/ji, there are still 91
parameters to estimate in translog (2)! To simplify the estimation and avoid
the potential for multicollinearity arising from so many second-order terms,
the translog must be further restricted. Since the focus here is on the
separability of male and female labor with respect to the other inputs, on:y
the interactions across input sets N, and N, are considered, i.e., only the
interaction parameters yMkand YFk,k= 1,. . ., 10, are estimated. These para-
meters are sufficient to test conditions (3) and (4). This simplification
amounts to imposing SS with respect to any partition of Nt.r7 The
In (L,) x In (&) interaction in N 1 is maintained for the pressnt, resulting in a
production function containing 33 second-order terms altogether, ~~clud~~g
the 12 squared terms.
Ordinary Least Squares estimates of this restri
function for total output on the sampk of 1,034
first column of rabie 4. All of the inputs, save equipment, have positive point
estima:=ls for their marginal products at the geometric

16Note. the farm characteristics 2nd dummies are not inputs as such.
“FUSS (1977) similarly resiks a rime-inptit
major categories of labor, capital, etc.
Table 4
Translog production function: Sequentially restricted (selected parameters).
___-
Tra.nslog Translog
Independent __.. -.-_ _-._ Independent
variable I !! variable 1. . II

(M) Adult male labor 0.180 0.187 (10)x(10) 0.017 0.018


(0.046) (0.045) (0.004) (0.004)
(F) Adult female labor 0.082 0.087 Of: ~(1) 0.012 0.012
(0.043) (0.042) (0.021) (0.012)
(1) Land 0.224 0.223 (M) x (2) - 0.024 - 0.023
(0.021) (0.020) (0.015) (0.014)
(2) Equipment - 0.002 - 0.005 (M) x (3) - 0.060 -0.061
(0.014) (0.013) (0.018) (0.018)
(3) Insecticide 0.007 W) x (4) 0.022 0.024
(0.025) (8::) (0.016j (0.016)
(4) Fertilizer 0.003 0.015 (M) x(5) -0.000 0.003
(0.027) (0.015) (0.018) (0.018)
(5) Transport 0.044 0.074 (M) x (6) 0.026 0.026
(0.030) (0.016) (0.014) (0.014)
(6) Livestock 0.052 0.05 1 (M) x (7) 0.003 0.002
(0.013) (0.012) (0,014) (0.013)
(7) Hired labor 0.117 0.121 (‘M)x@) -0.004 -0.003
(0.016) (0.015) (0.009) (0.009)
(8) Teenage labor 0.017 0.016 (M) x (9) --0.001 0.000
(0.008) (0.008) (0.010) (0.010)
(9) Child labor 0.006 0.011 (M)x(lO) - 0.024 - 0.025
(0.013) (0.009) (0.011) (O.OI0)
#(iPi FXE azi~ls 3.052 0.053 (F) x!l) - 0.087 - 0.088
(0.009) (0.009) (0.026) (0.025)
(M) x(M) 0.051 0.053 (0 ~(2) 0.018 0.017
(0.021) (0.020) (0.018) (0.017)
W) x(F) 0.05 1 0.054 (F) x (3) -0.010 -0.01 I
(0.030) (0.025) (0.022) (0.022)
(1)x(l) - (i-‘) x (4) -0.016 -0.015
(8:E) (0.020) (0.020)
(2) x (2) - 0.002 - (0 x (5) 0.045 0.044
(0.005) (0.020) (0.020)
(3) x (3) 0.026 0.029 (F) x (61 0.013 0.013
(0.010) (0.009) (0.015) (0.015)
(4) x (4) 0.005 - (F) x(7) - 0.006 - 0.006
(OM!O) (0.016) (0.015)
(5) x (5) 0.009 - (F) x(RI - 0.007 c 0.006
(0.008) fO.010) (0.010
(6) x (6) -0.001 _ (F) x (91 ._ 0.004 - 0.004
(0.014) (0.011) (0.011)
(7) x(7) 0.014 0.015 (F)x(!Oi 0.012 0.012
(0.004) (0.004) (0.012) (0.011)
(8) x (8) 0.002 - (M)xtF) O.MR
(0.006) (0.032)
(9) x (9) 0.003
(0.00X)
RZ 0.530
H.G. Jacoby, Producticiry of men and women 279

Table 5
Test statistics for strong separability of male and
female labor with respect to other inputs. individua!‘y
and jointly.”
~ ~_~
F-value for H o:;‘_~=;.Flr=O
--- -__
Input (k) Translog I I Translog I I I
Land -6.5 - 7.4 ___--
Equipment 1.4 -
Insecticide 7.2 7.8
Fertilizer 1.2 -
Transportation 2.9 3.8
Livestock 3.5 3.6
Hired labor 0.1 -
Teenage labor 0.3 -
Child labor 0.0 -
Farm animals 2.9 4.2
Joint restriction 2.8 5.0
“The Y!/, critical F-value for the individual input
rest&ions (2 degrees of freedom) is 3 for both trans-
logs II and III. The critical value in the case of the
joint test is F(20,987)= 1.57 for translog II and
F( 10,987) = 1.88 for translog 111.

though several (including equipment) are not statistically signifkant.’ 8 The


marginal productivities of adult male and female !abor are significantly
positive, and are diminishing at the geometric mean. Since the coefficient on
!n(LM) x In (~5~) is essentially zero, men and women are complementary, in
the sense that an increase in the hours of one raises the marginal product of
the other.” The stripped down version of the translog in column 2 of table
4 excludes the second order terms which are neither significant (at the 50/6
level) nor directly involved in the tests of SS or WS.‘O
Condition (3) dictates a joint F-test of twenty linear restrictions to test the
hypothesis of SS with respect to partition P,,. Using the translog II
estimates from the second column of table 4, the P-value for this joint test is
2.8, as shown in table 5. SS can thus be strongly rejected at the 5:;
significance level. However, in the interest of further reducing the
parameters, SS can be imposed with respect to a subset of IV;,
cnlumn cf tab!c 5 dis$~yc tt;c r-values , .- the tests of ;I,&~
= T~,~.=O
individually. This restriction cannot be rejected at the YO level
of the ten cases. Mindful of the possibility of m~~ticolli~ear~ty

181rrigation and the education of the household head also have cismifiirandy
c psi?Iv~ dkrs on
farm output iri a’ll iranslog specifications. Age of head has no efiect.
j3Boisvert (19821 gives the relevan! forrn~~~~sfor I
‘“The squared term in female labor i\ rr::aine 1 does not quite ac
sigmficance in column I.
280

__._^. .._____...“._.~.. ...” _“_ . - .-


Translog
~__ ~~___.-._
Independent
variable III IV variable 1V
- ._-P _.~
(M) Adult male MN

IF) Adult female labor

(1) Land

(23 Equipment

(3) Insecticide

14) Fertilizer

(5) Transport

I:6) Livestock

(7) Hiwl labor

(8) Teeriage labor

(9) Child labor

( 10) Farm animals

{M) x(M3

(FI XlFI

R’ 0.513
-~-
“Standard erwrs in parentheses.

interaction terms and seeing that the F-values for t


farm animals inputs are very nearly sig~i~caut~ only e of t seven non-
rejs?r;rtedrestrictions 2:~ imposed (fcr k=equipment, fertilizer, and the three
other labor inputs). The resu!ting production fun is then estimated
(translog III in the first column of table 6), and the i idual SS restrictions
re not imposed are re-tested. All five of t
at the 5yP lewd, according t-c!the serif!
aintaining non-Se~arabiIity with
the next step is to see whether trans
~~~ctio~. The ost c~~v~~ie~t w2y
sexual division of E,~!Ywin the Zxu7+x2 Sierra. MCWeholds with tittte land
to supplement their incomes wit non-farm activities. Thereiore, in near
es households men spend mo 9f their time off the farm in market
work, Heavingwomen responsible for relatively more af the cro
mcluding th ier chores involving agricultural implements.
tion in the between male and female labor by farm s
from this shifting task composition, although observed variation in the
may simply be a consequence of the difficulty in measurinlg the contribution
‘CLL
to oi;:p~t of cc&z activities, s-uzh as those engaged iz by women on Lo.=* Ee
ing for field hands).

see how the elasticity of substitution (ES) as well as the

eometric means
er than that allowe;_t by C-
282 H.G. jacoby, Productivity of men and women

Table 7
Sensitivity analysis of translog production function.

SDeciiication MRT ES
(1) Translog IV 1.68 1.84
(2) Translog I 1.81 1.80
(3) Translog IV with selectivi$ correction 1.75 2.0
(4) Translog I with selectivity correction 1.92 1.89
(5) Translog IV with constants of translation = 0.1 1.73 1.88
(4) Translog IV with constants of translation = 10 1.61 1.79
(7) Translog IV with constants of translation proportiona! to mean input level 1.76 1.88
(8) Translog IV with %LKWK =O.l 2.44 2.69
(9) Trans!og IV with %LP’STK = 0.3 1.37 1.63
(IO) Translog IV with family labor inputs augmented by housework hours 2.84 0.77
(11) Translog IV with family labor inputs augmented by ten percent of 2.80 0.69
housework hours
-_ ~~~~__ -_.

unchanged. Thus, within a range of animal inputs containing about two-


thirds of the farms in the sample, the ratio of male to female marginal
productivity nearly doubles.
In the case of land, a smaller range of variation is considered,2’ Increasing
farm size from the geometric mean of 1.5 hectares to 2.5 hectares gives an ES
of 4.9@and an MRT of 2.82, while an equivalent one hectare reduction in
farm size drops the ES to 1.37 and the MRT below unity, to 0.92. Thus, farm
size has an enormous effect both on the slope of the isoquants between male
and female labor and on the percentage rate of change of the slope with a
percentage change in these factor proportions. These findings cast doubt on
pcrfectiy substitutable farm labor and demcnsfrate the impc??:rHcceof the
sexua:l division of labor, as well as the inappropriateness of constant elasticity
production functions, in the Peruvian highlands.

5. Sensitivity analysis
This section reports the results of sensitivity tests of the final version of
translog arrived at in table 6. The robustness of these final estimates can be
assessed by looking at how the estimated MRT and ES between male and
female labor changes with modifications to the maintained hypotheses. Table
7 summarizes the estimates of the MRT and ES for a variety of specifications
to be described below. The estimated MRT and ES from translog IV at the
geometric mean, reiterated in the first row of the table, have standard error;
of approximately 0.84 and 0.54, respectively. As the second row of tile table
shows, estimates from the less restrictive translog I arc very similar,
,
“At larger farm sizes the marginal product of female labor t :cornes negative, not an
uncommon problem ;:: the estimation C”
! f!trihle functional forms:
H.G. Jacoby, Productiaity of men und women 283

indicating that the restrictions imposed in the previous section are fairly
innocuous.
One concern explessed earlier was the possibility of selection bias resulting
from choosing the sample of households based on whether at least one male
and one female worked on the family farm. This problem is closely related to
the issue of input endogeneity. Unobserved inputs, such as management
ability, soil quality and climate, may well be correlated with observed inputs.
Similarly, selection bias arises if the farm work participation decision of
males or females is conditioned on these unobserved inputs. With the limited
number of good instruments available in the PLSS and the large number of
endogenous variables in the translog, input exogeneity tests will have very
low power.2i Testing for sample selectivity, however, using a Neckman
(1980) two-step procedure, is a feasible strategy.
The household’s decision to work both its male and female members on
the family farm is arrived at by comparing labor productivity on and off the
farm (off-farm activities include housework, non-farm self-employment and
wage work). Thus, among the explanatory variables included in the discrete
choice probability of sample selection are the fixed farm inputs, mean
characteristics of adult males and females in the household (age, schooling
and marital status), housing characteristics (number of rooms, lightirlg source
and water source dummies, etc.), and number of children in the household.
The probit estimates (not reported) on the full sample of 1,549 farm
households show that, of the farm inputs, only farm size significantiy affects
the probability of sample inclusion, with larger farms more likely to be
included. Age effects are very strong, with older men and wamen more likely
to stay on the family farm. Also, the presence of children under the age of six
significantly increases the likelihood that both men and worneTawork on the
farm.
When the resulting inverse Mills’ ratio is inserted into the final version of
the translog, its estimated coefficient is -0.13, with an (uncorrected)
standard error of 0.09. Furthermore, the estimated parameters of the
production function, and most importantly those associated with male and
female labor, are virtually unaffected.23 Minimal changes in the
ES are recorded in the third row of table 7. For completeness, the
of table 7 shows that selectivity is not a problem in dhe less restrictive
translog I either (the t-value on the inverse Mills” r tio is slightly lower than
it is in translog IVj. Thus, it is safe to say t at the outcome of tk

22.41?hougha Wt-Hausmzn test rejects rhe joint exogenaity of Lhe !il.re variable inputs in Ihe
much more restrictive C-D specliication (not reported here), the Two Stage Least square!;
estimates (using village means of household characteristics and individual wage rates as
instruments) are quite imprecise.
23For example, the coeficient on the male labor-farm animals interaction term
-0.018 to -0.0117. while the female labor- land interaction coeffkient drop% frro
- M!7Z!, we!! within their respective st;iii&ld crrorri.
separability iests performed in the previous section are not affected by
sample selectivity.
Consider now the choice of the constants of translation, added to the
input levels to avoid taking logarithms of zero. So far in the production
functions estimated, a common value of unity has been arbitrarily set. When
common constants of 10 and 0.1 are used in the final version of the translog,
coefftcients on some of the inputs, such as livestock inputs and transpor-
tation, change noticeably. However, the parameters of interest are quite
stable, leading to only very small changes in the estimates of the MRT and
ES. Since there are large differences in the mean values of the inputs,
constants of translation proportional to the mean of the corresponding input
are also tried; the input with the largest mean having a constant of one
added to it. Again, table 7 shows no significant changes in the estimated
MRT and ES.
Another somewhat arbitrary ingredient to the baseline estimates is the
choice of 0.2 as the fraction of the value of livestock (Y$VSTK) regarded as
part of the year’s output. It does not seem unrcnsonable to suppose that the
‘gross return’ on an asset such as livestock is in the neighborhood of 20”/,.
However, the translcrg production function is reestimated with ,“;;LVSTK set
at 0.3 and 0.1. Not surprisingly, the coefficient on livestock inputs increases
with %LVSTK, but the change is not very dramatic statistically. As would
be expected, based on the finding in section 3 that women tend to specialize
in livestock production, the estimated MRT between male and female labor
falls with %LK?TK, though by not much more than a standard deviation as
%LVSTK is raised from 0.1 to 0.3. Coefficients on the second-order terms in
the translog also change only very marginally.
A problem inherent in surveys of indiviilwall tienc allocation in rural areas
of developing countries is the sometimes hazy distinction between farm work
activities and housework activities. Not only might these two sectors be
difficult to distinguish in the minds of survey respondents, but, in reality,
housework time may jointly produce agricultural output and ‘!lome=
produced commodities’. Deere (1982) argues that ‘one of the principal
problems in measuring women’s agricultural participation [in the Peruvian
Sierra] is that women must often combine participation in field activities
with responsibilities for meal preparation [for field hands] as well as child
and animal care’ (p. 804).
The proposition that women’s estimated contribution to agricultural
output is biased due to joint production can be examined using data from
the PLSS, which provides information on the hours of housework performed
during the past seven daysz4 When extrapolated to a yearly figure, the
24Thequestionnaire prompts the respondent with the following examples of housework:
‘cleaning the house, Preparing meals far your family, doing laundry, bvying food or clothes,
s?rvising children’s howwork., etc.’ [Grootacrt and hrriaga& (1986, p, 53)].
H.G. Jacoby, Producriuicy 01 melt and women 285

amount of housework done by adult females, who also work on the farm25
in the average household is 1,828 hours, almost equai to the 1,940 hours of
farm work they do. On the other hand, teenagers and children (who work on
the farm) spend about half as much time in housework as in farm work, and
adult males only 20”/, as much time.
To check whether the male to female marginal productivity ratio might be
overestimated by excludi ng housework, hours spent doing housework are
first simply added to the farm labor input for each category of family worker.
Reestimating the translog production function yields a MRT of 2.84,
substantially higher than the baseline estimate. A similar resuh is obtained
when only ten percent of housework hours are added to farm hours (see the
last row of table 7). Housework hours do not seem to be at all productive on
the farm, and by raising the relative labor input of females end up actually
increasing the MRT. Thus, poor accounting of women’s time, at least on this
dimension, does not seem to be a major cause of the male-female producti-
vity gap.26

6. Summary and impkatispns


The conclusion that men contribute a greater share to farm output and are
more productive at the margin than women seems a fairly robust result of
this production function analysis, though the extent of the difference depends
on how output is measured (in particular, how livestock is accounted for),
how the female labor input is measured (namely, whether housework is
included), and especially on the ‘levels of other farm inputs that the
household employs. One can speculate that the reason for the productivity
difference is that women are more casual farm workers than men. In the
Peruvian Sierra, women spend more time on average in housework and non-
farm business activities than in farm work. Indeed, women may sort into
these activities because they are innately more productive in them than men.
Over the life-cycle, such productivity differences are accentuated by the
acquisition of sector-specific human capital by both men and women.
The same process of specialization according to comparative advantage
and investment in specific human cap.., i+ql occurs within farm work itself, as
women perform different agricultural tasks than men. Women may be more
productive weeders than men. for example, while men coul
productive plowers. If plowing is an activity which contributes
margin to output than weeding does, or if it is simply easier to
time spent plowing than the more diffuse time s
% is assumed that joint production is only an issue
housework of those who report no farm work is not i~c~~de~ i
261nterestingly, exactly the opposite results were obtained
namely. the estimated MRT fell substantially with the a~~~?i~~ao
286 H.G. Jacoby, Productivity of men and women

that male labor is more productive. With labor input data that aggregates
over farm tasks, it is difficult to address questions of specialization of tasks.
However, the evidence presented in section 2 indicates that women spend
relatively more time in caring for livestock than men do. If anything, this
finding suggests that it may be inappropriate to aggregate male and female
labor inputs in production function analyses.
The estimates of the translog productIon function reinforce the notion that
the farm labor inputs of men and women are inherently different, and cannot
be aggregated. The rejection of separability of male and female labor with
respect to two of the physical inputs, farm animals and land, implies that
certain agricultural interventions could change the price of male time relative
to female time. For example, a land reform could have the effect of lowering,
on average, the marginal productivity of women relative to men in agricul-
ture, as women in previously near landless households substitute out of
normally male dominated, high productivity, tasks. Whether such a shift
would have any effect on, say, child bearing decisions is still an open
question.

References
Bardhan, Pranab K., 1973, Size, productivity, and the returns to scale: An analysis of farm-level
data in Indian agriculture, Journal of Political Economy 81, no. 6, 1370-l 380.
Barnum, Howard N. and Lyn Squire, 1979, An econometric application of the theory of the
farm househoid, Journal of Development Economics, no. 6, 39-102.
Boisvert, Richard N.. 1982, The !ranslog production hmction: Its properties, its several
interpretations and estimation problems, Unpublished manuscript (Department of Agricul-
tural Economics, Cornell University, Ithaca, NY).
Deere, Carmen D., 1982, The division of labor by sex in agriculture: A Peruvian case study,
Economic Developmen! and Cul?ura! Change 39, no. 4. 795-8 11.
Deere, Carmen D., 1983. The allocation of familial labor and the formation of peasant
household income in :he Peruvian Sierra, in: M. Bovinic, M. Lycette and W. McGreevey,
eds., Women and poverty in the Third World (Johns Hopkins University Press, Baltimore,
MD).
Denny, Michael and Melvyn Fuss, 1977, The use of appruItimation analysis to test for
separability and the existence of consistent aggregates, American Economic Review 67, no. 3,
404-418.
Figueroa, Adolfo, 1984, Capitalist development and the peasant economy in Peru (Cambridge
University Press, Cambridge, MA).
Fuss, Melvyn A., 1977, The demand for energy in Canadian manufacturrng, Journal of
Econometrics 5, no. 1, 89-116.
Grootaert, C. and A.-M. Arriagada. 1986, The Peruvian living standards survey: An annotated
questionnaire, Unpublished manuscript (The World Bank, Washington, DC) July.
Heckman, James J., 1980. Sample selection bias as a specification error with an application to
the estimation of labor supply functions, in: James P. Smith, ed., Female labor supply:
Theory and estimation (Princeton University Press, Princeton, NJ).
Huffman. Wallace E., 1976. The productive value of human time in U.S. agriculture, American
Journal of Agricultural Economics, Nov., 672-683.
Jamison,Dean and Lawrence J. Lau, 1982, Farmer education and farm efftciency (Johns
Hopkins University Press, Baltimore, MD).
H.G. Jacoby, Producticity of men and women 287

Lau, Lawrence J., Wuu-Long Lin and Pan A, Yotopoulos, 1978, The linear Logarithmic
expenditure system: An application to consumption-leisure choice, Econometrica 46,843~868.
Laufer, Leslie A., 1985, The substitution between male and female labor in rural Indian
agricultural production, Discussion paper no. 472 (Economic Growth Center, New Haven.
CT).
MaCurdy, Thomas E. and John l-i. Pencavel, 1986, Testing between competing models of wage
and employment determination in unionized markets, Journal of Political Economy 94, no.
3, S3-S39.
Rosenzweig, Mark R., 1980, Neoclassical theory and the optimizing peasant: An econometric
analysis of market family labor supply In 2 devdoping country, Quarterly Jourmli of
Economics 95,31-B
Rosenzweig, Mark R. and Kobert Evenson, 1977, Fertility, schooling and the economic
contribution of children in rural India: An econometric analysis, Econometrica 45,
1065-1079.
Sato, K., 1967, A two-level CES production function, Review of Economic Studies 34, no. 2,
201-218.
Strauss, John, 1984, Joint determination of food consumption and production in rural Sierra
Leone, Estimates of a household-tirm model, Journal of Development Economics 14, 77-103.

You might also like