Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/publicpolicyexteOObert
OEWEV
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
AND EXTENDED FAMILIES:
EVIDENCE FROM SOUTH AFRICA
PUBLIC POLICY
Marianne Bertrand
j
Sendhil Mullainathan
Douglas Miller
Working Paper 01-31
March 2001
Room
E52-251
50 Memorial Drive
Cambridge, MA 02142
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http ://papers. ssm. com/abstract=xxxxxx
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
AND EXTENDED FAMILIES:
EVIDENCE FROM SOUTH AFRICA
PUBLIC POLICY
Marianne Beitrand
Sendhil Mullainathan
Douglas Miller
Working Paper 01-31
""
March
2001
Room
E52-251
50 Memorial Drive
Cambridge,
MA 021 42
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http ://papers. ssm coiTi/abstract=xxxxxx
.
MASSACHUSElTSlISTrnJTr
OF TECHNOLOGY
LIBRARIES
J
Public Policy and Extended Families:
Evidence from South Africa
Marianne Bertrand (University of Chicago GSB,
Sendhil Mullainathan
and
NBER)
(MIT and NBER)
Douglas Miller (University of California
March
CEPR
at Berkeley)*
27, 2001
Abstract
How
are resources allocated within extended families in developing countries?
this question,
we use a unique
social experiment:
To
investigate
the South African pension program.
Under
that program, the elderly receive a cash transfer that represents roughly twice the per capita
We
how
working-age individuals
drop in the working hours of the prime-age individuals
in these households when elder women reach 60 years old or elder men reach 65, the respective
ages for pension eligibility. We also find that the drop in labor supply is much larger when the
pensioner is a woman, suggesting an imperfect pooling of resources. The allocation of resources
among prime-age individuals depends strongly on their absolute age and sex as well as on their
relative age. The oldest son in the household reduces his working hours more than any other
prime-age household member. The large labor supply response we observe raises important
issues for the design of social policy programs in developing countries and also lead us to be
wary of any model of intra-household allocation of resources that does not fully account for the
endogeneity of earned income.
African income.
ask
living with these elderly.
We
this transfer affects the labor supply of
find a sharp
*We axe extremely grateful to Abhijit Banerjee, Anne Case, .Angus Deaton, Esther Duflo, Jon Gruber, Michael
Kremer, Jonathan Morduch, and Jim Poterba. We have also benefited from feedback by seminar participants at the
MIT Public Finance Lunch, Princeton Development Workshop, Har\-ard-MIT Development Seminar, and the NBER
Summer Institute 2000. Miller acknowledges financial support from the National Science Foundation's Graduate
Fellowship Program, e-mail: marianne.bertrand@gsb.uchicago; mullain@mit.edu; dlmiller@econ.berkeley.edu
Introduction
1
In
many
developing countries, large extended families often
suggest sharing of other resources, most notably money.
social policies
may produce unexpected
effects.
A
depend on the sharing
will
transfer targeted at one
we attempt
family members. In this paper,
families.^
Under
is
To do
this
this,
program,
a very large lump
households.^
the flow of
we
sum
over the age of 60 and
in the
end benefits
between the targeted group and other
how
resources are transferred in extended
of the transfer
cleanly thair
eventually reaches family
men
over the age of 65 receive what in practice
cash transfer, roughly twice the average per capita income in African
The magnitude
makes
it
a useful "experiment," permitting us to track
more marginal changes would
members other than the pensioners?
transferred and which family
To address these
to understand
prevalent,
demographic group may
Who
same house.
is
investigate a unique policy in South Africa: the old age pension program.'
women
money more
rules in place
Shared housing may
such resource sharing
If
eventually find itself in the pockets of relatives living in the
from the transfer
live together.
members
questions,
receive
most of
Does the pension money
allow.
If yes,
how much
of the cash
is
it?
we study the labor supply
of relatives living with pensioners.
We
focus on labor supply for two reasons. First, typical household survey data do not usually allow one
to directly
measure the transfers each family member
receives.
Our data
diture data measure consumption at the household, not individual level.
A
is
no exception. Expen-
Only a few consumption
examined resource transfers in the close family (husband and wife or parent and young
Lundberg and Pollak (1996) provide a survey. In the close family, one can reasonably assume that resource
transfers do take place, for example between parents and young children. The rare evidence we have on resource
transfers in the extended family comes from the U.S. (Altonji et al, 1992) and suggests that resource transfers are not
large in that case. It is an open question whether such finding generalizes to developing countries where the extended
large literature has
child).
family often live under a
common
roof.
"In theory, the transfer program
is
means-tested but this has
quite low relative to the test. See Section
2.
little effect in
practice for Africans
whose income
is
items are exclusive enough that they can be matched to a specific sex or age group. ^ Leisure time,
however,
can
in
is
a good that can easily be assigned (Chiappori 1992).
infer (at least partly)
how the pension money
is
By examining
labor supply,
we
allocated to the various prime-age individuals
a household.
Second, a labor supply response would most clearly underline the unexpected outcomes caused
by family redistribution. Since the social pension targets a group who by and large
the labor force, and conditions mainly on an unalterable variable, age.
little effect
on labor supply
in the economy.'*
occurs, aggregate labor supply
reduce their hours of work.
direction
may
How
if
we might expect
to have
intrahousehold redistribution
large such
an aggregate
effect will
be directly depends on the
One
article
articles hint that the
pension
may
well have affected rela-
mentions that "the impact of pensions on communities with a high
unemployment was huge,
as multi-generation households
formed a constellation around the
"five children,
who
when they can
find
person receiving the pension" (Ngoro, 1998). Another describes a pensioner's
also live with
him
in his
But none has a
work.
it
and strength of redistribution flows inside households.
labor supply.
rate of
the other hand,
already out of
as the working-age individuals that live with pensioners
fall
Anecdotal evidence and newspaper
tives'
On
is
two-bedroom
full-time job"
flat,
contribute to the family income
(Caelers, 1998).
Of
course, such stories
do not constitute
causal evidence, but they provide a backdrop for our statistical work below.
To
only
identify the
men
eff'ect
over 65 and
of the pension,
women
we
will exploit the non-linearity in
over 60 can in theory receive a pension.
Our
pension
eligibility rules:
strategy uses this sharp
Subramanian and Deaton (1991) use expenditures on adult goods such ;\s alcohol and tobacco to study discrimination based on children's gender. Browning et al. (1994) use expenditures on men's and women's clothing to stydy
sharing rules between couples.
A
this
labor supply effect might arise because the pension increases the expected future income of the young.
affect all the young equally, not merely the relatives of pensioners as our results suggest.
would
But
rise in
income when an elder crosses the threshold age to identify the pension's
effect.
suggest that the pension dramatically reduces the labor supply of the working age
household.^
One
The
members
results
of the
sees a clear discontinuity exactly at the age-eligibility frontier, with labor supply
for presence of
dropping at the age of 60
female elders and 65 for presence of male elders in the
household. Roughly speaking, the age of the elderly does not seem to affect labor supply except at
these discontinuous points.
Despite this striking discontinuity, one
still
worry that these results are driven by the
through some channel other than the pension. Most notably,
elders' age operating
of ailments
may
and diseases among the elderly displayed a similar discontinuity
could reflect the fact that working age family
elderly living with them.
To
and our basic dataset
is
are staying at
deal with concerns such as these,
The major expansion
Population Census.
members
from 1993.
If
of the old age pension
the labor supply response
not by other factors associated with living with elderly people,
to be as large in the 1991 data. Indeed,
find an effect that
is less
we
when we
is
first
if
the prevalence
in age,
home
our findings
to take care of the
turn to data from the 1991
program took place
after 1992
truly driven by the pension
we would not expect
this response
replicate our analysis in that earlier period,
than a fourth as large as
in the 1993 data.
sickness patterns changed so dramatically over a two year period.
It is
We
do we find that reported sickness exhibits any non-linearities
we
in age.
we
hard to believe that
also directly control for
reported sickness in our basic 1993 data and find this does not change our original results.
check,
and
As
Nor
yet another robustness
exploit the fact that certain local authorities in South Africa have de facto equalized
eligibility across sexes
ineligible otherwise).
by extended the pension to males between 60 and 65 years old (who are
We
find that in the regions practicing equalization, relatives' labor supply
""We find effects on both hours worked and on the work or not work margin. It is also important to note that our
measure of employment includes all forms of employment (regular, casual and self-employment).
also responds to the presence of
The
tests
males between 60 and
above suggest there
is
65.
a real effect of the pension, but
"leisure" time?
Perhaps we are observing a
more
measure than regular forms of employment.
difficult to
shift to casual or
casual working hours actually declines and the level of self
level of
home production
activities,
is
it
truly an increase in
farm employment, which might be
We find
no evidence
for this.
employment does not change
Reported
at
all.
such as agricultural crop production or livestock production,
does not change either.^ Alternatively, could we be simply picking up on migrating behavior?
pension
more
may make
likely to
and family
out.
and
find
size
move
the unemployed more likely to
move
We
The
also find
no evidence
in
for this.
The
with the pensioners or the employed
We
directly study migration patterns
no sign that the pension significantly affected those variables either.^
All of our findings suggest that
some pension money did
allowing them to reduce their work. But
how
in fact
go to working age relatives,
exactly was the pension redistributed?
We
find that
absolute age, relative age, and sex are important determinants of resource flows in that they affect
the strength of the labor supply response. Holding family composition constant, we find that the
marginal rand of pension income going to a female pensioner reduces labor supply more than the
marginal rand of pension money going to a male pensioner.
reduce their labor supply
received by the elderly.
less
than working-age men
Working hours drop more
We
for each
also find that working-age
women
marginal rand of pension money
as the age of the prime-age family
increase. Finally, controlling for the differential effect of the pension
members
by sex and age, we find that
the eldest prime-age male in the household reduce his labor supply more than other prime-age
^A
that prime-age individuals are investing more in human capital. We find no evidence for
it is the older relatives of the pensioners, not the school age one.s, that show the
largest drops in working hours.
related possibility
this either.
As we
will
is
see below,
While changes in family composition seem an intuitive response to the pension, the relative recency of the
program, at least in its most generous form, could explain the absence of such changes.
'
household member.
As a whole,
^
men and
these results paint a picture of significant resource transfers, with
especially
middle aged men being the biggest recipients, and female pensioners being the biggest contributors.
These findings have broader implications
for empirical
and theoretical work on
families. First, unlike
the previous evidence for developed countries, resource transfers are large within extended families
in
South Africa. Second, the impact of the pensioner's sex on the flow of resources suggests that
common
preference models of the family (in which
common
utility function)
may
not
fit
one would have expected very
If
family as maximizing one
well for these extended families. In such models, the source
of the pension income should not matter.^
unitary theories of the family.
we can view the
Third, our findings raise interesting issues for non-
earned income (and hence labor supply)
little
affects
negative (or even a positive) labor supply
related, the large labor supply response
we observe
casts
redistribution of earned income, since labor supply
is
bargaining power,
eflfect.
Finally,
and
doubt on empirical strategies that focus on
likely to
be endogenous to the redistribution
process.^
The
and
paper
is
organized as follows. Section 2 describes in detail the historical origin
institutional features of the
data.
its
rest of the
Our
South African social pension program.
central finding of a labor supply response to the pension
robustness to alternative interpretations
A
is
studied in Section
5.
is
Section 3 describes the
presented in Section 4 and
In Section
6,
we attempt
to
few papers have tried to assess whether the demographic characteristics of the income recipient matter in
More specifically, some researchers have studied whether money in the hands of women leads to
resource allocation.
hands of men. For example, Schultz (1990) suggests that money in
fertility. Thomas (1990) shows that female sources
of unearned income are associated with improvement in child health. Unlike our work, male and female labor incomes
in most of tho.se previous studies result from different economic activities. They might thus vary with respect to
their riskiness or liquidity, which makes the isolation of a gender effect very difficult. Moreover, one worries about
the endogeneity of income in all of these cases.
Previous research typically abstracts away from labor supply responses. It often focuses on samples of people
working full time (see, e.g., Browning et al., 1994), which raises possible issues of selection. For example, relative
income may appear to play a central role in defining how resources are shared because relative income determines
which families have both parties working.
a different resource allocation than
the hands of
women
rather than
money
men
in the
leads to stronger decline in
integrate this finding into existing models of the family. In the light of these models, Section 7 uses
the distribution of the labor supply response to study the direction and strength of the resource
flows between the different household
members.
We
offer
concluding remarks in Section
8.
The Old Age Pension Program
2
We now
describe the South African old age pension program.
historical
found
in
Additional information about the
background, institutional features, and practical implementation of this program can be
Lund
(1993),
Van der Berg
(1994) and
Although the social pension program
intended for White South Africans only.
in
Case and Deaton (1998).
South Africa dates back to the 1920s,
The
it
was
historically
process of disintegration of the apartheid regime in
the late 1980s and early 1990s led to pressures for more racial parity in pension eligibility and
benefits.
The major reforms
of the pension
program
with the introduction of superior technologies
in the
for African
households took place after 1992,
pension delivery system
(in part to
access to remote areas) as well as with the equalization of both the means-test
improve
and the pension
benefits levels across racial groups.
Eligibility for
of 60
pension receipt
is
determined primarily by one's age. Only
and men over the age of 65 are
local authorities
men between
to exploit this fact later on
The
eligible for the state social pension. In practice
have been equalizing the pension
a non trivial share of
is
eligibility
age between
over the age
though, some
men and women. Hence,
60 and 65 years of age report receiving a pension.
when we analyze
state social pension
women
We
will try
the effect of the pension.
means-tested.
The means-testing
excluded from the pension while most. Africans are entitled to the
is
such that most Whites get
maximum
benefits.
Case and
Deaton (1998) show that 14% of White women and
take up rates
among
African
The South African
women and men
social pension
is
7%
are 80
of
White men report receiving the pension;
and 77
%
respectively.^"
The maximum
a very generous program.
benefit in 1993,
the year the survey used in this paper was conducted, was 370 rand per month. This
large
amount
relative to Africans' average
income
level.
370 rand
is
it
is
an extremely
about half the average African
household income and more than twice the median per capita income
pension transfers are so large,
is
among
Africans. Because the
reasonable to expect that intrahousehold redistribution could
lead to significant behavioral responses, such as a reduced willingness to participate in the labor
force
among
the family
members
that are not originally targeted by policy makers.
We
proceed to
empirically analyze these responses in the next sections.
Data and Summary
3
The primary data
survey
is
set
Statistics
used in this paper
is
the Integrated Household Survey of South Africa.^^. This
the result of a cooperation between the
Research Unit
(SALDRU)
at the University of
World Bank and the South African Development
Cape Town.^^
It
households and was conducted during the second half of 1993.
different racial groups;
only.
The survey
We
respectively).
It is also
women and
elder white
men
random sample
classifies
will focus
Indeed, as we mentioned earlier, the means-testing of the pension
share of elder white
%
White, Coloured, Indian and African.
consists of a
is
of 9000
people into four
on African households
such that only a small
report receiving any pension transfer (14 and 7
While a larger share of Coloured and Indian South- Africans do report receiving
worth mentioning that the means-testing formula does not take into account income from other family
the elderly (Case and Deaton, 1998). Hence, there are no direct incentives in the program design for
members than
family dissolution or migration,
"The
The
1991 Population Census data used to check the robustness of our
main
paper can be directly downloaded
http ;// www. worldbankorg/html/prdph /Isms/country /za94/za94home. html.
database
used
in
the
results will
from
the
be described
World
in section 4.
Bank
webpage:
the pension, their participation rate
is still
well below the African levels (Case
Moreover, the prevalence of the multi-generation households we wish to study
among any
Africans than
households.
A
and Lund,
of the other racial groups (see Ardington
To better focus on such extended
families,
three-generation household
is
we
is
and Deaton, 1998).
much
larger
among
1994).
will further restrict ourselves to three-generation
defined as one containing at least a child, a parent of
the child and a grand-parent of the child. Note that this restrictions also reduces the heterogeneity
in
our sample. Without
from their
elders.
elders, this
it,
pension ineligible households would also include individuals living away
Since such individuals would clearly be different from those living with their
can introduce a selection
bias.
The
restriction to three-generation households
other hand guarantees that the age of the elderly
is
on the
our only source of variation.
For these households, we will study the labor supply of working-age individuals between 16
and 50 years
old.
We
use the conservative 50 year old
maximum
age in order to avoid any
of the pension that could arise because people expect to get the pension themselves soon.
than a third of prime-age individuals
Moreover, a large proportion of
women
in
our original sample
live in
over 60 years of age and of
effect
More
three-generation households.
men
over 65 years of age
live in
a three-generation household. This fact has previously been noted by Case and Deaton (1998).^^
and method, the
In content
dard Surveys.
We will use
survey closely resembles the World Bank's Living Stan-
information on a wide variety of household and individual characteristics.
the data on labor force participation and employment status, working hours, migration,
health status,
as well as
It collects
SALDRU
home production
on a
set of
The dependent
activities,
intrahousehold relationships and pension income receipt,
standard individual, household, and geographic characteristics.
variable used in most of our regressions
is
weekly working hours for working-age
'^As naturally expected, households that contain eligible elderly but which
smaller (a little less than 4 persons on average) and older.
much
;u-e
not throe-generations are on average
individuals.
The survey
many hours
did
to all forms of
— work
asks the fohowing question for each person of 16 years of age or more:
last
week?".
It is
"How
important to note that the working hours question relates
employment. Regular wage employment (self-employed professionals), casual wage
employment, self-employment
in agriculture
and other forms of employment and self-employment
are hence covered by the survey question.
Our
analysis will also sometimes use a
labor supply.
dummy
variable for
employment status
Again, the employment status variable refers to
exclusively regular employment.
supply response,
it
all
measure of
as a
forms of employment and not
While employment status provides a cruder description of labor
will allow us to
perform a comparison with the 1991 Population Census, which
does not contain a working hours question but does ask about present work status.
We
will also briefly
unemployment
document whether any change
or labor force participation status.
employed are asked
if
in
employment status
Individuals that report not being currently
Individuals out of the labor force are then asked
previous week.
We
We
use answers
and not
in the labor
they have been looking for work during the previous week.
to these two questions in order to classify people as employed, unemployed,
force.
a change in
reflect
why they
did not look for work in the
define discouraged workers as the subset of individuals out of the labor force
that did not look for work because they thought there was "no jobs or work available."
Table
1
presents
means and standard deviations of the main
individuals between the ages of 16 and 50
who
live in
variables of interest for African
three generation households.
Because our
identification of the pension impact will eventually rely on the presence or not of age-eligible persons
in the household,
we
also present these
means and standards deviations separately
that contain at least one elderly person that
is
age-eligible
households that do not.
10
(woman
over 60,
for
man
households
over 65) and
We
want to highlight several interesting
facts in
Table
First,
1.
sample are employed. The employment rate among men
this
among women
is
21%. Average working hours,
are not employed,
8%
are
6.3, are also
only
23%
of the individuals in
26%
while the employment rate
very low.
Of the remaining 77% that
is
unemployed and 21% are discouraged, leaving roughly 48% out of the
labor force and not discouraged. This dramatically low employment and high discouragement and
unemployment
rates
among prime-age African
individuals
a characteristic of the South African
is
labor markets that has already been well-documented in previous work.
Second, on background characteristics, the differences between eligible and ineligible households
For example, there
are small.
is
only limited differences in education, or in the geographical
distribution across rural and urban areas. Age-eligible households appear a
versus
(9.1
8.5).-''*
One
noticeable difference in Table
households report being sick more often.
among
1
is
One might argue
little
bigger on average
that prime-age individuals in eligible
that sickness
is
in fact a
luxury good
those African households and that this might be looked at as an outcome of the social
pension.
Third, on employment status and working hours, the difference between the two household
types
is
dramatic.
Raw
differences in
The econometric work below
employment
will translate these
rates are
more than three percentage
raw differences
points.
into estimates of the effect of the
pension.
Table
and
1
shows some other interesting
.34 eligible
men,
facts.
The average
for a total of 1.24 eligible
eligible
household has
.9 eligible
women
members. Most of the pension income, therefore,
comes through a woman and many households have more than one pensioner. Pension income, in
'""a similar exercise performed on all prime-age individuals, not only those living in 3-generation households produces dramatic differences on such vaiiables. This underlines the importance of focusing on 3-generation households
only.
11
these households also accounts for
more than a quarter
of the total household income. This allows
us to stress one more time the generosity of the social pension program.
Basic Results
4
We
begin in Table 2 by comparing the labor supply of working-age individuals that
eligible elderly relative to those that
do not. Each regression
the pension variable, a quartic in individual age, a
8th grade, 14 province dummies, a rural area
dummy, a female dummy, household
16
and
and 21 and 22 and
18, 19
for
of household
whether the individual completed
area
dummy, a metropolitan
members between
24.^^ Table 2 considers the effect for
16 and 50 years old. In this table and
with age-
in Table 2 includes, in addition to
dummy, an urban
number
size,
dummy
live
and
5,
6
area
and
15,
both men and women between
the tables that follow, standard errors are corrected to
all
allow for correlation in outcomes within household clusters.
Columns
employment
(1) to (3)
status.
use working hours as the dependent variable while columns (4) to (6) use
Columns
continuous pension income.
(1)
We
and
estimate the basic
(4)
find that
OLS
more pension income
regressions of labor supply on
significantly reduces
both working
hours and employment rates.
The simple OLS
results,
comes from the age of the
however, are not only exploiting the variation
elders.
biased by endogenous take up or
means
test
is
By using information on
eligibility.
low but some elderly do
pension are different from those
who do
fail
The
in
pension receipt that
actual pension receipt, they
may be
take up rates are high, but not complete, and the
to get the pension.
not, the
OLS
If
those
who
actually receive the
estimate will be biased. To address
this,
we
'^The completion of matric (10th grade) is another important determinant of employment and unemployment
among South African men and women. Our results are unaffected if we use the completion of 10th grade
instead of the completion of 8th grade as a control for educational attainment.
probabilities
12
examine the
(5),
we
effect of
pension
eligibility rather
than of actual pension receipt. In columns
directly use the age-eligibility criterion as the source of variation.
We
(2)
and
find a similar negative
compared
labor supply response in the households that have at least one age-eligible person
to the
households that do not.
This ehgibihty measure cannot easily be transformed into a meaningful economic measure (such
To ease economic
as an elasticity with respect to pension benefits).
interpretation,
we instrument
the amount of pension benefits received by a household with the number of age-eligible
age-eligible
women
in that household.
Columns
(3)
and
(6)
men and
contain this specification where
we
instrument the continuous pension income variable with the number of females over 60 years old
and the number of males over 65 years old
with columns
the
number
coefficient
and
(3)
(6) (not
of age-eligible
on number of
and one cannot
in the household.
men
women
over 60 years and
number
reject the null hypothesis that these
coefficients in
columns
(1)
and
(4).
large are these effects?
of
two
men and women have
columns
men
over 65 years old are very similar
coefficients are the
similar take-up rates.
(3)
and
The
(6) are
same
at standard
The instrumental
even more negative than
Each extra hundred rand of pension income reduces
weekly labor supply of prime-age individuals by about
How
women and
are very significant determinants of monthly pension income.
variable (IV) coefficients on pension receipt in
OLS
first-stage regressions associated
reported here) show that both the number of age-eligible
confidence levels and hence that the
the
The
For simplicity,
let's
1.7 hours.
^^
assume that the pension
is
split across all
working age household members equally.'^ Since there are 4.7 working age people in the average
One
implication
verified in the
is
that household income net of the pension declines when pension income increases. This can be
level data by studying the effect of pension income on non-pension total household income.
household
In the IV specification, we found that non-pension income goes down by about 1.05 rand for each extra rand of
pension money. Note that Jensen (1998) explores another channel, the decline in remittance income, through whicli
the social pension can affect non-pension income.
'^We
will see later
that equal sharing
among
all
prime-age individuals does not occur
13
in practice.
household, the coefficient of —17.07 in Table 2 suggests that a 1000 rand change in individual
income reduces hours worked by —17.07 *4.7. Average individual income (computed as household
income divided by number of working age people
in the household)
272 rands. Average hours,
is
conditional on working, equals 41.4.^^ Scaling by these gives us an elasticity of hours to income of
— 17.07*4.7*^^ =
— .55. These
1999 for
—.53. For employment,
elasticities are large if
US numbers).
the elasticities would
If
we can compute a
viewed as pure income
similar elasticity of —.099*4.7* ^|^
effects (see
we assumed that the pension were
become even more
the very low employment rates in the
negative. ^^
first place.
One
6.
We
will
show
need not represent a pure income
Such low employment rates make the marginal
An
alternative interpretation
in that section that the elasticity of labor
effect in
likely
is
supply to the pension
bargaining models of household resource allocation.
Table 3 presents results from these regressions estimated separately on prime-age African
and women.
We
employment
rates
find that
more pension income
among prime-age
hours for women, although the
Moreover, there
B, column
(4).
is
males.
significantly reduces
More pension income
effect is smaller in
is
for
men
with
(-.01
less
(2)
and
(5),
scale on hours conditional on working becau.se
working
versus -.015).
no apparent adjustment of female labor supply on the extensive margin
Looking at columns
men
both working hours and
also associated
magnitude than
in
Panel
the only labor supply variable that does not appear
to be significantly affected by the presence of eligible elderly
'*We
is
Men and Women
Effect on
4.1
more household members,
reason for the large magnitude
return to search quite small, lowering in effect the cost of leisure.
discussed in Section
Imbens, Rubin and Sacerdote
over
split
=
we
is
again female employment status.
will consider the effect
on the work-not-work decision
separately.
'
that
How reasonable
women re.spond
is
the Eissumption that working-age people receive the
less,
suggesting that
improved the anthropometric status of
men
girls
full
pension income''
under
5,
suggesting that some of
14
We
will see
below
show that the social pension
the pension income is spent on children.
get a disproportionate share. Duflo (1999)
While the point estimate
in
columns
(3)
and
(6),
Let's separately
(and assuming
— .66
front,
for
we
negative,
compute
13.27
not statistically significant. In our preferred specification
elasticities for
* 4.7 *
find elasticities of
As
it is
the effect on hours worked
men and women
men and
respectively.
in
is
much
is
larger for
^| =
—.43 for
-.201*4.7*^ =
elasticities.
This
supply (43.7 versus 39.6 for hours and
men
is
is
women on
-.98
for
the hours dimension.
the
we
scale
employment
for
women,
by average employment and
because males and females are quite similar
in their labor
versus .26 for employment). These numbers suggest that
.21
not solely due to their greater employment but likely comes from
money.
upon
studied in more details the nature of the drop in employment for men.
asked whether the "missing" working
men have
they have dropped out of the labor force.
We
entered a phase of
request,
More
we have
specifically,
unemployment
or
we found no
also
we
whether
found no difference in unemployment probabilities
and non-eligible households. Rather, the missing working men appear to have
the labor force. .Moreover,
=
men. Also, note that the difference
In regressions not reported here but available from the authors
eligible
On
* 4.7 * ^^-y
men and -.023*4.7*^ = -.14
persists even after
their receiving a larger fraction of the pension
between
women(1.3).
for
earn similar incomes), we find an elasticity of —22.48
magnitude between men and women
the greater response by
than
(2.2)
both groups. Using calculations similar to those above
before, these elasticities are quite large for
hours worked to form
men
left
sign that the social pension increases the probability of
discouragement.'^'^
discouragement variable truly captures people's pessimistic beliefs about their chance of finding a job, there
no reason to think that the social pension should affect the discouragement rate. In fact, if the pension has any
impact on labor maiket conditions, one would expect such an impact to be positive and lead to the creation of new
^°If the
is
working opportunities.
15
•
'
Possible Confounding Effects
5
There are several reasons to be cautious
leisure of
income increased
in interpreting these results as
working age people
m
showing that the pension
In this section,
the household.
we
investigate
other possible interpretations.
Household
Direct Effect of the Presence of Elderly in a
5.1
A
primary concern
ferent
is
households are systematically difthat individuals living in pension-eligible
from individuals
in ehgible
living in pension-ineligible households.
For example, the prime-age livmg
in non-eligible households. Furhouseholds are slightly younger than their counterparts
average larger.
thermore, pension eligible households are on
conceivable that prune age
It is
work, or
qualified for work, less willing to look for
living with older individuals are less
other
way
are biased:
We
less likely to find
we would
work.
If this
m
were the case, then our estimate of the pension's
men
some
effects
unobserved differences.
attribute to the pension the effect of these
problem.
take several approaches to address this
First,
we
exploit the non-linearity in
from these
to better isolate the pension effect
pension receipt as a function of the elder's age
a specific form for these non-lineanties:
confounding factors. The pension program rules predict
the presence of a
have large
woman
effects.
age thresholds
if
older than sixty years old or of a
There are no obvious reasons
man
older than sixty-five years old should
at these
to expect such specific non-lineanties
what our estimates are capturing
is
two
a general impact of the presence of elderly
members.
people on the labor supply of younger household
Table 4 presents the results of this exercise
a three-generation household.
We examine how
for
in
our sample of prime-age individuals living
working hours
16
for that
group are affected by the
presence of elderly in different age groups. To start,
labor supply of living with eligible vs.
that equals
We
there
1 if
a
is
we simply
We
non-eligible elderly.
woman between
50 and 60 or a
contrast the impact on prime-age
construct a
man between
dummy
new
variable
50 and 65 in the household.
find that the presence of a non-eligible elderly in the household has neither a statistically nor
On
an economically significant impact on prime-age working hours.
we have
the other hand, as
already demonstrated before, living with an eligible elderly has a dramatic effect on working hours.
The
rest of this table further refines this finding.
Column
includes as regressors, in addition to the usual
(2)
in each of the following four
clearly
show a negative
list
of controls, the
number
age categories: 50-55, 55-60, 60-65, and 65 and over.
effect of
The
of people
coefficients
the presence of elderly between 60 and 65 and an even more
negative effect of the presence of elderly over 65 years old.
On
the other hand, the presence of
between 50 and 55 and 55 and 60 seems to have neither an economically nor a
elderly people
impact on working hours among prime-age individuals. Moreover, the
statistically significant
bottom of Table 4
statistics at the
test
clearly reject that the pre-eligibility coefficients are equal to the
post-eligibility coefficients.
While these
elder's age has
results provide
some compelling evidence, one may
an independent, non-linear
effect.
them
The second
first
strategy
glance,
This
may
we use
tlii.s
list
Table 4
in
tries to deal directly
input in
to
home
both care
care, they
for
likely to
have
care.^^
The
SALDRU
sick or injured over the past
men reduce their work more than women do." If
ought to reduce their work hours more. One could liowever claim
the elderly and work, while men will either work or take care of the elderlv.
story seems inconsistent with the fact that
women provide the main
that women are expected
home
with this problem.
any household member that has been
effect of the
cause prime-age individuals
to reduce their labor force participation in order to provide
survey asks respondents to
^'At
worry that the
Most notably, the very old are more
health problems and to require some assistance at home.
living with
still
17
two weeks,
''including people
who have some form
of
permanent
injury, disabihty, or ailment."
survey then asks respondents to report the nature of the main and second
We
ailment.
data
set,
problem.
A
17%
that about
we
To do
we
regressed for the entire
dummy
variables for age
SALDRU
is
60 and the people over 60.
Hence,
occur before the age of pension
men and women), we found
being sick for the people between 55 and
any age discontinuity exists
if
as the
While people between
elderly."-
true for both
statistically significant difference in the probability of
same non-linearity
data the probability of having some
and sex groups of
50 and 55 appeared healthier than people over 55 (this
We
of health
in that age group. In regressions
investigated whether health problems display the
this,
health problem on 10
no
SALDRU
very high fraction of those (22%) suffer from high blood pressure. Asthma, diabetes
not reported here,
rule.
some kind
of all people over 50 years old report
and rheumatic heart disease are other important forms of ailment
pension
illness, disability, or
are particularly interested in the health status of the elderly. In the entire
we found
The
in
health status,
appears to
it
eligibility.
then included the number of elders in the household that report health problems into our
employment regression
(each sick elder
is
in
column
(3) of
Table
4.
While the
coefficient
on health problems
associated with an hour less of work) and marginally significant,
the pension coefficients.
The same
discontinuity pattern as in colunui (2)
is
it
is
negative
does not affect
observed, suggesting
that health status of the elderly does not drive our results.
Columns
break
down
(4)
the
and
(5) replicate
the specifications in columns
number of people over 65
and number of people over
70.
The
into
results are
are:
woman between
50 and 55,
and
(3) respectively
but further
two groups: number of people between 65 and 70
imchanged. The coefficients on
belcjw the eligibility threshold are not statistically significant
^"The 10 dummies
(2)
woman between
over 70, and the erjuivalpnt for men.
18
from
55 and CO,
0.
The
all
the age categories
coefficients
woman between
on
60 and 65,
all
the
women
age categories above the
Our
threshold are statistically negative. Moreover, the test statistics
show that we can
in the notes of Table 4
and above
eligibility
reject the hypothesis of equality of the coefficients
below
eligibility threshold.
third strategy exploits regional differences in
how
the pension program
certain areas, authorities deviated from the eligibility rule that
women
is
are eligible at an earlier age
than men, a rule they viewed as "unfair". They de facto extended the pension to
and 65 years
old.^'^ If
men between
60
our results are in fact due to the pension, we would expect that in the regions
that deviated from the
official rule,
the
number
of
men between
60 and 65 should affect labor supply.
As one might expect from the informal nature of the extension, we cannot
on which areas extended, but we can compute a proxy
the fraction of households with
that report receiving
implemented. In
men between
in
our data.
We
get administrative data
can compute, by province,
60 and 65 years old and no other age eligible elderly
some pension income. ^^ This
fraction ranges from
in the
most obedient
province to .66 in the least obedient province.
In
column
men between
(6),
we use
this proxy.
60 and 65 in the household.
by sex categories. The results are
different
of
from
We
interact this fraction with a
We
striking.
further break
None
down
all
dummy
for the
from the
eligibility rule) is
from the
eligibility rule
old
of
the age groups of column (4)
of the pre-eligibility coefficients are statistically
zero. All the post 65 years old coefficients are significantly negative.
number of men between 60 and 65 years
number
(i.e., its
not statistically different from
effect in the
0.
The
direct effect
provinces that do not deviate
interaction term between deviation
and number of men between 60 and 65 years old
Finally, 10 out of the 12 tests statistics reject the
The
is
negative and significant.
assumption of equality between
pre-eligibility
^ As Case and Deaton
(1998) report, the age differential in pension eligibility is technically iinconstitutional ind
under revision at the central government level. Certain local authorities may have already gone ahead with age
equalization by 1993.
Remember that we do not observe pension income at the individual level but only at the household level.
'''
19
coefRcient
we
and
post-eligibility coefficient.
control for the
number
Column
(7)
shows that the exact same results hold once
of elderly with health problems in the household. These results suggest
that the extension did in fact correlates with the pension's estimated effect, bolstering the case
that
we
Our
are not capturing spurious effects of age.
final
attempt to account
a completely different approach.
eligibility prior to the
we turn
for
confounding factors cissociated with age of the elderly takes
We
examine the relationship between employment and pension
bulk of the reform of the social pension program in South Africa. To do
to the 1991 Population Census, another cross-sectional household survey that
out prior to the major extension of the social pension to African households.
earlier,
this,
was carried
As we mentioned
while the process of racial equalization of the social pension has been under way since the
early 90s in South Africa,
benefits levels achieved,
it is
only after 1992 that the means-tests were unified, the racial parity in
and new technologies introduced to improve the benefits delivery system.
it
was administered
we
are in fact finding
Thus, while the 1991 Census was not exactly administered prior to the reform,
at a time
when
the pension was far less generous and accessible to Africans. If
a causal effect of the pension, the effect of living with eligible elderly ought to be
much
smaller in
the 1991 data.
Unfortunately, constructing a sample comparable to the
no easy
task.
The
first
generation households.
major hurdle
is
Because the variable detailing relationship
insure that son/daughter, parent and
partner) live in the
same household.
three kinds of family relatives.
extract from the Census
is
that the Census data does not allow us to isolate three-
cannot insure that son/daughter, parent and grandparent
we can
SALDRU
We
live in
to
the
head
all
rather crude,
same household.
some other family members
therefore select
is
we
Instead,
(other than the parent's
the households that include these
Not surprisingly, these households are on average younger than
20
we
the households
generations.
We
selected in the
SALDRU
data where we can
select
on the presence of three
therefore decide to operate an additional cut of the data
and
force the
majcimum
age in the selected households to be over 50 years old. To ensure comparability, we redefine our
SALDRU
extract along exactly similar selection rules.
Another major limitation of the Census data
The only labor supply
variable
we observe
that
does not ask questions on working hours.
it
employment
is
on employment status were only important
is
^^
for
men, we
status.
restrict
Given that our previous findings
both extracts to prime-age males
only.
Simple summary
statistics testify to the fact that the
with household size (9.07 in
educational attainment.
Census data
in
SALDRU
versus 9.16 in Census), average age (26.4 versus 26.9) and
The employment
(.26 versus .33).
This
two samples are very similar, such as
may
rate in the
SALDRU
reflect deteriorating
data
is
however lower than
economic conditions during
in the
this
time
South Africa.
Table 5 compares the impact of the presence of an age-eligible elderly in the two samples,
controlling for the usual vectors of individual
area type.
in
We
find
characteristics as well as indicators for
no evidence that the large negative employment
effects
found in the 1993 data
pension eligible households (-6.8%) are present in the 1991 data. While there
effect associated
that
with the presence of an age eligible elderly in 1991 (which
some limited pension program was already
size as
In
the 1993
summary, the
'^A smaller issue
these 4 regions
in place then), the effect
is
is
some negative
not surprising given
is less
than a quarter
in
effect.
is
results in this section provide multiple pieces of evidence all of
which suggest
that the 1991 Census does not survey people living in the then four independent homelailds
(Transkei, Bophuthatswana, Venda,
in
and household
and
SALDRU
SALDRU sample.
Ciskei) while the
when we form our comparable
21
survey does.
We
therefore exclude people living
that
we
are in fact identifying a causal effect of the pension
and not an independent
effect associated
with hving with elderly people. ^^
5.2
Is
the Labor Supply Response Real?
We now tackle
a different concern. Taking as given that
of the pension,
we ask what
effect
we
we
are in fact identifying
are identifying. Perhaps
we
some causal
effect
are capturing other behavioral
changes induced by the program, rather than a decrease in labor supply.
5.2.1
One
Change
in
Type
of
Work
behavioral response might be a change in the nature of work. Access to pension income might
allow prime-age individuals to change the type of work they do. Prime-age individuals might
be able to leave their unrewarding regular jobs and instead start working locally
setting, or start
working on the family farm. For example, the pension income may
in
now
a more casual
literally
provide
seed capital for farm production at home.
Remember, however, that our
(casual, self) forms of
definition of
employment.
One might
precisely or frequently reported than regular
employment might appear
to
employment includes both regular and non-regular
be a drop
still
worry that,
employment.
in total
say, casual
employment
is
less
Thus, a substitution towards casual
employment
if
some part
of that substitution
is
not reported.
We
present two tests to assess the importance of such a potential source of bias.
We
start
by
^^Another robustness check of our finding would consist in assessing whether the size of the labor supply response
depends on the size of the household. Indeed, if the labor supply response truly represents a sharing of the pension
money between the various household members, one would expect that response to be smaller as the pension has
to be split among a larger number of household members. In regressions not reported here but available from the
authors, we have found that the labor supply respon.se to the marginal rand of pension money does indeed decrease
as the total size of the household increases.
22
focusing on prime-age males for
A
of Table
we decompose
6,
employment and a change
this
whom we
have observed a drop in employment
drop into a change
in self-employment.
We
in regular
In Panel
rate.
employment, a change
find that both regular
in Ccisual
and casual employment
rates are negatively affected by the presence of social pension income, while the self-employment
rate
unchanged. This contrasts with the expected rise in casual or
is
The
have expected under the changing type of work hypothesis.
employment
In Panels
self
employment one might
significant reduction in Ccisual
especially hard to reconcile with this hypothesis.
is
B and
C,
we
investigate
more thoroughly the seed
capital story.
The SALDRU survey
reports information at the household level both on agricultural activities and livestock. In Panel
of Table
is
more
6,
we
collapse our original data set at the household level
intense
among
activity
the pension eligible households, controlling for household characteristics and
for the vector of geographic
dummy
and ask whether farming
B
dummies.
We
consider three different measures of farming activity: a
variable for the production of any kind of crop in the household, the
number
of liters of milk
obtained from the herd over the previous week, and the number of eggs obtained from the poultry
over the previous week.
eligible households.
The
We
also investigate
whether the farming capital stock
specific capital variables
we consider
is
are the value of tractors
equipment as well as the value of mechanized farm equipment and pumps.
We
among
higher
the
and farming
find
no evidence
that either farming outputs or farming inputs are higher in the eligible households.
This result
holds true
These
when we
restrict
our sample to the subset of households living in rural areas (Panel C).
results are inconsistent with the idea that prime-age individuals are leaving regular factory
jobs to start working on the family farm.
The
results in Table 6 contradict the idea that the
more than a
shift
measured reduction
towards more poorly measured forms of employment.
23
in
employment
is
nothing
Migration
5.2.2
Another behavioral response could be migration. The pension may induce the more employable
prime-age individuals to move out and the
less
employable ones to move
in.
The pension may
reduce pressure on the prime-age to financially care for the elderly and increase the pressure on the
elderly to financially care for the prime-age.
The measured
labor supply response might then reflect
a change in household composition rather than a change in labor supply of any given individual
While such a reshuffling
in the household."^
behavioral response in
itself, it is
in
household composition would be an interesting
important to separate
it
from an increase
in individual leisure
time.
To
investigate this issue,
we
first
examine the education of individuals
living in pension-eligible
households. If selective migration of the least employable were driving our results,
the pension eligible households to show lower levels of education."* In Table
on the same
both the
dummy
list
7,
we
we might expect
regress education
of regressors as before (expect of course for educational attainment).
OLS and IV
for at least
specifications.
The
three dependent education variables
We
present
we consider
are a
completing 4th grade, at least completing 8th grade, and at least passing the
Matric (which corresponds roughly to finishing high school). The results suggest that higher pension
income
is
insignificantly (though negatively for the
related to education.
The magnitude
appears to be no change
We now more
in
two higher educational attainment variables)
of the coefficients
is
quite small.
On
education at
least, there
household composition.
directly study differences in migration patterns between eligible
and non-eligible
^^The short time frame between the bulk of the pension extension and the survey may be a practical hindrance to
It is only after 1992 that the most dramatic extension of the social pension to African
households took place. This would mean only about a year for the changes in household composition to take place.
^'Simple regressions of employment on education show that, as might have been expected, education strongly
predicts employment with the better educated more likely to be employed.
the migration interpretation.
24
households.
There are two major types of migration decisions worth considering
may
working-age person
formally be
member
of a household but decide to spend
more time
time away from the household once the pension income becomes available. ^^ While the
survey reports employment status for
reports are less accurate or
the household. This
is
all
for the
especially problematic
employable people to migrate out part of the
we
or less
SALDRU
household members, one might be concerned that those
maybe missing
In the first two columns Table 8,
First, a
here.
if
people that spend a significant time away from
the presence of a pension income allows the
more
year.'^'^
see that the average
number
of
months spent away from
the household by working age individuals, while negatively affected by pension income in the
OLS
The
point
regression (column
1),
does not appear to be statistically related to
it
in the
IV
regression.
estimates are also quite small, certainly not of the magnitude required to generate the observed
drop
in labor supply.
The second type
The
SALDRU
of migration decision consists of formally leaving or joining a given household.
data set allows us to assess whether the household composition was affected by the
presence of age-eligible elderly. First, we use as a dependent variable a
1 if
(3)
and
(4) of
Table
8.
The
coefficients are insignificant
not of the magnitude required to generate our findings (that
employment, such small
While
in
migration would barely
this indicates that there
of out migration.
"
variable that equals
the prime-age individual reports having "move(d) here during the past 5 years."
columns
in
dummy
The nature
is little in
aff"ect
migration,
pool of resources are considered
In fact, our data set indicates that
household, suggesting that this
is
is, if all
results are
also extremely small, again
the in migrants were at zero
average employment in the household).
we
still
need to examine the possibility
of the data precludes us from directly tracking leavers, but
it
does
month in the household and contribute to or share in a
members of the household.
as much as 85 % of the people in our sample spend all their time in the
All individuals that spent at least 15 days of the last
common
and
The
not a major issue in this sample.
25
provide us with family
size.
Since
we have found
that in-migration
is
not changing, then for
migration to explain our results, family size ought to be decreasing significantly with the pension.
In columns (5) and
(6),
we
regress the
are again statistically insignificant
number
and tiny
of prime-age people on pension income.
in
The
results
magnitude, as before not large enough to explain
the significant labor supply response. Together with the in migration findings, these results suggest
that selective migration does not drive our results. While migration
find that the pension did not aff'ect
As a whole, the
it,
is
an interesting outcome, we
perhaps because of the recency of the program.
findings in this section strongly suggest that the pension did have a causal effect
and that we are not merely measuring an independent
effect of elder age.
They
also suggest that
the pension led to a drop in labor supply and did not only change the type of work performed or
the composition of the family.
Models of Collective Labor Supply
6
Common
6.1
Our
Preference (Income Pooling) Models
results so far provide
some evidence of a
redistribution of the pension towards prime-age
workers in the household. In this section, we push the analysis a step further and ask whether the
South African experiment can teach us more about how resources are allocated and collective labor
supply decisions
made
within these extended families.
There are several prominent theories about resource allocation within households. Probably
the simplest one assumes that households can be described as jointly maximizing a single utility
function (Samuelson, 1956).'^^ Consider the following setup.
A
family
is
composed of
2 individuals.
^'The common preference approach can be motivated either through the assumption of a family consensus, such
Samuelson (1956), or through the assumption of altruistic behavior, such as in Becker's "rotten kid" theorem
(Becker, 1974 and 1981). In this paper, wc will not try to distinguish between these two models
as in
26
1
and
2.
Individual
1 is
the working age person in the household. Individual 2
household. Both individual
Individual
working,
levels,
1.
1
/i,
and individual 2 receive some non-labor income,
1
can also allocate some of his time
will
be devoted to leisure.
C\ and C2 respectively.
T
to
working
Both individuals
1
for a
and
y-[
wage of w\
.
is
the elderly in the
and
j/2
The time spent not
2 also have private
Let's normalize the price of one unit of private
T—
Under the common preference approach, C\, C2 and working hours
respectively.
li
consumption
consumption to
are derived from
the maximization of a single utility function for the whole family under the whole family budget
constraint:
max
f/(Ci,Co,/i)
wi
s.t.
One simple way
many hours
to think
to work,
li
+
Ci
+
62
<wiT + yi+y2
about this maximization
is
that the family collectively decides
and then completely pools labor and non-labor income to
how
jointly purchase
individual consumption.
A
central result from the
common
preference approach
is
that
money
is
money. Whoever gets
the marginal dollar of non-labor income will affect neither the choice of private consumption levels
nor
leisure.
Mathematically, this means that
^ = ^.
It is
important to note the
this result. It holds even in the presence of differential altruism
between
generalitj' of
different individual family
members.
A
second feature of these models
viewed as a pure income
about the
effect in
utility function,
is
that the consumption responses to non-labor income can be
the joint utility function L'(Ci,C2,0-
With simple assumptions
one can translate this labor supply response into the individual income
effect.
27
A
uals
As
third interesting feature of these
who
get
models
relates to
who
more resources are those who receive the greatest weight
this function
is
resources. Individ-
in the joint utility function.
a black box, greater weight could be interpreted in several ways. Altruism be-
tween the family members may be such that one specific individual
Alternatively, greater weight
6.2
more
gets allocated
may
arise
"cared for" more by others.
is
because this specific individual sets the agenda.
Bargaining Models
Another strand of the literature on intra-household redistribution of resources
that families can be reduced to a single optimizing agent.
members have
distinct preferences
Most often the bargaining
the different parties.
(1981), or
consists of a Pareto efficient process, such as a
all
Manser and Brown
Several authors, such as
Lundberg and Pollak
assumes that household
it
and studies how bargaining between them
Chiappori (1992) produces a
allocates resources.
Nash bargaining between
(1980),
(1993), have developed cooperative
intra-household resource allocation.
includes
Instead,
rejects the idea
McElroy and Horney
Nash bargaining models of
far
more general model that
Pareto-efficient bargaining models.
In our setup,
we model such bargaining
as a
Nash bargain.
So, decisions will be a solution to
the following program:
max
[U, [h
wi
s.t.
li
,
+
C,
)
C\
- U,f
+
-
[f/2(C2)
C2 < wi (T -
Uo]
/i)
+
1-a
yi
+1/2
Here U\ and U2 respectively represent the threat points of individual
threat points are key to the model.
household members. The higher
is
They capture
someone's
1
and individual
These
the "threat" the individual has over other
utility in the threat point, the
28
2.
higher
is
their utility
in the
a
is
Nash bargaining
solution,
larger, individual 1 captures
Many
a captures
the relative bargaining power of individual
more of the
surplus.
of the bargaining models
assume that threat points are determined by the
members can achieve
that household
in case of household dissolution (see e.g. McElroy, 1990). In
go their separate ways multiplied by a factor 5
1
=
=
SUoiyo) and Ui
when by
himself.
5Ui{r ,wi{T —
l*)
it
matters
will influence the
+ yi)
<
1,
where
1
—
5 of
/* is
the optimal leisure choice for individual
who
gets the money.
First, the strong fungibility result
The income
why
1
—a
go to individual
different people's
Second,
effect.
As
A
it is still
money
is
2.
from before no longer
controlled by the various household
bargained outcome. Non-labor income to individual
each person's resources are pooled.
and a fraction
individuals
which indicates the cost of separation. So
consumption pattern than non-labor income to individual
that
if
'^'^
Several results follow in this model.
holds:
When
level of utility
the context of this literature, the threat point in our simple set up can equal the utility
t/2
1.
fraction
2.
a
A
1
may
result in a different
simple way to view this model
of these resources go to individual
Since only a fraction of the resources are pooled,
1
is
it is
is
1
clear
spent diiferently.
true that the labor supply response to non-labor income y,
far as individual
members
concerned, he gets 6 of
own income and a
is
a pure income
of the shared income.
His labor supply adjusts exactly because of this rise in income.
Finally, the bargaining
resources.
The higher
model here
offers
an equally vague answer
as to
who
captures more
^'^
the bargaining power, the greater the resources that are received.
Other models assume that threat points Eire internal to the "marriage" ajid are determined by the level of utility
members can achieve if they stop cooperating with each other when making resource allocation
decisions (see, e.g. Lundberg and Pollak, 1993).
For simplicity, this simple set up abstracts from altruism. The addition of altruism would alter the interpretation
of "bargaining power".
If individual 2 cares about 1, then 1 will receive more resources, holding constant the
bargaining power, q. This should be kept in mind in interpreting the results below.
^^
that household
29
Endogenous Bargaining Power
6.3
A
complication to the standard bargaining models would be to allow the threat point to depend
on the labor supply
choice.
In our setup, the simplest
allow a, the bargaining power of individual
then have two stages. In the
first stage,
1,
to
way
to incorporate this effect
depend on the
individual
1
leisure choice
/j,
stage, the
Nash
say a{li).
implication of that model would closely match the implication of the exogenous threat model
but with one important wrinkle. The labor supply decision
how
The game would
would choose h. In the second
solution would arise but with person Ts bargaining power being a function of
The
li.
would be to
is
now
subsidized or taxed depending on_
labor supply affects bargaining power. Suppose increasing hours worked increases bargaining
power.
Then
This could happen simply because working people have greater voice in the household.
individuals actually face a subsidy to working, since the
more they work, the bigger part
of
the total pie they capture. This effect means that, relative to the exogenous bargaining power case,
individuals will adjust their labor supply less in response to the pension.
sufficiently strong, individuals
income
effect
may be
may work
In fact,
harder in response to the pension.
totally offset as individuals
if
the effect
is
In other words, the
attempt to work hard to capture more of pension
income.
Alternatively, suppose that increasing hours worked decreases bargaining power.
occur, for example,
if
those
who
This might
are around the house a lot can effectively capture a lot of bargaining
power. In this case, one gets a tax to work, not a subsidy to work. Working harder reduces one's
share of the
the income
pie.
When
effect,
the pension increases the pie, one
but also because one wants to stay
In other words, the income effect
is
may work
home
hard, not only because of
to capture a greater fraction of the pie.
now complemented by an
30
less
additional "pie-grabbing" effect.
Relation to Results So Far
6.4
Both common preference and bargaining models are
consistent with our findings so
placed into the
Men
shock.
income
Labor supply decreases because of
split.
In the simplest bargaining models,
effect.
some
If
pie.
labor supply, then the drop in labor supply
we
the substantial labor supply responses
of the pension
money
is
this exercise demonstrates,
results.
both
we observe overestimates the pure income
observe, this last case seems
more
likely,
effect.
common
model, as we have noted, the labor supply response should not depend on
through before
perfectly, the
it
enters the
money
common
is
completely pooled,
pool.
In
effect.
Given
though
this of
all
these cases,
'^''
preference and bargaining models can explain
Yet one key prediction allows them to be separated. Under the
the pension income. Since the
partly to totally
hand bargaining power decreases with
decrease their labor supply more because of their greater bargaining power.
many
placed in the
is
In other words, in this case our estimated elasticities
course depends purely on prior beliefs about the magnitude of the income
As
is
income
this positive
bargaining power increases with labor supply, this income effect
actually understate the true income effect. If on the other
men
be made
Labor supply of the working-age household members decrease again because of an
pool.
by a desire to capture more of the
offset
all
decrease their labor supply more because they are given greater weight in the household
utility function.
communal
they can
Under the common preference model, the pension money
far.
communal pool and
sufficiently flexible that
it
who
doesn't matter
In the bargaining model, since
magnitude of the labor supply response may
well
common
exactly
preference
is
whose hands
money
is
receiving
it
passes
not pooled
depend on who receives the pension.
Previous tests of complete pooling typically do so by asking whether an individual's consumption
^"^Or, as noted above, because other family members show greater altruisui towards prime-age
prime-age women.
.31
men than towards
own income.
correlates with her
assumption of such a
test
Perfect pooling would suggest that
it
should not.
The working
the exogeneity of income. Yet our results so far clearly highlight the
is
limitation of this assumption. '^^ If labor supply responds to other people's income, whatever income
people earn
may
itself result
pooling hypothesis even
from redistribution.
when
It is
easy to see that one might then reject the
true or accept pooling even
when
untrue.
Distribution of Effect
7
Test of Income Pooling
7.1
The
social pension
mon
program
in
South Africa provides a rather unique experiment to separate com-
preference and bargaining models of the family.
while in theory means-tested,
African households.
is
in practice
As we mentioned
earlier, the social pension,
mainly a lump-sum transfer
for the relatively
poor
Hence, the pension transfer, especially when instrumented for by the pres-
ence of age-eligible elderly, does not depend on earned income or other possible choice variables
in the household resource allocation decision.
One can
therefore test for income pooling by asking
whether pension transfers made to elderly women have the same
than pension transfers made
to elderly
That
effect
(in
money going
to elderly
women may
on prime-age labor supply than pension income going to elderly men.
table showed that the co(>lficients on
be systematically larger
on prime-age labor supply
men, holding family composition constant.
In fact, the findings in Table 4 already suggest that pension
have a larger negative
effect
number of women above
absolute value) than the coefficients on
eligibility
number
threshold tended to
of meii al)c)ve eligibility
'^Another key problem is that different individuals' incomes may have different time series properties.
'^While unearned income might be less subject to this endogeneity concern, it is often unclear what constitutes the
bulk of unearned income. But to the extent tliat it is correlated with earned income flows or household expenditures,
it would also be endogenous to the resource allocation process.
32
Column
threshold.
hours of both
(1)
of Table 9 confirms this earlier finding. In that column,
men and women on
we
regress working
the standard set of geographic, individual, and family controls.
We add as regressors the number of women above 60 years old and the number of men above
The
old.
coefficient
on number of
number of eligible men.
number
error in the
It is
women
eligible
is
more than twice
important to note that such
of eligible men.
Even
if
we account
as large as the coefficient on
diff'erence is not
for the fact that
due to any measurement
some men between 60 and
65 also receive pension income in certain South African provinces (column
number
of
men
over 60 years of
over 65 years old
age."^^
of prime-age males,
(column
(3)).
is still
females,
still
number
In other words, the fact that female pension
holds once
we
the coefficient on
control for the
of male kids and
money reduces
male pension money cannot be explained away by any systematic
sex composition of the non-elderly in households with eligible
eligible
(2)),
only half the size of the coefficient on number of
Finally, note that this finding
number of prime-age
65 years
number
women
number
of female kids
labor supply more than
diflference in the
women compared
number and
to households with
men.
This
first
finding in Table 9 appears inconsistent with pooling of resources within the household
and builds some preliminary support against a common preference model of collective labor supply
here.
The marginal
dollar of pension
more than the marginal
finding
is
not conclusive.
income received by an elderly
dollar of pension
It
income going to an elderly
money
does not account
woman may
woman
received by an elderly
for the possibility that the
have to be distributed among a
reduces labor supply
man. However,
this first
marginal rand of pension
diff'erent set
of individuals
then the marginal dollar of pension money going to an elderly man. Of primary concern to us
^^Note that
in this
is
Table the coefficient on the interaction term between number of men between 60 to 65 and
is insignificant, though negative. The rise in standard error is due to the fact that we are
deviation from eligible rule
not separately including each of the age categories.
33
the possibility that old
women might have a
lower weighting than old
function. In such case,
and assuming that households with
than households with
eligible
the
common
men
preference model.
(a very likely event),
We
man
whether
number
(1)
over 50 years old).^*
it
women
in
a household utility
have more elderly
our finding could
still
women
be reconciled with
deal with this concern by restricting our sample to the set of
households that have exactly one elderly
(a
eligible
men
woman
(a
woman
over 50 years old) and one elderly
man
In this subset of households, the marginal rand of pension income,
comes from a female pensioner or a male pensioner
of elderly of each sex. In
on that subset of households.
column
(4)
We
find a
still
of Table
much
9,
we
will
be reallocated among a fixed
column
replicate the specification of
stronger negative labor supply response to
female pension money than to male pension money. The marginal rand of pension income going
to a female pensioner reduces labor supply by about three times as
much
as the marginal rand of
pension income going to a male pensioner. The coefficient on number of age-eligible
less precisely
men
is
however
estimated. ^^
While we have forced the number and sex composition of the elderly to be the same
restricted sample, one could further worry that the
number and sex composition
systematically varies with the sex of the pensioner.
We
have checked our data
tematic difference. In the subset of households with exactly one
50,
we found no
statistically significant differences in the
woman
number and
in this
of the non-elderly
for
any such
sys-
man
over
over 50 and one
sex of prime-age individuals
between the household with female pensioners and the households with male pensioners. However,
we found
that the households with female pensioners had slightly
more kids than the household
^^There are few households that contain two or more elderly of both sex.
''^Differences in life span provides an alternative interpretation of the between male and female pensioners. Since
women live longer (and get the pension earlier), the present value of the income shock is larger when the pensioner
While plausible, this does not seem to be the case as seen in Table 4. Even a 70+ year old female
is a woman.
pensioner results in twice the effect of a 65+ year old male pen.sioner.
34
with male pensioners. This
to be split
common
among
last difference
suggests that the marginal rand of pension
more individuals when the pensioner
slightly
is
a
woman. Hence, under
preference model, this could only lead the coefficient on eligible
absolute value than the coefficient on eligible
utility function.
This
is
men
exactly the opposite of
if
money needs
women
to
be smaller
what we
In
find.
summary, our
rand of pension income has drastically different
ply depending on whether that income
is
we ensure that these households only
differ
in
children receive any weighting in the family
results in Table 9
strongly point towards rejecting the income pooling hypothesis for African extended families.
find that a marginal
the
received by a
man
or a
effect
We
on prime-age labor sup-
woman. This
still
holds true
when
through the sex of their pensioners and not through
the number, age and sex composition of their members.
Who
7.2
Benefits
From the Old Age Pension?
Testing for income pooling
hold.
is
only
way
One might more broadly want
of resources. Is pension
money evenly
to look at
to ask
who
distributed
how money
is
distributed inside of the house-
are the biggest beneficiaries of the reallocation
among
all
the prime-age individuals in the ex-
tended family or are certain family members able to reduce their labor more than others? Table
10 answers these questions. This table estimates our standard regression of hours worked on the
pension variable, but
For the sake of space,
As the
it
now
we mostly report IV
reduces female labor supply.
Hence, prime-age
Column
(1)
to benefit less
male labor supply more than
of Table 10 reiterates that fact.
pension on female labor supply
women seem
characteristics.
specification results in this table.
results in Table 3 already suggested, the pension reduces
effect of the social
demographic
interacts the pension variable with several
is
about half of the
from the
35
social pension
effect
We
find that the
on male labor supply.
than prime-age men. Does
this effect
the
depend on the sex of the pensioners? In column
number
of eligible
women and
the
find that the presence of an additional
different effect
number
of eligible
male pensioner
on male and female labor supply.
On
effect of
we
men
in a
interact tlie female
dummy
in the household. Interestingly,
than working-age males. In other words,
the pension on prime-age males occurs only
On
we
household does not have a statistically
when
the pensioner
is
a female.
Another potential determinant of who benefits the most from the pension transfer
tional attainment.
with
the other hand, an additional female pensioner
in a household benefit working-age females relatively less
the greater
(2),
is
educa-
the one hand, one might believe that individuals with higher educational
attainment have higher outside options, which could increase their threat points when bargaining
over resources with other family members.
individuals with the lowest market wages
is
positively correlated with
On
may
the other hand, at a given level of redistribution,
up
give
market wage, we would
workers reduce their labor supply the most.
The
their job
in this case
first.
If
educational attainment
expect to see the least educated
differential effect of the
pension by education
group thus appear to be a mostly empirical question, which we investigate
we
in
columns
(4).
In column
25%
of prime-age individuals in our sample that have not completed fourth grade.''"
(.3),
isolate individuals
least
difference in labor supply response
the people that have not. All in
completed fourth grade. In column
between the people that have at
all, it
and
with very low educational attainment. There are about
We
individuals that have not completed fourth grade reduce their labor supply by about
than the individuals that have at
(3)
appears that
it is
among
least
(4),
find that
50% more
however, we find no
completed the Matric and
that the very least skilled that the
labor siipply response was the strongest, probably because these individuals face very unattractive
labor market options to start with.
That
fraction
is
roughly identical for
men and women.
36
Columns
and
(5)
(6)
ask
how
the age of the working-age individuals in the extended family
One
supply response.
affects the size of their labor
pension depresses labor supply more as working-age
clearly sees in
men
money and age
a quadratic relation between pension
One
More
precisely,
effect of
it is
we allow
for
whether the
effect of
age
age appears mostly linear and
we want
absolute or relative age that matters in intra-household
to focus
on the special position that eldest sons are
single-handedly receive more pension
money than
know from previous
men
results that older
is
age and sex. In column
we
household members.''^ The result
in
men
How
in
in the
column
(7)
resources. Here, however,
we
find that eldest
household
we already
we want
effect of
to
absolute
and by
sex,
supports the view that oldest sons are redistributed
men reduce
for the direct effect of
their labor supply
household and about 70% more than
in the
more
Clearly,
at least
household reduces his labor supply more than other
extended families. After accounting
resource distribution,
other
are allocated
members?
in the
control for the differential effect of the pension by age
and then ask whether the oldest man
more resources
other household
man
anything special about being the oldest man, beyond the
(7),
that the social
(6),
column
anecdotally believed to hold inside of their family. Does the oldest prime-age
assess whether there
(5)
old.''^
could further wonder whether
redistribution.
In
in order to assess
peaks at any point over the range of working ages. The
does not peak before 50 years
get older.
column
women
own age and sex on
by about 50% more than
in the household.
can we understand the results above in light of the bargaining models of household resource
allocations?
Two
in redistribution
interpretations are possible. First, one could by
we observe
to differences in the bargaining
and
large attribute the differences
power parameter
a.
Men
respond more
"These results Eilso cut against the view that reduced labor suppl}' was used to get education.
The sample size is slightly smaller here than in the basic regression because we have excluded from the sample
all the households that have only one prime-age individual. Only about
have only one prime-age individual.
37
2%
of the households in our basic
sample
because they have more power inside of their household. That the male-female differential
when
and so
is
picture.
When
is
quite suggestive of a situation in which dominant males captures
the pensioner
is
male, prime age males' ability to capture resources
is
a
the male-female differential in labor supply response.
The
largest
women
the pensioner
resources.
is
oldest
The age
is
diminished
results are in line
with this
male seems most capable of capturing household resources.
Second, one could rely on differential altruism to explain the patterns observed in Table
Perhaps pensioners care more about males. To
the female pensioners
who
care the most about males.
strongest towards their oldest working-age kids.
particularly intuitive,
it
our results, one would have to argue that
fit
Even
it is
Moreover, pensioners' altruism should be
if
this pattern of altruism
does not seem
nevertheless provides another lens for interpreting our findings.
Conclusion
8
With improving health conditions and lengthened
many governments
will
in place in
different living
more advanced
expectancy
in
many developing
own
in
countries,
for the
needs
Before simply replicating the type of programs that have been
nations,
pohcymakers
in those countries will
arrangements could interfere with their social objectives.
often live on their
ones.
life
soon have to introduce full-fledged social programs to provide
of a growing elderly population.
put
10.
have to consider how
While the elderly may
developed countries, multi-generation households prevail in developing
The South African pension program
programs when extended family
provides a
way
to understand the effects of such targeted
links are strong.
The South African government's
state pension
living conditions for older individuals that are
program was introduced
no longer active
38
as a
in the labor force
way
to improve
and that do not
have access to a private pension. The vast majority of the older-age Africans in South Africa are
participating in this pension program. This paper provides
some evidence
that, in practice, at least
part of the cash transfers that are targeted to the pensioners by the South African government
end up
in the
hands of a group that was not the originally targeted one: working-age individuals
that cohabit with the pensioners.
We
reduce their labor supply when they
redistribution, a
program targeted
find that African individuals
live
at a
between 16 and 50 years old
with pension beneficiaries. Hence, because of intra-family
group that
is
out of the labor force unexpectedly altered
the labor supply of a non-targeted group.
Moreover, we were able to relate this labor supply response to standard theories of intra-
household resource allocation and collective labor supply choice. The different impact of male and
female pension
money on
labor supply leads us to believe that a
common
family labor supply cannot adequately describe our results and that
takes place within these families.
prime-age
man
in a family,
set of bargaining
of bargaining
In general, older prime-age men. and in particular the oldest
appear to be the biggest beneficiaries from the pension. Within the
men have
relatively
more bargaining power
other family members. In the end though, one
What
some amount
models of intra-household allocation of resources, one could interpret
as evidence that these
and by the many
preference model of
tjuestions that are left
is
If so,
or are being cared for
more by
struck by the flexibility of these interpretations
unanswered by the existing models of family bargaining.
determines bargaining power or altruism?
power or are cared about more?
this finding
why?
If not,
Is it
a generic fact that older males have more
when do they? These questions
are outside the
scope of most of these models. But answering them seems crucial for understanding the process of
resource allocation within families.
39
Bibliography
Altonji, Joseph,
cally
Fumio Hayashi and Laurence
Kotlikoff, 1992, "Is the
Extended Family
Linked? Direct Tests Using Micro Data," American Economic Review,
v.
Altruisti-
82, no. 5, pp.
1177-1198.
Ardington, Libby and Frances Lund, 1994, "Pensions and Development; the Social Security System
as a
Complementary Track
to
Programs of Reconstruction and Development," Centre
for
Social and Development Studies, University of Natal. Durban.
Becker, Gary, 1974. "A Theory of Social Interaction," Journal of Political Economy,
v.
82, no. 6,
pp. 1063-1094.
Becker, Gary, 1981,
A
Cambridge,
Treatise on the Family.
MA: Harvard
University Press.
Browning. Martin, Francois Bourguignon, Pierre-Andre Chicappori and Valerie Lechene, 1994, "In-
come and Outcomes: A Structural Model
Economy,
v.
Caelers, Di, 1998,
Case,
102, no.
6,
of Intrahousehold Allocation," Journal of Political
pp. 1067-1096.
"Noah Saves Khayelitsha's Old
Anne and Angus Deaton,
Economic Journal,
v.
Folk," Independent,
June
16.
"Large Cash Transfers to the Elderly in South Africa,"
1998,
108, no. 450, pp. 1330-1361.
Chiappori, Pierre-Andre, 1992, "Collective Labor Supply and Welfare," Journal of Political Econ-
omy,
v.
100, n. 3, pp. 437-67.
Duflo, Esther. 1999. "Child Health
and Household Resources
in
South Africa: Evidence from the
Old Age Pension Program." Working paper, Massachussetts Institute of Technology.
Imbens, Guido, Donald Rubin and Bruce Sacerdote, 1999, "Estimating the Effects of Unearned
Income on Labor Supply, Earnings, Savings, and Consumption: Evidence from a Survey of
Lottery Players,"
NBER
Working paper
#
7001, March.
Jensen, Robert. 1998, "Public Transfers, Private Transfers and the 'Crowding Out' Hypothesis:
Evidence from South Africa," John F. Kennedy School of Government, Harvard University.
Lund, Frances, 1992, "State Social Benefits
46
in
South Africa," International Social Security Review,
(1). 5-25.
Lundberg, Shelley and Robert Pollak, 1993, "Separate Spheres Bargaining and the Marriage Market,"
Journal of Political Economy,
v.
101, no. 6, pp. 988-1010.
Lundberg, Shelley and Robert Pollak, 1996. "Bargaining and Distribution
of
Economic
in Marriage,"
Perspectives, v. 10, no. 4, pp. 139-158.
Manser, Marilyn and Murray Brown, 1980, "Marriage and Household Decision Making:
gaining Analysis." International Economic Review,
Mc
Elroy, Marjorie,
v.
21, no.
1,
A
Bar-
pp. 31-44.
and Mary Jean Horney, 1981, "Nash-Bargained Household Decisions: Towards
a Generalization of the Theory of
Mc
Journal
Elroy, Marjorie, 1990,
Demand," International Economic Review,
22, 333-349.
"The Empirical Content of Nash-Bargained Household Behavior," Jour-
40
nal of
Human
Resources, 25, 559-583.
Ngoro, Blackmail, 1998a, "Retiring to Life in Shackland," Independent, July 20.
Samuelson, Paul, 1956, "Social Indifference Curves," Quaterly Journal of Economics,
v.
70, no.
1,
pp. 1-22.
Schultz, Paul, 1990, "Testing the Neoclassical
of Human
Resources,
v.
Model of Family Labor Supply and
Thomas, Duncan,
of
Van
Human
v.
14, pp.
Journal
25, no. 5, pp. 599-634.
Subramanian, Shankar and Angus Deaton, "Gender Effects
Sarvekshana,
Fertility,
in
Indian Consumption Patterns,"
1-12.
1990, "Intra-Household Resource Allocation:
Resources, v. 25, no.
4,
An
Inferential Approach," Journal
pp. 635-664.
der Berg, Servaes, 1994, "Issues in South African Social Security," mimeo,
41
The World Bank.
TABLE
1
Descriptive Statistics:
African Individuals, 16 to 50 Years Old
Three-Generation Households
Means and Standard
Sample:
All
Deviations"
HHs
Non Age-
AgeEligible
HHs
Eligible
HHs
Mean
S.D.
Mean
S.D.
Mean
S.D.
Age
27.5
9.3
27.5
8.7
27.5
9.9
Employed
.229
.420
.212
.409
.246
.431
Hours Worked
6.32
16.37
3.21
12.51
9.45
19.00
Unemployed
.079
.270
.087
.282
.071
.256
Discouraged
.211
.408
.232
.422
.191
.393
4th Grade or More
.754
.431
.748
.434
.760
.427
8th Grade or More
.348
.477
.338
.473
.360
.480
Matric or More
.130
.336
.128
.335
.132
.338
Household Size
8.81
3.62
9.13
3.88
8.50
3.30
Rural
.683
.465
.707
.455
.660
.474
Urban
.166
.372
.152
.359
.180
.384
Metro
.151
.358
.141
.348
.161
.367
Sick
.065
.246
.073
.261
.056
.2.30
1325
1833
1318
1246
1333
2272
207
275
371
277
42
142
Variable:
Total Income
Pension Income
Number
of Elig.
Females
.454
.526
.906
.377
Number
of Elig.
Males
.169
.383
.338
.485
Sample Size
"Notes: Sample
3169
6326
is
composed
of set of African individuals
3157
between 16 and 50 years old
that live in a three-generation household. All variables are from the World
survey,
August-December 1993.
42
Bank/SALDRU
TABLE
2
Effect of Old Age Pension Income on
Working Hours and Employment Status
of 16 to 50 Year Old Africans"
Hours Worked
Dependent Variable:
OLS
IV
OLS
OLS
IV
(1)
(2)
(3)
(4)
(5)
(6)
-12.32
—
-17.07
-.053
-.099
(1.78)
(.022)
(.035)
—
—
(1.18)
HH
Eligibility
Dummy
OLS
Specification;
Pension Income* 1000
Employment
—
Dummy
-6.401
(.580)
FemeJe
Age
-2.666
-2.629
-.068
-.069
-.069
(.447)
(.452)
(.452)
(.012)
(.012)
(.012)
-6.526
-6.732
-6.585
-.394
-.395
-.394
(3.640)
(3.709)
(3.611)
(.090)
(.090)
(.090)
.407
.412
.404
.022
.022
.021
(.187)
(.190)
(.185)
(.005)
(.005)
(.005)
-.010
-.010
-.010
-.0004
-.0004
-.0005
(.004)
(.004)
(.004)
(.0001)
(.0001)
(.0001)
Age"* 1000
8th Grade or
More
(.011)
-2.552
Age^
Age^
—
-.043
.082
.081
.080
.0036
.0036
.0036
(.032)
(.032)
(.032)
(.0008)
(.0008)
(.0008)
1.485
1.262
1.520
.064
.062
.064
(.466)
(.468)
(.469)
(.012)
(.012)
(.012)
.126
.123
—
.192
.193
—
R"
"Notes:
1.
Standard errors are
effects within
in parentheses.
SALDRU
household
Standard errors are corrected to allow
2.
Sample
3.
Other covariates included in regression
size in all regressions
is
4.
In the
eligible
IV
aire
size,
14 province indicators, 3 metro indicators
number
of household
members aged
0-5,
and 22-24.
income is instrumented with the number of agehousehold and the number of age-eligible men in the house-
specification, pension
women
group
6326.
(urban, rural and metro), household
6-15, 16-18, 19-21,
for
clusters.
in the
hold.
43
TABLE
3
Effect of Old Age Pension Income on
Working Hours and Employment Status
16 to 50 Year Old Africans"
Panel A: Males
Specification:
Pension Income*1000
HH
Eligibility
Dummy
Employment
Hours Worked
Dependent Vanable:
D ummy
OLS
OLS
IV
OLS
OLS
IV
(1)
(2)
(3)
(4)
(5)
(6)
-15.13
-22.48
-.098
-.201
(1.72)
(2.72)
(.034)
(.056)
—
—
—
-8.703
(.849)
—
.166
.163
—
-.086
(.018)
234
.234
Panel B: Females
Dependent Variable:
Specification:
Pension Income*1000
HH
Eligibility
Dummy
Ho'urs
Employment
Worked
Dummy
OLS
OLS
IV
OLS
OLS
IV
(1)
(2)
(3)
(4)
(5)
(6)
—
-10.29
-13.27
-.018
(1.11)
(1.73)
(.029)
—
—
—
-4.810
(.646)
.107
-.023
(.043)
—
-.013
(.014)
—
.100
.m
—
.178
"Notes:
1.
Standard errors are in parentheses. Standard errors are corrected to allow
effects within S.A.LDRU household clusters.
2.
Sample
3.
4.
size
is
2532
in
Panel
A
and 3794
m
for
group
Panel B.
Other covariates included in regression are a quartic in age, a dummy variable for
having completed at least 8th grade, 14 province indicators, 3 metro indicators
(urban, rural and metro), household size, number of household members aged 0-5,
6-15, 16-18, 19-21, and 22-24.
In the IV specification, pension income
eligible
women
in
the household and the
is
instrumented with the number of age-
number
hold.
44
of age-eligible
men
in
the house-
—
)
TABLE
4
Effect of the Presence of Elderly on
Hours Worked by
16 to 50 Year Old African Individuals
Dependent VarmMe:
Hours Worked
Eligible Elderly
in
HH
(1)
(2)
(3)
(4)
(5)
(6)
(7)
-6.79
—
—
—
—
—
—
—
—
—
—
—
(.64)
Non-Eligible Elderly
in
-.46
HH
(.63)
Number
HH
(n5055)
Number
of 50 to 55
Females
in
Number
of 50 to 55
Males
HH
in
Number
Persons
—
of 50 to 55
Persons in
HH
(n5055f)
(n5055m)
of 55 to 60
HH
in
(n5560)
Number
of 55 to 60
Females
in
Number
of 55 to 60
Males in
HH
Number
of 60 to 65
Persons in
Number
HH
(n5560f)
(n5560m)
HH
in
Number
'
HH
(n6570f)
of 65 to 70
Males in HH (n6570m)
Number of Persons
HH Older than 70 (n70p)
Number of Females
in
HH
Older than 70 (n70pf
Number of Males
in HH Older than 70 (n70pm)
in
—
—
—
-.23
-.09
-.22
-.08
—
(.70)
(.71)
(.70)
(.71)
—
—
—
—
—
—
—
—
—
—
—
—
—
—
—
—
-2.54
-2.41
-2.53
-2.41
(.58)
(.58)
(.57)
(.57)
—
—
—
—
—
—
—
—
(n6065m)
Older than 65 {n65p)
Number of 65 to 70
Persons in HH (n6570)
Number of 65 to 70
in
-.21
(.50)
—
HH
Females
-.40
(.50)
—
n6065m*Deviation from
Elig. Rule in Region
Number of Persons
in
-.22
(.50)
—
(n6065)
of 60 to 65
HH
-.42
(.50)
—
Females in HH (n6065f)
Number of 60 to 65
Males
—
Number of HH members Over
With Health Problems
R"
50
—
—
—
—
—
—
—
—
—
-5.37
-5.21
(.51)
(.53)
—
—
—
—
—
—
—
—
—
—
—
—
—
—
-1.05
—
—
-5.17
-5.03
(.70)
—
—
(.72)
-5.49
-5.31
(.56)
(.57)
—
—
—
.124
.125
"Notes: see next page.
45
—
—
—
-1.04
-.65
-.49
(.67)
(.67)
-.23
-.03
(.96)
(.95)
—
.124
,125
—
-.19
-.04
(1.04)
(1.04)
-.62
-.50
(.86)
—
(.87)
-2.95
-2.81
—
(.94)
(.93)
-1.16
-1.03
(1.41)
(1.41)
-7.47
-7.43
(3.86)
—
—
(3.88)
-6.67
-6.52
—
;
(.86)
(.88)
-3.85
-3.74
(1.19)
(1.20)
-7.47
-7.28
—
—
(.72)
(.71)
-2.87
-2.76
(1.13)
(1.14)
—
(.69)
(.69)
.119
—
—
—
-.94
(.68)
.129
.130
Notes;
1.
StaJidaxd errors are in parentheses. Standard errors are corrected to allow for group effects within
household
2.
3.
Sample
size in all regressions
Other covariates included
is
6326.
in regression are a quartic in age,
a
dummy
for sex, a
dummy
for
of household
members aged
0-5, 6-15,
that do not deviate) to
5.
size,
number
men between
60 and 65 and no
a social pension in the region. This variable ranges from
(in the regions
"Deviation from Eligibility Rule in Region"
eligible elderly that are receiving
completion of at
metro indicators (urban, rural and metro), household
16-18, 19-21, and 22-24.
least 8th grade, 14 province indicators, 3
4.
SALDRU
clusters.
is
the fraction of households with
.67.
Tests of Equality of Coefficients
Column 2: n5055=n6065 (p=.004),
Column 3: n5055=n6065 (p=.003),
Column 4: n5055=n6065 (p=.004),
Below and Above
Eligibility
Threshold.
n5055=n6.5p (p=.000), n5560=n6065 (p=.009), n5560=n65p (p=.000).
n5055=n65p (p=.000), n5560=n6065 (p=.009); n5560=n65p (p=.000).
n5055=n6570 (p=.000), n5055=n70p (p=-000), n5560=n6065 (p=.009),
n5560=n6570 (p=.000), n5560=n70p (p=.000).
Column 5: n5055=n6065 (p=.003), n5055=n6570 (p=.000), n5055=n70p (p=.000), n5560=n6065 (p=.009),
n5560=n6570 (p=.000), n5560=n70p (p=.000).
Column 6: n5055f=n6065f (p=.020), n5055f=n6570f (p=.000), n5055f=n70pf (p=.000), n5055m=n6570m
(p=.011), n5055m=n70pm (p=.045), n5560f=n6065f (p=.030), n5560f=n6570f (p=.000), n5560f=n70pf (p=.000),
n5560m=n6570m
Column
(.023),
n5560m=n70pm
(p=.100),
n6065m=n6570m
(p=.131),
n6065m=n70pm
(p=.379).
n5055f=n6065f (p=.018), n5055f=n6570f (p=.000), n5055f=n70pf (p=.000), n5055m=n6570m
(p=.009), n5055m=n70pm (p=.037), n5560f=n6065f (p=.029), n5560f=n6570f (p=.000), n5560f=n70pf (p=.000),
7:
n5560m=n6570m
(p=.022),
n5560m=n70pm
(p=.098),
46
n6065m=n6570m
(p=.129),
n6065m=n70pm
(p=.378);
TABLE
5
Effect of Presence of Age-Eligible Elderly
on Employment of
16 to 50 Year Old African Males"
Dependent Variable:
Employment Status
Data Set
HH
Population Census 91
is:
Eligibility
Dummy
-.016
-.068
(.026)
-.317
-.728
(.012)
(.180)
Age'
.020
.040
(.0006)
(.009)
-.0004
-.0009
(.00001)
(.0002)
Age'' * 1000
8th Grade or
.0034
.0073
(.0001)
(.0016)
More
R'
Sample Size
Survey 93
(.002)
Age
Age^
SALDRU
.034
.039
(.002)
(.024)
.260
.171
358394
1697
"Notes:
1.
2.
Standard errors are
in parentheses.
Samples are composed of set of African males between 16 and 50 years old that live
household where: 1) a son or daughter is present, 2) another family member
other than household head's partner is present, and 3) the maximum age in the
household is over 50 years old.
in a
3.
Other covariates included
household
size,
number
in regression are
of household
a quartic in age, geographic controls,
members aged
22-24,
47
0-5, 6-15,
16-18, 19-21,
and
TABLE
6
Old Age Pension and Alternative Sources of Work"
Coefficient on Pension
Panel A: Males
in
Income*1000
3-Generation
HHs
OLS
Specification;
Dependent Variable:
Regular Employment
Casual Employment
Self
Employment
Panel B: All 3-Generation
Dependent Variable:
Crop Production in HH (Yes=l; No=0)
Number
Obtained from Herd
of Eggs Obtained from Poultry
Value of Mechanised Farm Equipment and
Pumps
Rands)
Value of Tractors and Farming Vehicles
(in Rands)
(in
Panel C: Rural 3-Generation
Dependent Variable:
Crop Production in HH (Yes=l; No=0)
Number
of
Obtained from Herd
Eggs Obtained from Poultry
Value of Mechanised Farm Equipment and
(in Rands)
Value of Tractors and Farming Vehicles
(in
-.099
(.055)
-.050
-.085
(.018)
(.024)
-.004
-.009
(.013)
(.030)
OLS
IV
-.015
.059
(.035)
(.067)
-.190
-.414
(.291)
(.578)
-1.095
-.483
(.952)
(2.410)
-22.91
-11.02
(11.89)
(31.75)
-69.91
.035
(106.91)
(237.48)
HHs
OLS
Specification:
Litres of Milk
-.022
(.036)
HHs
Specification:
Litres of Milk
Pumps
Rands)
"Notes: see next page.
48
IV
IV
-.023
.062
(.045)
(.090)
-.250
-.612
(.368)
(.774)
-1.409
-.486
(1.180)
(3.236)
-30.88
-20.11
(16.12)
(42.14)
-100.53
-37.86
(138.18)
(322.76)
Notes:
1.
Each
2.
Sample
coefficient in the table belongs to a separate regression.
in
(sample
Panel
size
is
A
is
2532).
the set of
Sample
all
in
African males between 16 and 50 years old living in a 3-generation households
Panel
B
is
the set of
all
three-generation households with at least one individual
between 16 and 50 years of age (sample size is 1849). Sample in Panel C is the set of
households with at least one individual between 16 and 50 years of age (sample size
3.
4.
Standard errors are
household clusters.
Other covariates included
metro), household
in Panel
5.
in parentheses.
A
size,
Standard errors are corrected to allow
in all regressions are 14
number
are a quartic in age
of household
and a
dummy
for
all
is
group
rural three-generation
1305).
effects
within
SALDRU
province indicators, 3 metro indicators (urban, rural and
members aged
0-5, 6-15, 16-18, 19-21,
and 22-24. Also included
variable for having at least completed the 8th grade.
IV specification, pension income is instrumented with the number
and the number of age-eligible men in the household.
In the
49
of age-eligible
women
in the
household
TABLE
7
Old Age Pension Income on
Education Level
of 16 to 50 Year Old Africans
Effect of
'^
+
Matricp +
4th Grade
Specification:
OLS
IV
OLS
IV
OLS
IV
(1)
(2)
(3)
(4)
(5)
(6)
Pension Income*1000
R^
8th Grade
+
Dependent Variable:
.022
.005
.023
-.042
-.016
-.021
(.025)
(.041)
(.031)
(.044)
(.019)
(.030)
.129
—
.154
—
.092
—
"Notes:
1.
Standard errors are
effects within
2.
3.
4.
Sample
in
parentheses. Standard errors are corrected to allow for group
SALDRU
household clusters.
size in all regressions
is
6326.
Other covariates included in regression are a quartic in age, a dummy for sex, 14
province indicators, 3 metro indicators (urban, rural and metro), household size,
number of household members aged 0-5, 6-15, 16-18, 19-21, and 22-24.
In the
eligible
IV
specification, pension
women
in
income
the household and the
is
instrumented with the number of age-
number
hold.
50
of age-eligible
men
in the house-
TABLE
8
Old Age Pension and Selective Migration"
Months Away
from HH
Dependent Variable:
Specification:
Pension Income*1000
Number
Migrated
in
HH
in
of 16 to 50
HH
OLS
IV
OLS
IV
OLS
IV
(1)
(2)
(3)
(4)
(5)
(6)
-.757
-.316
-.012
.029
-.200
-.272
(.196)
(.340)
(.012)
(.027)
(.157)
(.256)
.111
—
.073
—
.068
R^
"Notes:
1.
"Months Away from HH"
is
the
number
of
months the
HH member
spent away
1 if the
months; "Migrated in HH" is a dummy variable that equals
person "move(d) here during the past five years."
in the last 12
2.
Standard errors are
effects within
3.
Sample
in
in parentheses.
SALDRU
columns
Standard errors
cire
corrected to allow for group
household clusters.
(1) to (4) is
composed of the
between 16 and 50
6326). Sample in
that have at least one member
set of African
years old that live in a three-generation household (Sample size
columns (5) and (6) is the set of 3-generation HHs
betwen 16 and 50 years old (Sample size is 1951).
is
4.
Other covariates included in all regressions are 14 province indicators and 3 metro
indicators (urban, rural and metro). Also included in columns (1) to (4) are a
quartic in age, a dummy for sex, a dummy for having completed at least 8th grade,
household size, number of household members aged 0-5, 6-15, 16-18, 19-21, and
5.
In the
22-24.
eligible
IV
specification, pension
women
in the
income
is
instrumented with the number of age-
household and the number of age-eligible
hold.
51
men
in the house-
TABLE
9
Old Age Pension and Pooling of Resources"
Dependent Variable:
Hours Worked
Sample:
All
Number
of
Women
Over 60
Number
of
Men
Over 65
Number
Number
of
of
Men
Men
Prime-Age
of
Gen HHs
(1)
(2)
(3)
(4)
-5.13
-5.13
-3.89
(.58)
(.58)
(.57)
(1.44)
-2.32
-2.55
-2.54
-.71
(.87)
(.87)
(.88)
(1.47)
60 to 65
60 to 65*
-1.13
-1.12
(1.46)
(1.43)
-5.31
-5.10
(4.06)
(4.05)
—
Women
.089
Between 16 and 50
Number
of
(.19)
Men
-.14
Between 16 and 50
Number
of
Between
Number
of
Between
Prime-Age Living With Exactly
One Old Woman And One Old Man
-5.02
Deviation from Elig- Rule
Number
in 3
(.22)
Women
-.41
and 16
(.17)
Men
.01
and 16
(.18)
.122
.120
.118
.120
"Notes:
1.
StJindard errors are in parentheses. Standard errors are corrected to allow for group
effects within
2.
The sample
in a
in
SALDRU
columns
household clusters.
(1) to (3) is
three-generation households.
the original sample of prime-age Africans living
The sample
individuals that live with exactly one
50 years of age. Sample Size
column
3.
4.
is
6326
woman
in
in
column
(4) is the subset of those
over 50 years of age and one
columns
(1) to (3);
sample
size
man
is
over
1471 in
(4).
Other covariates included in all regressions are 14 province indicators and 3 metro
indicators (urban, rural and metro). Also included in columns are a quartic in age,
a dummy for sex, a dummy for having completed at least 8th grade, household size,
number of household members aged 0-5, 6-15, 16-18, 19-21, and 22-24.
"Deviation from Eligibility Rule in Region" is the fraction of households with men
between 50 and 65 and no eligible elderly that are receiving a social pension in the
region. This variable ranges (in the regions that do not deviate) to .67.
52
TABLE
10
Old Age Pension Income on
Working Hours of 16 to 50 Year Old Africans:
Effect of
Distribution of Effect"
Dependent Variable:
Hours Worked
(1)
Pension Income*1000
Pension Income* 1000*
Female
Number
of
Women
(5)
(6)
(7)
(3)
(4)
-21.04
-14.09
-15.55
25.48
39.35
20.79
(2.51)
(1.85)
(1.63)
(3.45)
(12.37)
(3.66)
(2)
9.05
7.10
(2.21)
(2.47)
Over 60
-6.98
(.81)
Number
of
Men Over
-2.73
65
(1.03)
Number
of
Women
Over 60*
3.23
Female
Number
(.75)
of
Men Over
65*
.71
Female
Pension Income*1000*
-7.42
4th Grade or Less
(3-02)
Pension Income*1000*
-.07
Matric or More
(3.93)
Pension Income*1000*
-1.53
-2.54
-1.49
Age
(.14)
(.91)
(.14)
Pension Income* 1000*
.02
Age^
.02)
Pension Income* 1000*
Oldest Prime-Age
Man
-9.08
in
HH
(5.00)
"Notes:
1.
Standard errors are
effects within
2.
3.
Sample
size in
in
parentheses. Standard errors are corrected to allow for group
SALDRU
columns
household clusters.
1
to 6
6326. Sample size in column 7
is
is
6189.
Other covariates included in regression are a quartic in age, a dummy for sex, a
dummy for having completed at least 8th grade, 14 province indicators, 3 metro
indicators (urban, rural and metro), household size, number of household members
aged 0-5, 6-15, 16-18, 19-21, and 22-24. Columns 3, 4 and 7 also respectively include
a dimimy for "4th grade or less" a dummy for "matric or more" and a dummy for
,
"oldest prime-age
4.
man
in
the HH."
columns except column 2 represent IV results. In the IV specifications, pension
income as well as the interactions of pension income with the other variables of
All
interest are instrumented.
53
561^5
nio
Date Due
MIT LIBRARIES
3 9080 02246 2102
iililiii
..„