Effects of family planning and health services on women’s welfare: evidence on dowries and...

Effects of family planning and health serviceson women’s welfare: evidence on dowriesand intra-household bargaining in Bangladesh

Christina Peters

Received: 18 June 2009 / Accepted: 30 June 2010 / Published online: 11 July 2010

� Springer Science+Business Media, LLC 2010

Abstract This paper demonstrates how the availability of family planning and

maternal and child health services alters the structure of intra-household bargaining.

The overall welfare gains from such programs are likely to be large, but when

women obtain access to services only through marriage, some of these gains may be

partially offset by changes in their bargaining power and in the dowries that they

pay their husbands. I examine these marriage market effects using a family planning

and health services program in rural Bangladesh, finding that compared to women

without program access, women in the treatment area are 35% less likely to be able

to make purchases without permission from their husbands or another household

member. Moreover, a difference-in-difference specification confirms that women

pay 14% higher dowries in order to obtain husbands with access to the program. The

fact that adjustments are made both before and within marriage suggests that marital

contracts in rural Bangladesh are negotiated along multiple margins.

Keywords Marriage � Intra-household bargaining � Family planning

JEL Classification D13 � J12 � J13

1 Introduction

Programs to limit fertility and provide health care to mothers and children promise

to improve socioeconomic outcomes in developing countries. Evaluations of

plausibly randomized family planning programs in Bangladesh and Colombia

document large declines in fertility (Koenig et al. 1992, Miller 2010). Recent work

C. Peters (&)

Department of Economics, Metropolitan State College of Denver, Campus Box 77;

PO Box 173362, Denver, CO 80217-3362, USA

e-mail: [email protected]

123

Rev Econ Household (2011) 9:327–348

DOI 10.1007/s11150-010-9100-7

also finds secondary effects on household behavior, including increased human

capital investments and reduced child labor supply (Sinha 2005, Joshi and Schultz

2007). This paper demonstrates a further avenue through which family planning and

health services programs can affect the welfare of individual members of the

household: access to a program can alter marriage market behavior and change the

dynamics of post-marriage intra-household bargaining.

Although the overall welfare gains from family planning and health services

interventions are likely to be large, I observe the important result that when men

control access to the program, these gains may be partially offset for women in other

aspects of their lives. Using a supply and demand model for marriage (Becker 1973,

Grossbard-Shechtman 1993), I predict that the program will decrease the share of

the marital surplus going to the wives of husbands living in program areas. This

prediction is tested on the treatment and comparison groups of a plausibly

randomized family planning program in Bangladesh, revealing that household

adjustments to the program appear to occur not only within marriage through

changes in bargaining weights but also before marriage through a dowry payment.

Using multiple measures of female-decision making power within the household,

probit models find that women who reside within the program treatment area (i.e.

treated women) are 35% less likely than women in the comparison area (i.e.

untreated women) to be able to make certain purchases without permission from

their husbands or another household member. Furthermore, a difference-in-

difference model indicates that the families of women who marry after the program

begins pay 14% higher dowries in order to secure husbands from the treatment area.

I estimate these results using the cross-sectional Matlab Health and Socioeco-

nomic Survey (MHSS), administered in 1996 to a random sample covering 4,364

households1 in the nearly 150 villages of Matlab district, a rural agricultural area of

Bangladesh 55 km southeast of Dhaka (Rahman et al. 1999). Since 1977, a maternal

and child health and family planning program (the MCHFP) has been operating in

randomly chosen but contiguous geographic blocks within Matlab. This paper

confirms that the services provided by the program not only reduce fertility rates in

the treatment area when compared with the control area, but they also enhance the

health and survival rates of infants and children. Because geographic location

determines eligibility for the program and women typically move in with their

husband’s family upon marriage, wives obtain access to program services only if

their husbands reside in the treatment area.2 Thus, the MCHFP endows men with a

new asset to bring to marriage that is highly desired by women—the enhanced child

quality and reduced child quantity provided by the program.

A supply and demand model of marriage predicts that a couple will revise their

contract in light of the husband’s new endowment to increase his share of the

surplus at the wife’s expense. Much of the previous literature on marriage has

implicitly assumed that this change will occur along only one margin, such as

1 These households cover 2,687 baris (approximately 33% of the total number of baris). A bari is a group

of households living and working closely together.2 Even if a woman grows up in the treatment area, she loses access to the program if she marries a man

outside the treatment area.

328 C. Peters

123

changes in dowry (Anderson 2000) or spousal bargaining power in the household

(Manser and Brown 1980, McElroy and Horney 1981). However, there is no ex-ante

reason to restrict changes to only one margin of adjustment. This paper follows the

observations of Becker (1973) and Grossbard-Shechtman (1993) by allowing the

possibility that a couple on the post-program marriage market may in fact choose to

adjust along multiple margins, partially increasing the dowry payment while

simultaneously giving decreased weight to the wife’s preferences over the allocation

of household resources. My empirical results suggest just this mix—separating the

data sample by period of marriage finds that women in post-program marriages (i.e.

those marriages occurring after the program begins) both increase their dowry

payments to treatment area men and exhibit decreased bargaining power within the

marriage. Women in pre-program marriages show a similar bargaining power

differential but no differences in dowry payments, since dowry changes cannot

occur for women who married before the program begins. Meanwhile, there are no

significant differences in the level of independence over purchases for women who

are ineligible for direct participation in the program, including unmarried women in

their fertile years and divorced women.

Despite the many previous studies arguing that the comparison area villages are

an appropriate control for the treatment villages, there may be concern that these

results merely reflect pre-existing or unobservable differences between the two

groups. Although I am limited to a cross-sectional sample of women and cannot

look directly at pre-program outcomes, descriptive statistics for two pre-program

cohorts indicate few substantial demographic differences between pre-program

women living in the treatment and comparison areas (these cohorts include women

within the dataset who are past menopause at the start of the MCHFP, and women in

a pre-program census of the Matlab population). However, like all other studies

using this data, it is ultimately impossible to fully account for unobservable

characteristics. Despite this limitation, the availability within this dataset of explicit

measures of control over household purchases provides a unique advantage over

previous studies of exogenous shocks to the marital contract, which have been

forced to infer changes in bargaining power through more commonly observed

variables such as labor supply (e.g. Chiappori et al. 2002, Oreffice 2007).

Moreover, when estimating changes in dowry payments to treatment area males,

I can apply a difference-in-difference specification to control for unobservables. The

data records retrospective information for all couples on dowries paid in the year of

marriage (as reported by the wife interviewed separately from her husband), which

allows me to find the change in dowry payments over the pre and post-program

periods for the treatment area, after accounting for the corresponding change in the

control area. While there are no significant differences in either the levels or trends

in dowry payments paid to treatment and comparison area men in pre-existing

marriages, treatment area men in post-program marriages receive dowry payments

that are 1,192 taka higher than their untreated counterparts (a 14% difference).

Difference-in-difference estimates confirm that treatment area males begin receiv-

ing higher dowries than untreated males in 1977, the first year of the program.

Combined with the first set of results on bargaining power, this suggests that many

couples respond to shocks along multiple margins of adjustment. These findings

Effects of family planning and health services 329

123

caution that when family planning and health services programs are administered

such that men control access, they may appropriate the increase in marital surplus

due to the program and consequently induce some unintended welfare changes for

women. Despite these changes, however, Matlab women are still likely to

experience substantial gains overall from having access to the MCHFP.3

The next section describes the data setting, the Matlab family planning and health

services program, and its first-order impacts. I then provide a theoretical framework

that illustrates the comparative static effects of the program endowment on female

bargaining power and dowry payments, and present my empirical results.

2 Data setting and the Matlab maternal and child health and family planningproject

Sponsored by the International Centre for Diarrheoal Disease Research in

Bangladesh, the Maternal and Child Health and Family Planning Project began in

1977. The program administrators designated the treatment and control areas of the

program by dividing Matlab district into six geographic blocks chosen at random.

For logistical reasons of managing the program, the blocks were then merged into

treatment and control groups based on contiguity of the treatment blocks. This

method carries the added benefit of minimizing program spillovers, as long as there

are not increasing returns to scale in program treatment externalities. Four

contiguous blocks form the treatment group (89,000 people in 70 villages), while

the other two blocks (85,000 people in 79 villages) are the control group.4 These

geographic divisions remain unadjusted through 1996 (the year of the survey data

used in this paper). A 1974 census indicates that the Matlab area is demographically

homogenous before the project begins (Fauveau and Chakraborty 1994, Joshi and

Schultz 2007), and Sinha (2005) confirms this finding using 1996 data. Although the

large sample size of the 1974 census results in significant differences between the

future treatment and control groups across several indicators, Table 1 shows most of

these differences to be small in magnitude.

Descriptive statistics in 1996 for the pre-program female cohort (women already

past menopause at the inception of the MHCFP) also show few significant

differences between women living in the treatment and comparison areas (see

Table 2). The pre-program cohort from the treatment area have an average of .2

3 Another potential explanation for the empirical results of this paper is that treatment area boys who

receive child health services grow up to be healthier and more desirable on the marriage market. Their

higher quality enables them to extract larger dowry payments and more bargaining power within their

marriages. However, only two women in my dataset have husbands born after the program begins (who

would have received services from birth), and my empirical results on bargaining power are robust to the

exclusion of the youngest cohorts of women (whose husbands, although born before program inception,

may still have been children at the time and received some health benefits). Thus, my results preclude this

alternate explanation.4 Spillovers occur if women from comparison areas travel to nearby treatment village clinics for services.

Although any spillovers should only understate the true effects of the MCHFP project, the main results of

this paper are robust to limiting the sample to border villages, where the potential for spillovers is greatest

(i.e. treatment villages that share a border with a comparison village, and vice versa, see Table 7).

330 C. Peters

123

fewer male child deaths, yet this statistic may be entangled with a program effect,

since these families could still have taken advantage of intensive child health

services despite no longer being fertile. When asked to recall any employment

beyond housework twenty years earlier (1976, the year before the MCHFP begins),

pre-program treatment area women were 14 percentage points less likely than

control women to report having participated in outside work. This trend continues

through 1996, with labor force participation rates for fertile married treatment area

women remaining 5 percentage points lower than comparison area rates, so I control

for outside work in all probit models for differences in bargaining power.5

In the treatment villages, local community health workers visit all fertile, married

women in their households fortnightly, offering a range of contraceptives,

immunizations and vitamins for pregnant women, and child nutritional and health

advice.6 In 1982, half the treatment area also began receiving expanded services of

prenatal care and immunizations for all women and children, with these services

later being diffused through the remaining treatment villages. Since 1989,

community health workers have been providing the entire treatment group with

comprehensive immunization services, nutritional education, help with childhood

dysentery diseases, and extensive maternity care (Fauveau and Chakraborty 1994).

Table 1 Descriptive statistics from a 1974 pre-program census

1974 Census Treatment Control Difference

Male years of education 2.14 (3.63) 1.93 (3.45) .21*** (.03)

Female years of education 1.35 (.02) 1.45 (.02) -.09*** (.01)

Household head years of education 2.82 (3.38) 2.85 (3.21) -.04 (.05)

Farming household .40 (.49) .37 (.48) .02*** (.01)

Agricultural labor household .19 (.39) .19 (.39) .00 (.00)

Married females with jobs (%) .04 (.20) .03 (.18) .01*** (.00)

Percentage currently divorced .02 (.13) .02 (.13) -.00 (.00)

Percentage muslim .79 (.40) .88 (.32) -.09*** (.00)

Households 13,916 13,035

Total Obs. 38,780 31,560

Years of education measured for individuals over 14 years. Percentage of females with jobs measured for

women between 14–45 years. Farming and agricultural labor households indicates occupation of

household head. Data from 1974 ICDDR,B census of the Matlab population

Standard deviations in parentheses (standard errors in parentheses for difference). *** indicates t-test

significant at 1% level

5 I define labor force participation for women as claiming some job other than housework as their

primary activity over the past month. Nearly half the women who currently work cite rearing hens and

ducks as their job, with a further third husking paddy. There do not appear to be any differences in 1974

between labor force participation rates of married women in the treatment and comparison areas (see

Table 1). I am unsure why the 1974 census results vary from the labor force results shown by the smaller

MHSS sample in 1976 and 1996; consequently, I am careful to control for it.6 Initial health services included tetanus toxoid immunizations for pregnant women, neonatal vitamins,

maternal and child nutritional advice, and oral rehydration for diarrheal diseases.


123

Ta

ble

2F

irst

-ord

eref

fect

so

fM

CH

FP

and

des

crip

tiv

est

atis

tics

from

the

MH

SS

for

the

pre

-pro

gra

mco

ho

rtan

dcu

rren

tly

fert

ile

mar

ried

wo

men

Pre

-pro

gra

mco

ho

rtF

erti

lem

arri

edw

om

en

Tre

atm

ent

Contr

ol

Dif

fere

nce

Tre

atm

ent

Contr

ol

Dif

fere

nce

Pan

elA

:d

emo

gra

phic

var

iab

les

Age

atfi

rst

mar

riag

e12.7

1(3

.87)

13.0

1(3

.11)

-.3

0(.

35

)1

5.8

2(3

.21

)1

5.6

5(3

.23

).1

7(.

13

)

Ag

eat

firs

tb

irth

18

.10

(4.4

2)

18

.80

(4.7

8)

-.7

1(.

46

)1

9.0

9(2

.96

)1

9.0

0(2

.98

).0

9(.

12

)

Ob

s.1

93

21

71

,23

91

,250

Yea

rsof

educa

tion

.56

(1.5

0)

.77

(1.7

3)

-.2

1(.

27

)2

.09

(2.8

8)

2.0

6(2

.78

).0

4(.

11

)

Had

job

in1

97

6.8

2(.

38

).9

6(.

19

)1

4*

**

(.0

5)

Curr

entl

yhas

job

.58

(.50)

.51

(.50)

07

(.08)

.69

(.46)

.74

(.44)

-.0

5**

*(.

02

)

Ear

ned

inco

me

42

9.8

2(1

52

1.7

4)

25

0.9

7(1

06

9.3

3)

17

8.8

5(2

04

.57

)1

38

9.4

5(1

07

09

.41)

92

0.5

9(6

38

2.2

6)

46

8.8

6(3

53

.03

)

Val

ue

of

hel

pfr

om

oth

ers

90

.35

(23

2.1

2)

75

.71

(20

7.2

1)

14

.64

(35

.51

)1

25

4.7

3(8

43

4.5

2)

63

6.8

5(2

19

6.7

2)

61

7.8

9*

*(2

46

.59

)

Ho

use

ho

ldin

com

e2

49

44

.68

(27

17

7.8

1)

32

142

.54

(40

58

2.1

5)

-7

19

7.8

7(5

99

2.2

6)

47

31

9.3

6(7

57

66

.19)

42

624

.35

(55

811

.16)

46

95.0

2*

(26

65.7

9)

Lan

do

wn

edb

yh

ou

seh

old

10

3.5

5(1

25

.85

)1

37

.67

(18

4.2

1)

-3

4.1

2(2

7.3

1)

14

9.9

2(1

19

7.8

7)

13

5.1

5(8

99

.70

)1

4.7

6(4

2.4

4)

Sp

ou

sey

ears

of

edu

cati

on

3.9

2(4

.11

)3

.56

(3.9

0)

.36

**

(.1

6)

Sp

ou

seea

rned

inco

me

16

26

4.0

6(2

49

74

.4)

16

754

.75

(31

810

.29)

-4

90

.70

(11

47

.03

)

Val

ue

of

hel

pre

ceiv

ed

fro

mo

ther

sb

ysp

ou

se

11

37

.02

(87

20.0

5)

19

28.3

2(2

92

17

.07)

-7

91

.31

(86

5.9

1)

Ob

s.5

71

06

1,2

39

1,2

50

Pan

elB

:fa

mil

yp

lann

ing

/hea

lth

var

iab

les

Ever

use

dco

ntr

acep

tion

.11

(.31)

.09

(.28)

.02

(.03)

.92

(.28)

.71

(.45)

.20***

(.01)

Curr

ent

use

r.7

6(.

43

).5

4(.

50

).2

2**

*(.

02

)

Sp

acin

gb

etw

een

bir

ths

2.9

6(1

.07

)2

.83

(1.0

4)

.13

(.1

1)

3.1

7(1

.16

)2

.77

(1.0

9)

.40

**

*(.

05

)

Per

cen

tag

eh

avin

ga

stil

lbir

th.1

7(.

38

).1

7(.

38

).0

0(.

04

).0

9(.

29

).1

2(.

32

)-

.03

**

(.0

1)

Nu

mb

ero

fso

ns

that

hav

ed

ied

1.0

7(1

.19

)1

.27

(1.2

4)

-.2

1*

(.1

2)

.29

(.5

9)

.34

(.6

5)

-.0

5**

(.0

3)

Nu

mb

ero

fd

aug

hte

rsth

at

hav

ed

ied

1.1

0(1

.14

)1

.08

(1.1

0)

.02

(.1

1)

.29

(.5

9)

.40

(.7

0)

-.1

1**

*(.

03

)

332 C. Peters

123

Ta

ble

2co

nti

nu

ed

Pre

-pro

gra

mco

ho

rtF

erti

lem

arri

edw

om

en

Tre

atm

ent

Contr

ol

Dif

fere

nce

Tre

atm

ent

Contr

ol

Dif

fere

nce

Co

mp

lete

dfe

rtil

ity

*7

.23

(2.5

8)

7.1

0(2

.55

).1

3(.

25

)7

.02

(2.5

0)

7.2

3(2

.55

)-

.22

*(.

12

)

Ob

s.1

93

21

71

,27

51

,270

Pre

-pro

gra

mco

ho

rtin

clu

des

wo

men

pas

tm

eno

pau

sein

19

77

(beg

inn

ing

of

pro

gra

mp

erio

d);

som

eo

bse

rvat

ion

sin

this

coh

ort

are

mis

sin

ged

uca

tio

nan

din

com

e

info

rmat

ion

,re

sult

ing

inth

esa

mple

size

dif

fere

nti

al.

Com

ple

ted

fert

ilit

ym

easu

red

for

wo

men

hav

eco

mp

lete

dth

eir

fert

ilit

ycy

cle

by

19

96

Sta

nd

ard

dev

iati

on

sin

par

enth

eses

(sta

ndar

der

rors

inp

aren

thes

esfo

rd

iffe

ren

ce).

**

*in

dic

ates

t-te

stsi

gn

ifica

nt

at1

%le

vel

;*

*in

dic

ates

t-te

stsi

gn

ifica

nt

at5

%le

vel

;

*in

dic

ates

t-te

stsi

gn

ifica

nt

at1

0%

lev

el


123

In contrast, women in the control villages have access only to the government-

sponsored health program, which began in 1965 but remains much less intensive

than the MCHFP program. While women treated by the MCHFP receive regular

home visits from local female trained health workers, women in the control group

receive very infrequent visits from government workers and must travel to the

nearest government clinic to obtain contraceptive and health services. These clinics

are often dirty and unsterile, and counseling is not typically done in private (Piet-

Pelon and Rob 1997, Joshi and Schultz 2007). Moreover, restrictions on female

mobility outside the bari severely limits their access to the services available at

these clinics. Purdah (female seclusion) ensures that the MCHFP area with its

in-home services becomes something that women will actually pay to marry into,

rather than simply trading a longer travel time to services for the relatively lower

dowries required by the control area.

Descriptive statistics in Table 2 highlight the first-order effects of the Matlab

family planning and health services program on the fertile married women in the

data sample. Panel A shows that most demographic differences between the

treatment and control areas remain small even into 1996,7 though treatment area

women do appear to receive significantly greater amounts of transfers from other

households (measured in both monetary and in-kind transfers), resulting in a total

household income that is nearly 5,000 taka higher than comparison households (an

11% differential).8 However, family planning and health variables from Panel B

show marked differentials. Women living in the program area have significantly

higher rates of contraceptive use and fewer child deaths and stillbirths. In addition,

they space their births by nearly 5 months more on average than control area

women, and treatment area women who have completed their fertility cycle have

had .2 fewer children than women from comparison villages (a 3% differential,

although it is likely to increase as women who have been treated for longer periods

begin to complete their fertility cycle). Several other studies share these findings of

differential declines in both fertility rates as well as infant, child, and maternal

mortality rates (Fauveau and Chakraborty 1994).

The impact of the MCHFP on child health is further illustrated through a 1996

vaccination rate that is 31 percentage points higher (t-test significant at the 1%

level) for treated 0–5 years olds than for untreated ones. Treatment area children

under 15 also miss .2 fewer days of school due to illness in the month of the

survey (t-test significant at the 10% level). Overall, these statistics point to

7 This sample, which is used in the empirical specifications, excludes the 22 women reporting

employment in ICDDR,B hospitals (the MHCFP sponsor), whose average 1996 earned income is 47 times

the average earned income of other women and therefore represent clear outliers.8 Treatment area women in both pre and post-program marriages receive larger transfers than control

area women, though the difference is much larger and significant for post-program marriages. However, it

is uncertain whether these differences existed before the MCHFP began or in fact resulted from the

program (as an effort to partially offset the negative program effects on treatment area women). The

empirical specifications control for unearned income, finding that it is significantly positively associated

with increased bargaining power over household resources (results not shown, from probit models in

Table 7). This result is shared by Anderson and Eswaran (2009). Thus, even if it was not controlled for,

the fact that treatment area women have higher levels of unearned income should only bias the empirical

results upwards toward zero.

334 C. Peters

123

substantial first-order impacts of the program on both family planning and maternal

and child health trends.

3 Theoretical framework

By virtue of being born in a treatment area village, husbands can attract wives with

the promise of contraceptive access and of raising their future children under the

MCHFP. Thus, the nature of Matlab as a patrilocal society grants treatment area

men an increased endowment through the program. A supply and demand model of

marriage provides a guide to predicting the effects of this endowment on female

power within the household (Becker 1973, Grossbard-Shechtman 1993).9 In this

model, marriage produces gains beyond the utility achieved when single. This

surplus is divided between husband and wife according to a pre-negotiated contract

that depends on individual and societal characteristics such as income earned by

each spouse, availability of fertility control technology, or sex ratios (Grossbard-

Shechtman 1993, Grossbard-Shechtman 1984, Oreffice 2007).

Holding everything else constant, women will be more willing to enter a marriage

contract as their share of the surplus increases (creating an upward sloping supply

curve for women to marriages, Becker 1973). Similarly, all else constant, treatment

area men will have a lower quantity demanded for marriages in which women receive

a higher share of the surplus. The initial allocation of marital output can be shown in

Fig. 1 as occurring at equilibrium e0. At that point, marriages are contracted between

women and treatment area men according to the division of marital output such that

women receive share Zfpre�program; and men receive the rest.10

The introduction of the family planning and health services program provides

males with an extra endowment that increases the payoff to the female of matching

with a treated male. This result implies that the new endowment given to treatment

area males should be represented by an increase in their outside options on the

marriage market (i.e. an expansion in their pool of potential mates). This desirability

can be illustrated through an increase in the supply of women searching for

treatment area men (see Fig. 1). The Matlab data is consistent with such an

increased supply, suggesting that treatment area males are in fact able to attract

mates from a larger pool after the program begins. Between the pre and post-

program periods, the prevalence of inter-area marriages (e.g. marriages between a

treatment area person and a control person) increases from 7% to 16% (t-test

significant at the 1% level, this sample includes all marriages between 1975 and

1976 for the pre-program period and 1978–1982 for the post-program period).

Using retrospective census information on birthplaces and marriage years, a

difference-in-difference specification further finds that treated males on the post-

9 An alternative framework that would predict a similar result is the collective model of marriage

(Chiappori 1992). In that model, bargaining weights determined by distribution factors govern intra-

household resource allocations, with the outcome being dependent on the marital threat point (the

allocation that occurs if couples cannot agree).10 If this figure illustrates aggregate supply and demand on the Matlab marriage market for treatment area

men, then e0 represents the allocation for the average husband and wife on that market.


123

program marriage market are 37% less likely than treated males on the pre-program

market to choose a wife from within the treatment area (a 35 percentage points

decrease significant at the 1% level, see the marginal effects from Table 3).11

As the supply of women increases, a competitive marriage market will force down

the share of the marital surplus going to women in equilibrium to e1 (see Fig. 1). The

intuition behind this result is simple. Due to the increase in total utility that follows

from participation in the program, women will place a premium on being able to

marry treatment area males. They should thus be willing to sacrifice some of the

overall marital surplus in order to obtain (or hold onto) these men.12 Moreover, this

adjustment should be able to be made not only for couples who get married after the

program begins, but also for those who were married prior to the start of the program.

For those couples in pre-existing marriages, the new program endowment given to the

husband makes him more attractive to other women, thereby increasing his

opportunities outside of marriage if he chooses to divorce. Even though only 7% of

my data sample ever gets divorced, this threat may still be relevant for the Matlab

population (it is just rarely reached because women will tolerate many unfavorable

changes within their own marriages before turning to divorce).13

0e

1e

fpreS f

postS

fpostz

fZ(Share of Output going to Female)

fprogrampostS −

Supply of Women to Treatment Area Men

fprez

fprogrampreS −

fD

Fig. 1 Female share of marital output

11 Coefficient and standard error estimates for all probit models reported in this paper are available upon

request.12 Marriages in Matlab are usually arranged by the couple’s families, with over 75% of marriages in my

sample being arranged directly by the couples’ parents. For simplicity, I assume fully benevolent parents

who, despite arranging the marriage, do not consider any utility other than their child’s when making

decisions.13 Due to limited education and the widespread practice of purdah, Matlab women have few

opportunities outside of marriage (Bhuiya et al. 2005). When divorce does occur, it is often initiated by

the husband rather than the wife. The exogenous shock of the family planning and health services

program will not be enough to trigger divorce among Matlab women, because the gains to marriage for

them remain so large (Weiss 2001).

336 C. Peters

123

In their development of a supply and demand model for marriage, Becker (1973)

and Grossbard-Shechtman (1993) suggest that the marital surplus may be divided in

several ways, such as dowry payments, variations in labor supply, share of

consumption expenditures, or spousal power in the household. However, despite

the potential for multiple margins of adjustment, it has been common in much of the

literature on marriage to assume that couples will negotiate along only one such

margin at a time. For instance, the intra-household bargaining literature focuses on

differences in spousal power in the household (recent examples of bargaining models

related to fertility decisions include Oreffice (2007), Chiappori and Oreffice (2008),

and Rasul (2008)). In these types of models, the introduction of the family planning

and health services program would be expected to result in lower bargaining power for

treatment area women relative to the power of women in the comparison area.

Alternatively, a strand of the development literature often assumes that the extra

marital surplus attained through the program endowment will instead be transferred

before marriage through a dowry payment (e.g. Anderson 2000, Arunachalam and

Naidu 2008). This literature argues that the marriage market can completely offset

any potential change within marriage due to the MCHFP endowment, since women

will pay increased dowries. However, forcing all changes to occur solely along one

dimension may be too restrictive. There is no ex-ante reason to believe that

adjustments cannot occur along multiple margins simultaneously—for instance,

women may choose to pay only some portion of that increased dowry in return for

some smaller change in bargaining power. Furthermore, this option could be chosen if

utilities are only partially transferable between husband and wife.

4 Empirical estimation

A supply and demand model for marriage predicts that the MCHFP will decrease

the share of marital output given to the wives of treatment area husbands relative to

Table 3 Difference-in-difference for effect of MCHFP on the pool of mates for treatment area males

Dependent variable—wife from treatment area

Marginal effects

Treatment area .85*** .85***

Post-program marriage 19*** -.19***

Treatment area 9 post-program marriage -.35*** -.35***

Demographic controls? No Yes

R2 .43 .43

Obs. 9,635 9,604

Dependent variable is whether husband chooses a wife from the treatment area. Robust standard errors are

clustered by program treatment block. *** indicates significance at 1% level. All models estimated using

probit, and marginal effects reported are calculated at the mean of the explanatory variables. Demo-

graphic controls include: age at marriage, age-squared, spouse age at marriage, spouse age-squared, years

of education, and spouse years of education. Sample includes all marriages in the Matlab area between

1975–1976 (pre-program) and 1978–1982 (post-program), from ICDDR,B


123

comparison area wives. I therefore expect to empirically observe adjustments along

one or more margins within marriage, taking the form of increased dowry payments

to treated men and/or lower bargaining power of the women who marry those men.

For treatment area wives in pre-program marriages who can only adjust along one

margin (because dowry payments were already made), their bargaining power over

household resources should decline relative to control area wives in exchange for

retaining their treated husbands. Women in post-program marriages, however, may

either choose to adjust fully along one margin or to adjust along both.

After restricting the sample to only those observations that report having received

some form of positive dowry (33% of marriages), a difference-in-difference tobit

model is used to estimate differential changes in the amount of dowry payments

between program groups. This hedonic model determines dowry as a function of the

program endowment of the male, the year of marriage (to account for inflation), the

interaction of the endowment with marriage year, control variables for age and

education of both partners, and an indicator for whether the father of the bride or of

the groom was wealthier at the time of marriage.

Probit models then estimate the bargaining power of the female in a given match,

using all married women surveyed by the MHSS dataset who are fertile for some

period after the first year of the program.14 Female decision-making power is

measured as the percentage of women who do not need permission from their

husband or another household member to make daily bazaar, betel leaf, sari, or

children’s clothing purchases.15 Descriptive statistics suggest that women living in

the program area are on average 5.5 percentage points less likely to be able to

independently purchase any of these items during the year of the survey (a

differential of 39%, see the third column of Table 4).

Table 4 Descriptive statistics for bargaining power over purchases (currently fertile married women)

Able to make purchases Treatment Control Difference

Items at daily bazaar .13 (.33) .18 (.39) -.05*** (.01)

Betel leaf .19 (.39) .25 (.43) -.07*** (.01)

Saris for self .11 (.32) .16 (.37) -.05*** (.01)

Children’s clothing .11 (.31) .16 (.37) -.05*** (.01)

Obs. 1,630 1,569

All variables measuring income or value of help received are measured in taka

Standard deviations in parentheses (standard errors in parentheses for difference). *** indicates t-test

significant at 1% level

14 A few women are missing spousal information, so I lose observations depending on the control

variables used.15 I focus on these specific purchases because they are considered large purchases in the Matlab setting,

following Williams (2005). The data also includes information on small purchases (kerosene or cooking

oil for the family, bangles or soap for their own use, and sweets or ice cream for children). However, the

significant differences that I find in probit models do not extend to the ability to make small purchases,

likely because the amount of resources required are simply too small to matter in daily life. In addition,

Williams (2005) observes that several papers have argued that the ability to make small purchases is not

adequately reflective of female power in rural Bangladesh.

338 C. Peters

123

Because the family planning and health services program was an unanticipated

exogenous shock for already married couples in the treatment area (pre-program

marriages), reverse causality is not a concern in those regression results.16 However,

there may be concern about selection for women who marry after the program

begins (post-program marriages), since women who choose to marry treatment area

men may be systematically different than women who marry into the comparison

area. This selection can work in two directions—low-quality, low-health women

may select into marriages with treatment area men in order to obtain health services

for themselves and their future children (causing a downward bias in the regression

results), or it may be the case that treatment area men take advantage of their new

program endowment to secure higher-quality women on multiple dimensions

(causing an upward bias).17 In order to account for either type of selection, the

regressions control for observable characteristics of the individual that may be

related to both her health and quality on the marriage market (including the

percentage of married years she has been fertile, total births, years of education,

current age, age-squared, dowry, whether she has a job, whether she is Muslim,

earned income over the past year, and unearned income as measured by the value of

help received from family or friends in the past year). The regressions also control

for quality of the individual’s spouse, including spousal years of education, current

age, age-squared, earned income, and unearned income. Finally, they control for the

socioeconomic status of the individual’s household in the form of household

income, household income-squared, and the amount of land owned by the

household head (income and land variables are logged and scaled in tens of taka

or decimals). Further variables control for the female’s relationship to her household

head (wife, daughter, mother, son’s wife, or self) and whether or not she is currently

co-residing with her husband. Regression errors are clustered by the six program

blocks.

5 Results

5.1 Dowry payments

For women who married before the MCHFP began, there are no significant

differences between the amounts of dowries paid to treatment or control area males

(see the first row of Table 5 and the pre-program marriage results from column 2 of

Table 6). Column 1 of Table 6 shows that there are also no significant pre-existing

trends in dowry payments across the treatment and control areas for pre-program

marriages, as indicated by the coefficient on the interaction between treatment status

16 The exception to being exogenous to female choices occurs if there are program spillovers with

untreated women seeking services, but as discussed earlier, these exceptions should only bias the results

towards zero.17 I do not find evidence for this second type of selection; the quality of the match achievable by

treatment area males does not appear to change along observable dimensions outside of dowry after the

program begins. Difference-in-difference regressions show no differential changes in the age, education,

or relative wealth of spouses chosen by treatment and control area males.


123

and marriage year for pre-program marriages. If anything, the pre-program trend in

dowry payments finds men from the treatment area receiving lower dowries than

control men, which means that the significant positive program effect found in the

difference-in-difference specification for all marriages is distinguished not only

from the zero effect but from its pre-existing trend as well.

Looking at post-program marriages, column 3 of Table 6 finds that treated men

obtain dowries that are 1,192 taka higher than untreated men (this coefficient

estimate represents a 14% difference over the mean payment of 7,855 taka,

Table 6 Effects of the MCHFP on dowry received by males

Dependent variable—value of dowry recevied

Pre-program marriages Post-program

marriages

All marriages

(1) (2) (3) (4) (5) (6)

Treatment area 2029.87

(7037.20)

-609.82

(776.81)

1191.55**

(565.17)

-7385.27

(4699.46)

-8542.36*

(5075.32)

-9901.53*

(5340.12)

Marriage year 106.43

(71.90)

88.24

(53.26)

435.57***

(62.09)

244.44***

(43.20)

245.48***

(46.29)

236.45***

(49.27)

Treatment area 9

marriage year

-39.04

(103.45)

97.05*

(55.55)

111.65*

(59.93)

124.52**

(63.03)

Demographic

controls?

Yes Yes Yes Yes Yes Yes

R2 .02 .02 .01 .01 .02 .01

Obs. 102 102 560 672 599 557

Sample Any

marriage

Any

marriage

Any marriage Any

marriage

First

marriage

No migrants

Dependent variable is value of dowry paid to male in taka and is conditional on having received a positive

dowry. Standard errors in parentheses. *** indicates significance at 1% level; ** indicates significance at

5% level; * indicates significance at 10% level. All models estimated using tobit. Demographic controls

include: age at marriage, age-squared, spouse age at marriage, spouse age-squared, years of education,

spouse years of education, dummy for whether father-in-law wealthier than father at time of marriage.

Sample with no migrants includes only males who live in their birth village in 1996

Table 5 Mean value of dowries received by males

Value of dowry received

Treatment Control Difference

Pre-program marriages 3593.55 (3635.1) 3586.70 (5649.52) 6.84 (927.84)

Obs. 55 47

Post-program marriages 8408.23 (8444.05) 7302.94 (6434.71) 1105.29* (632.06)

Obs. 271 289

Dowry value is in taka and is conditional on having received a positive dowry

Standard deviations in parentheses (standard errors in parentheses for difference). * indicates t-test sig-

nificant at 10% level

340 C. Peters

123

significant at the 5% level). The difference-in-difference specification confirms that

men in the treatment area begin receiving higher dowries than untreated men in

1977, the first year of the family planning and health services program (significant at

the 10% level, see column 4 in Table 6). This result is robust to restricting the

sample to first marriages only, since second marriages may be systematically

different (see column 5 of Table 6).

I also re-estimate the difference-in-difference specification using a falsification

test modeled after Aghion et al. (2008), which replaces the actual program treatment

and comparison groups with false ones. The first false group consists of northern

vs. southern villages, while the second group divides households on either side of a

flood protection embankment built in the area. In both cases, there is no evidence of

a statistically significant program effect where I do not expect to see it. Moreover,

when the false program groups are included as control variables alongside the real

groups, the effect of the actual MCHFP on dowry payments is always greater than

the effect of any false program.18 These results provide further support in favor of

the MCHFP as the driving force behind changes in dowry payments to treatment

area males.

5.2 Female bargaining power

Probit models with demographic controls indicate that fertile, married women living

in the treatment area are 5 to 7 percentage points more likely than comparison

women to require permission for purchases (these marginal effects represent a 35%

differential from the mean and are significant at the 1% level, see Table 7).19 These

results remain robust to excluding the youngest cohorts (women born in 1970 or

after) from the sample (results available in Appendix 1). I also decompose the effect

of treatment access by the time of marriage for all women fertile during some period

of the program. Treatment area women in both pre-existing and post-program

marriages are between 4 and 7 percentage points less likely than comparison area

women to be able to make purchases independently of their husbands (a 33%

differential from the mean, significant at the 5% level; results available in Appendix 1).

These findings suggest that the program does indeed provide an increased

18 For these tests, the t-statistic for the actual program coefficient is always greater than the t-statistic for

any false program. When the actual and false program groups are both included in the same regression,

the actual program coefficient is always significant at the 10% level, and the false program coefficient is

never significant.19 I focus on measures of decision-making power over purchases because they are reflective of the

allocation of resources within the household. There are several other measures of female empowerment

(Williams 2005), including freedom of movement outside the bari and preferences over modesty in the

public sphere (i.e. the use of head coverings or burqas). However, because these measures may be

governed largely by individual preferences or perceptions of status rather than control over the allocation

of household resources, the MCHFP should not necessarily induce any relative changes in these

measures. Accordingly, probit models using these measures as dependent variables do not show strong

differences between treated and untreated women. Moreover, the MCHFP has a statistically significant

effect on the measures of bargaining power used in this paper (i.e. the specifications with controls in

Table 7) even under a Bonferroni correction for false discovery rates (the Bonferroni correction in this

case requires a p-value below .0053 to retain significance at the 10% level; this number is based on the

identification of 19 different empowerment outcomes within the MHSS data).


123

Tab

le7

Pro

bit

mo

del

sfo

rb

arg

ain

ing

po

wer

ov

erp

urc

has

es(c

urr

entl

yfe

rtil

em

arri

edw

om

en)

Dep

enden

tvar

iable

—ab

leto

mak

eth

efo

llow

ing

purc

has

e

Baz

aar

item

sB

etel

leaf

Sar

isK

id’s

cloth

es

ME

ME

ME

ME

ME

ME

ME

ME

ME

ME

ME

ME

ME

ME

ME

ME

Tre

atm

ent

area

-.0

5***

-.0

5***

-.0

7***

-.0

5**

-.0

7***

-.0

7***

-.0

9***

-.0

7***

-.0

5***

-.0

5***

-.0

7***

-.0

5***

-.0

5***

-.0

5***

-.0

6***

-.0

4**

Dem

ogra

phic

contr

ols

?

No

Yes

Yes

Yes

No

Yes

Yes

Yes

No

Yes

Yes

Yes

No

Yes

Yes

Yes

All

vil

lages

or

bord

ervil

lages

only

?

All

All

Bord

erA

llA

llA

llB

ord

erA

llA

llA

llB

ord

erA

llA

llA

llB

ord

erA

ll

Incl

ude

wom

en

whose

husb

ands

mig

rate

d?

Yes

Yes

Yes

No

Yes

Yes

Yes

No

Yes

Yes

Yes

No

Yes

Yes

Yes

No

R2

.01

.04

.08

.05

.01

.04

.06

.04

.01

.04

.09

.05

.01

.04

.07

.05

Obs.

3,1

99

2,4

78

816

1,9

59

3,1

99

2,4

78

816

1,9

59

3,1

99

2,4

78

816

1,9

59

3,1

99

2,4

78

816

1,9

59

Obse

rvat

ions

are

mar

ried

fem

ales

fert

ile

in1996.

Abord

ervil

lage

isdefi

ned

asa

trea

tmen

tar

eavil

lage

bord

erin

ga

contr

ol

area

vil

lage,

or

vic

ever

sa.

Wom

enw

hose

husb

ands

hav

enot

mig

rate

d

incl

ude

all

wom

enw

ith

husb

ands

curr

entl

yli

vin

gin

the

husb

and’s

bir

thvil

lage

Dep

enden

tvar

iable

isw

het

her

fem

ale

nee

ds

per

mis

sion

from

husb

and

or

oth

erfa

mil

ym

ember

topurc

has

eth

eit

em.R

obust

stan

dar

der

rors

are

clust

ered

by

pro

gra

mtr

eatm

ent

blo

ck,an

dm

argin

al

effe

cts

report

edar

eca

lcula

ted

atth

em

ean

of

the

expla

nat

ory

var

iable

s.***

indic

ates

signifi

cance

at1%

level

;**

indic

ates

signifi

cance

at5%

level

.A

llin

com

ean

dla

nd

ow

ned

var

iable

sar

e

logged

and

inte

ns

of

taka

(ten

sof

dec

imal

sfo

rla

nd).

Dem

ogra

phic

contr

ols

incl

ude

tota

lnum

ber

of

bir

ths,

whet

her

fem

ale

curr

entl

yhold

souts

ide

job,

fem

ale

age,

fem

ale

age-

sq,

mal

eag

e,m

ale

age-

sq,fe

mal

eyea

rsof

educa

tion,

mal

eyea

rsof

educa

tion,

whet

her

adow

ryis

pai

dat

tim

eof

mar

riag

e,fe

mal

eea

rned

inco

me

(ln),

mal

eea

rned

inco

me

(ln),

fem

ale

val

ue

of

outs

ide

hel

pre

ceiv

ed

(ln),

mal

eval

ue

of

outs

ide

hel

pre

ceiv

ed(l

n),

house

hold

inco

me

(ln),

house

hold

inco

me-

sq(l

n),

land

ow

ned

by

house

hold

(ln),

per

centa

ge

of

mar

ried

yea

rssp

ent

fert

ile,

Musl

imvs.

Hin

du

house

hold

,co

resi

den

thouse

hold

,re

lati

onsh

ipto

house

hold

hea

d

342 C. Peters

123

endowment to the male and influences not only dowries but also bargaining within

marriage.

I further restrict the sample to treatment villages sharing a border with a control

village and vice versa (e.g. border villages), observing that the magnitude of the

difference between the relative bargaining power of treatment and comparison area

females becomes even greater with this sample (see Table 7). This finding is

consistent with the program endowment effect— Matlab women yield bargaining

power to obtain or hold onto treated men. Those women living near the program

border search for mates on a marriage market with a mixed supply of treated and

untreated males. As a result, they are relatively more likely to be stuck with an

untreated man when compared with women living in the far reaches of the treatment

area (where all men on their local marriage market are treated). Thus, border women

should be willing to sacrifice more for treated men than their neighboring treatment

area women near the edges of Matlab.20

5.3 Robustness tests

Since program treatment status is a village-level indicator, village fixed-effects

cannot be used to control for unobservable village-specific characteristics. However,

the results remain robust to controlling for several observed village characteristics.21

Another concern may be that men migrate after the program begins in order to

obtain permanent access to services, either as children, or as adults in order to make

themselves more attractive marriage partners. However, 80% of men in the sample

currently live in the same village of their birth, and a sample restricted to only these

non-migrant men does not change the difference-in-difference results for dowry

payments (see column 6 of Table 6). The results on bargaining power are also

robust to a sample of the wives of these non-migrant men (see Table 7). This

suggests that the empirical results are not biased due to endogenous sorting of

households or men into the treatment area.22

20 This result distinguishes my model from that of Arunachalam and Naidu (2008), who explain the

increase in dowry payments paid to treatment area men after the Matlab program begins by assuming that

women are pre-paying for a decrease in the price of contraception. However, this explanation would also

predict that such compensation should be similar throughout the program area, rather than varying based

on village location. Thus, the variance in bargaining power according to treatment village location is

inconsistent with their model. This result also precludes another potential explanation for the empirical

results of this paper- that dowry and bargaining power adjust in order to account for possible increases in

female life expectancy due to improved maternal health services from the program. If these adjustments

can be entirely attributed to changes in life expectancy, then the differences in bargaining power should

be uniform throughout the program area, regardless of village location (due to the fact that increases in

life expectancy should likewise be uniform throughout the program area).21 Results available in Appendix 1. Village controls include travel time in minutes to the nearest large

market, travel time in minutes to the nearest small market, the proportion of households with electricity,

whether the village is protected by the Meghna-Dhonogoda flood embankment, and dummy variables for

whether the village has a credit institution, irrigation for crops, some type of cottage industry, or some

type of other industry (including a mill, factory, or workshop).22 Since only two women in the full sample have husbands born after the program begins, there is no

concern about endogenous sorting before the husband’s birth (i.e. that his parents moved into the

treatment area before he was born, in order to ensure his access to program services).


123

A final concern may be that the empirical results are also consistent with a

scenario where men choose younger wives after the introduction of family planning

and birth control. This occurrence could create a glut of women on the marriage

market, which might then result in increased dowry payments and lower bargaining

power in marriage. However, descriptive statistics from the first row of Table 2

show that the average age of marriage for women has increased over time and is not

significantly different across the treatment and comparison areas. Furthermore, a

difference-in-difference specification using the full sample of marriages does not

find any evidence that treatment area men area shift toward marrying younger

women after the program begins, compared to comparison area men (results

available in Appendix 2).23

5.4 Further evidence: unaffected cohorts

If the family planning and health services program is driving the observed changes

in female decision-making power within the household, then this effect should be

absent for those groups of women who are not impacted by the program.

Accordingly, estimates for the sample of unmarried women who are in their fertile

years in 1996 show no differences between the power of treatment and control area

women over resources (see Table 8). Some members of a second cohort of

ineligible women (those who are currently separated or divorced) may potentially

still participate in the child health portion of the program (regardless of fertility

status), making expected differences in bargaining power harder to predict.

However, if marriage to a treated man is the key reason for the decreased bargaining

power of women, then any woman outside of that state should not show such a

change. Although the magnitudes of the marginal effects indicate that separated and

divorced women in the treatment area are less likely than control women to

23 Finally, although control variables account for selection of women into post-program marriages with

treatment area men along observable measures, there may exist some remaining concern regarding

potential selection on unobservables. A potential instrument for a selection model must be correlated with

a female’s choice to marry a treatment area male while remaining independent of her subsequent

bargaining power within that marriage, which is itself a function of her outside options for any future

marriage. Since a plausible instrument does not exist, I create propensity scores (Rosenbaum and Rubin

1983) for all women in post-program marriages that estimate their probability of marrying a man living in

the treatment area, based on a probit model with the following regressors: age at marriage, age at

marriage squared, education, marriage year, religion, and age at menarche. I keep only those observations

with propensity scores in the area of common support [.28, .80], which eliminates comparison women

who are most unlike treatment area women in their observed covariates. Following Robins and Rotnitzky

(1995)) and Hirano and Imbens (2002), the inverse of each observation’s probability of marrying into the

treatment area becomes weights in the post-program marriage regressions (the probability of marrying

into treatment for those in the treatment area is the estimated propensity score, and the probability of

marrying into treatment for those in the comparison area is one minus the estimated propensity score). I

balance the mean propensity scores between the treatment and comparison area within 8 blocks, and

bootstrap the standard errors from the weighted regression results. Even after including these propensity

score weights, women who marry men in the treatment area are still on average nearly 5 percentage points

less likely than women in the comparison area to make purchases (significant at the 1% or 5% level in

each regression). Thus, it does not appear that any bias resulting from unobserved selection into post-

program marriages with treated men is likely to be large enough to fully account for the difference in

bargaining power observed between treated and untreated women.

344 C. Peters

123

independently make purchases, these differences are not statistically significant.24

This result is consistent with the endowment effect of the MCHFP being the

underlying cause of decreased power for treated cohorts.

6 Conclusion

The Matlab family planning and health services program has reduced fertility rates

and improved child quality. Using a supply and demand model of marriage based on

the work of Becker (1973) and Grossbard-Shechtman (1993), I outline how this

program endowment alters the division of the marital surplus. In order to obtain

access to program services, women in post-program marriages are willing to pay

dowries to treatment area men that are 14% higher than payments to men in the

comparison area, and this result is corroborated through a difference-in-difference

specification. The program also decreases female independence over purchases by 5

to 7 percentage points when compared to women in the comparison area (a

difference of 35%). The observation of marginal adjustments to the program

Table 8 Probit models—unaffected cohorts

Dependent variable- able to make the following purchase

Bazaar items Betel leaf Saris Kid’s clothes

ME ME ME ME

All unmarried women (ages 13–45)

Treatment area -.00 -.00 -.00 .01

R2 .50 .55 .43 .52

Obs. 246 246 246 246

Currently separated and divorced women

Treatment Area -.04 -.04 -.01 -.02

R2 .57 .62 .58 .56

Obs. 47 51 47 47

These specifications include the full range of controls used in earlier specifications, with the exception of

spousal and dowry information. About half of unmarried sample is comprised of never-married

women,with widows, separated, and divorced women comprising the rest

Dependent variable is whether female needs permission from husband or other family member to pur-

chase the item. Robust standard errors are clustered by program treatment block, and marginal effects

reported are calculated at the mean of the explanatory variables. All income and land owned variables are

logged and in tens of taka (tens of decimals for land). Demographic controls include total number of

births, whether female currently holds outside job, female age, female age-sq, male age, male age-sq,

female years of education, male years of education, whether a dowry is paid at time of marriage, female

earned income (ln), male earned income (ln), female value of outside help received (ln), male value of

outside help received (ln), household income (ln), household income-sq (ln), land owned by household

(ln), percentage of married years spent fertile, Muslim versus Hindu household, coresident household,

relationship to household head

24 Lack of statistical significance may be a result of the small sample size (40–50 observations), so

caution must be used when interpreting this result.


123

endowment both before and within marriage suggests that marital contracts are

negotiated along multiple margins. Because these shifts appear to stem from

unintentionally determining treatment access based on location of the husband, the

results of this paper underline the importance of carefully targeting whether the

husband or wife will control access to such a program. With its current design,

Matlab women may experience some unintended effects on dowries and bargaining

power, though the welfare gains of the program itself may remain large.

Acknowledgments I thank Terra McKinnish, Randall Kuhn, Jennifer Lamping, Tania Barham, and

especially A. Mushfiq Mobarak and Murat Iyigun for valuable discussion and suggestions. All errors

remain my own.

Appendix 1

See Table 9.

Table 9 Alternative specifications and robustness checks

Dependent variable—able to make the following purchase


ME ME ME ME

Only women born before 1970

Treatment area -.05* -.07*** .04*** -.04**

Demographic controls? Yes Yes Yes Yes

R2 .03 .03 .03 .03

Obs. 2,051 2,051 2,051 2,051

Pre-program marriages

Treatment area -.04** -.07** -.05*** -.04***


R2 .04 .02 .03 .03

Obs. 1,793 1,793 1,793 1,793

Post-program marriages

Treatment area -.05*** -.06*** -.05*** -.04***


R2 .06 .05 .07 .07

Obs. 1,351 1,351 1,351 1,351

Full sample with extra controls for village-specific characteristics

Treatment area -.04*** -.06*** -.05*** -.04***


R2 .07 .06 .07 .07

346 C. Peters

123

Appendix 2

See Table 10.

Table 9 continued

Dependent variable—able to make the following purchase


ME ME ME ME

Obs. 2,105 2,105 2,105 2,105

Observations for pre-program marriages are women married before 1977 and fertile for some period after

1977; observations for post-program marriages are women married after 1977 and fertile for some period

after 1977. Observations for sample with village characteristics and non-migrants are married females

fertile in 1996. All Specifications include the full set of control variables used in Table 7. Village-specific

characteristics controlled for include: travel time in minutes to the nearest large market, travel time to the

nearest small market, proportion of households with electricity, whether the village is protected by the

Meghna-Dhonogoda flood embankment, whether the village has irrigation for fields, whether the village

has any industry or any cottage industry, and whether the village has a credit institution

Dependent variable is whether female needs permission from husband or other family member to pur-

chase the item. Robust standard errors are clustered by program treatment block, and marginal effects

reported are calculated at the mean of the explanatory variables. *** indicates significance at 1% level;

** indicates significance at 5% level; * indicates significance at 10% level. All income and land owned

variables are logged and in tens of taka (tens of decimals for land). Demographic controls include total

number of births, whether female currently holds outside job, female age, female age-sq, male age, male

age-sq, female years of education, male years of education, whether a dowry is paid at time of marriage,

female earned income (ln), male earned income (ln), female value of outside help received (ln), male

value of outside help received (ln), household income (ln), household income-sq (ln), land owned by

household (ln), percentage of married years spent fertile, Muslim vs. Hindu household, coresident

household, relationship to household head

Table 10 Effects of the MCHFP on wife’s age at marriage

Dependent variable—wife’s age at marriage

Treatment area -.34 (.57)

Marriage year .14*** (.00)

Treatment area 9 marriage year .01 (.01)

Demographic controls? Yes

R2 .49

Obs. 3,540

Dependent variable is wife’s age at marriage. Observations are all males who have married. Robust

standard errors in parentheses are clustered by program treatment block. *** indicates significance at 1%

level. Model estimated using OLS. Demographic controls include: age at marriage, age-squared, years of

education, spouse years of education, dummy for whether father-in-law wealthier than father at time of

marriage


123

References

Aghion, P., Burgess, R., Redding, S., & Zilibotti, F. (2008). The unequal effects of liberalization:

Evidence from dismantling the license Raj in India. American Economic Review, 98(4), 1397–1412.

Anderson, S. (2000). The economics of dowry payments in Pakistan. Mimeo: Tilburg University.

Anderson, S., & Eswaran, M. (2009). What determines female autonomy? Evidence from Bangladesh.

Journal of Development Economics, 90(2), 179–191.

Arunachalam, R., & Naidu, S. (2008). The price of fertility: marriage markets and family planning inBangladesh. Working Paper.

Becker, G. (1973). A theory of marriage: Part I. The Journal of Political Economy, 81(4), 813–846.

Bhuiya, A., Chowdhury, A., Momen, M., & Khatun, M. (2005). Marital disruption: Determinants and

consequences on the lives of women in a rural area of Bangladesh. Journal of Health, Population,and Nutrition, 23(1), 82–94.

Chiappori, P. A. (1992). Collective labor supply and welfare. The Journal of Political Economy, 100(3),

437–467.

Chiappori, P. A., Fortin, B., & Lacroix, G. (2002). Marriage market, divorce legislation, and household

labor supply. Journal of Political Economy, 110(1), 37–72.

Chiappori, P. A., & Oreffice, S. (2008). Birth control and female empowerment: An equilibrium analysis.

Journal of Political Economy, 116(1), 113–140.

Fauveau, V., & Chakraborty, J. (1994). Family planning and maternal and child health services in Matlab.

In V. Fauveau (Ed.), Matlab: women, children, and health. ICDDR,B Special Publication No. 35.

Grossbard-Shechtman, A. (1984). A theory of allocation of time in markets for labour and marriage. TheEconomic Journal, 94(376), 863–882.

Grossbard-Shechtman, S. (1993). On the economics of marriage: A theory of marriage, labor, anddivorce. Boulder: Westview Press.

Hirano, K., & Imbens, G. W. (2002). Estimation of causal effects using propensity score weighting: An

application to data on right heart catherization. Health Outcomes and Research Methodology, 2,

259–278.

Joshi, S., & Schultz, T. (2007). Family planning as an investment in development: Evaluation of aprogram’s consequences in Matlab, Bangladesh. Yale University Economic Growth Center

Discussion Paper No. 951.

Koenig, M., Rob, U., Khan, M., Chakraborty, J., & Fauveau, V. (1992). Contraceptive use in Matlab,

Bangladesh in 1990: Levels, trends, and explanations. Studies in Family Planning, 23(6), 352–364.

Manser, M., & Brown, M. (1980). Marriage and household decision-making: A bargaining analysis.

International Economic Review, 21(1), 31–44.

McElroy, M., & Horney, M. (1981). Nash-bargained household decisions: Toward a generalization of the

theory of demand. International Economic Review, 22, 333–349.

Miller, G. (2010). Contraception as development? New evidence from family planning in Colombia. TheEconomic Journal, 120(545), 709–736.

Oreffice, S. (2007). Did the legalization of abortion increase women’s household bargaining power?

Evidence from labor supply. Review of Economics of the Household, 5(2), 181–207.

Piet-Pelon, N., & Rob, U. (1997). Quality of care: The progress in the Bangladesh family planningprogram. Report Prepared for the Population Council, Bangladesh.

Rahman, O., et al. (1999). The Matlab health and socioeconomic survey: Overview and user’s guide.

DRU-2018/1, RAND, Santa Monica CA.

Rasul, I. (2008). Household bargaining over fertility: Theory and evidence from Malaysia. Journal ofDevelopment Economics, 86(2), 215–241.

Robins, J., & Rotnitzky, A. (1995). Semiparametric efficiency in multivariate regression models with

missing data. Journal of the American Statistical Association, 90(429), 122–129.

Rosenbaum, P., & Rubin, D. (1983). The central role of the propensity score in observational studies for

causal effects. Biometrika, 70, 41–55.

Sinha, N. (2005). Fertility, child work, and schooling consequences of family planning programs.

Economic Development and Cultural Change, 54(1), 97–128.

Weiss, Y. (2001). Family theory: Economics of marriage and divorce. In N. Smelser & P. Baltes (Eds.),

International encyclopedia of the social and behavioral sciences. Amsterdam: Elsevier.

Williams, J. (2005). Measuring gender and women’s empowerment using confirmatory factor analysis.

IBS Working Paper 01, University of Colorado at Boulder.

348 C. Peters

123

Effects of family planning and health services on women’s welfare: evidence on dowries and...

Documents

Transcript of Effects of family planning and health services on women’s welfare: evidence on dowries and...