Marriage Market Effects of a Wealth Shock in Bangladesh

A. Mushfiq Mobarak*
Yale University

Randall Kuhn
University of Denver

Christina Peters
Metro State College of Denver

Abstract
Conditions of marriage such as dowries and consanguinity influence women’s subsequent life
outcomes. However, research on the determinants of these conditions is largely descriptive. This
paper uses a wealth shock from the construction of a flood protection embankment in rural
Bangladesh coupled with data on the universe of all 52,000 marriage decisions between 1982 and
1996 to examine changes in marital prospects for protected households after embankment
construction relative to unprotected households living on the other side of the river. First we use
two-sided matching models with difference-in-difference specifications to describe the changes in
the marriage market, and show that protected households commanded larger dowries, married
into wealthier families, and became less likely to marry biological relatives. The marriage market
becomes more segregated by wealth, but the positive wealth shock does not allow women to
delay marriage or reduce spousal age gaps. The same family is 40% less likely to marry a
younger child to a cousin after the wealth shock, compared to their older child who married prior
to the embankment construction. Second, we try to understand the structural changes that led to
this drop in consanguinity, and find that liquidity-constrained households use within-family
marriage (where one can promise ex-post payments) as a form of credit to meet up-front dowry
demands, and the wealth shock relaxed this need for taking an adverse biological risk.

JEL Codes: O1, J12, O13

Keywords: Marriage, Embankment, Flood Protection, Consanguinity

November, 2009

* Corresponding Author: ahmed.mobarak@yale.edu or 203-432-5787.

Address: 135 Prospect Street, P.O. Box 208200, New Haven, CT 06520-8200
1. Introduction

Across the world, a woman’s marital prospects have important implications for

her subsequent life outcomes. Characteristics of the bride and her family at the time of

marriage in conjunction with the characteristics of her spouse and his family determine

the conditions of marriage such as dowries, marrying biological relatives, and age at

marriage.1 These conditions in turn affect socio-economic outcomes for the woman and

her children, including the likelihood that she will have to endure domestic violence, her

social status in her husband’s home, her school attainment, health status, and her control

over reproductive choices.2 Marrying a cousin or uncle, a surprisingly common practice

around the developing world, can decrease the amount of dowry required, but increases

the risk of genetic diseases among offspring.3

Although the literature on the consequences of marriage is large, the evidence on

the determinants of the conditions of spousal matching is mostly qualitative or descriptive

(e.g. Fruzzetti, 1982; Huq and Amin, 2001). A few studies account for multiple co-

varying determinants of marital prospects, and use cross-sectional regressions on

relatively small samples of survey data from rural India to show that older, taller, more

educated grooms of high caste living in areas with a larger supply of potential brides

command larger dowries, and that spouses mate assortatively in age and education.4 The

fact that families can offer compensating differentials along many unobservable

dimensions in order to secure a desirable match is a significant challenge to empirically

1
Dalmia, 2004; Rao, 1993; Foster 1998
2
Jahan, 1991; Tiemoko, 2001; Bloch and Rao, 2002; Wickrama and Lorenz, 2002; Jensen and Thornton,
2003; Suran et al, 2004; UNICEF 2005; Field and Ambrus, 2008
3
In the mainly Muslim countries of West, Central, South Asia and North Africa, marriage between close
relatives account for between 20 and 50 percent of all unions. Cousin marriage appears to be a social norm
in Pakistan, Afghanistan, Iraq, Yemen, where about 50% of marriages are between first cousins. Caldwell
et al., 1983; Bittles, 1994; Bittles 2001; New York Times 2003, BBC 2005.
4
Rao, 1993; Deolalikar and Rao, 1998; Dalmia and Lawrence, 2001; Dalmia, 2004

1
identifying the precise determinants of marriage outcomes in these studies. We exploit

the construction of a flood protection embankment in rural Bangladesh coupled with pre

and post embankment data on 33,000 marriages in treatment and control villages to

examine how a plausibly exogenous change in certain households’ wealth manifests itself

in marriage market outcomes such as dowries, socio-economic status of the spouse, age

at marriage, and consanguinity (i.e. marrying biological relatives).

The flood protection embankment in rural Bangladesh that we study induced a

discrete improvement in socio-economic conditions for families living on the

embankment side of the river relative to the opposite bank that remained unprotected.

The major effect of the embankment was to extend the crop growing season, thereby

increasing relative wealth for households on the protected side, though it also may have

reduced flood risk exposure. We investigate differential changes in the conditions of

marriage for protected households using panel data on the entire universe of marriages

across a fourteen-year pre and post-embankment period.

Our paper is constructed in two parts. The first part uses stylized two-sided

matching models of the marriage market along with difference-in-difference

specifications to describe changes in the marriage market following the wealth shock, and

documents changes in dowries, spousal socio-economic status, and a drop in marriages

between biological relatives. In the second part, we explore structural changes that led to

this drop in consanguinity. We find that liquidity-constrained households facing dowry

demands used consanguinity as a costly solution for their lack of access to credit to make

the up-front dowry payment.

2
 For the first part, the model predicts that the protected are likely to secure better

matches only along characteristics that are complementary. For example, if the wealth

that a man and woman bring in to a marriage are complementary inputs in generating

marital surplus, then the protected would in general choose to (and be able to) marry into

wealthier households. If age at marriage is not a complementary input, protection will

not necessarily change spousal age or age gaps, since the protected are not willing to pay

relatively more than the unprotected for this characteristic.5 Furthermore, if the

embankment’s primary contribution is to lower flood risk exposure, then we should

observe negative assortative matching in protection post-embankment. The unprotected

have the largest marginal gain from bonding with a protected family, and are therefore

willing to pay the most to secure that match. A corollary is that the protected should

receive larger dowries.

Difference-in-difference specifications that explore changes in marriages post-

embankment show that individuals from protected households experienced a 3 percent

higher likelihood of marrying into wealthier households (as measured by land ownership)

post embankment relative to those that remained unprotected. The embankment had a

larger wealth effect on farmers, and triple difference (pre/post, un/protected, by

occupation) results confirm that agricultural households drive this change. Protected men

command larger dowries following embankment construction, but there are no significant

changes in age at marriage. We also do not find evidence of assortative matching in

protection status in either direction. These results provide valuable evidence on how a

wealth shock affects marriage markets in a developing society. The marriage markets

5
However, if age happens to be a relevant consideration for spousal choice, its effect will get capitalized
into the dowry transfer, and thus age may bear some relationship to protection status if the embankment
changes the ability to pay dowry.

3
becomes increasingly segregated (beneficiaries of the environmental shock command

larger dowries, marry wealthy families and crowd out others). Norms regarding the

proper age at marriage and spousal age gaps appear quite inelastic even for those who

receive the large positive shock to wealth. This first part of the paper documents general

equilibrium effects of an infrastructure commonly constructed in flood-exposed

developing countries.

The second part of the paper turns to changes in consanguinity – the practice of

marrying biological relatives – which is surprisingly common in many parts of

Bangladesh, in neighboring India and Pakistan, and more generally in the developing

countries of Asia and Africa.6 Such marriages impose adverse biological risk on children

in the increased likelihood of receiving two copies of a deleterious gene from parents,

which manifests itself in larger child morbidity and mortality rates (Bittles and Makov,

1988; Bittles, 2001; Shah et al., 1998).7 Social scientists have a limited understanding of

why so many couples accept these risks, and the wealth shock associated with the

embankment provides a unique opportunity to gain further insight into this practice.

In difference-in-difference estimates using household fixed effects, protected

households show a 3.3 percentage point larger drop (a 40 percent decrease at the mean) in

the likelihood of forming consanguineous unions following the construction of the

embankment relative to unprotected households living on the other side of the river.

Multiple mechanisms can link flood protection to consanguinity prevalence, but our

6
In the mainly Muslim countries of North Africa, West and Central Asia, and in large parts of South Asia,
marriage between close relatives account for at least 20 percent to 50 percent of all unions, with a further
2.8 billion people in countries where 1 percent to 10 percent of marriages are between biological relatives
(Bittles et al., 2001; Hashmi, 1997; Khan, 2001; http://www.consang.net).
7
Shah et al. (1998) conclude that when two first-cousins marry, they are 18 percent more likely to
experience the death of one of their children before the child is 5 years old. Bittles (2001) reports that child
morbidity rates are between 1 percent and 4 percent higher for consanguineous children. Selection issues
are generally not adequately addressed in any of these studies.

4
ancillary evidence indicates that marrying consanguineously reduces the need for dowry

payments which are often difficult for brides from poor families to make, and this dowry

constraint is relaxed for protected households following embankment construction.8

Since a bride’s parents often do not have either cash on hand or access to credit to make

the up-front dowry payment, they use within-family marriage (where it becomes possible

to promise ex-post payments) as a form of credit. Triple difference results by gender

show that it is protected women who show the larger drop in consanguinity (rather than

protected men), which is consistent with this dowry-credit constraint story, since the

dowry payment is uni-directional from women to men. The contract theory model in Do,

Iyer and Joshi (2009) formalizes this story, and we bring credible evidence to bear on this

mechanism by taking advantage of an exogenous environmental shock.

In the next two sections, we describe our data and present evidence that the

embankment created plausibly exogenous variation in wealth. Section 4 presents two-

sided matching models of the marriage market to explore how this is expected to affect

marriage market behavior. Section 5 empirically describes the changes in the marriage

market, and finally section 6 tries to understand the structural changes that led to the large

drop in consanguinity.

2. Data and Setting

Since 1963, the International Centre for Diarrhoeal Disease Research, Bangladesh

(ICDDR,B) has conducted periodic socioeconomic censuses and recorded all vital events

8
Other possible explanations for which there is less support in the data: (a) Supply Side - unprotected
households are possibly “forced into” consanguineous marriages as their external marriage market
prospects decline in relative terms, (b) Demand Side - unprotected households may seek mutual insurance
by forming a cross-generational bond with another household through consanguinity.

5
of residents in 148 villages in Matlab district. Residents of this rural area are mostly poor

and landless, subsisting off fishing, agricultural labor, and sharecropping. The Meghna-

Dhonagoda River runs through the middle of the study area, and the Water and Power

Development Authority in Bangladesh used external donor funds in 1987 to construct a

65 km embankment along the northwest bank that prevents water overflow and provides

systems for pumped drainage and irrigation along the waterway (Strong and Minkin,

1992). The embankment was breached during abnormally high floods in 1987 and 1988,

after which it was strengthened and resealed in 1989. Consequently, in the empirical

specifications, the pre-embankment period becomes 1982-1986, and the post-

embankment period covers 1989-1996.

Using ICDDR,B’s data, we observe all 33,000 marriages (or 52,000 marriage

decisions) of Matlab residents between 1982 and 1996, and we can merge census data

from 1982 to these marriage files. We know the age at marriage of each individual and

their spouse, any biological relationship between them (i.e. consanguinity status), their

wealth, occupation, and location (and therefore embankment protection status). The vital

events database also allows us to examine fertility, mortality and migration patterns at

monthly intervals for the full sample of households over the entire sample period.

We supplement these data with retrospective dowry information, cropping

practices, and land value information reported in the 1996 Matlab Health and

Socioeconomic Survey. The MHSS cross-sectional dataset covers a random sub-sample

of over 4000 households from Matlab and asks all respondents to recall dowries

exchanged during past marriage transactions (thus covering marriages both before and

after embankment construction). We use the estimated cash value of dowries as reported

6
by women who are interviewed separately from their husbands. Appendix table A1 lists

the variables used in this study and their definitions.

3. What the Embankment Construction Helps Identify

The key identifying assumption in our statistical analysis is that the construction

of the Meghna-Dhonogoda embankment induced measurable changes in socio-economic

status for people resident on the northwest side of the Dhonagoda river in Matlab district,

relative to people resident on the other side of that river (see figure 1). Before turning to

any marriage outcomes, in this section we explore: (a) whether the embankment

construction can be treated as exogenous (i.e. uncorrelated with other coincident changes

in events and conditions), and (b) what exactly were the socio-economic changes the

embankment effectuated – changes in wealth or risk exposure – that ultimately resulted in

changes in the marriage market.

The embankment was not designed as a scientifically random experiment. We

thus explore in what ways the event falls short of the ideal of a “natural experiment” or

strictly “exogenous” change. One potential concern is that people resident on the south-

east bank of the river are not an appropriate control group for the “treated” households,

since the placement of the embankment on the northwest bank may itself signal some

pre-existing differences between the two groups.

During fieldwork interviews we conducted in 2006, Matlab residents indicated

that the embankment was placed on the north-west side mainly because drainage was

worse in that bank prior to embankment construction. Further anecdotal evidence from

first-hand witnesses observe complaints of residents that the project was coordinated by

7
politicians in conjunction with the largest local landowners who actually live in Dhaka

(Briscoe, 1998; Kabir 2004). Neither of those groups are part of our sample since they do

not live in Matlab district. However, this may indicate possibility of other forms of

preferential treatment for the protected side of the river, and we therefore delve into the

program evaluation and anthropological literatures to examine these issues.

It is possible that agriculture practices are systematically different across the two

banks due to drainage differences, although program evaluations of the embankment

project report that the geographic areas experienced similar weather patterns, households

grew similar crops and reported similar incomes, and demographic distributions were

nearly identical. We find no mention of any other significant differences in technology

or other construction such as bridges that may have happened to coincide with

embankment construction (Strong and Minkin, 1992; Thompson and Sultana, 1996;

Briscoe, 1998).9 Descriptive statistics from our own data show that the protected and

unprotected groups are similar prior to construction along most observable dimensions

(see Table 1). Due to the large sample size, the marriage outcomes of interest are often

statistically different across the two groups prior to completion of the embankment, but

the magnitude of those differences remains very small.

One exception is the baseline consanguinity rate, which is appreciably larger prior

to embankment construction on the protected side. However, since our difference-in-

difference estimates rely on relative changes in marriage outcomes, the most important

concern is whether the pre-existing temporal trends were different across the two groups.

9
Since Matlab is part of a demographic surveillance area run by the International Centre for Diarrheal
Disease Research (ICDDR, B), the population has been exposed to some development programs. Our
results are robust to controlling for exposure to the largest of these programs, the Matlab Maternal and
Child Health and Family Planning program and the Safe Motherhood Initiative.

8
We show that protected and unprotected trends in the marriage outcomes of interest prior

to embankment construction were not statistically different, and the magnitudes of the

differences are very small when compared with the actual embankment effect estimates

we present later (see table 2). In fact, the (statistically insignificant) point estimate shows

that if anything, consanguinity on the embankment side was increasing relative to the

unprotected side (which also means that the relative decrease in consanguinity on the

protected side post embankment construction that we will demonstrate will be

statistically different from both the zero effect and from its pre-existing trend). Any

relative changes in trends post-embankment that our difference-in-difference estimates

uncover are therefore not merely a continuation of pre-existing differences in trends.

We also conduct falsification exercises on our DID estimates (modeled after

Aghion et al. (2008)), which replace the indicator for the actual year of embankment

construction with every possible false embankment year of the sample. In other tests, we

replace the variable for protection status (indicating which side of the embankment a

household is on) with false embankment locations of northern vs. southern villages and

the treatment and control groups of the Matlab Maternal and Child Health and Family

Planning Program, an experimental program present in the area. These tests for false

times and locations all show that statistical impacts are absent in cases where we should

not observe them (e.g. there are no statistical differences in behavior across two sub-

periods other than that of embankment construction), and that the actual embankment

effect typically trumps the false embankment effects.10

10
The t-statistic for the actual program coefficient is greater than the t-statistic for any false program in 100
percent of specifications for consanguinity and 92 percent of specifications for marrying a wealthy spouse
(i.e. one owning an above average amount of land). When the actual and false program groups are both
included in the same regression, the actual program coefficient is significant at the 10 percent level in 100

9
 Thus, though the embankment may not be a perfectly clean experiment, we have

searched for possible threats to the embankment as a source of identification. The results

support its use as a plausibly exogenous source of variation in a difference-in-difference

setting across the protected and unprotected banks of the river. It is a useful source of

variation to examine the effects of changes in wealth and risk exposure on marriage

outcomes. Furthermore, the embankment is an infrastructure projects whose effects on

the social and economic lives of the village residents are inherently interesting to

examine. We next explore the relative changes in wealth and risk exposure it induced.

3.1 Impacts of the Embankment on Income and Risk

Frequent flooding in Matlab destroys crops and induces volatility in household

income, and the embankment provides security by extending growing seasons and

increasing overall farm incomes for protected agricultural households. During fieldwork

we conducted in December 2006, Matlab residents often reported that the primary effect

of the embankment was to increase the number of crop cycles from only one per calendar

year to two or three.11 Consistent with our informal interviews, Matlab data from 1996

indicates that protected rice farmers enjoy almost one extra growing season per calendar

year compared to farmers on the other side of the river (t-test significant at the 1 percent

level). Thompson and Sultana (1996) mention that the largest effects of flood protection

projects should be on monsoon crops, and accordingly, our data shows that protected

farmers grow 2-3 times more Aman and Aus paddy (the two varieties grown during the

monsoon) per decimal of cultivated land compared to unprotected farmers (t-tests

percent of specifications for consanguinity and 54 percent of specifications for marrying a wealthy spouse.
The false program coefficient is only significant in 20 percent of consanguinity specifications and is never
significant for marrying a wealth spouse.
11
Previously, water would encroach on the agricultural land for up to 3-5 months per year.

10
significant at the 5 percent level). Meanwhile, the embankment had no discernible effect

on yields for the dry season Boro paddy.

Protected farmers should have experienced a substantial wealth effect from these

large increases in rice yields. We measure this effect directly by using principal

components analysis to construct asset indices for household wealth in 1982 and 1996

(see table 3).12 Compared to unprotected households, protected households experience a

greater increase (significant at the 1 percent level) in asset ownership between 1982 (pre-

embankment) and 1996 (post-embankment). Moreover, this change is almost entirely

driven by farmers.13 A hedonic regression of the value of land using cross-sectional 1996

data reveals that post-embankment, the unit price of land is over 3000 taka higher per

decimal on the protected side.14 Table 4 shows that the variance of assets across

households within a village (which would be linked to changes in risk exposure) does not

differentially change across the protected or unprotected banks. Consistent with the

fieldwork findings, the mean wealth effect of the embankment thus appears to dominate

changes in variance.

In addition to this wealth effect, the embankment also may have reduced the

health risk posed by stagnant floodwaters. Using DID specifications to test the

12
The index measures household ownership of any combination of the following assets: a radio, a watch or
clock, a bicycle, cows, and a hurricane lamp. Data is taken from Matlab DSS (Demographic Surveillance
System) 1982 and 1996 household censuses. We compare asset indices for only those households with
information in both years. Occupation of household head is kept as reported in 1982.
13
Both landowners and tenant farmer benefited in significant ways, which indicates that the extended
growing season on the protected side probably increased both the productive capacity of land as well as the
demand for agricultural labor. From Table 3 it may appear that the impact on tenants was larger than on
landowners, but this may be due to ‘top coding’. The index measures ownership of particular assets, and is
thus ‘capped’ when an individual reports owning all such items. Since the index values for relatively
wealthier landlords are more likely to be capped even pre-embankment, the magnitude of the change in
their index value is likely to be smaller.
14
Control variables for this regression include total land area under irrigation, distance to nearest market,
travel time to nearest market, whether the village has a credit institution, whether the village participates in
the MCHFP program, and controls for area under cultivation and cost of cultivation by crop type.

11
embankment’s effect across different sex-age categories, we find a differentially greater

decline for protected households than unprotected households in mortality rates from

diarrheal diseases and other water-borne infectious diseases.15 Specifically, adults

protected by the embankment are less likely to die from diarrheal diseases (including all

adults between 30-35, females between 45-50, and males between 55-60), and protected

adult males between 30-35 are less likely to die from infectious diseases.16

In summary, while the embankment may have affected both the mean and

variance of outcomes (wealth and risk exposure), both our fieldwork and the data analysis

are strongly suggestive the wealth effect is likely the most salient change and the

dominant embankment effect for protected Matlab residents. As figure 2a indicates, the

embankment is not a soaring barrier that can protect residents from the gushing

floodwaters that are an enduring risk to life and property periodically faced by rural

Bangladeshis. It is a more modest barrier designed to protect agricultural fields from

seasonal variation in water levels that render those fields in-arable during the monsoons.

The data on cropping cycles, agricultural yields, land values, wealth by occupation all

indicate that the embankment performs this limited function well, and bestows a positive

wealth shock on protected households.

4. Effects of the Embankment in Stylized Models of the Marriage Market

The wealth and risk effects outlined above may each have very different

manifestations in the marriage market. Highly stylized two-sided matching models can

15
Mortality rates are standardized to the sex and age category population shares of 1983, and we control for
exposure to the Matlab Maternal Child Health and Family Planning Program. Results available upon
request.
16
As we would expect, for causes of death not directly related to the embankment (such as accidents), we
see very small differences in mortality rates between protected and unprotected individuals.

12
help us understand what effects we expect to observe on a variety of marriage outcomes.

Most existing models of the marriage market describe matching functions under the

assumption that each person brings a single trait to marriage (e.g. Browning, Chiappori

and Weiss 2003, Anderson 2007, Weiss 1997), which makes generating predictions on

assortative matching relatively simple (e.g. Bergstrom 1997, Siow 1998, Anderson 2000).

We take the following approach to describe matches in a situation where multiple

attributes of protection can affect outcomes:

1. We derive analytical predictions on matching in a simplified transferable-utility

marriage market where each person has only two discrete attributes –

embankment protection status and wealth status. This model helps clarify the

basic economic intuition behind the dynamics of a more complicated market.

2. We then relax the assumptions on transferable utility, endow each person with

multiple continuous characteristics, and simulate stable matches in a larger

marriage market characterized by search frictions.

4.1 A Transferable Utility Model of the Marriage Market

Since marriages in Matlab are typically arranged by the families of the groom and

bride, we assume that preferences of the bride and her family are grouped together, as are

the preferences of the groom and his family. Males and females on the marriage market

are indexed by m and f. Each potential spouse has two relevant characteristics: the level

of wealth (wf or wm), which can be either high (H) or low (L), and the embankment status

(ef or em), which can be either protected (P) or unprotected (U). Each marriage produces

an output zfm, and there exists a medium of exchange (such as a dowry payment, which

we denote dfm) that can be used to transfer utilities from the bride to the groom. This

13
assumption (e.g. Weiss 1997, Siow 1998, Anderson 2007) simplifies the matching

problem by allowing each person to use zfm in comparing the gains across different types

of matches and against the payoff from remaining single. dfm regulates the division

between spouses, so that each person’s decision is conveniently split into: (1) choose the

match that maximizes the surplus generated from marriage, and (2) choose a value of dfm

to split that surplus.

The groom’s payoff from the marriage is dfm, while the bride’s payoff is zfm -

dfm.17 We assume the following general form for zfm: z fm = max(e f , em ) ⋅ Π ( w f , wm ) .

This formulation reflects the fact that embankment protection increases the productivity

of land by extending the crop season, which is the principal component of wealth in rural

Bangladesh. The embankment also protects from flood risk, and it is most important to

have at least one side of the newly joined families be protected, an idea embodied in the

max(ef, em) function.18 A woman may gain from starting to live under embankment

protection after marriage, and conversely, an unprotected groom’s family may gain from

forming a marital bond with a protected family where they can take refuge during a flood.

Variables ef, em (which can take on values P and U) and wf, wm (with values H or

L) are all assumed to be strictly positive so that greater wealth can be valuable even in the

∂Π ∂Π
absence of protection. P>U, H>L, and Π f = and Π m = are positive, so that
∂w f ∂wm

protection and greater wealth are both positive characteristics in the marriage market.

17
Another reasonable and important formulation of the distribution of gains from marriage assumes Nash
bargaining between spouses over the marital surplus (e.g. McElroy and Horney 1981, Rasul 2008), and
extends to other forms of bargaining (Manser and Brown 1980). The bargaining solution in turn has
implications for spousal choice (McElroy 1990), which is the focus of our analysis here.
18
The max(ef, em) function is an admittedly strong conceptualization of risk mitigation, but the idea is to
model the embankment as a substitute characteristic, and the particular form (i.e. the max function) is
chosen for convenience. The intuition we develop holds for any form of a substitute characteristic.

14
Further, we will focus on the case where there are gains to marriage: Π fm > 0 . All

couples with any wealth gain from being married relative to remaining single

when Π fm > 0 , and for the unprotected there are additional gains from marrying into a

family protected by the embankment.

Our task is to uncover a stable set of matches for the four types of men and

women in this marriage market, such that no married person would rather be single and

that no two people, married or single, would prefer to form a new union. Stability implies

a participation constraint for each woman which specifies that her payoff from marriage

must be as large as her payoff from remaining single:

max(e f , em ) ⋅ Π ( w f , wm ) − d fm ≥ e f ⋅ Π ( w f ,0) . Similarly, the participation constraint for

each man requires d fm ≥ em ⋅ Π (0, wm ) . For stable matches, a set of incentive

compatability constraints must also be satisfied for each person that specify that the

payoff from the chosen match is larger than under alternate matches:

max(e f , em ) ⋅ Π ( w f , wm ) − d fm ≥ max(e f , en ) ⋅ Π ( w f , wn ) − d fn , ∀n ≠ m , and

d fm ≥ d gm ∀ g ≠ f .

Since there are only four types of each gender, the above represents three

incentive compatibility constraints for women and a further three for men. A final

market clearing condition stipulates that for a match of type f and type m to be feasible in

the aggregate, the supply of these types must be equal.

4.2 Solution to the Transferable Utility Model

Under transferable utility and a unique output measure zfm associated with each

marriage, the stable assignment is the set of matches that maximizes total output over all

15
possible assignments.19 It is easy to verify that under complementarity ( Π fm > 0 ), the

only stable set of matches is where type (P, H) get matched to type (U, H) of the opposite

gender, while (P, L) and (U, L) also form bonds. In other words, we observe positive

assortative matching in wealth, but negative assortative matching in protection status.

In order to illustrate why these matches are optimal, it is useful to derive the result

assuming a market structure where the women can bid for the men and are the residual

claimant of the marital surplus generated (the results are analogous when men bid). The

maximum willingness to pay for a (P, H) man by each type of woman is as follows:

PH
By a (P, H) woman, WTPPH = P ⋅ Π ( H , H ) − P ⋅ Π ( H ,0)

By a (P, L) woman, WTPPLPH = P ⋅ Π ( L, H ) − P ⋅ Π ( L,0)

PH
By a (U, H) woman, WTPUH = P ⋅ Π ( H , H ) − U ⋅ Π ( H ,0)

By a (U, L) woman, WTPULPH = P ⋅ Π ( L, H ) − U ⋅ Π ( L,0)

PH
Since P>U, WTPUH > WTPPH
PH
and WTPULPH > WTPPLPH . This is because a protected

man offers greater value added to an unprotected woman than he does to a protected

woman, and the unprotected woman will therefore be willing to outbid the protected

PH
woman. Also, WTPUH > WTPULPH when Π mf > 0 . Under complementarity in the

husband’s and wife’s wealth, a wealthy woman gains greater surplus from a wealthy man

than does a low wealth woman, and will therefore be willing to outbid her. Thus the (U,

H) woman can outbid all other types of women in order to match with a (P, H) man.

The above implies that a (P, H) man will be feasible for a (U, H) woman. For this

match to occur in equilibrium, we also need to demonstrate that the (U, H) woman wants

19
This result is derived in Weiss (1997), pp. 100-101, and in Browning, Chiappori and Weiss (2005),
chapter 8.

16
the (P, H) man – that a marriage to this man generates more surplus for her than a

marriage to any other man. If the (P, H) - (U, H) match is surplus maximizing, then we

can find a transfer dfm such that the (U, H) woman and (P, H) man are better off under

this match than under any other pairing. This is easily established, as we can use the

assumptions P>U, Π m > 0 , Π f > 0 , and Π fm > 0 to show that WTPUH

PH PL
exceeds WTPUH ,

UH UL
WTPUH , and WTPUH . In other words, a protected, high-wealth woman’s desire for an

unprotected high-wealth man exceeds her desire for any other type of man.

Analogous arguments establish that (P, H) type women have the highest

willingness to pay for (U, H) type men, and achieve the largest surplus from those

matches. So for both men and women, all matches are of the form (P, H) - (U, H). Once

all these (P, H) - (U, H) men and women are paired up, the remaining (U, L) women in

the market place the highest bid for (P, L) men (their surplus maximizing choice). So the

remaining matches for both men and women are of the form (P, L) - (U, L).20

The general result highlighted by this model is that we should observe positive

assortative matching in men’s and women’s characteristics that are complements (such as

wealth) and negative assortative matching in characteristics that are substitutes (such as

protection status). Although the transfer payments from wives to husbands are not

precisely pinned down in the general model (the participation and incentive compatibility

constraints only place upper and lower bounds on the feasible values of dfm), we can also

predict changes in dowries following embankment construction under specific market

structures, such as the case where women bid for men in a multi-unit English auction

20
These are results for a monogamous society with equal numbers of men and women of each type. Note
that H-type women can typically outbid L-type women, and if there are an excess of H-type women (over
H-type men) in the market, then we will observe some (U, H) women marrying (P, L) men (and (P, H)
marrying (U, L)), which will in turn force some L-type women to remain single.

17
setting. If there are multiple (U, H) women bidding for the same (P, H) man, the women

would compete away the entire surplus generated by this man, and dowry payments

would increase with protection status after embankment construction, since the man’s

contribution to the total marital surplus increases with his protection status.

4.3 Embankment Effects in a Simulated Gale and Shapley (1962) Matching Model

We now relax a number of the restrictive assumptions made in the model outlined

above and simulate the dynamics of matching in a more general model. Potential spouses

can offer compensating differentials along multiple dimensions in order to secure a

desirable match. For example, a family could make up any deficiency in its relative

wealth position by offering their candidate at the age most desirable by the opposite sex,

or accepting a candidate of a less desirable age. Thus, we now endow each candidate

with a continuous characteristic that is complementary to embankment protection (such

as the amount of land or wealth), another continuous characteristic relevant to spousal

choice which is neither a complement nor a substitute to protection (e.g. age at marriage),

a discrete protection status, and an idiosyncratic attractiveness parameter.

A male m’s payoff from marrying a female f is postulated to be:

[ ]
s mf = max(e f , em ) ⋅ Π ( w f , wm ) + α − β (a f − a *f ) 2 + ε mf (1)

e is embankment protection status, w is wealth, a is age, a *f (a constant) is the most

desired female age at marriage from a man’s perspective, ε mf is the idiosyncratic pair-

specific attractiveness parameter that measures male m’s preference for female f, and α

and β are constants. Greater wealth and protection status are considered attractive

characteristics, and wealth is complementary to protection (e.g. the embankment extends

the crop growing season). The insurance benefits of the embankment make the

18
husband’s and wife’s protection status substitutes. Candidates are penalized if their age

at marriage differs from some optimal age at marriage. Female f has an analogous

scoring function over each male m that she uses to evaluate which proposal to accept:

[ ]
s mf = max(e f , em ) ⋅ Π ( w f , wm ) + α − β (a m − a m* ) 2 + ε mf (2)

With a total of M men and F women on the market, we can use (1) and (2) to

define an M x F matrix of scores over all men and women. Since we cannot describe

analytical solutions to the matches that occur, we simulate the matches by endowing 2500

men and 2500 women with a distribution of wealth, age, protection and attractiveness

characteristics. We assume that initially each individual gets an independent draw on

wealth from a truncated normal distribution over positive support, a draw on age from a

uniform distribution (on support 16-22 for women with an optimal age at marriage, a *f of

19, and on support 21-27 for men, with am* =24), and a draw on preferences for each

individual of the opposite gender from a normal (0,1) distribution. Half of all men and all

women are randomly assigned to each bank of a river with an embankment on only one

side. We add search frictions to this model by assuming that individuals are more likely

to see (and propose to) others on the market who are physically closer to them. The Gale

and Shapley (1962) algorithm identifies the stable set of matches in this market.21

Results of the matching simulation show that for both protected men and women,

the wealth (land) distribution of spouses they match with shifts to the right following

embankment construction (see figure 3). Conversely, the wealth distribution of spouses

shifts to the left for men and women on the other bank of the river (who remain

21
In the Gale and Shapley (1962) algorithm, men propose their most preferred woman, and the woman
holds on to the most attractive man while rejecting the rest. The men then propose to their next best option,
and so on …, producing a stable set of matches where no two man and woman not paired to each other
through the algorithm would be better off by contracting that marriage.

19
unprotected) following embankment construction. Thus, the protected are able to secure

wealthier spouses at the expense of the unprotected. Individuals residing on the two sides

of the river are in direct competition in the marriage market, and this result comes about

because (a) the protected have an extra desirable characteristic to offer, so their offers are

more likely to be accepted and (b) due to complementarity in protection status and land,

they are more likely to extend offers of marriage to higher-wealth individuals.

For age at marriage, where no such complementarity exists, figure 4 shows that

there no clear trend to indicate that the protected are better able to secure partners at the

“optimal” age, or reduce spousal age gaps. Complementarity in inputs is therefore key to

understanding the potential effects of the embankment on the variety of possible marriage

outcomes. The model also exhibits negative assortative matching in the substitute

characteristic – protection status. Within-bank marriages are less likely to occur after

embankment construction, even with cross-bank search frictions (figure 5).

Although dowries are not well defined in this non-transferable utility model,

figure 6 plots the surplus accruing to matched men and women if the marital surplus

(over the payoff from remaining single) is divided between spouses according to Nash

bargaining. The distribution of surpluses shifts to the right for both protected men and

women. Since the dowry payment would be a positive function of the difference between

the man’s surplus and woman’s surplus, when protected men (women) marry unprotected

women (men), dowry payments increase (decrease). This result would also be predicted

by a model where the embankment is thought to shift spousal threat points (McElroy

1990). For protected-protected matches, the prediction on dowry payments is not clear.

20
 Thus, our simulated matching model generates predictions consistent with the

results of our analytical model:

1. If the embankment mitigates risk, then we should observe more cross-river

marriages after embankment construction. Search frictions across the river may

dampen this effect.

2. Positive wealth or health benefits conferred by the embankment that are

complementary across spouses should lead to positive assortative pull by

protection status. Characteristics independent of the embankment (such as age)

should remain unaffected.

3. In general, the level of dowry received by protected men should increase

following embankment construction.

5. Empirical Results

5.1 Basic Estimation Strategy

Our difference-in-difference set-up compares the marriage market outcomes for protected

households following embankment construction to their pre-embankment outcomes, after

differencing out the corresponding change in unprotected household outcomes.22 We

include household fixed effects where possible, which controls for household-specific

unobservable preferences such as heterogeneous attitudes toward risk. In this case, each

observation becomes a household experiencing at least one marriage before and at least

one marriage after embankment construction (e.g. for two different daughters).23 It is

22
We aggregate time of marriage into 2 periods, pre and post embankment construction, which avoids
inconsistent standard errors due to possible serial correlation in outcomes (Bertrand et al., 2004).
23
We check whether this is a selected sample in the sense that households with multiple marriages may
have higher fertility than other households. However, there are no substantial pre-embankment differences

21
important to note that the empirical results presented next do not constitute a direct test of

the models developed in the previous section, but jointly the theory and the empirical

results help us sensibly describe the changes in the marriage market following

embankment construction.

5.2 Household Responses to Changes in Risk Exposure

We first examine whether marriage outcomes respond to the embankment in ways

that are consistent with the construction reducing exposure to flood risk. For risk averse

individuals the embankment may lower their demand for mitigating risk through other

channels (e.g. marrying daughters into geographically distant households not subject to

the same weather patterns or planting different crops, a la Rosenzweig and Stark, 1989).

We would then expect to see changes in female migration patterns for marriage following

embankment construction. Difference-in-difference results from table 5 find no evidence

of such behavior, in that female marriage migration rates into households outside their

own villages or outside the Matlab area do not differentially change following

embankment construction for women from protected households.

Further, the theoretical model showed that if risk mitigation benefits of the

embankment were an important consideration, then we would expect to see negative

assortative matching in protection status. The last two columns of table 5 (and the last

column of table 10) show no evidence of assortative matching by embankment protection

in either direction.24 In general, households are much more likely to marry others who

in family size between the fixed effects sample and the full sample (see Appendix table A2). The average
age of males in the fixed effects sample is a few years younger than in the full sample, but we estimate our
fixed effects models both with and without controls for gender and birth order of the observation, plus the
gender and number of their siblings.
24
Given the occupational distribution in villages across Matlab district, households living in southern
villages would match with households likely engaged in a different occupation when marrying across the
river, but the same is not true in the more homogenous northern villages. The last two columns of table 5

22
are located closer to them (see the coefficients on “protected” in table 5), an indication of

search frictions in the marriage market, but this propensity to marry close does not

differentially change after embankment construction. This highlights the possibility that

empirical results on assortative matching based on cross-sectional data may be

uninformative, since it is difficult to separate out search frictions from true assortative

matching in cross-sectional data. Our panel data, which allows us to control for both

“protected” and “post*protected” (labeled “embankment”) helps resolve the issue.

In table 6, we do find that protected households become 4-5 percentage points less

likely to marry into different occupations following embankment construction, an

indication that some households use the marriage market to diversify income risk. This

result may merely indicate assortative matching in wealth rather than response to risk if

the correlation is driven by rich protected farmers marrying other rich protected farmers

following embankment construction. However, the last two columns indicate otherwise;

the correlation is entirely driven by households engaged in non-farming occupations.

In summary, the embankment does not differentially change households’

propensity to diversify and hedge against risk through marriage along most dimensions,

which is further evidence that changes in risk exposure is not the dominant effect of the

embankment. Non-farmers on the protected side (who did not experience the agricultural

wealth shock) are less likely to occupationally diversify through marriage. This could be

a response to lowered risk exposure, or because non-farmers find it more difficult to

marry into now-wealthier farming households around them.

show that households in the north and south do not behave any differently in terms of marrying across the
river following embankment construction, another indication of no differential change in risk mitigation
behavior.

23
5.3 Household Responses to the Wealth Shock

If protected households become wealthier following embankment construction

and can offer “protection” as a desirable marriage market characteristic, then the

matching model predicts that such households would seek out better spousal

characteristics that are complementary to their own in producing marital surplus (e.g.

socio-economic status), but not necessarily characteristics which are not complementary.

Since land is the primary asset for Matlab households, we measure a spouse’s socio-

economic status according to the amount of land owned by the head of the household in

1982.25 We also look for effects on changes on age at marriage and spousal age gaps.

Protected households become 3 percentage points more likely to marry into

households with above average amounts of land after embankment construction relative

to the unprotected (a 10 percent increase from their pre-embankment likelihood of 32

percent, see table 7). The direction of this change is robust to including household fixed

effects, although the effect becomes statistically weaker when the identification comes

from multiple marriages before and after embankment construction in the same

household.26 Furthermore, a triple difference in table 7 confirms that the effects are

entirely driven by protected farmers, who experienced the largest gain in assets between

1982 and 1996 and are the primary beneficiaries of embankment construction. Protected

farmers (who form 58 percent of our sample) exhibit a 5.3 percentage point increase in

25
Our data observes the amount of land owned by a household only if it lies within the surveillance area, so
this specification cannot include any household marrying outside Matlab. The marriage migration results
discussed earlier do not find evidence of differential post-embankment marriage migration rates across
protected and unprotected households (see Table 5), so a sample excluding these migrants should not yield
biased estimates of the embankment effect.
26
This is not altogether surprising, since the quality of the first child’s marriage has an important effect on
the likelihood of finding quality spouses for subsequent children’s marriage, which lends some inelasticity
to the quality of marriages across siblings within the same family (which is the type of variation the fixed
effects models use).

24
the propensity to marry into wealthy households (a 16 percent increase at the mean)

following embankment construction relative to unprotected farmers, whereas the effect

among non-farmers is essentially zero.

In order to establish that our results are not crucially dependent on data from the

years farthest after embankment construction, we include additional specifications that

eliminate the last three years of the sample (the post-embankment years then become

1989-1993). This limited sample does not alter the results (in fact, makes them stronger),

supporting embankment construction as the cause of a discrete one-time change in socio-

economic conditions for the protected.

Consistent with theory, we find no effect of the embankment on either age at

marriage or spousal age gaps, which we interpret as independent (i.e. not complementary

or substitute) characteristics in the marriage market. The coefficient signs are indicative

that both protected men and women are able to differentially delay marriage, but these

age effects are statistically indistinguishable from zero (see table 8).

Tobit models in table 9 regress each woman’s report of dowry payment from the

1996 MHSS data as a function of husband’s embankment protection status. Comparing

the “Protected Husband” and “Year*Protected” coefficients, we see that protected men

start receiving larger dowries than unprotected men in 1989 or 1990 (the beginning of the

post-embankment period) for all specifications.27 The “premium” that protected men

command in the years after embankment construction is quite larger - roughly 40% of the

average value of dowries in the entire sample.

27
For example, in the first specification in Table 9, the coefficient on “Year*Protected”, +128.6, multiplied
by 1989, equals +255785, which just exceeds the coefficient on “Protected Husband” (-255708)

25
 We check that our results are not generated by the endogenous sorting of

households around the embankment after construction. 5 percent of our sample migrates

to another area in Matlab during the post-embankment period for a reason other than

marriage, and re-estimating the models without these migrants does not qualitatively

change the results (see Appendix table A3).28 In addition, our results are robust to

exclusion of marriage observations that end in divorce and to the exclusion of non-

Muslims, who follow different marriage customs and face a narrower market.

6. Effects on Consanguinity

Although the biological and genetic risks for the offspring of the union of

biologically close relatives are well understood in the scientific community,

consanguineous marriages remain common practice in much of the developing world

(Grant and Bittles, 1997; New York Times, 2003). While the rates of consanguinity are

falling over time in both protected and unprotected households in Matlab, table 10 shows

the drop is much larger among protected households after the embankment is built. In the

difference-in-difference, protected households show a 2.5 percentage point greater

decrease in the likelihood of marrying a biological relative (i.e. second cousin or closer)

after the embankment over and above the change among the unprotected. In the

household fixed effects specification controlling for gender and birth order effects, we

find that the same family is 40% (about 3 percentage points) less likely to marry a

28
Strong and Minkin (1992) also analyze migration data before and after construction to conclude that
there exist no real changes in outmigration rates for either the protected or unprotected areas.

26
younger child to a biological relative after they are protected by embankment, than their

older child who married prior to the embankment construction.29

There are several possible conceptual links between the embankment and rates of

consanguinity. First, if consanguinity is a desirable marriage outcome based on cultural

or religious preferences, then protected households experiencing a positive wealth shock

from the embankment may become more able to attract (or pay for) such marriages. This

theory is rejected by our difference-in-difference results showing differential decreases in

consanguinity following embankment construction rather than increases. Other possible

motivations for consanguinity are consistent with these results. For instance, if

consanguinity is an inferior marriage outcome, protected households, with the additional

attractive characteristic they offer on the marriage market, may be more likely to avoid

this outcome. Consanguinity may also be a response to risk exposure if households

prefer to marry cousins in order to form robust inter-generational bonds with the

extended family.30

Finally, it has been postulated that households marry within the family in South

Asia in order to avoid large dowry payments at the time of marriage (Caldwell et al.,

1983; Bittles, 1994; Do, Iyer and Joshi 2009). This last link must be a little more

complicated, since one must explain why a rational male would forego larger dowry

payments in the outside market in order to marry his female cousin. Do, Iyer, and Joshi

(2009) develop a model predicting that dowry and consanguinity act as substitutes to

mitigate a marriage market failure— in a patrilocal marriage market, the bride’s family

has no incentive to continue investing in the couple once they are married and may

29
Fixed effect logit models produce similar results.
30
Do, Iyer, and Joshi (2009) present empirical evidence rejecting these alternative theories of
consanguinity (as a preferred cultural outcome, or as a risk mitigation measure).

27
instead choose to free-ride on investments by the groom’s family. The authors outline an

optimal tradeoff between dowries (pre-marital transfers) and consanguinity (close family

ties) that can help overcome this problem. One implication of their model is that in the

presence of tight credit markets, dowries will become more costly relative to

consanguinity. In this case, we should expect to see higher levels of consanguinity

among poorer families, since credit constrained households who cannot borrow to pay

dowries at the time of marriage may use consanguinity as a way to delay payments. The

promise to pay over a longer period is more credible when made within the family.

Consistent with this hypothesis, while the MHSS data shows that the dowry transfer at

the time of marriage is much smaller in consanguineous unions (see table 9),

conversations with Matlab residents during our fieldwork indicated the total amount of

effective dowry transfer over the course of the marriage may not be any different.

In order to identify these effects further, we look for additional ancillary evidence

in favor of the credit constraint story above as a motivation for consanguinity that

explains its link to the embankment. The embankment, by increasing wealth on the

protected side, relaxed the liquidity constraint (in the sense that these now wealthier

households had more dowry to offer at the time of marriage), taking away this important

motivation for marrying within the family. We first show that the dowry transfer at the

time of marriage is almost 50% lower at the mean in consanguineous unions (see table 9).

Second, we show that the drop in consanguinity following embankment construction is

much larger among protected females than among protected males (see table 11). Since

dowries are paid by the brides’ families (and not the grooms’), then we can take

advantage of a triple difference by gender, since we would expect consanguinity rates to

28
drop among protected females relatively more than males if the dowry payment and

credit constraint story is correct. The same protected family is up to 4.7 percentage

points less likely to marry their younger daughter to her cousin after embankment

construction relative to her older sibling, while for males, this drop is only about half as

large and not significantly different from zero in the fixed effects specifications.

7. Conclusion

Although the placement of the embankment may not be entirely random, it

provides plausibly exogenous variation to examine the determinants of conditions of

marriage through more rigorous empirical analysis than had been possible previously in

the large literatures on marriage in sociology and in economics. The first part of the

paper documents the following changes in marriage markets following wealth gains that

accrue to a subset of Matlab residents:

1. There is increasing segregation in the marriage market in terms of spousal wealth.

Members of farming households who benefit from the wealth shock are differentially

more likely to marry into wealthy households, and non-farmers living on the

protected side of the river find it increasingly difficult to marry into the now-

wealthier farming households.

2. Men from these protected (wealthier) households start commanding larger dowries.

3. Neither women nor men from protected households are able to delay marriage. Nor

are they able to marry younger spouses. Norms regarding age at marriage appear

much more inelastic compared to the quicker changes in spousal socio-economic

status that we document.

29
 The second part of the paper establishes that the practice of marrying biological

relatives appears to be closely linked to the institution of dowry. Rural Bangladeshi

households engage in consanguinity in response to the absence of credit for paying

dowries at the time of marriage. With the high prevalence of consanguinity in South

Asia, Middle East and North Africa it is important to understand the underlying socio-

economic drivers of this practice. The triple difference created by the embankment

construction (by gender, protected side of river, pre/post) allowed us to bring credible

empirical evidence to bear on this important question for the first time. The high child

morbidity and mortality effects of consanguinity reported in the literature imply that

liquidity constraints and lack of access to credit impose yet another costly burden on poor

households in developing countries through their marriage market choices.

Finally, our paper documents the general equilibrium changes associated with an

infrastructure project in disaster mitigation. Evaluation of such projects typically focus

on direct impacts on ecosystem equilibrium, agricultural practices and incomes, and

health (e.g. Haque and Zaman, 1993; Myaux et al., 1997; Paul, 1995; Thompson and

Sultana, 1996), and we show that indirect general equilibrium changes can be quite

substantial, and need to be taken into account in program evaluation.

30
Figure 1: Matlab Surveillance Area, Figure 2a: The Embankment: Not Very High Figure 2b: Protected Bank from the Top of the
the River and the Embankment and Reinforced with Sandbags Embankment: Agricultural Fields Very Close to
the Embankment

The light colored (green) polygons are

villages, the blue (double) line is the river,
and the red line is the embankment

31
Figure 3: Land Distribution of Spouses of Protected and Unprotected Men in the
Simulated Gale-Shapley Marriage Market

32
Figure 4: Age Distribution of Spouses for Protected Men and Women in the
Simulated Gale-Shapley Marriage Market
Panel A: Spousal Age Gap

Panel B: Age at Marriage

33
 Figure 5: Assortative Matching in Protection Status in the Simulated Model

Low Search Friction in Marriage Market:

qb = 0.90 (probability of seeing a partner on the same bank)
qs = 0.50 (probability of seeing a partner on the opposite bank)

Before Embankment After Embankment

Same Side of River 64% 30%
Matches
Across the River Matches 36% 70%

With Greater Search Friction in Marriages Across the River: (qb = 0.95, qs = 0.30)

Before Embankment After Embankment

Same Side of River 78% 42%
Matches
Across the River Matches 22% 56%

Figure 6: Nash Bargaining Surplus for Men and Women Pre and Post Embankment

34
 Table 1: Descriptive Statistics
Full Sample (ICDDR Matlab Demographic Surveillance Area)

Before Embankment After Embankment

Mean: protected unprotected protected unprotected
Percentage of Consanguineous Marriages .081** .067** .044*** .059***
(.01) (.00) (.00) (.00)
Female Age at Marriage 19.02 19.04 19.90 19.84
(.06) (.05) (.06) (.04)
Male Age at Marriage 26.93 25.71 27.33*** 25.96***
(.15) (.10) (.12) (.08)
Household Land Owned (1982) 11.12*** 10.28*** 11.44*** 10.64***
(.23) (.15) (.19) (.13)
Land Owned by Spouse's Household (1982) 10.82** 9.89** 11.13*** 10.22***
(.31) (.31) (.27) (.20)
Percentage of Marriages to Spouse from outside Matlab .478*** .576*** .512*** .597***
(.01) (.00) (.01) (.00)
Percentage of Marriages to Spouse from outside Village .864*** .880*** .882*** .895***
(.00) (.00) (.00) (.00)
Total Marriage Observations 5225 11050 8438 17896
Standard errors in parentheses. *** indicates t-test significant at 1%, **indicates significant at 5% level. Pre indicates 1985-1986
for consanguinity and 1982-1986 for all other variables. Post indicates 1989-1996.

MHSS 1996 Sample

Pre Post
Mean: protected unprotected protected unprotected
Percentage of Men Receiving Dowry .11 .12 .42 .43
(.01) (.01) (.04) (.02)
Value of Dowry Received by Men 576.08 575.27 4077.90 4193.57
(83.88) (46.82) (487.01) (378.62)
Value of Dowry Received by Men in 389.47 524.73 5708.33* 3658.67*
Consanguineous Marriages (113.34) (109.60) (1310.88) (561.12)
Total Marriage Observations 1119 2543 182 422
Standard errors in parentheses. * indicates t-test significant at 10% level. Pre indicates 1982-1986; Post indicates
1989-1996. Data taken from 1996 MHSS.

35
 Table 2: No Pre-existing Differences in Trends in Marriage Outcomes by Protection Status

Spouse Owns Above

Consanguineous Spouse from Different Spouse from Outside
Average Amount of
Marriage? Village? Matlab?
Land?
Coeff. ME Coeff. ME Coeff. ME Coeff. ME
Protected -2.553 -.254 -2.579 -2.893 -1.525
-.656 -.772 -.549
(8.164) (2.223) (1.784) (1.414)
Year -.280*** -.037 .001 .012 .002
.000 .003 .001
(.056) (.016) (.012) (.010)
Protected * Year .031 .031 .034 .015
.004 .011 .001 .006
(.096) (.026) (.021) (.017)
Land Owned -.002 .003*** .002***
-.000 .001 .001
(.002) (.001) (.001)
Above Avg. Amount of Land .144*** .052
(.035)
Age at Marriage

R-squared .011 .003 .002 .006

Total obs. 7130 6199 14938 14946
Sample Pre-Embankment Years (1982-1986) Only

Standard errors in parentheses. *** indicates t-test significant at 1% level. All models estimated using probit. All specifications include pre-
embankment years 1982-1986 only. Marginal effects calculated at means of explanatory variables.

36
 Table 3: Effect of the Embankment on Wealth

Asset Indices

Pre-embankment Post-embankment Difference

Protected Unprotected Protected Unprotected Protected Unprotected
-.05*** .05*** .38 .38 .43*** .34***
Farmers
(Landowners) (.02) (.01) (.02) (.02) (.02) (.02)
Obs. 3377 7270 3377 7270 3377 7270
-.48 -.44 .03*** -.21*** .50*** .23***
Farmers
(Tenant) (.03) (.02) (.04) (.03) (.05) (.03)
Obs. 552 1179 552 1179 552 1179
.07*** .16*** .33*** .42*** .26 .26
Non-Farm
Occupations (.03) (.02) (.02) (.02) (.03) (.02)
Obs. 1530 3207 1530 3207 1530 3207
-.06*** .03*** .33 .33 .39*** .30***
All
Occupations (.01) (.01) (.01) (.01) (.02) (.01)
Obs. 5459 11656 5459 11656 5459 11656
Standard errors in parentheses. *** indicates t-test significant at 1% level; ** indicates significance at 5% level. Asset index constructed using principal components factor analysis, and measures
household ownership of any combination of the following assets: radio, watch or clock, bicycle, cows, and hurricane lamp. Data taken from Matlab DSS 1982 and 1996 household censuses.

37
 Table 4: Changes in the Variance in Assets across Households within Village
Following Embankment Construction
Within-Village Variance in Assets
Pre-embankment Post-embankment
Protected .87 .91
(.04) (.04)
Obs. 32

Unprotected 1.02 .96

(.04) (.02)
Obs. 93
Each observation is a village in Matlab, and the table reports the variance in assets across households
within a village. Only villages with >80% of land on one side of the embankment or the other are included.

Table 5: Location of Spouse does not Change Differentially after Embankment

Spouse from Marrying Across the River
Spouse from Different
Outside Matlab Village Northern Villages Southern Villages
Coeff. ME Coeff. ME Coeff. ME Coeff. ME
Protected -.247*** -.075*** -.015 1.06*** 1.28***
-.098 .381 .470
(.021) (.027) (.051) (.046)
Post .053*** .082*** .009 -.040
.021 .016 .003 -.015
(.015) (.020) (.041) (.040)
Embankment .031 .002 .012 .028
.012 .000 .004 .011
(.027) (.034) (.066) (.060)
R-squared .006 .002 .116 .175
No. of obs. 42609 42596 7662 8596

Table 6: …but Unprotected Households More Likely to Marry into Different Occupations
Marrying into Different Occupation Marrying into Different Occupation
- Full Sample - Farmers/Non-Farmers
Household
Diff-in-Diff FE Farmers Non-Farmers
Coeff. ME Coeff. Coeff. ME Coeff. ME
.150*** .131 .187***
Protected .051 .006 .074
(.041) (.080) (.052)
-.019 .014 .067 .078
Post -.001 .003 .031
(.031) (.026) (.064) (.040)
-.124** -.056 .06 -.128*
Embankment -.044 .001 -.051
(.053) (.037) (.100) (.067)
R-squared .080 .003 .016 .001
Total obs. 12338 3112 5983 6355
Standard errors in parentheses. *** indicates significance at 1% level; ** indicates significance at 5% level; *
indicates significance at 10% level. All models estimated using probit. Marginal effects calculated at means of
explanatory variables.

38
 Table 7: Protected Households Differentially More Likely to Marry Into Wealthier Households after Embankment
Construction

Spouse Land
Dependent Variable: Ownership Spouse Owns Above Avg. Land
Ex. Top 5% Full Ex. Last 3 years
Sample: Full Landowners (1982 - 1996) (1982 - 1993) Farmers Non-Farmers
Coeff. ME Coeff. ME Coeff. ME Coeff. ME
.871** .851** .062* .064* 0.001 .139**
Protected 0.022 0.023 0 0.048
(0.41) (0.41) (0.035) (0.035) (0.044) (0.055)
0.273 0.208 -0.025 -0.023 -0.034 -0.022
Post -0.009 -0.008 -0.012 -0.008
(0.317) (0.317) (0.027) (0.029) (0.035) (0.042)
Embankment 0.082 0.035 .080* .108** .143** 0.007
0.029 0.039 0.053 0.003
(Protected*Post) (0.528) (0.528) (0.044) (0.048) (0.058) (0.07)
.073*** .103***
Land Owned
(0.008) (0.013)
Above Avg. Amount .202*** .115*** .159*** .237***
0.073 0.041 0.058 0.084
of Land (0.022) (0.025) (0.028) (0.041)
R-squared 0.006 0.005 0.006 0.003 0.004 0.007
Total obs. 15649 15047 15649 12659 9075 6425

Standard errors in parentheses. *** indicates significance at 1% level; ** indicates significance at 5% level; * indicates significance at 10% level. DID indicates difference-in-difference; ME indicates
marginal effects (calculated at means of explanatory variables); FE indicates fixed effects. DID Above Avg. Land estimated using probit; all other models estimated using OLS.

39
 Table 8: The Embankment has Weaker Effects on Age at Marriage

Dependent Variable: Age at Marriage Spouse Age at Marriage Age Gap

Females Males Female Spouses Male Spouses Male Spouses
DID FE DID FE DID FE DID FE DID ME FE
Protected -.028 .215 -.067 -.051 -.056*
-.019
(.091) (.182) (.095) (.127) (.033)
Post .801*** 1.423*** 1.248*** 1.267*** .675*** .846*** .159* .632*** .009 .024*
.003
(.066) (.096) (.131) (.201) (.068) (.120) (.089) (.159) (.023) (.013)
Embankment .086 .116 .158 .391 .012 -.095 -.013 -.257 .046 .003
.015
(.116) (.169) (.231) (.363) (.119) (.212) (.156) (.274) (.040) (.022)
Age at Marriage .321*** .343*** .676*** .659*** -.051*** -.008***
-.017
(.004) (.009) (.008) (.021) (.002) (.002)
R-squared .010 .023 .008 .010 .310 .316 .224 .204 .018 .012
Total obs. 24349 9914 18255 6781 17362 6145 22479 8680 22749 8680
Number of groups 4037 2804 2575 3602 3602
Standard errors in parentheses. *** indicates significance at 1% level; * indicates significance at 10% level. DID Age Gap estimated
using probit; all other models estimated using OLS. Fixed Effects models include household fixed effects.

Table 9: Husbands from Families Protected by the Embankment Command Larger

Dowries after 1989:
Dependent Variable: Dowry paid by Women
Protected Husband -255708.2***-244390.2***-242281.9***
(83419.0) (83769.6) (83753.4)
Year 733.9*** 738.9*** 740.6***
(25.8) (26.0) (26.0)
Year * Protected Husband 128.6*** 122.9*** 121.8***
(42.2) (42.3) (42.3)
Age at Marriage -544.9*** -547.6*** -560.1***
(55.9) (56.2) (56.3)
Consanguineous Marriage -2346.8***
(596.1)
Consan. (Maternal Relative) -2521.7**
(1128.2)
Consan. (Paternal Relative) -1698.9*
(949.9)
R-squared .045 .046 .046
Total obs. 5699 5674 5674
Standard errors in parentheses. *** indicates significance at 1% level; ** indicates significance
at 5% level; * indicates significance at 10% level. All models are tobit and use 1996 MHSS data.

40
 Table 10: Rates of Consanguinity Drop in Protected Households
Consanguinity Protected Spouse
DID DID FE FE DID
Coeff. ME Coeff. ME Coeff. ME
Protected .089* .090* 1.194***
.011 .011 .437
(.047) (.047) (.034)
Post -.070** -.069** -.010* -.014** -.015** -.016
-.008 -.008 -.006
(.032) (.034) (.006) (.006) (.007) (.029)
Embankment -.236*** -.228*** -.028***-.033*** -.030** .011
-.025 -.024 .004
(.056) (.059) (.010) (.011) (.012) (.044)
Land Owned -.001 -.001 .001**
-.000 -.000 .000
(.001) (.001) (.001)
Male .039 .038
(.031) (.034)
Oldest Child -.005 -.015
(.009) (.012)
Youngest Child .007 -.002
(.009) (.011)
Male * Oldest Child -.007 .010
(.016) (.021)
Male * Youngest Child -.033*** -.020
(.012) (.015)
Further Controls:
Number of Brothers No Yes Yes
Number of Sisters No Yes Yes
Sample Years 82-96 82-93 82-96 82-96 82-93 82-96
R-squared .004 .004 .001 .001 .001 .150
Total obs. 31189 23463 14707 12230 9258 16302
Number of groups 5436 4675 3721
Standard errors in parentheses. *** indicates significance at 1% level; ** indicates significance at 5% level; * indicates significance
at 10% level. DID indicates difference-in-difference; ME indicates marginal effects (calculated at means of explanatory variables);
FE indicates fixed effects. DID estimated using probit; FE estimated using OLS. Fixed Effects models include household fixed
effects.

41
 Table 11: The Change in Consanguinity Rates by Gender

Males
Consanguineous Marriage
DID DID FE FE FE
Coeff. ME Coeff. ME
Protected .067 .067
.008 .008
(.072) (.072)
Post -.079* -.085* -.005 -.028* -.027
-.009 -.010
(.048) (.051) (.011) (.015) (.017)
Embankment -.204** -.173* -.032 -.021 -.025
-.022 -.019
(.085) (.090) (.020) (.024) (.026)
Land Owned -.002 -.002
-.000 -.000
(.001) (.001)
Oldest Child -.022 -.025
(.019) (.025)
Youngest Child -.043*** -.037**
(.014) (.016)
Sample Years 82-96 82-93 82-96 82-96 82-93
R-squared .004 .004 .001 .005 .006
Total obs. 13446 10071 4259 3161 2321
Number of groups 1837 1383 1037
Females
Consanguineous Marriage
DID DID FE FE FE
Coeff. ME Coeff. ME
Protected .109* .109*
.013 .013
(.063) (.063)
Post -.062 -.056 -.005 -.006 -.006
-.007 -.007
(.043) (.045) (.009) (.010) (.011)
Embankment -.262*** -.272*** -.043*** -.047*** -.042**
-.027 -.028
(.074) (.079) (.015) (.016) (.018)
Land Owned -.000 -.000
-.000 -.000
(.001) (.001)
Oldest Child -.004 -.012
(.012) (.016)
Youngest Child .003 .003
(.011) (.013)
Sample Years 82-96 82-93 82-96 82-96 82-93
R-squared .005 .005 .001 .001 .001
Total obs. 17743 13392 6186 5403 4034
Number of groups 2685 2367 1817

42
References
Aghion, P., Robin Burgess, Stephen Redding, and F. Zilibotti. 2008. “The Unequal
Effects of Liberalization: Evidence from Dismantling the License Raj in India,”
American Economic Review 98(4): 1397-1412.
Anderson, Siwan. 2000. “The Economics of Dowry Payments in Pakistan,” mimeo,
Tilburg University
Anderson, Siwan. 2007. “Why the Marriage Squeeze Cannot Cause Dowry Inflation,”
Journal of Economic Theory 137(1): 140-152.
Becker G. 1991. Treatise on the Family, Harvard University Press, Cambridge, MA
Bergstrom, Theodore. 1997. “A Survey of Theories of the Family,” 21-74 in M.
Rosenzweig and O. Stark eds, Handbook of Population and Family Economics:
Amsterdam, Elsevier Science.
Bertrand, M., Duflo, E. and S. Mullainathan. 2004. “How Much Should We Trust
Differences-In-Differences Estimates?” Quarterly Journal of Economics 119 (1):
249-275.
BBC 2005. “The Risks of Cousin Marriage,” BBC Newsnight Published 2005/11/16
Bittles, A.H. 2001. “Consanguinity and its Relevance to Clinical Genetics,” Clinical
Genetics 60: 89-98.
Bittles, A.H. 1994. “The Role and Significance of Consanguinity as a Demographic
Variable,” Population and Development Review 20(3): 561-584.
Bittles, A.H. and J.V. Neel. 1994. “The Costs of Human Inbreeding and their
Implications for Variations at the DNA Level,” Nature Genetics 8: 117-121.
Bittles, A.H. Mason, W.M., Greene, J., and N. Appaji Rao. 1991. “Reproductive
Behavior and Health in Consanguineous Marriages,” Science 252: 789-794.
Bittles, A.H. and U. Makov. 1988. “Inbreeding in Human Populations: Assessment of the
Costs,” in Mating Patterns, C. Mascie-Taylor and A. Boyce (ed.), Cambridge:
Cambridge University Press.
Bloch, F., Rao, V. and S. Desai. 2004. “Wedding Celebrations as Conspicuous
Consumption: Signaling Social Status in Rural India,” Journal of Human
Resources 39(3): 675-695.
Bloch, F. and V. Rao. 2002. “Terror as a Bargaining Instrument: A Case-Study of Dowry
Violence in Rural India,” American Economic Review 92(4): 1029-1043.
Briscoe, J. 1998. “Against the Flow: Reflections on Two Decades of Development in a
Bangladeshi Village,” unpublished paper
Browning, M., Chiappori, P.A., and Weiss, Y. 2003. “A Simple Matching Model of the
Marriage Market,” mimeo, University of Chicago.
Browning, M., Chiappori, P. and Y.Weiss (2005). Family Economics, Cambridge
University Press, Cambridge, UK, forthcoming.
Caldwell, J., Reddy, P., and P. Caldwell. 1983. “Causes of Marriage Change in South
India,” Population Studies 37(3): 343-361.
Dalmia, S. 2004. “A Hedonic Analysis of Marriage Transactions in India: Estimating
Determinants of Dowries and Demand for Groom Characteristics in Marriage,”
Research in Economics 58: 235-255.
Dalmia, S. and P. Lawrence. 2001. “An Empirical Analysis of Assortative Mating in
India and the U.S.,” International Advances in Economic Research 7(4): 443-458.

43
Deolalikar, A. and V. Rao. 1998. “The Demand for Dowries and Bride Characteristics in
Marriage: Empirical Estimates for Rural South-Central India,” in Gender,
Population, and Development, M. Krishnaraj, R. Sudarshan, A. Sharif (ed.),
Oxford University Press.
Do, Quy-Toan., S. Iyer, and Shareen Joshi. 2009. “The Economics of Consanguineous
Marriages,” Working Paper.
Dyson, T. and M. Moore. 1983. “On Kinship Structure, Female Autonomy, and
Demographic Behavior in India,” Population and Development Review 9(1): 35-
60.
Field, E. and Ambrus, A. 2008. “Early Marriage, Age of Menarch, and Female Schooling
Attainment in Bangladesh,” Journal of Political Economy 116(5): 881-930.
Foster, Andrew 1998. “Marriage Market Selection and Human Capital Allocations in
Rural Bangladesh,” mimeo, University of Pennsylvania.
Fruzzetti, L. 1982. The Gift of a Virgin: Women, Marriage and Ritual in a Bengali
Society, Rutgers University Press.
Gale, D. and L.S. Shapley (1962). “College Admissions and the Stability of Marriage,”
The American Mathematical Monthly, 69 (1): 9-15, January.
J. C. Grant and A H. Bittles (1997). “The Comparative Role Of Consanguinity In Infant
And Childhood Mortality In Pakistan,” Annals of Human Genetics 61: 143-149.
Grossbard-Schechtman, A. 1984. “A Theory of Allocation of Time in Markets for Labor
and Marriage,” Economic Journal 94: 863-882.
Haque, C. and M. Zaman. 1993. “Human Responses to Riverine Hazards in Bangladesh:
A Proposal for Sustainable Floodplain Development,” World Development 21(1):
93-107.
Hashmi, M. 1997. “Frequency of Consanguinity and its Effect on Congenital
Malformation—A Hospital Based Study,” Journal of the Pakistan Medical
Association 47: 75-78.
Huq, L. and S. Amin. 2001. “Dowry Negotiations and the Process of Union Formation in
Bangladesh: The Implications of Rising Education.” Paper presented at the
Annual Meetings of the Population Association of America.
Hurt, L., Ronsmans, C., Campbell, O., Saha, S., Kenward, M. and M. Quigley. 2004.
“Long-Term Effects of Reproductive History on All-Cause Mortality Among
Adults in Rural Bangladesh,” Studies in Family Planning 35(3): 189-196.
Hussain, R., and Bittles, A.H. 1999. “Differentials in Age at Marriage, Contraceptive Use
and Fertility in Consanguineous Marriages in Pakistan,” Journal of Biosocial
Science 31: 121-138.
Jahan, R. 1991. “Hidden Wounds, Visible Scars: Violence Against Women in
Bangladesh,” in Structures of Patriarchy: State, Community and Household in
Modernising Asia, B. Agarwal (ed.), Kali for Women, New Delhi.
Jensen, R., and R. Thornton. 2003. “Early Female Marriage in the Developing World,”
Gender and Development 11(2): 9-19.
Kabir, M. 2004. “Development or a Disaster Forgotten? The Case of Meghna-
Dhonogoda Irrigation and Flood Control Project.”
Khan, Nizam (2001). “Effects of Parental Consanguinity on Fetal Loss and Infant
Mortality: Evidence from Bangladesh,” Working Paper, University of Colorado at
Boulder.

44
Manser, M. and M. Brown (1980). “Marriage and Household Decision-Making: A
Bargaining Analysis,” International Economic Review 21 (1): 31-44, February.
McElroy, Marjorie (1990). “The Empirical Content of Nash-Bargained Household
Behavior,” The Journal of Human Resources 25 (4): 559-583, Autumn.
McElroy, Marjorie and M. J. Horney (1981). “Nash-Bargained Household Decisions:
Toward a Generalization of the Theory of Demand,” International Economic
Review 22(2): 333-349, June.
Myaux, J.A., Ali, M., Chakraborty, J., and A. de Francisco. 1997. “Flood Control
Embankments Contribute to the Improvement of the Health Status of Children in
Rural Bangladesh,” Bulletin of the World Health Organization 75(6): 533-539.
New York Times (2003). “The Struggle for Iraq: Traditions; Iraqi Family Ties
Complicate American Efforts for Change,” September 28, Sunday, by John
Tierney.
Paul, B. 1995. “Farmers’ Responses to the Flood Action Plan (FAP) of Bangladesh: An
Empirical Study,” World Development 23(2): 299-309.
Rasul, Imran. 2008. “Household Bargaining over Fertility: Theory and Evidence from
Malaysia,” Journal of Development Economics 86: 215-241.
Rao, V., 1993. “The Rising Price of Husbands: A Hedonic Analysis of Dowry Increases
in Rural India,” Journal of Political Economy 101: 666-677.
Rosenzweig, M. and O. Stark. 1989. “Consumption-Smoothing, Migration, and
Marriage: Evidence from Rural India,” Journal of Political Economy 97: 905-
926.
Shah, G., Toney, M., and B. Pitcher. 1998. “Consanguinity and Child Mortality: The
Risk Faced by Families,” Population Research and Policy Review 17: 275-283.
Siow, Aloysius, 1998. “Differential Fecundity, Markets and Gender Roles,” Journal of
Political Economy 106 (2): 334-354.
Strong, M. and S. Minkin. 1992. “The Demographic, Health, and Nutritional Impacts of
the Meghna-Dhonagoda Embankment,” Bangladesh Flood Action Plan FAP 16
Environmental Study, Special Studies Program. ICDDR,B and ISPAN.
Thompson, P.M. and P. Sultana. 1996. “Distributional and Social Impacts of Flood
Control in Bangladesh,” The Geographical Journal 162(1): 1-13.
Tiemoko, R. 2001. “The Gender Age Gap: Marriage and Rights in the Cote d’Ivoire,”
Development 44(2): 104-106.
Weiss, Yoram (1997). “The Formation and Dissolution of Families: Why Marry? Who
Marries Whom? And What Happens Upon Divorce,” 81-124 in M. Rosenzweig
and O. Stark eds, Handbook of Population and Family Economics: Amsterdam,
Elsevier Science.
Wickrama, K. and F. Lorenz. 2002. “Women’s Status, Fertility Decline, and Women’s
Health in Developing Countries: Direct and Indirect Influences of Social Status
on Health,” Rural Sociology 67(2): 255-277.

45
 Appendix

Table A1: List of Variables

Marriage Outcome Definition
Value of Dowry Total estimated cash value of dowry paid to husband in Taka
Consanguinity Individual married 1st, 2nd, or other cousin
Spouse Land Owned Land owned by head of spouse’s household in 1982
(measured in decimals)
Spouse Above Average Land Spouse’s household owns more than 10.347 decimals of land
(avg. land owned by households of spouses chosen within the
sample period)
Age at Marriage Age of individual at time of marriage
Spouse Age at Marriage Age of spouse at time of marriage
Spouse Age Gap Indicator for whether or not male spouse is more than 10 years
older than the female marriage observation
Protected Spouse Group status of spouse (time-invariant indicator for whether or
not spouse’s household is protected by embankment)

Explanatory Variables Definition

Protected Group status of individual
Post Indicator equal to 1 if marriage year between 1989-1996 and 0 if
marriage year between 1982-1987
Embankment Embankment effect (interaction of Protected and Post)
Land Owned Land owned by head of individual’s household in 1982
Above Average Land Owned Household owns more than 10.744 decimals of land (avg. land
owned by households of individuals getting married within the
sample period)
Farmer Indicator equal to 1 if household head is a farmer and 0
otherwise

46
 Table A2: Descriptive Statistics for Fixed Effects Sample

Pre-embankment Post-embankment
Mean: Protected Unprotected Protected Unprotected
Percentage of Consanguineous Marriages .09*** .06*** .05 .05
(.01) (.00) (.00) (.00)
Female Age at Marriage 18.86 18.97 19.98 19.92
(.11) (.10) (.10) (.07)
Male Age at Marriage 24.21 24.20 25.76 25.64
(.17) (.13) (.13) (.09)
Household Land Owned (1982) 11.03 10.25 11.13** 10.35**
(.43) (.29) (.33) (.20)
Land Owned by Spouse's Household (1982) 10.69 10.14 11.24* 10.28*
(.66) (.47) (.48) (.32)
Percentage of Marriages to Spouse from outside Matlab .47*** .59*** .51*** .59***
(.01) (.01) (.01) (.01)
Percentage of Marriages to Spouse from outside Village .87** .89** .89 .90
(.01) (.01) (.01) (.00)
Total Marriage Observations 1233 2699 2626 5672
Standard errors in parentheses. *** indicates t-test significant at 1%, **indicates significant at 5% level, * indicates significant at
10% level. Pre indicates 1985-1986; Post indicates 1989-1996. Sample taken from fixed effects estimation for consanguinity.

Family Size in 1982

Pre-embankment Post-embankment
Mean: Protected Unprotected Protected Unprotected
DID Sample 6.69 6.61 6.85 6.82
(.07) (.04) (.03) (.02)
Obs. 2288 4842 7741 16318
Fixed Effects Sample 6.76 6.71 6.90 6.90
(.09) (.06) (.06) (.04)
Obs. 1152 2498 2458 5353
Standard errors in parentheses. Pre indicates 1985-1986; Post indicates 1989-1996. Samples
taken from difference-in-difference and fixed effects estimations for consanguinity.

47
 Table A3: Robustness to Excluding Migrants, Divorces and Hindus
Without Migrants Without Marriages Ending in Divorce Muslims Only
Dependent Variable: Consanguinity Above Avg. Land Consanguinity Above Avg. Land Consanguinity Above Avg. Land
DID FE DID FE DID FE DID FE DID FE DID FE
Coeff. ME Coeff. ME Coeff. ME Coeff. ME Coeff. ME Coeff. ME
.090* .062* .103** .027 .062 .063*
Protected .011 .022 .012 .010 .008 .023
(.047) (.035) (.049) (.039) (.049) (.036)
-
-.077**
- -.013** -.027
- -.001
-.066**
- .015** -.020
- .026
-.069**
- -.017** -.016
- -.003
Post
(.032)
.009 (.007) (.027)
.009 (.023) (.033)
.008 (.007) (.029)
.007 (.026) (.033)
.009 (.007) (.029)
.006 (.026)
- - - - -
.234***
- .036*** .082*
.007
.237***
- .029** .114**
-.012
.235***
- -.030** .070
.020
Embankment .029 .041 .026
(.056)
.024 (.011) (.045) (.033) (.058)
.024 (.011) (.048) (.037) (.057)
.026 (.012) (.047) (.036)
-.001 - -.001 - -.001* -
Land Owned
(.001) .000 (.001) .000 (.001) .000
Above Avg. .199*** .195*** .176***
Amount of Land
.072 .070 .064
(.023) (.024) (.023)
.040 -.104 .036 -.110 .040 -.113
Male
(.031) (.092) (.033) (.121) (.032) (.094)
-.004 .023 -.002 .009 -.006 .017
Oldest Child
(.009) (.036) (.010) (.038) (.011) (.038)
.008 -.014 .005 .012 .013 -.033
Youngest Child
(.009) (.030) (.009) (.033) (.010) (.033)
-
Male * Oldest -.090 -.101
-.008 .106* -.014 -.004
Child
(.017) (.058) (.017) (.062) (.019) (.063)
- - -
Male * Youngest -.033 -.048 -.010
.034*** .026** .042***
Child
(.013) (.041) (.013) (.044) (.014) (.045)
Further Controls:
Number of
Brothers Yes Yes Yes Yes Yes Yes
Number of Sisters Yes Yes Yes Yes Yes Yes
R-squared .005 .001 .006 .000 .004 .000 .006 .001 .005 .001 .005 .001
Total obs. 29909 11821 15013 4272 29561 11268 13810 3614 27461 10149 13753 3763
Number of groups 4481 1805 4427 1586 3863 1597

Standard errors in parentheses. *** indicates significance at 1% level; ** indicates significance at 5% level; * indicates significance at 10% level. DID indicates difference-in-difference;
ME indicates marginal effects (calculated at means of explanatory variables); FE indicates fixed effects. DID Consanguinity and Above Avg. Land estimated using probit; all other models
estimated using OLS. Fixed Effects models include household fixed effects. Migrants are defined as any individual migrating within or outside Matlab for a reason other than marriage during
the post-embankment period (1989-1996). "Divorce" is defined as any marriage ending in divorce by the end of 1996. We don't have information for 1992 divorces, so those are not accounted for in the robustness check.

48