WPS8217
Policy Research Working Paper 8217
The Socioeconomics of Fish Consumption
and Child Health in Bangladesh
Susmita Dasgupta
Mohammed Golam Mustafa
Tapas Paul
David Wheeler
Development Research Group
Environment and Energy Team
&
Environment and Natural Resources Global Practice Group
October 2017
Policy Research Working Paper 8217
Abstract
Child malnutrition in Bangladesh exceeds WHO’s thresh- share of fish. This suggests that increased income and edu-
old for public health emergencies. Using more than 36,000 cation impart a “status bias” toward eggs and meat, even
records from several waves of the Bangladesh Demographic though they are more expensive and less beneficial than
and Health Survey, the research focuses on the socioeco- fish for child health. In addition, mothers’ individual pref-
nomic determinants of household consumption of all erences for different animal-source foods, and the seasonal
animal-source foods; the socioeconomic determinants of availability of fish during the pre- and post-partum peri-
fish consumption, given its importance in the Bangladeshi ods have large effects on child mortality and significant
diet; and the impact of observed consumption patterns on effects on resistance to several common childhood illnesses.
mortality and resistance to infectious diseases for children These findings highlight the importance of programs to
in their first years of life. Better maternal education and increase supply of fish, maternal nutrition education and
family economic status significantly increase the level of ani- more public health programs to promote fish consumption.
mal-source food intake, but they decrease the consumption
This paper is a product of the Environment and Energy Team, Development Research Group and the Environment and
Natural Resources Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its
research and make a contribution to development policy discussions around the world. Policy Research Working Papers
are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at sdasgupta@worldbank.org.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team
The Socioeconomics of Fish Consumption and Child Health in Bangladesh
Susmita Dasgupta*
Lead Environmental Economist, Development Research Group, World Bank, United States
Mohammed Golam Mustafa
Scientist, World Fish, South Asia
Tapas Paul
Lead Environmental Specialist, World Bank, India
David Wheeler
Senior Fellow, World Resources Institute
This research was conducted under the South Asia Water Initiative - Sundarbans Landscape.
* Authors’ names in alphabetical order.
Corresponding Author: Susmita Dasgupta. Email: sdasgupta@worldbank.org;
Tele phone: 1-202-473-2679; Fax: 1-202-522-2714
We are thankful to Dr. Shakuntala Thilsted and Dr. Randall Brummett for their review of the paper.
Keywords: animal-source foods; fish; maternal nutrition; child mortality; child health; Bangladesh
JEL Codes: I120; I140; Q220
1. Introduction
Malnutrition is a primary source of health damage for children in poor countries, accounting for 45%
of under-five mortality and 50% of years lived with disability for children aged four and under (Golden et
al. 2013). Many children and adults in Bangladesh suffer from malnutrition, which is also responsible for
estimated economic losses of around $US one billion per year (Howlader et al. 2012). Bangladeshi fish
consumption is critical in this context, because fish are the largest source of high-quality protein (Roos et
al. 2007, Kent 1987), as well as micronutrients that are difficult to obtain from plant-based foods in
sufficient amounts (Murphy and Allen 2003, Michaelsen et al. 2009, Sandström et al. 1989).
Micronutrient intakes are enhanced by whole consumption of small indigenous fish, which is common for
poor households (Bogard et al. 2015b). This is particularly true for young children who eat small meals
(Bogard et al. 2015a). Protein from fish and other animals also increases absorption of nutrients from
dietary plant consumption (Neumann et al. 2001).
Extensive research has documented the health benefits of fish-intensive diets. Rimm and
Mozaffarian (2006) find a 36% reduction in mortality risk from heart disease; Zhao et al. (2015) show that
individuals who consume at least 60 grams of fish per day have 12% lower mortality than their
counterparts. In a global study, Lim et al. (2012) find that 1.4 million deaths in 2010 can be attributed to
deficiency in omega-3 fatty acids from fish. From the same database, Ezzati and Riboli (2013) calculate
that diets low in fish and seafood account for about 1% of total disability-adjusted life years (DALYs).
The stakes for children are particularly high, since micronutrient deficiencies can lead to increased
mortality risk, damaged cognitive development and reduced immunity to disease (Golden et al. 2013).
Fish consumption plays multiple roles in health improvement, including higher concentrations of
bioavailable minerals and vitamins; essential fatty acids and animal protein (Bogard et al. 2015b, Wheal et
al. 2016); and vitamin B12, found only in animal food sources, which supports physical development,
2
various brain functions, and nervous system maintenance (FAO and WHO 2004). Small fish, commonly
consumed by the poor in Bangladesh, are a particularly important source of calcium in diets that are often
deficient in milk and milk products (Hansen et al. 1998), as well as zinc and iron (Thilsted et al. 2016).
Pre-partum consumption of omega-3 fatty acids from fish is associated with lower risk of pre-term
delivery (Imhoff-Kunsch et al. 2012), as well as higher likelihood of children’s normal cognitive
development and fine motor skills (Hibbeln et al. 2007).
Several assessments for Bangladesh have established statistical associations between fish market
conditions, household socioeconomic characteristics, fish consumption and measures of child
undernutrition.1 In aggregative cross-section and time-series studies, Dey et al. (2010) and Toufique
(2015) find that fish consumption is highly responsive to income and fish prices. Grant (2016 - Figures
8.5 and 8.6) finds a higher incidence of stunting in poor households, and for children with less-educated
mothers. However, it also finds significant stunting in wealthy households. NIPORT (2016) reports a
similar finding: Infant and young child feeding (IYCF) practices improve with maternal education and
wealth status, but remain unsatisfactory for 1/3 of children in the highest wealth quintile. After
investigating traditional beliefs in very poor households, Choudhury and Ahmed (2011) suggest that
improved IYCF practices may require targeted health behavior education.
In summary, extensive nutritional research has established a significant link between fish
consumption and child health, and numerous aggregative studies have indicated that both fish
consumption and child health vary with maternal education and household economic status. However,
even relatively well-educated and wealthy mothers can have deficient IYCF practices. And more precise
1
These studies typically use WHO-recommended comparative physical measures based on reference populations or
international standards : stunting (low height/length for age); wasting (low weight-for-height/length); and malnutrition
(circumference of a child's upper arm, or a measure of weight for height).
3
impact assessment in Bangladesh has been hindered by the relative scarcity of individual-level
multivariate analyses.
In this paper, we attempt a more precise quantification using a household panel database constructed
from five Bangladesh Demographic and Health Surveys (2000, 2004, 2007, 2011, 2014). We also revisit
the role of maternal beliefs in child nutrition and health first investigated by Choudhury and Ahmed
(2011), extending the research to potentially-perverse status beliefs associated with income and higher
education.
The remainder of the paper is organized as follows. Section 2 introduces our panel database, with a
particular focus on its temporal and spatial components. Sections 3, 4 and 5 develop and estimate models
for assessing children’s intake of animal-source foods, the fish share of that intake, and the impact of fish
consumption on health status. We summarize and conclude the paper in Section 6.
2. Data and Estimation Strategy
We construct our panel from georeferenced survey data collected by the Bangladesh Demographic
and Health Surveys for 2000, 2004, 2007, 2011 and 2014 (NIPORT 2001, 2005, 2009, 2011, 2016). We
include data for 2,263 DHS clusters, 27,827 households and 36,491 individuals, distributed across years as
indicated in Table 1.
Table 1: DHS survey statistics by year, 2000-2014
DHS
Year Clusters Households Individuals
2000 341 4,847 6,813
2004 361 4,999 6,900
2007 361 4,626 6,150
2011 600 6,882 8,742
2014 600 6,473 7,886
Total 2,263 27,827 36,491
4
We focus on four sets of variables: animal-source foods and fish in children’s diets; their gender, age
and measured health characteristics; their mothers’ education and income status; and spatially- and
temporally-varying determinants of fish supply and consumption norms. All five DHS surveys ask
mothers whether, during the previous 24 hours, their children have consumed any animal-source foods --
eggs, poultry, fish, and other meats. DHS 2011 and 2014 also separate fish from other animal sources.
Our modeling exercise exploits all of this information in a two-part analysis. We investigate the
determinants of consumption of animal-source foods using all five DHS surveys, while using DHS 2011
and 2014 to focus on fish consumption and its impact on child health.
2.1 The Spatial Dimension
We overlay the map of Bangladesh with a 0.25-degree grid to permit controls for spatial variation in
fish supplies and consumption norms. As Figure 1 shows, this grid resolution allows for significant
spatial differentiation while ensuring ample cluster representation within most populated cells.
Figure 1: DHS cluster locations by grid cell
5
2.2 The Temporal Dimension
De Graff (2003) documents the annual “pulse” of Bangladesh’s inland fisheries, with a sharply-
higher catch during September - December as monsoon floods reach their peak and begin receding. By
implication, a complete analysis of fish consumption and child health should incorporate seasonal
fluctuations in fish supply. Single-year DHS surveys do not permit this, since they are implemented in a
subset of months. However, as Table 2 shows, full monthly effects can be incorporated into the five-year
composite panel. Survey periods are November-April (DHS 1999-2000); January-May (DHS 2004);
March-August (DHS 2007); July-December (DHS 2011); and June-November (DHS 2014). As the Total
column shows, combining the surveys provides ample observations for all 12 months. This permits a full
representation of seasonal effects in our exercise, combined with yearly controls to account for annual
fluctuations in flood extent.2
Table 2: Distribution of DHS surveys by month and year
Year
Month 1999-2000 2004 2007 2011 2014 Total
January 1,596 1,860 0 0 0 3,456
February 1,521 1,226 0 0 0 2,747
March 694 1,389 479 0 0 2,562
April 3 1,416 1,501 0 0 2,920
May 0 1,009 1,448 0 0 2,457
June 0 0 1,126 0 127 1,253
July 0 0 1,206 1,786 1,914 4,906
August 0 0 390 1,201 2,066 3,657
September 0 0 0 1,388 2,063 3,451
October 0 0 0 1,797 1,315 3,112
November 1,551 0 0 1,411 401 3,363
December 1,448 0 0 1,159 0 2,607
Total 6,813 6,900 6,150 8,742 7,886 36,491
2
While the rationale for incorporating monthly effects seems compelling, we acknowledge the possibility of countervailing
factors. For example, recent research has studied the impact of flooding in Dhaka on diarrheal disease (Schwartz et al. 2006)
and mortality more generally (Thiele-Eich et al. 2015). Our assessment of disease incidence in Section 5 controls for this
factor.
6
2.3 Modeling Approach
Studies for Bangladesh have found significant pairwise associations between children’s protein
intake and their mothers’ education and economic status (Dey et al. 2010, Toufique 2015, Grant 2016,
NIPORT 2016). Dey et al. (2010) and Toufique (2015) have shown that fish consumption in Bangladesh
responds significantly to seasonal variations in fish supply. Building on these findings, we use our DHS
panel database to explore the determinants of consumption of animal-source foods as well as fish and their
impacts on children’s health. Our econometric models incorporate annual and monthly fixed effects to
control for temporal variations in fish supplies and disease vectors, and spatial fixed effects to control for
regional variations in fish supplies, environmental disease factors, and consumption norms that affect
intake of animal-source foods.
3. Determinants of Intake of Animal-Source Foods
3.1 Model Specification
All five DHS surveys [2000, 2004, 2007, 2011, 2014] include questions on children’s intake of total
animal-source foods (fish, meat, eggs). Drawing on the full panel database, we specify and estimate the
following model:3
1 = + + + + + + +
where
Pi = Intake of animal-source foods by child i during the previous 24-hour period (1 if yes; 0 if
no)
Eij = Mother’s years of education [1 none; 2 primary (1-6); 3 middle (7-9);
4 secondary (10-12); 5 post-secondary (12+)]
Wik = Household wealth status [five groups by DHS household possession count]
Ai = Child age
Ml = Survey month
Ym = Survey year (1999-2000 consolidated to 2000)
Cin = Dummy variable for grid cell of survey cluster (see Figure 1)
3
We exclude gender from this specification because prior experimentation has revealed that animal-source food intake is
gender-neutral after accounting for the variables in model (1).
7
εi = Random error term
3.2 Estimation
We estimate (1) by probit, using the 13,997 observations with non-missing values for all model
variables. Table 3 reports their joint distribution by quintile for household asset ownership and mother’s
years of education. The sample populates all cells of the table, although observations are relatively sparse
for extreme pairs (e.g. [asset quintile 5, mother’s education none]; [asset quintile 1, mother’s education
post-secondary]).
Table 3: Intake of animal-source foods regression sample, DHS 2000-2014
Mother’s Years of Education
Household
Asset Primary Middle Secondary Post-
Quintile 0 (1-6) (7-9) (10-12) Secondary Total
1 1,351 749 135 28 7 2,270
2 1,736 2,623 1,260 290 28 5,937
3 404 1,106 1,092 444 67 3,113
4 128 456 706 496 164 1,950
5 38 85 222 268 114 727
Total 3,657 5,019 3,415 1,526 380 13,997
We report our results in Table 4, excluding estimates for the 191 grid cell dummy variables. To
prevent total collinearity, our specification drops dummy variables for mother’s education [0 years],
household asset quintile 1, survey year 2000 and survey month January. For each category (education,
etc.), estimates should therefore be interpreted as increments from partial effects for the excluded variable
levels.
We find high significance for the variables in each category. The effect of mother’s education
increases steadily by schooling level, while the impact of asset ownership increases rapidly for quintiles 2
and 3 and more slowly for quintiles 4 and 5. We approximate the age effect with a log-cubic
8
Table 4: Probit results - child animal-source food intakea
Mother's Education Parameter Month Parameter
(0 Excluded) Estimates (January Excluded) Estimates
Primary 0.153 February -0.047
(4.24)** (0.66)
Middle 0.422 March 0.048
(10.05)** (0.64)
Secondary 0.496 April -0.021
(9.18)** (0.27)
Post-Secondary 0.637 May -0.045
(7.02)** (0.55)
June 0.091
Asset Quintile (Q1 Excluded) (0.87)
Q2 0.142 July 0.176
(3.09)** (2.17)*
Q3 0.366 August 0.267
(6.80)** (3.06)**
Q4 0.407 September 0.361
(6.42)** (3.95)**
Q5 0.415 October 0.257
(4.99)** (2.82)**
November 0.334
Log Age -3.331 (4.21)**
(8.82)** December 0.279
[Log Age]**2 3.061 (3.34)**
(13.99)**
[Log Age]**3 -0.539
(14.12)** Constant -3.459
(11.39)**
Year (2000 Excluded)
2004 0.435
(7.77)**
2007 0.187
(2.76)**
2011 0.108
-1.73
2014 0.145
(2.08)*
a
Grid location results excluded for brevity
Observations: 13,987
Absolute value of t statistics in parentheses [* significant at 5%; ** significant at 1%]
9
specification, which indicates that, ceteris paribus, intake probability of animal-source foods increases
rapidly after the first months of life and then passes an inflection point toward an asymptote around 24
months. Our monthly results are consistent with significantly-expanded fish supplies in the period
September - December (per De Graaf (2003)). Controlling for monthly effects, we find that fish supply
was particularly ample in 2004 and somewhat greater than the 2000 supply in the other survey years.
3.3 Temporal Impacts: Mother’s Education and Economic Status
We use the probit results to predict monthly animal-source food intake probabilities for 23-month-
old children in 2014. Figure 2 presents monthly results by mother’s education group for a household in
the lowest asset quintile. Figure 3 presents monthly results by asset quintile for a mother with zero years
of education.
The figures highlight two particularly-striking features of the results. First, as previous research has
suggested, both mother’s education and household economic status have large impacts on the probability
of child intake of animal-source foods. For households in the lowest asset quintile (Figure 2), the median
difference in intake probability of animal-source foods for children of the most highly-educated and least-
educated mothers is 24%.4 For mothers with no education (Figure 3), the median difference in intake
probability of animal-source foods for children in the lowest and highest asset quintiles is 16%.5 Table 5
provides evidence on the joint impact of mothers’ education and economic status. In January, child
animal-source food intake probability for mothers with no education in the lowest asset quintile is 38%.
In contrast, the probability is 78% for mothers with post-secondary education in the highest asset quintile -
- an increase of 40%.
4
This is the median probability difference over 12 months for mothers with no education and those with post-secondary
education.
5
This is the median probability difference over 12 months for mothers with no education in the lowest-asset quintile and those
in the highest-asset quintile.
10
Figure 2: Predicted child animal-source food intake probability vs. mother’s educationa
a
Mothers in the lowest asset quintile
Figure 3: Predicted child animal-source food intake probability vs. household asset ownershipa
a
Mothers with no formal education
11
Table 5: Mother’s education, household asset ownership and
child animal-source food intake probabilitya
Mother’s Education (Years)
Household
Asset Primary Middle Secondary Post-
Quintile 0 (1-6) (7-9) (10-12) Secondary
1 0.38 0.44 0.55 0.58 0.63
2 0.44 0.50 0.61 0.63 0.69
3 0.53 0.59 0.69 0.72 0.76
4 0.55 0.60 0.70 0.73 0.77
5 0.55 0.61 0.71 0.73 0.78
a
Predictions computed for January
The second striking feature of the results is the seasonal impact of fish supply on child nutrition.
Table 6 displays predicted animal-source food intake probabilities by education and asset group for
September, a peak catch month, and February, after the floods have receded. For mothers with no
education, children’s September animal-source food intake probability in the lowest asset quintile (53%)
matches the February probability for children in the highest asset quintile. In the lowest asset quintile,
children’s September intake probability for mothers with no education (53%) matches the February
probability for mothers with middle school education (7-9 years). Taken together, the results in Table 5
highlight the critical importance of expanded fish supply for children’s animal-source food intake in the
poorest, least-educated households.
Table 6: Economic and education status vs animal-source food intake probability:
Seasonal comparison
Mother’s Education: 0 Years
Household Asset Quintile
Month 1 2 3 4 5
February 0.37 0.42 0.51 0.53 0.53
September 0.53 0.58 0.67 0.68 0.69
Mother’s Household Asset Status: Quintile 1
Mother’s Education (Years)
Primary Middle Secondary Post-
Month 0 (1-6) (7-9) (10-12) Secondary
12
February 0.37 0.43 0.53 0.56 0.62
September 0.53 0.59 0.69 0.71 0.76
3.4 Spatial Variations
We have estimated the animal-source food intake regression with controls for 191 grid cells overlaid
on the Bangladesh map (Figure 1). To assess the spatial dimension, we use our probit results to predict
animal-source food intake probabilities for children in the lowest asset ownership quintile whose mothers
have no formal education.6 Figure 4 presents monthly distribution statistics for grid cell predictions, from
the 5th percentile to the 95th. The results reflect Figures 2 and 3 in revealing substantial variation
between the peak and off-peak catch seasons. However, they also suggest that the seasonal catch
advantage varies widely by location. Across 191 grid cells and 12 months, the median difference between
predicted animal-source food intake in the 5th and 95th percentile cells is 36%. This is nearly as great as
the difference attributable to shifting a mother’s education and economic status from the lowest to the
highest quintiles in Table 5.
Figure 4: Animal-source foods intake probability: percentiles by month
6
We perform these predictions for 2014.
13
We also find that local variations exhibit significant spatial clustering. Figure 5 plots predicted
animal-source food intake probabilities in September for children in the lowest household asset quintile
whose mothers have no education. For a clear representation of regional clustering, we color-code
probabilities in two categories: P [intake of animal-source foods] ≤ 60% (blue) and P > 60% (red). Figure
5 reveals two clear clustering patterns: (1) Most of the red probabilities (higher animal-source food intake)
are west and south of the Padma and Jamuna Rivers, and most of the blue probabilities (lower animal-
source food intake) are in the east; (2) the coastal region is dominated by red probabilities. We conclude
that, other things equal, local supply conditions convey a notable animal-source food intake advantage to
children in the red-dominated regions.
Figure 5: Spatial distribution of animal-source food intake probabilities (September):
Children in lowest-quintile households [education, assets]
14
4. The Share of Fish in Consumption of Animal-Source Foods
All five DHS surveys ask mothers whether, during the previous 24 hours, their children have
consumed any animal-source food -- eggs, poultry, fish, and other meats. In Section 3, we have used the
full data set to explore the determinants of total consumption of animal-source foods. In this section, we
extend the analysis to fish consumption using the source-specific information in DHS 2011 and 2014.
As Table 7 shows, understanding the determinants of fish consumption is particularly important for
the poor in Bangladesh. The table summarizes DHS survey responses on children’s consumption of fish,
eggs and meat during the 24 hours prior to the survey date. For each child, we code each animal-source
food as 1 if it was consumed and 0 otherwise. We focus on cases with positive consumption (total coded
value > 0) and compute the fish share of animal sources by dividing the fish consumption code [0,1] by
total coded value. The computed ratio can be interpreted as the share of fish in animal-source foods,
conditional on non-zero consumption.
Table 7 reports mean and median shares by asset quintile. The results highlight the particular
importance of fish as a animal-source food for the poorest children. In household asset quintile 1, the
mean and median fish shares are 66.7% and 100%, respectively.
Table 7: Share of fish in children’s animal-source foods
by household asset class
Mean Median
Asset Fish Share Fish Share
Quintile (%) (%)
1 66.7 100.0
2 57.5 50.0
3 52.6 50.0
4 41.3 50.0
5 35.0 33.3
15
To assess the determinants of fish shares more fully, we specify a form of equation (1) that
incorporates the years and months covered by DHS 2011/2014.7 Prior experimentation has revealed that
child age is not significant in this case, so we have excluded it.
2 = + + + + + +
where
Fi = Share of fish in animal-source foods {Fish consumed [1,0] / Sum (protein consumed (fish
[1,0], eggs [1,0], meat [1,0])}
Eij = Mother’s educational status [1 none; 2 primary; 3 middle; 4 secondary; 5 post-secondary]
Wik = Household wealth status [five groups by household possession count]
Ml = Survey month
Ym = Survey year
Cin = Grid cell of survey cluster (see Figure 1)
εi = Random error term
For each surveyed child, the variable F can take on four values: 0, .33, .50, 1. The appropriate
estimator in this case is fractional logit, since the dependent variable is appropriately bounded by [0,1] and
includes intermediate values. Table 8 reports our results, with t-statistics computed using robust standard
errors.8 For brevity, we exclude the dummy variable estimates for 191 grid cells.
Our results indicate a high degree of statistical significance for mother’s education, household
economic status and the temporal variables. Ceteris paribus, we find increasingly-negative impacts on the
fish share as mother’s education and household economic status increase. In this case, the limitation of
DHS 2011/2014 to June - December hinders comparison between months of peak fish catch and off-peak
months. However, the September - December estimates are significantly higher than the June estimate,
and higher than the July estimate as well.
7
As Table 2 shows, survey data collection was restricted to the period June - December for DHS 2011/2014.
8
We employ Stata glm [link(logit), family(binomial), robust]
16
Taken together, the seasonal results in Tables 4 and 8 suggest that children’s fish intake, and
therefore animal-source food intake, increase with the pulse in fish supply during September - December.
And, as Tables 6 and 7 show, this increase is particularly important for poor households. Figures 6 and 7
highlight two important facets of these results: the impact of the peak-season fish catch (Figure 6) and the
joint impact of economic status and mother’s education (Figure 7).
Table 8: Fractional logit results - share of fish in animal-source foodsa
Mother's
Education Parameter Month Parameter
(0 Excluded) Estimates (June Excluded) Estimates
Primary -0.319 July 1.385
(2.71)** (3.14)**
Middle -0.518 August 1.815
(4.19)** (4.15)**
Secondary -0.726 September 1.777
(5.23)** (4.04)**
Post-Secondary -1.053 October 1.899
(5.28)** (4.24)**
November 1.565
Asset Quintile
(3.48)**
(Q1 Excluded)
Q2 -0.666 December 1.48
(2.14)* (3.22)**
Q3 -0.707
(2.25)* Constant -0.308
Q4 -1.081 (0.46)
(3.38)**
Q5 -1.175
(3.54)**
Year
(2011 Excluded)
2014 -0.203
(2.59)**
a
Grid location results excluded for brevity
Observations: 3,201
Absolute value of t statistics in parentheses [* significant at 5%; ** significant at 1%]
17
Figure 6: Monthly fish share in intake of animal-source foods
by household asset groupa
a
For mothers with no formal education
Figure 7: Fish share in intake of animal-source foods
by household asset group and mother’s educationa
a
For October 2014
Figure 6 reflects Figure 3, with the focus narrowed from animal-source foods to fish. The peak-
season catch sharply increases children’s fish intake, which translates to a sharp increase in the likelihood
of intake of animal-source foods. Unfortunately, as Figure 7 shows, the potential health benefits of
expanded fish supply are attenuated by both income and education effects. As we note in the Introduction,
18
extensive research has shown that fish consumption conveys more health benefits than consumption of
other animal sources. Nevertheless, our results for Bangladesh clearly indicate a “status bias” favoring
non-fish animal-source food consumption as mothers’ education and household economic status increase.
The education result seems particularly unfortunate, with two possible interpretations: Either current
school instruction pays little attention to the health benefits of fish consumption, or the effect of this
instruction is overwhelmed by the higher status assigned to non-fish protein sources. In any case, our
results for both education and economic status suggest a powerful and perverse “status goods” effect on
fish consumption. We believe that these findings have potentially-important implications for health
education policy in Bangladesh.
5. Child Health Impacts
Using DHS 2011 and 2014, we investigate the incremental health impacts of consuming fish and
other animal-source foods (eggs, meat) using an estimation model that includes the following elements:
(i) Secular declines in mortality and morbidity risk from improved public health services.
(ii) The socioeconomic, temporal and spatial determinants of consumption of animal-source foods
incorporated in models (1) and (2): mother’s education; household economic status, seasonal variations in
fish supply, and spatial variations in fish supply and consumption norms. After incorporating these
variables (and gender) into the estimation exercise, we also include the animal-food sources themselves.
Ceteris paribus, the direct estimates for these sources can be interpreted as indicators of individual
preferences. Our approach assumes that these preferences are constant through time.
(iii) Maternal fish consumption in the immediate pre-partum and post-partum periods. Although we
do not have direct evidence9, basic economics suggests that mothers’ fish consumption should also
9
We are unable to include direct survey information on maternal fish consumption because the relevant questions were
excluded from DHS 2011 and 2014.
19
respond to expanded supply in the peak-catch period. If children’s disease resistance is positively related
to maternal nutrition during the late gestation and early nursing periods, then we should observe better
health outcomes for children born during the peak-catch period.
Our general health impact model is specified as follows:
3 = + + + + + + + +
+ + + + +
where Hi = Health indicator
Deceased (1 if occurred; 0 otherwise),
Recent diarrhea (1 if occurred; 0 otherwise)
Recent fever (1 if occurred; 0 otherwise)
Recent cough (1 if occurred; 0 otherwise)
fi = Child fish intake during the previous 24 hours (Yes = 1; 0 otherwise)
ei = Child eggs intake during the previous 24 hours (Yes = 1; 0 otherwise)
mi = Child meat intake during the previous 24 hours (Yes = 1; 0 otherwise)
Fi = 1 if female; 0 if male
Ei = Years of mother’s education
Wi = Household asset count
Si = Child size at birth (included in mortality risk estimation)
Y2014 = 1 if survey year is 2014; 0 if year is 2011)
Mj = Dummy variable for survey month j
Cik = Dummy variable for location grid cell k (see Figure 1)
Pi = Dummy variable for birth period (1 if September - December; 0 otherwise)
BYim = Dummy variable for birth year m
5.1 Mortality Risk
Table 9 reports our results for child mortality risk. We focus particularly on two elements: (1)
different types of animal-source food consumption which, as noted above, should be interpreted as
reflecting individual preferences after controlling for all previously-identified socioeconomic, temporal
and spatial factors; (2) birth during the peak-catch season (September - December), when mothers’ fish
intake should increase.
We find a large, negative, highly-significant impact for fish consumption preference. The other two
animal-source foods also have negative estimated impacts, but the parameters are smaller in absolute
20
value and fail to meet the standard criteria for statistical significance. Per our prior expectation, we also
find a large, negative, highly-significant impact for birth during the peak-catch period.
We find that female children have lower mortality rates than male children, ceteris paribus.
Mother’s education, family asset ownership and child size at birth do not have significant effects.
Reported child deaths yield a higher mortality rate for the 2014 survey, but survey month has no impact on
the results. In contrast, the birth-year results are strong and declining as public health services improve.
To assess their quantitative significance, we use our probit results to estimate differential mortality
rates within a centrally-located grid square with typical population density (Figure 8, square outlined in
light blue). We perform the prediction for a mother with no education who resides in a household whose
asset count is zero; the mean child size at birth for this education and asset ownership group; survey year
2014; birth year 2012; the first six survey months (dummy values 0 for the survey months included in the
result); and zero consumption preference for eggs and meat.
We produce four predictions: with and without fish consumption preference and peak-catch-season
birth, respectively. Table 10 presents our results, which suggest substantial impacts. The estimated
mortality rate is 72.3 per 1,000 for children born in this locale during the off-peak-catch season to mothers
with no fish preference (beyond the structurally-determined factors, per model (1)). With no change in
fish preference, the estimated mortality rate declines to 32.4 per 1,000 for a child born during the peak-
catch season. In contrast, children born to mothers with fish preference have predicted mortality rates of
31.5 per 1,000 in the off-peak season and 12.3 in the peak-catch season.
In summary, our results suggest a remarkable 83% decrease in the child mortality rate (from 72.3 to
12.3) produced by two elements related to fish consumption. We believe that this should be viewed as an
outer-bound estimate, because our interpretation depends on the following assumptions: (1) the fish
consumption result reflects individual preferences, not other factors that cannot be observed;
21
Table 9: Probit results - child mortality ratea
Parameter Parameter
Protein Intake Estimates Survey Month Estimates
Fish -0.400 July 0.12
(3.62)** (0.24)
Eggs -0.186 August -0.022
(1.44) (0.04)
Meat -0.244 September 0.173
(1.48) (0.33)
October 0.217
Characteristics (0.41)
Female -0.278 November 0.262
(3.10)** (0.48)
Mother’s Education 0.021 December 0.576
(1.45) (1.01)
Household Assets -0.021
(0.69) Survey Year 2014 2.809
Size at Birth 0.081 (7.98)**
(1.42)
Birth Year
Peak-Catch Birth -0.388 2006 5.347
(3.96)** (9.33)**
2007 5.032
Constant -6.389 (9.54)**
(6.92)** 2008 5.081
(9.66)**
2009 4.806
(9.01)**
2010 3.963
(7.63)**
2011 2.927
(6.71)**
2012 2.127
(5.47)**
2013 0.836
(2.15)*
a
Grid location results excluded for brevity
Observations: 4,761
Absolute value of t statistics in parentheses [* significant at 5%; ** significant at 1%]
22
Figure 8: Area selected for predictive assessment of child mortality
Table 10 : Fish consumption elements and child mortality rate
(per thousand)
Fish Off-Peak Peak-Catch
Preference Birth Birth
0 72.3 32.4
1 31.5 12.3
(2) individual preferences have remained constant over the period covered by our database; (3) our result
for birth during the peak-catch season reflects increased fish supply, not other unobserved factors.
Despite these cautionary notes, we believe that our results support the following proposition:
Children’s mortality risk in Bangladesh is substantially lower if they are born during the peak-catch
season, to mothers with strong fish consumption preference.
23
5.2 Childhood Illness
We also consider the impact of animal-source food intake on the incidence of childhood illness, as
reported in the DHS surveys. Our exercise tests the impact of consumption of animal-source foods,
particularly fish, on resistance to common childhood infections. We focus on the reported occurrence of
diarrhea, fever and coughing immediately before the survey date.
As before, we consider the roles of fish consumption preference and fish supply during the birth
period, after controlling for the socioeconomic, temporal and spatial determinants of fish intake during the
survey period. We perform probit estimation on a stacked panel data set drawn from DHS 2011 and 2014.
The dependent variable is the occurrence of illness, defined as diarrhea, fever or coughing. Our
independent variables replicate those in the child mortality model (Table 9). We also introduce dummy
variables to allow for variation in mean incidence rates across illness categories.
Our results (Table 11) indicate that child health status is strongly affected by socioeconomic status:
Both mother’s education and household asset ownership have negative, highly-significant effects on
illness incidence. Our results also suggest that female children have lower illness incidence, ceteris
paribus. After controlling for the socioeconomic, temporal and spatial determinants of consumption of
animal-source foods, we find that fish consumption preference has a negative, significant impact on
illness. As before (Table 9), we find no significant impact for the intake of other animal-source foods.
And again, we find a lower incidence rate for children born during the peak-catch period (September -
December).
To assess the quantitative significance of these results, we replicate our simulation strategy for child
mortality. We use the probit results reported in Table 11 to predict incidence rates for children with birth
year 2012 whose mothers reside in the grid square identified in Figure 8; have no formal education;
belong to a household with zero tabulated assets; and have zero consumption preference for eggs and
24
meat. We produce four predictions: with and without fish consumption preference and peak-catch season
birth, respectively. Table 12 presents the results.
Table 11: Probit results - child illness ratea
Parameter Parameter
Protein Intake Estimates Survey Month Estimates
Fish -0.066 July -0.18
(2.48)* (1.46)
Eggs 0.005 August -0.303
(0.19) (2.45)*
Meat -0.042 September -0.218
(1.17) (1.74)
October -0.147
Characteristics (1.16)
Female -0.084 November -0.283
(3.86)** (2.19)*
Mother’s Education -0.018 December -0.483
(5.26)** (3.53)**
Household Assets -0.016
(2.09)* Birth Year
2009 0.241
Peak-Catch Birth -0.071 (3.82)**
(2.63)** 2010 0.292
(7.19)**
Constant 0.036 2011 0.146
(0.19) (3.65)**
Diarrhea Adjustment -1.27 2012 0.093
(41.54)** (1.57)
Fever Adjustment 0.084 2013 0.177
(3.60)** (4.60)**
a
Grid location results excluded for brevity
Observations: 18,008
Absolute value of t statistics in parentheses [* significant at 5%; ** significant at 1%]
25
Table 12: Child illness incidence by fish preference
and birth season (%)
Birth Month
Max
Off- Peak- Change
Peak Catch (%)
Diarrhea
Fish
Preference
0 8.7 7.6
1 7.7 6.7 -23.0%
Fever
Fish
Preference
0 49.8 46.9
1 47.1 44.3 -11.0%
Cough
Fish
Preference
0 46.4 43.6
1 43.8 41.0 -11,6%
For diarrhea, the predicted incidence rate is 8.7% for children born during the off-peak season to
mothers with no fish preference (beyond the structurally-determined factors, per model (1)). With no
change in fish preference, the rate declines to 7.6% for a child born during the peak-catch season. For
mothers with fish preference, the incidence rates are 7.7% and 6.7% respectively for children born in the
off-peak and peak-catch seasons. In comparison with the worst case (no fish preference, off-peak birth),
the best case (fish preference, peak-catch birth) has an incidence rate 23.7% lower.
Fever and cough have much higher mean incidence rates than diarrhea, but similar patterns of
variation with fish preference and season of birth. From worst to best case, the incidence rates for fever
and cough fall by 11.0% and 11.6% respectively.
26
6. Summary and Conclusions
In this paper, we have analyzed 36,491 case histories drawn from individual and household records
in the Bangladesh Demographic and Health Surveys for 2000, 2004, 2007, 2011 and 2014. We address
three related questions: What are the socioeconomic determinants of consumption of animal-source
foods? Among animal-source foods, what are the socioeconomic determinants of the fish share? What
are the impacts of consumption of animal-source foods (fish, meat, eggs) on child mortality and resistance
to infectious disease? The paper focuses particularly on fish consumption and its health effects for
children in their first years of life. Critical factors examined include maternal education, family economic
status, child gender, and temporal/spatial variations in fish supply.
We find that maternal education and family economic status are highly-significant determinants of
animal-source food intake, along with annual, seasonal and local-area fish supply conditions. After
controlling for these factors, we find that children’s animal-source food intake is gender-neutral. Our
results for the fish share of animal-source food consumption assign significant roles to family economic
status, maternal education and seasonal and spatial variations in fish supply. The importance of fish
supply for poverty alleviation is highlighted by another finding: The share of fish in consumption of
animal-source foods is highest for the poorest households, and declines significantly with income. By
implication, programs that increase fish supply and lower fish prices will have a significant impact on
intake of animal-source foods, particularly for the poorest households.
We also find a highly-significant, negative relationship between maternal education and the share of
fish in animal-source food consumption. In the lowest income groups, mothers shift strongly toward meat
and eggs as their education level increases, suggesting a perverse “status bias” against fish despite their
lower cost and health advantages.
27
We assess the implications for child health by estimating the impact of fish consumption on the child
mortality rate and the incidence of three common disease symptoms: diarrhea, fever and coughing. After
controlling for socioeconomic, temporal and spatial determinants of animal-source foods and fish
consumption, we focus on two additional factors: (1) mothers’ individual preference for fish consumption;
and (2) maternal fish consumption during the immediate pre- and post-partum periods. Our results
suggest that both factors have large effects on child mortality risk and significant effects on childhood
disease symptoms.10
In summary, our results highlight the critical role played by fish supply in maternal and child
animal-source food consumption, the fish share of that consumption, and children’s health status in
Bangladesh. They suggest that programs which promote fish supplies during the off-peak season may
significantly improve child health outcomes, particularly in the poorest, least-educated households. Even
for those households, however, we find that schooling has a strong and perverse “status bias” effect: As
mothers’ formal education increases, they shift away from fish toward other sources of animal-source
foods that are both more costly and less healthy. This suggests a potentially-important role for public
education about the benefits of fish as an animal-source food.
10
Identification of components in fish with desired effects on child nutrition and health is a subject for
future research.
28
References
Bogard, J.R., Hother, A.-L., Saha, M., Bose, S., Kabir, H., Marks, G.C., Thilsted, S.H. 2015a. Inclusion
of small indigenous fish improves nutritional quality during the first 1000 days. Food Nutr. Bull., 1–14.
Bogard, J.R., Thilsted, S.H., Marks, G.C., Wahab, M.A., Hossain, M.A.R., Jakobsen, J., Stangoulis, J.
2015b. Nutrient composition of important fish species in Bangladesh and potential contribution to
recommended nutrient intakes. J. Food Comp. Anal. 42, 120–133.
Choudhury, Nuzhat and Syed M Ahmed. 2011. Maternal care practices among the ultra poor households
in rural Bangladesh: a qualitative exploratory study. BMC Pregnancy and Childbirth 2011, 11:15.
De Graaf, G. 2003. Dynamics of floodplain fisheries in Bangladesh, results of 8 years fisheries monitoring
in the Compartmentalization Pilot Project. Fisheries Management and Ecology, 10: 191–199.
Dey, A. K., M. M. Bose and M. Alam. 2008. Country Case Study: Development and Status of
Freshwater Aquaculture in Bangladesh, World Fish Center, Bangladesh.
Dey, M.M., M.F. Alam and M. L. Bose. 2010. “Demand for Aquaculture Development:
Perspectives from Bangladesh for Improved Planning,” Reviews in Aquaculture, 2: 16-32.
Ezzati, M., Riboli, E., 2013. Behavioral and dietary risk factors for noncommunicable diseases. N. Eng. J.
Med. 369, 954–964.
FAO, WHO, 2004. Vitamin and Mineral Requirements in Human Nutrition: Report of a Joint FAO/WHO
Expert Consultation, second ed. Food and Agriculture Organization, World Health Organization,
Geneva, Switzerland.
Golden, Christopher D., Edward H. Allison, William W. L. Cheung, Madan M. Dey, Benjamin S.
Halpern, Douglas J. McCauley, Matthew Smith, Bapu Vaitla, Dirk Zeller& Samuel S. Myers. 2016.
Nutrition: Fall in fish catch threatens human health. Nature, Vol. 534.
Grant, James P, School of Public Health and National Nutrition Services. 2016. State of food security and
nutrition in Bangladesh 2015. Dhaka, Bangladesh: James P Grant School of Public Health and National
Nutrition Services.
Hansen, M., Thilsted, S.H., Sandström, B., Kongsbak, K., Larsen, T., Jensen, M., Sørensen, S.S.,
1998. Calcium absorption from small soft-boned fish. J. Trace Elem. Med. Biol. 12, 148–154.
Hibbeln, J.R., Davis, J.M., Steer, C., Emmett, P., Rogers, I., Williams, C., Golding, J. 2007. Maternal
seafood consumption in pregnancy and neurodevelopmental outcomes in childhood (ALSPAC study): an
observational cohort study. Lancet 369, 578–585.
29
Howlader, S.R., S. Kavita, B. Ferdousi, P. Dipika, E.A. Sommerfelt and K.Tara. 2012. Investing
in Nutrition Now: A Smart Start for Our Children, Our Future. Estimates of Benefits and Costs of a
Comprehensive Program for Nutrition in Bangladesh, 2011– 2021, PROFILES and Nutrition Costing
Technical Report. Washington, DC: Food and Nutrition Technical Assistance III Project (FANTA), FHI
360.
Imhoff-Kunsch, B., Briggs, V., Goldenberg, T., Ramakrishnan, U., 2012. Effect of n - 3
long-chain polyunsaturated fatty acid intake during pregnancy on maternal, infant, and child health
outcomes: a systematic review. Paediatr. Perinat. Epidemiol. 26, 91–107.
Kent, G. 1987. Fish, Food and Hunger: The Potential of Fisheries for Alleviating Malnutrition,
Colorado, USA: West View Press.
Lim, S., Vos, T., Flaxman, A., Danaei, G., Shibuya, K., Adair-Rohani, H., Amann, M. 2012. A
comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor
clusters in 21 regions, 1990–2010: a systematic analysis for the Global Burden of Disease Study 2010.
Lancet 380, 2224–2260.
Michaelsen, K.F., Hoppe, C., Roos, N., Kaestel, P., Stougaard, M., Lauritzen, L., Mølgaard, C., Girma,
T., Friis, H. 2009. Choice of foods and ingredients for moderately malnourished children 6 months to 5
years of age. Food Nutr. Bull. 30, 343–404.
Ministry of Health and Family Welfare (MOHFW) [Bangladesh]. 2014. Health, Population and Nutrition
Sector Development Program (2011-2016). Revised Program Implementation Plan (PIP), Volume I.
Dhaka, Bangladesh: Planning Wing, MOHFW.
Murphy, S. P. and L. H. Allen. 2003. “Nutritional Importance of Animal Source Foods,” The Journal of
Nutrition, 33:3932S–3935S.
National Institute of Population Research and Training (NIPORT), Mitra and Associates, and ORCMacro.
2001. Bangladesh Demographic and Health Survey 1999-2000 . Dhaka, Bangladesh and Calverton,
Maryland [USA]: National Institute of Population Research and Training, Mitra and Associates, and ORC
Macro.
National Institute of Population Research and Training (NIPORT), Mitra and Associates, and ORC
Macro. 2005. Bangladesh Demographic and Health Survey 2004. Dhaka, Bangladesh and Calverton,
Maryland [USA]: National Institute of Population Research and Training, Mitra and Associates, and ORC
Macro.
National Institute of Population Research and Training (NIPORT), Mitra and Associates, and Macro
International. 2009. Bangladesh Demographic and Health Survey 2007. Dhaka, Bangladesh and
Calverton, Maryland, USA: National Institute of Population Research and Training, Mitra and Associates,
and Macro International.
30
National Institute of Population Research and Training(NIPORT), Mitra and Associates, ICF
International. 2011. Bangladesh Demographic and Health Survey 2011. Dhaka and Calverton, Md,
USA: NIPORT/Mitra and Associates/ICF International, 2013.
National Institute of Population Research and Training (NIPORT), Mitra and Associates, and ICF
International. 2016. Bangladesh Demographic and Health Survey 2014. Dhaka, Bangladesh, and
Rockville, Maryland, USA: NIPORT, Mitra and Associates, and ICF International.
Neumann, C., D. M. Harris and L.M. Rogers. 2001. “Contribution of Animal Source Foods in
Improving Diet Quality and Function in Children in the Developing World,” Nutrition
Research, 22(1):193-220.
Rimm, E.B., Mozaffarian, D., 2006. Fish intake, contaminants, and human health: evaluating the risks and
the benefits. J. Am. Med. Assoc. 296, 1885–1899.
Roos, Nanna, M.A. Wahab, Mostafa A. Hossain, S. H. Thilsted and H. Shakuntala. 2007.
“Linking Human Nutrition and Fisheries: Incorporating Micronutrient-dense, Small Indigenous Fish
Species in Carp Polyculture Production in Bangladesh,” Food and Nutrition Bulletin, 28(2 Suppl):
280-93.
Sandström, B., Almgren, A., Kivistö, B., Cederblad, Å., 1989. Effect of protein level and protein source
on zinc absorption in humans. J. Nutr. 119, 48–53.
Schwartz, Brian, Jason Harris, I. Khan Ashraful, Regina Larocque, David Sack, Mohammad Malek, Abu
Faruque, Firdausi Qadri, Stephen Calderwood, Stephen Luby and Edward Ryan. 2006. Diarrheal
Epidemics in Dhaka, Bangladesh, During Three Consecutive Floods: 1988, 1998, and 2004. Am J Trop
Med Hyg. 2006 Jun; 74(6): 1067–1073.
Thiele-Eich, Insa, Katrin Burkart and Clemens Simmer. 2015. Trends in Water Level and Flooding in
Dhaka, Bangladesh and Their Impact on Mortality. Int. J. Environ. Res. Public Health, 12: 1196-1215.
Thilsted, Shakuntala Haraksingh Thilsted, Andrew Thorne-Lyman, Patrick Webb, Jessica Rose Bogard,
Rohana Subasinghe, Michael John Phillips, Edward Hugh Allison. 2016. Sustaining healthy diets: The
role of capture fisheries and aquaculture for improving nutrition in the post-2015 era Food Policy 61
(2016): 126–131.
Toufique, Kazi Ali. 2015. Analysis of Fish Consumption and Poverty in Bangladesh. Dhaka: The
Bangladesh Institute of Development Studies.
Wheal, M.S., DeCourcy-Ireland, E., Bogard, J.R., Thilsted, S.H., Stangoulis, J.C.R., 2016.
Measurement of haem and total iron in fish, shrimp and prawn using ICP-MS: implications for dietary
iron intake calculations. Food Chem. 201, 222–229.
Zhao, L.G., Sun, J.W., Yang, Y., Ma, X., Wang, Y.Y., Xiang, Y.B., 2015. Fish consumption and all-cause
mortality: a meta-analysis of cohort studies. Eur. J. Clin. Nutr., 1–7.
31