Nutritional shocks in utero and its long-term effects : Ramadan fasting in Indonesia

Nutritional Shocks in Utero and its

Long-Term Effects

Ramadan Fasting in Indonesia

Yente van Roosmalen

11273070

Master’s Thesis (15 ECTS)

MSc in Economics – Development Economics

University of Amsterdam

Supervisor: Prof. Dr. Hessel Oosterbeek

15

August 2018

1

Statement of Originality

This document is written by the student Yente van Roosmalen who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

2

Table of Contents

1. Introduction ... 3

2. Data ... 7

2.1 Indonesian Family Life Survey ... 7

2.2 Descriptive Statistics ... 7

2.2.1 Pregnancy Panel ... 8

2.2.2 Employment Outcomes Panel ... 8

2.2.3 Cognitive Panel ... 10

3. Empirical Methodology... 11

3.1 Treatment variables: Exposure to Nutrition Shocks in Utero ... 11

3.1.1 Effects per Trimester... 13

3.1.2. Self-Reported Religiosity ... 14

3.2 Experimental Design and Procedures ... 15

4. Results ... 16

4.1 Pregnancy Panel... 16

4.2 Employment Outcomes Panel ... 18

4.3 Cognitive Panel... 20

5. Discussion ... 22

6. References ... 25

3

Abstract

Each year pregnant women fast during Ramadan, exposing individuals to mild in-utero nutrition shocks. In this thesis, data from Indonesian Family Life Survey is used to show that individuals who are exposed to in-utero exposure to Ramadan fasting, are negatively affected in short-term health outcomes and long-term cognitive and employment outcomes.

1. Introduction

The crux of the Fetal Origins Hypothesis (FOH) is that the time in utero is vital for life long human development. The fetus adapts to the situation in order to increase the likelihood of survival but that comes at the expense of some negative consequences in the long run. General growth and cognitive development are affected and individuals become more likely to have type 2 diabetes, heart and kidney disease (Chen, 2014). The developmental adaptations result in accelerated starvation for the mother as well as damage the fetus (Mirghani et al. 2005). Specifically the growth of a placental enzyme that governs cortisol is hampered, which results in a heightened exposure to the hormone cortisol that is in turn linked to certain diseases as well as a lower cognitive development (Dikensoy et al. 2009). In order to prove FOH it is necessary to research mild nutrition shocks rather than severe nutrition shocks. Severe shocks used in this context are for example the disaster in Chernobyl (Almond et al. 2009) and the Dutch Famine in 1944 (Ravelli et al. 1998). The more severe shocks impact economic growth and result in mortality which means that the more healthy population survive making it difficult to observe FOH (Chen, 2014). While severe shocks have contributed significantly to research, it is interesting to look at milder shocks since they affect an even more widespread population.

Fasting is part of a many cultures and has been practiced over the ages in all corners of the world. Christians fast during Lent, Hindus during Durga Puja Navartri, Jews during Yom Kippur and Muslims during Ramadan. Restriction of caloric intake is also practiced outside of the religious context, in the United States, one in four pregnant women skip meals to restrict weight gain (Chen, 2014). The impact of nutritional shocks in utero could be affecting a vast majority of people every year. In this paper, we explore this effect through the Ramadan fast observed by Muslims. Ramadan takes place during the ninth month of the Islamic calendar and is a 29-30 days strict fast which entails abstaining from (among others) drinking and eating from dawn till sunset. It is very much a communal affair where friends and family

4

spend more time together and reflect on Islamic teachings. Muslims are required to observe the fast once puberty is reached. One can be excluded from this if you are elderly, under the age of 12, menstruating, breastfeeding, pregnant or chronically ill. When people do not participate in fasting, they will have to make up later (by themselves) and oftentimes also donate something to charity or the poor. The exact rules concerning Ramadan differ per Islamic denomination and region. Research has found substantive evidence that pregnant women still fast regardless of being exempted from it (Joosoph et al., 2004). Around 70-90% of pregnant women do participate in Ramadan, a similar estimation can be found in Indonesia (van Ewijk, 2011).

Since three quarters of all pregnancies overlap with Ramadan each year, yearly 1.2 billion Muslims are exposed to in utero nutrition shocks, and when taking into account fasting outside of Islam the impact is even more widespread. What makes it even more relevant is that even though Ramadan fasting is embedded in social norms, problems resulting from exposure to Ramadan fasting in utero can be solved easier than severe shocks lik e the Dutch Famine since mothers (and their surroundings) are in control of managing these shocks. The need to disentangle the effects that the surroundings have on in utero nutrition choices is vital to understanding and designing effective policy measures (Majid, 2015). The most important paper’s views on in utero nutrition decision making are aggregated in this paper and can be found in appendix 7.1. The main reasons can be roughly divided in subcategories: Religious, costs of delay, social, health and other.

The FOH is being tested in the field of economics and has provided the most credible evidence yet. Almond and Mazumber (2011) pioneered measuring the long-term effects of in utero Ramadan fasting exposure using data from Michigan, Uganda, and Iraq. Their research shows that long-term effects of nutrition shocks are not only evident in more severe nutrition shocks such as famines, but also in milder nutrition shocks during fasting. The long-term effects they observe are lower birth weights and lower share of male births, higher probability of having disabilities and lower wealth. Almond et al. (2014) use English registry data on students from Pakistan and Bangladesh and finds lower math and reading test scores for 7 year olds.

Van Ewijk (2011) builds on Almond and Mazumber’s findings by looking at in-utero exposure to Ramadan fasting in a different country and context using the Indonesian Family Life Survey (IFLS). Van Ewijk’s research has a focus on health outcomes, which the dataset used can facilitate and has followed closely the medical theory and expected effects on

5

individuals exposed to Ramadan. He finds that general health is influenced by in utero exposure to Ramadan fasting, especially for older people. He confirms earlier research that the ratio of males born is lower, as is expected due to their increased vulnerability in utero. Specifically, exposure leads to a higher chance of developing symptoms that are likely to lead to heart disease, type 2 diabetes and kidney problems. One of his major contributions is on excluding potential drivers for alternative explanations: By using mother fixed effects he shows that unobservables which are time invariant do not drive the observed results. Combining this and comparing observed characteristics it seems unlikely that the observed results are driven by selective timing of pregnancy. Secondly, he uses date of birth rather than month of birth reducing noise. Besides van Ewijk’s quantitative analysis, he has interviewed doctors, midwives and other health workers in Indonesia to ‘shed some light on the local situation’.

Majid (2014) uses the 4th wave of the IFLS to study the effects of in-utero exposure to Ramadan fasting. He explores the fetal health origins hypothesis’ life cycle approach by measuring the impact on cognitive skills, child labour and labour supply behaviour of individuals exposed to in-utero Ramadan exposure. With the limitation that there is no actual data available on specifics in fasting behaviour, Majid (2014) was the first one to use self-reported religiosity as a proxy for actual fasting behaviour. This provides an indicative compliance to treatment. Majid shows parental socio-economic status is not related to exposure to fasting and therefore time-invariant unobservables do not seem to drive his results. Additionally, Majid looks at biological fixed effects for a sub-sample of adults. His results show that individuals exposed to Ramadan fasting work less hours as well as increase probability of being self-employed and finds that these effects are driven by more religious families. Cognitive tests are lower for those exposed to Ramadan and an increase in incidence of child labour is found.

This thesis builds on the aforementioned research using the most recent 5th wave of the Indonesian Family Life Survey. Using the self-reported religiosity as a proxy for compliance to treatment, I measure exposure to in utero Ramadan fasting in multiple points of the course of an individual’s life. Similar to earlier research, my analysis uses robust standard errors based on household id so that unobserved family and community variables will stay constant over time (van Ewijk, 2016). Three different subsets of the IFLS dataset are used. Starting out by using detailed information on pregnancy and childbirth, the actual duration of the pregnancy is known and can be used to calculate more exactly whether an individual was

6

exposed to Ramadan fasting in utero reducing erroneous allocation to exposure and trimester effects. Three other identification strategies are used: Firstly, the literature-based exposure variable identical to the one used by van Ewijk (2015). Secondly, the narrow exposure variable uses the average duration pregnancy in this dataset. Thirdly, takes into account a broad pregnancy duration of 280 days. Having three different treatment variables facilitates critical analysis of the data as well as how the methodology affects it.

The second data panel evaluates impact of exposure to Ramadan fasting in utero on the w-score. This score has only been included in the fifth wave of IFLS and is measured by IFLS researchers themselves. The w-score measures cognitive ability and is adapted to work for the low- education and income environment in Indonesia. The third data subset evaluates long-term employment outcomes similar to Majid (2015): Hours worked per week and self-employment. Besides using the most recent wave in IFLS which facilitates measuring whether the effects are the same over time, it also is of vital importance to replicate research in order to confirm external validity.

This study has found that exposure to Ramadan significantly impacts the relative size of Muslims at birth as well as the w-score. Both outcomes are especially affected for very religious Muslims who are exposed in their third trimester of pregnancy. Providing suggestive evidence that self-reported religiosity is a good indicator of adherence to fasting. These observations are found across all three treatment exposure variables, which increases the robustness of these findings. This research finds that Muslims who are exposed to Ramadan work 0.13 less hours per week but are not more likely to be self-employed.

The upcoming chapters of the paper are divided as follows: chapter 2 describes the data, chapter 3 discusses the methodology and treatment variables, chapter 4 presents the results, and finally chapter 5 provides the discussion.

7

2. Data

In the second chapter, elaborates on the dataset used as well as the reasons why it was chosen. After that the three panels are analysed and the described by evaluating their summary statistics.

2.1 Indonesian Family Life Survey

The Indonesian Family Life Survey (IFLS) is a multifaceted longitudinal survey carried out by the RAND corporation in Indonesia. The survey has been administered 5 times starting in 1993 with the most recent one being fielded in 2015. The IFLS contains information on community, household and individual level on economic, health and social indicators. The IFLS survey has high re-contact rates of around 86.9% in all 5 waves, this enhances data quality by lowering bias caused by non-random attrition1.

Indonesia is a useful choice for this study since the majority (88%) identify as Islamic, which leaves a substantial amount of non-Muslims who are not expected to fast and therefore can be used for falsification tests. The second benefit of choosing Indonesia for this research is that it is close to the equator resulting in a homogenous treatment over the years. The Islamic calendar differs from the Gregorian calendar with 10-11 days per year, which could result in change of sunlight hours per day if the country was not close to the equator.

2.2 Descriptive Statistics

Table 1 contains relevant summary statistics for Muslims and non-Muslims divided by treatment exposure, a binary variable based on whether an individual was exposed to Ramadan for a full month2. All panels were linked to a subset containing information on an individual’s religion and self-reported religiosity. For the purpose of this research some participants were excluded if their information was incomplete, for example those who do not know their date of birth or refused to answer which religion they ascribe themselves to.

Three subpanels are analyzed in this paper: Pregnancy, employment and cognitive outcomes. Each panel we looked at age, gender ratio and self-reported religiosity and panel specific outcomes.

1

The data can be downloaded from this website. More information on can be found and the user guides which contain detailed information: https://www.rand.org/labor/FLS/IFLS/study.html

2

Apart from the pregnancy panel, the literature-based exposure variable is used here. More elaborate description of treatment variables used can be found later in the thesis (in section 3.2).

8

2.2.1 Pregnancy Panel

The IFLS survey contains a huge amount of data on individuals’ wellbeing, their experience with healthcare and also describes the communal health services available to them. In book 4 of the IFLS survey, information on all pregnancies and was administered to all women from 15-58 years old. The child’s date of birth was collected and with that exposure to Ramadan could be calculated. Individuals with incomplete information or pregnancies ending in a miscarriage (due to incomplete information) have been excluded. For this panel pregnancy duration was measured, with this information the exact duration of pregnancy was known with which individualized treatment variables were created. With that information two variables were looked at: weight at birth and relative size. Both are subject to bias by faulty recollection as they are asked to parents and not verified by birth documents. Parents were asked what was their baby’s weight at birth and the answer is noted in kilograms. The additional question was asked whether the babies were weighted at birth, those who did not (151 observations) were excluded from this measurement. The question being asked for relative size was ‘what was the relative size of your baby at birth’. With the answers being: 1-Much smaller (0.67%), 2-Smaller (57.9%), 3-Similar (57.9%), 4- Bigger (25.88%), and 5- Much bigger (1.75%). The summary statistics on this panel show that exposed Muslims are slightly more religious, we do not observe that difference for non-Muslims which could influence the analysis. Not exposed Muslims are younger than exposed Muslims, the same is observed for non-Muslims. To account for age differences, age and its quadratic term (expressed in days) are added as controls in the regression.

2.2.2 Employment Outcomes Panel

In this panel long-term employment outcomes are looked at. Both dependent variables log hours per week and self-employed are analysed in earlier research (Majid, 2015). Log hours per week contains the normal amount hours individuals work per week in their primary job. Majid (2015) included Self-Employed in his analysis with the assumption that in the context of the study being self-employed is not something positive. In the low-income and education setting of the study if someone is self-employed it often means a more unreliable source of income. A downside is that also more successful entrepreneurs fall under this category. If an individual answered the question ‘In the last five years, have you been working for a salary’ with a no, the person is considered self-employed and the binary variable gets the value of 1. The summary statistics on this panel show that exposed Muslims are slightly older, the same situation is observed for non-Muslims. Not exposed Muslims are slightly

9

more likely to be self-employed which is not observed for non-Muslims. To account for age differences, child’s age and its quadratic term (expressed in days) are added as controls in the regression.

Table 1 Summary Statistics: Pregnancy & Employment outcomes panel

Muslims Non-Muslims

Exposed Not exposed Exposed Not exposed

Mean N Mean N T-Test Mean N Mean N T-Test

Panel A: Pregnancy Female 0.49 6729 0.49 3570 0.001 0.48 849 0.45 435 -0.06 (0.50) (0.50) (0.01) (0.50) (0.50) (0.06) Age child 8.01 6729 6.11 3570 -1.91 8.26 849 6.36 435 -1.90 (8.31) (5.36) (0.14) (8.31) (5.46) (0.39) Religiosity 2.91 6725 2.88 3568 -0.03 1.87 845 1.87 433 -0.008 (0.67) (0.66) (0.014) (0.63) (0.58) (0.04) Relative size 3.14 3605 3.13 2097 -0.01 3.20 437 3.17 227 -0.03 (0.69) (0.69) (0.02) (0.65) (0.60) (0.05) Birth weight 3.13 3510 3.13 2041 -0.004 3.30 400 3.31 203 0.01 (0.62) (0.66) (0.02) (0.71) (0.87) (0.07)

Panel B: Employment outcomes

Female 0.50 15438 0.49 7189 -0.01 0.50 1729 0.48 851 -0.02 (0.50) (0.50) (0.01) (0.50) (0.50) (0.02) Age 37.45 15438 36.83 7189 -0.63 38.45 1729 37.56 851 -0.89 (12.17) (11.45) (0.17) (12.32) (12.21) (0.51) Religiosity 2.87 15429 2.86 7186 -0.01 3.14 1720 3.14 848 0.01 (0.69) (0.69) (0.01) (0.66) (0.68) (0.03) Self-employed 0.51 15001 0.53 6983 0.02 0.50 1692 0.51 838 0.01 (0.50) (0.50) (0.01) (0.50) (0.50) (0.02) Log hours 3.66 3886 3.66 1792 -0.01 3.63 408 3.74 216 0.01 (0.77) (0.76) (0.02) (0.82) (0.65) (0.06)

Mean of each variable with the standard deviation in parentheses. T-scores are noted with their standard errors.

10

2.2.3 Cognitive Panel

The fourth wave of IFLS added an entire section on cognition which was built upon in IFLS5. FOH suggests that during in utero exposure fetal cognitive development can be hampered. The newly added w-score is an adaptive number series test specifically designed for low-income and education environment appropriate for the context of Indonesia. Additionally, it does not depend on the local language used and therefore is not dependent on belonging to a certain culture/ group. The test measures fluid intelligence and is comparable to the Raven’s test previously used in IFLS waves. Raven’s test is referred to as the best test available to measure general intelligence factor “g” (Majid, 2015). The test is divided in different blocks and stages. The first three questions are answered by all participants and depending on how many they got right, they are assigned to the next block that vary by difficulty. Each individual answers 6 questions out of 15 and the average of all questions is how the W-score is calculated, binary indicators were used to categorized a questions as answered correctly or not. More details on the number series test and the pre-testing stages in Mexico and Indonesia can be found in the IFLS 5 handbook.3 In table 2 we observe that both for Muslims and non-Muslims the ages are different. Again in order to account for age differences, age and its quadratic term (expressed in days) are added as controls in the regression.

3

https://www.rand.org/labor/FLS/IFLS/download.html On this website the IFLS data is available as well as two handbooks where more detailed information about the adaptive number series tests can be found.

11

Table 2 Summary Statistics: Cognitive Panel

Muslims Non-Muslims

Exposed Not exposed Muslims Exposed Not exposed Non-Muslims

Mean N Mean N T-Test Mean N Mean N T-Test

Panel C: Cognitive Female 0.52 23264 0.51 9610 -0.004 0.46 5348 0.46 2403 0.003 (0.50) (0.50) 0.006 (0.50) (0.50) (0.01) Age 37.72 23264 39.77 9610 2.05 41.10 5348 45.54 2403 4.43 (15.39) (17.31) (0.20) (17.86) (20.71) (0.49) Religiosity 2.88 20094 2.87 8151 -0.01 3.14 2213 3.15 958 0.14 (0.68) (0.68) (0.01) (0.66) (0.66) (0.03) W-score 508.11 20024 507.91 8116 -0.20 517.10 2201 517.36 952 0.25 (71.15) (71.80) (0.94) (71.40) (70.75) (2.75)

Mean of each variable with the standard deviation in parentheses. T-scores are noted with their standard errors.

3. Empirical Methodology

The third chapter starts out with describing the four different treatment variables used in this research, the overall exposure variables as well as the trimester and religiosity effects. After that the experimental design and procedures as well as its limitations are discussed.

3.1 Treatment variables: Exposure to Nutrition Shocks in Utero

Someone is classified as exposed if the two criteria are met: Firstly, identifying yourself as a Muslim as well as having their date of birth (DOB) suggesting that their gestation period overlapped with Ramadan month of that year. If the corresponding year’s Ramadan start and end dates are in between the individuals’ estimated date of conception (DOC) and DOB, someone is classified as exposed. Historical Ramadan dates are been retrieved from multiple websites to ensure that the dates are correct.4 Apart from the pregnancy panel, the pregnancy duration is an estimation and based on the average of 280

4

Historical Ramadan data is retrieved from: https://calendar.zoznam.sk/islamic_calendar-en.php?ly=2000 and

12

days after onset of last menstrual period while only 4% of pregnancies actually end on this date and 70% deliver within 10 days of their estimated due date (even if estimation is by ultrasound) (Jukic et al., 2013). Variation in pregnancy duration is problematic since individuals might be considered exposed to Ramadan when they’re actually not and vice versa, especially when trying to disentangle trimester effects. The variability of duration in pregnancy is also prevalent in the IFLS with a mean of 36.5 weeks and a standard deviation of 0.23. The different pregnancy durations, divided per religion and exposure, can be found in the appendix 7.2. In total 4 treatment exposure variables are used in this thesis: Pregnancy panel, literature-based, narrow and broad.

For the pregnancy panel duration of the pregnancy is known and with that a more precise DOC can be estimated. Which is done simply by subtracting the pregnancy duration from their DOB. Duration of pregnancy was answered both in months and weeks and was recoded to create a uniform unit of measurement.5 The different pregnancy durations, divided per religion and exposure, can be found in the appendix 7.2.

Majid (2015) and van Ewijk (2011) create the treatment exposure variable based on 266 days gestation period to derive the expected date of conception (Jukic et al., 2013). To compare results from earlier research to the most recent data available the exposure variable set by Majid and van Ewijk is included in this paper also. Majid (2015) and van Ewijk (2014) control for individuals born within 3 weeks of Ramadan which is intentionally large and controls for pregnancies up to 43 weeks. Also it keeps the individuals separate who are conceived during the celebrations after Ramadan. The first trimester does not exclude those exposed in the first 30 days of pregnancy to ensure that individuals are exposed to an entire month of Ramadan. Excluding a disproportionate amount of individuals is not necessary and may even affect comparability. A more serious bias would originate from individuals who are born having a pregnancy of shorter than 40 weeks, as a shorter pregnancy might even be a result of exposure to fasting. No solution to this problem has been suggested before. It could incorrectly classify someone as exposed while they are not leading to a lower exposure coefficient.

Based on the average duration of pregnancy found in this dataset, a new identification of Ramadan effect was added. A more narrow gestation period is used to allocate treatment effect based on the average pregnancy duration given in the Pregnancy panel, 256 days. There

5

It was checked and fixed wherever the duration pregnancy was incorrectly classified as months. E.g. Some pregnancies were suggested to have 40 months, here it is obvious that pregnancy duration was meant to be in weeks.

13

are two ways of measuring duration pregnancy: starting from first day of the last period or conception day. Conception day is used by Majid and van Ewijk and lasts two weeks shorter than starting from first day of last period. The IFLS handbook does not specify whether they asked for duration pregnancy based on ovulation or based on the last period. It seems more likely that the average duration pregnancy in IFLS survey was taken from the first day of the last period since that is the most reliable and commonly used method to estimate gestational age. 6 This leads to a gestation period which is on average two weeks shorter, therefore the narrow gestation period is estimated to be 256 days. This could lead to people erroneously being classified as not exposed. To compare results we add a fourth treatment exposure variable with a 280 days gestation period. This broad exposure variable most likely erroneously classifies people as exposed while they are not, resulting in a lower Ramadan exposure coefficient. For all exposure variables only individuals who are exposed to a full month of Ramadan are considered.

3.1.1 Effects per Trimester

The effects of exposure to Ramadan can differ depending on the gestational phase of the pregnancy. It is possible in the first trimester that women are not yet aware of being pregnant, as well as experiencing symptoms which could result in a decrease in appetite make it more likely that fasting would take place more in those days (van Bilsen 2016). Furthermore, more women indicated in their study while in a later trimester that it is more difficult to fast. This is the most given reason to abstain from fasting so, it could well be that there are simply less women fasting (or at least women are fasting less days) in the second and third trimester compared to the first. These concerns are not possible to solve as we do not have the information on whether individuals actually fasted, and if so how many days, hopefully in later research it can be looked further into.

Van Ewijk (2011) divided the treatment in 5 different categories: Conceived in Ramadan, 1st trimester, 2nd trimester, 3rd trimester and born in Ramadan. While Majid (2015) divided it in only the 3 trimesters: first trimester (1-89 days gestation), second trimester (90-178 days gestation) and the third trimester (179-266). In this paper we divide trimester effects in three categories. For the four different exposure variables the trimester effects are all slightly different. For the pregnancy panel, the trimester exposure variables are more reliable as the correct pregnancy duration is known. The first trimester is expected to start at the first day of the last period and therefore last 89 days. The broad exposure variable also assumes

6

14

that the first trimester lasts 89 days. Both the literature-based and the narrow exposure variable use a shorter trimester with 65 and 75 days respectively, both starting from day of conception.7 For all exposure variables but the literature-based one, allocation to trimester is strict: An individual is only considered to be exposed to Ramadan in the first trimester if the entire month of Ramadan took place while they were in their first trimester, this excludes individuals who are exposed to Ramadan in first as well as the second trimester. This will result in an underestimation in trimester exposure effects and will not include all individuals who are exposed to Ramadan.

The literature-based exposure trimester effects consider an individual exposed if Ramadan took place during their first trimester, either partially or fully. If Ramadan started in the first trimester an individual is considered exposed to Ramadan in the first trimester, regardless if they are exposed also in the second trimester. An individual can only be classified as exposed to one trimester. The span of the first and third trimester literature-based exposure variables are shorter and therefore might not include all individuals who are exposed. 8

3.1.2. Self-Reported Religiosity

One of the biggest limitations in this field of research is that there is no exact information on women’s fasting behaviour during Ramadan, treatment is allocated only based on the their religion and date of birth. According to van Bilsen et al. (2016), around 10-30% of pregnant women do not fast during Ramadan in Indonesia. Majid (2015) used self-reported religiosity as an indication of adherence to fasting in order reduce noise while estimating treatment allocation. The variable has been included in IFLS4 and the individuals were asked how religious they consider themselves to be with four possible answers: “not religious”, “somewhat religious”, “religious” and “very religious”. Higher self-reported religiosity is correlated with objective measurements such as frequency of prayers and participating in communal religious activities Gaduh (2011). Self-reported religiosity is used to show differential effects for individuals who are most likely to have fasted.

7

The literature-based first trimester exposure variable wants individuals who are exposed for an entire month to Ramadan and therefore is only 65 days long.

8

The reasons why Majid (2015) and van Ewijk (2014) exclude these individuals to control for individuals born within 3 weeks of Ramadan. The first trimester does not exclude those exposed in the 30 days since to ensure that individuals are exposed to an entire month of Ramadan

15

3.2 Experimental Design and Procedures

Since different treatment variables and panels are looked at, more than one econometric equation is used. For all models robust standard errors based on household id are included so that unobserved family and community variables will stay constant over time (van Ewijk, 2016).

𝑌_𝑖 =∝₁+ 𝛽₁ 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒_𝑥_𝑖+ 𝜕₁ 𝑋_𝑖+ 𝛾₁𝑀𝐹𝐸_𝑖+ 𝜀_𝑖 (1)

In the first model the dependent variable Yi is a continuous variable containing information on individual’s i outcome we want to measure. The ∝₁ depicts the constant term in this model and 𝛽₁ 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒_𝑥_𝑖 is the dummy variable 1 if exposed to in utero Ramadan fasting and 0 if it is not. The error term 𝜀_𝑖 includes random error. The model includes control variables in order to limit potential omitted variable bias which otherwise might be included in the error term. The variable month fixed effects ‘ 𝛾₁𝑀𝐹𝐸_𝑖’ is included in the model for all panels since it controls for the effect of parents influencing when in the year they will be pregnant (Majid, 2015). The calendar month fixed effects are 12 dummies for each month. ′𝜕₂ 𝑋_𝑖’ are the context specific additional control variables used which limits any potential bias that may influence the real effect of the exposure coefficient. For each subset the control variables used are specified underneath the tables. The exposure variable 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒_𝑥_𝑖 can be (a) based on the self-reported pregnancy duration, for the pregnancy panel. (b) the broad gestation period exposure variable, (c) the narrow gestation period, and (d) the exposure variable based on related literature.

𝑌_𝑖 =∝₂+ 𝛽₂ 𝑡𝑟𝑖𝑚𝑒𝑠𝑡𝑒𝑟_𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒_𝑖 + 𝜕₂ 𝑋_𝑖+ 𝛾₂𝑀𝐹𝐸_𝑖 + 𝜀_𝑖 (2)

When trimester exposure coefficients are estimated the second model is used. 𝛽₁ 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒_𝑖 would then be replaced with a trimester exposure dummy 𝛽₂ 𝑡𝑟𝑖𝑚𝑒𝑠𝑡𝑒𝑟_𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒_𝑖, a separate regression is run for each trimester and only on Muslim population. The self-reported religiosity coefficients only considers on the Muslim population as well.

The exposure coefficient depicts an underestimation of the true effect of exposure to Ramadan due to a variety of reasons. First of all, the average duration of pregnancy is used to assign treatment status (excl. pregnancy panel). There is a significant variability of pregnancy duration, also specifically in this dataset, leading to erroneous allocation to treatment. Specifically for the literature-based and narrow treatment allocation, those who are exposed to

16

Ramadan could be allocated as not exposed leading in an underestimation of the effect on cognitive and employment outcomes.

Secondly, we assume that when people are exposed to Ramadan in utero and they identify themselves as Muslim, we allocate them to the treated group. There is no information available on if they actually fasted, and if yes, how many days. A third downfall is due to the assumption that religion (as well as self-ascribed religiosity), is based on the participants themselves and not their parents. This information could change over time and leading to misclassification and lower estimates (Almond & Mazumder, 2011). Even though exposure to Ramadan is considered a mild shock in nutrition in utero, it could still affect mortality rate both during pregnancy and after. The estimates which are used are only gathered from those who have survived resulting in a more healthy population overall and making it more difficult to observe FOH.

4. Results

The results chapter starts out with the results found using the pregnancy panel. After that the employment and later the cognitive outcomes are analysed.

4.1 Pregnancy Panel

Table 3 shows that exposure to Ramadan fasting in utero has a small but significant effect on relative size for Muslims and not for non-Muslims. When unpacking that for trimester and religiosity effects, we see that for very religious individuals there is a significant difference at 1% level that relative size of a new-born is 0.32 smaller than those who are not exposed (Relative size is measured on a scale of 1 to 5).

In table 4 the trimester effects of individuals who classified themselves as very religious are unpacked. Table 4 shows suggestive evidence that in the third trimester Muslims who are exposed are rated 0.24 smaller at birth than Muslims who are not exposed at a 5% significance level. For baby weight there are no observed significant differences. It does not immediately imply that exposure to Ramadan does not affect birth weight, it could be due to underestimation of the coefficient or due to recollection bias of parents.

For this analysis only data entries on pregnancies which ended in a live birth were included. Unfortunately, those pregnancies ending in a miscarriage or still birth did not have sufficient observations and/or complete ‘date of birth’ in order to determine whether they have been exposed to Ramadan.

17

Table 3 Pregnancy Panel Results

(1) (2)

Relative Size Baby's weight in kg’s Muslims Exposed -0.06* (-0.03) 0.03 (-0.03) Observations 5701 5,550 Non-Muslims Exposed -0.15 (-0.10) 0.10 (-0.09) Observations 661 600 Trimester effects Exposed 1st trimester -0.02 (-0.03) 0.00 (-0.03) Exposed 2nd trimester 0.04 (-0.03) -0.05 (-0.03) Exposed 3rd trimester -0.05 (-0.04) 0.04 (-0.03) Observations 5701 5548 Religiosity Not religious -0.05 (-0.28) -0.05 (-0.20) Observations 157 149 Rather religious -0.02 (-0.07) 0.03 (-0.06) Observations 1288 1246 Somewhat religious -0.01 (-0.04) 0.01 (-0.04) Observations 3505 3426 Very religious -0.32*** (-0.10) 0.12 (-0.09) Observations 749 727

Control variables used: Birth-month fixed effects, multiples at birth, age and age2. Robust standard errors are clustered at household level in parentheses. Exposure dummy is 1 if an

individual is exposed to full month of Ramadan. *** p<0.01, ** p<0.05, * p<0.1

18

Table 4 Very Religious Muslims & Trimester Exposure Results

Exposed 1st trimester Exposed 2nd trimester Exposed 3rd trimester Observations Relative Size -0.095 0.104 -0.242** 749 (0.103) (0.101) (0.112)

Control variables used: Birth-month fixed effects, multiples at birth, age and age2. Robust standard errors are clustered at household level in parentheses. Exposure dummy is 1 if an

individual is exposed to full month of Ramadan. *** p<0.01, ** p<0.05, * p<0.1

4.2 Employment Outcomes Panel

Table 5 includes regressions on self-employment and log hours with three different treatment exposure variables (Further details on each can be found earlier in section 3.1.) The assumption is that exposure to Ramadan leaves individuals worse off and therefore they are more likely to be self-employed as in developing countries context that is considered a negative outcome. For overall exposure to Ramadan it seems that people are less likely to be self-employed, but that significant difference disappears after controlling for age. Meaning that any decreased likelihood in self-employment could be attributed to the difference in age between exposed and non-exposed Muslims. In earlier research, Muslims exposed to Ramadan were 3% more likely to be self-employed, that effect disappeared after using for household fixed effects though. (Majid, 2015).

Using the narrow exposure variable, non-Muslims who are exposed to Ramadan work less hours. Before controlling for age, the narrow treatment exposure was not significant. Across all three treatment exposure variables it can be seen that for very religious Muslims, those who are exposed to Ramadan work more hours per week. Additional regressions are run and show that very religious non-Muslims are not negatively impacted by exposure to Ramadan (is not included in the table). This implies that only very religious Muslims who are exposed to Ramadan work 0.13 more hours per week. No explanation for this has been found yet.

19

Table 5 Long-Term Employment Results

(1) Self-Employment (2) Log hours

Broad Narrow Literature Broad Narrow Literature Muslims Exposed -0.01 -0.01 -0.01 0.01 0.02 0.04 (0.01) (0.01) (0.01) (0.02) (0.02) (0.03) Observations 21,977 21,977 19,371 5,675 5,675 4,962 Non-Muslims Exposed -0.00 0.00 0.00 -0.01 -0.13* -0.12 (0.02) (0.02) (0.02) (0.07) (0.07) (0.07) Observations 2,523 2,523 2,216 621 621 537 Trimester effects Exposed 1st trimester -0.00 -0.01 -0.01 -0.01 0.00 0.02 (0.01) (0.01) (0.01) (0.03) (0.03) (0.03) Exposed 2nd trimester 0.00 0.01 0.01 -0.04 -0.01 -0.02 (0.01) (0.01) (0.01) (0.03) (0.03) (0.02) Exposed 3rd trimester -0.01 -0.01 -0.01 0.05* 0.03 0.05 (0.01) (0.01) (0.01) (-0.03) (0.03) (0.03) Observations 21,977 21,977 21,977 5,675 5,675 5,675 Religiosity Not religious 0.00 -0.03 -0.02 -0.15 -0.14 0.01 (0.04) (-0.04) (0.04) (-0.13) (0.12) (-0.16) Observations 674 674 589 208 208 181 Rather religious -0.01 -0.02 -0.02 -0.02 -0.01 -0.04 (0.02) (-0.01) (0.02) (0.04) (0.04) (-0.05) Observations 4,915 4,915 4,287 1,573 1,573 1,364 Somewhat religious -0.01 -0.01 -0.01 0.01 0.02 0.04 (0.01) (-0.01) (0.01) (0.03) (0.03) (-0.03) Observations 13,156 13,156 11,612 3,212 3,212 2,813 Very religious 0.00 0.00 0.01 0.127* 0.13* 0.131* (0.02) (0.02) (0.02) (0.07) (0.07) (0.08) Observations 3,232 3,232 2,883 682 682 604

Control variables used: Birth-month fixed effects, age and age2. For literature exposure does not include those conceived less than three weeks after the end of Ramadan. Robust standard errors are clustered at household level in parentheses. Exposure dummy is 1 if an individual is

exposed to full month of Ramadan. *** p<0.01, ** p<0.05, * p<0.1

20

4.3 Cognitive Panel

Table 6 shows the effect of the three Ramadan exposure variables on w-score. The overall effect of Ramadan exposure on Muslims seems to be significantly negative around -1.85 points. As a reference while interpreting the results the average w-score is 508.05, with a minimum of 299 and a maximum of 635. For both the broad and the narrow duration treatment variable there is a significant negative scoring. For the literature-based treatment variable the p-value is 0.1 percentage point away from being significant at 10% and seems likely to be negative as well. For the narrow treatment variable it seems that non-Muslims who are exposed to Ramadan are also significantly affected with having 5.02 points lower than those who are not exposed. Since the other two treatment variables do not observe this, it could be suggestive evidence that the pregnancies excluded in the narrow treatment variable are wrongly allocated. In column 2, the narrow treatment variable when divided per religiosity, non-religious and rather religious who are exposed to Ramadan seem to have a lower W-score. Across all three exposure variables the very religious Muslims have significantly lower w-scores as hypothesized. In table 7 the trimester effects for very religious Muslims are unpacked. There we see that for all three treatment variables third trimester exposure to Ramadan has significant negative effect on w-scores.

Table 6 Cognitive Panel Results

W-Score

(1) Broad (2) Narrow (3) Literature Muslims Exposed -1.90** (0.87) -1.85** (0.83) -1.43 (-0.92) Observations 28,133 28,133 28,133 Non-Muslims Exposed -2.81 (-2.55) -5.02** (2.43) -3.39 (-2.73) Observations 3,150 3,150 3,150 Trimester effects Exposed 1st trimester -0.44 (-1.02) -0.63 (1.03) -0.64 (-0.98) Exposed 2nd trimester -0.42 (-1.01) -1.38 (-1.00) -1.20 (-0.91) Exposed 3rd trimester -0.31 (-0.93) -0.73 (-1.08) -0.01 (-1.08) Observations 28,133 28,133 28,133 Religiosity

21 Not religious -8.33 (-5.82) -11.13* (-5.45) -6.85 (6.20) Observations 791 791 729 Rather religious -2.44 (-1.65) -2.96* (-1.57) -2.27 (-1.78) Observations 6,078 6,078 5,604 Somewhat religious -0.16 (-1.13) -0.41 (-1.09) -0.5 (-1.22) Observations 17,009 17,009 16,592 Very religious -6.86*** (-2.25) -4.34** (-2.12) -3.72* (-2.06) Observations 4,207 4,207 4,513

Control variables used: Birth-month fixed effects, age and age2. For literature exposure does not include those conceived less than three weeks after the end of Ramadan. Robust standard errors are clustered at household level in parentheses. Exposure dummy is 1 if an individual is

exposed to full month of Ramadan. *** p<0.01, ** p<0.05, * p<0.1

Table 7 Cognitive Panel: Very Religious Muslims & Trimester Exposure Results

W-Score

VARIABLES (1) Normal (2) Narrow (3) Literature

Exposed 1st trimester 2.17 2.68 -2.61 (2.57) (2.70) (2.75) Exposed 2nd trimester -2.66 -3.78 -3.40 (2.34) (2.49) (2.60) Exposed 3rd trimester -5.26* -6.85** -4.11* (2.84) (2.82) (2.43) Observations 4,207 4,207 4,207

Control variables used: Birth-month fixed effects, age and age2. The Literature exposure variable does not include those conceived less than three weeks after the end of Ramadan. Robust standard errors are clustered at household level in parentheses. Exposure dummy is 1

if an individual is exposed to full month of Ramadan. *** p<0.01, ** p<0.05, * p<0.1

22

5. Discussion

The natural experiment occurring during Ramadan fasting by pregnant Indonesian women is used to study short-term and lifelong effects. This paper provides suggestive evidence that mild nutrition shocks are measurable already immediately after birth. The pregnancy panel contains the self-reported duration of pregnancy and that is used to determine the Ramadan exposure variable for each individual. Muslims exposed to Ramadan seem to have a lower relative size of 0.06. After unpacking this for religiosity effects, very religious individuals there is a significant difference at 1% that relative size of a new born is 0.32 smaller than those who are not exposed. Looking at trimester effects, it can be observed that Muslims exposed in third trimester are 0.24 smaller (at 5% significance level). No suggestive evidence is found that the weight at birth is negatively affected by exposure to Ramadan. This does not immediately imply that exposure to Ramadan does not affect birth weight, it could be due to underestimation of the coefficient or due to recollection bias of parents. This provides suggestive evidence that the Fetal Origins Hypothesis (FOH) holds also in this context and could be a potential channel for the lower w-scores and employment outcomes. For cognitive outcomes a significant negative effect of Muslims exposed to Ramadan has been found. This seems to influence mostly very religious Muslims and when those individuals are exposed during their third trimester in-utero. Again supporting the assumption that self-reported religiosity is a good indicator of adherence to fasting. These observations are found across all three treatment exposure variables, which makes the conclusion seem more robust.

In this research, there is no evidence suggesting that exposure to in- utero fasting makes an individual more likely to be self-employed. This is in itself not concerning since this variable both encompasses poorer self-employed individuals as well as more wealthy and successful ones. Further research could look into how valid an indicator self-employment is for the purpose of this study. A different story is observed for hours worked: Muslims who are exposed to Ramadan work around 0.13 less hours per week than those who are not exposed. Strangely, very religious Muslims who are exposed to Ramadan work around 0.13 more hours per week than those who are not exposed. No explanation for this has been found yet.

For the narrow treatment variable, some surprising outcomes have been found. Non-Muslims’ w-scores are negatively affected when exposed to Ramadan. Also there is suggestive evidence that non-Muslims exposed to Ramadan work less hours than those who are not exposed. This could be attributed to the treatment exposure estimation technique, since the negative effect is not observed for the broad and literature-based treatment variables. It

23

was expected that pregnancies lasting longer than average would be allocated to the non-exposed category erroneously. However, that does not explain why this negative effect is observed for non-Muslims. Besides the estimation technique, it could be that non-Muslims are negatively affected by being exposed to Ramadan. This could be explained by the decreased access to healthcare during Ramadan fasting and the festivities afterwards. The decrease in access to services could be an explanation why also non-Muslim people are affected by the Ramadan. Further research could look into whether people born during Ramadan are negatively affected, in the appendix 7.3 one can find an initial look at how the dependent variables are influenced for the individuals born during Ramadan.

The narrow and broad exposure variables have their limitations but in combination with the literature-based variable has enabled a more thorough look into how delicate the allocation process is and how it might impact the observed results. It has also provided a more robust conclusion whenever all exposure variables show the same result. The self-reported pregnancy duration is a useful piece of data which should be taken into consideration when looking at the IFLS dataset in this context again.

This study has found the most impact of third trimester exposure on the dependent variables. The rigorousness of the trimester effects could partially explain why in this study there is no first trimester negative effects found, as hypothesized in literature and similar studies. Majid (2015) has found indicative evidence the employment outcomes are negatively affected by exposure to Ramadan. In this thesis, similar observations have not been found, which might be attributed to a change within the population over the years. So far I have not been able to find a satisfactory explanation. Further research, could verify my findings and look into what has caused this dissimilarity. Van Ewijk (2016) observed a significant less amount of males for those exposed to Ramadan, which is in congruence with literature. In this dataset no significant difference was observed in gender ratio. A further look into whether the gender ratio is exactly as expected by medical literature would be recommended for future studies, e.g. usually slightly more males are born than females.

Limitations of this study have been mentioned throughout the paper, some will be highlighted here in brief. Despite using different exposure variables, the pregnancy duration is still an unknown factor for the employment and cognitive panels. Leading to erroneous treatment allocation and potentially bias the exposure coefficient downwards.

24

Is Ramadan fasting really driving results found in this and other papers? It is possible that circumstances surrounding the month of Ramadan are being measured. This is a serious concern in this field of research as we have no information on actual fasting behaviour. Almond & Mazumder (2011) also indicate that there might be some other behavioural changes associated with Ramadan such as different sleeping pattern and dehydration- in this research it is also impossible to disentangle such effects without having more information available. Also both people who fast and those who do not, are stressed about possible negative health effects of fasting on both baby and mother, which might influence their pregnancy and later child outcomes negatively (van Bilsen et al.,2016 ).

In future research it would be interesting to link community health practices to Ramadan exposure. Since IFLS has earlier waves, the panel data could even be used to link quality of health care access at the time the individual was exposed. The information received during pregnancy has a big influence with regards to fasting behaviour (Lou and Hammoud, 2016). Additionally, a different exposure variable could be created for individuals potentially exposed to 37-40 weeks fasting , which is the pregnancy duration for around 18-21% individuals exposed to Ramadan. Besides the panels used, the IFLS dataset contains a lot more interesting variables to analyse, such as children’s testing scores and type of employment.

As supported by the results found in this paper, in utero exposure to Ramadan fasting have short- and long-term negative consequences. When one looks into the reasons why women decide to fast, it is clear that they are not aware of the legal exemption of them fasting or of the negative consequences if they do decide to fast. This issue seems to have a relatively easy solution: To spread the information and awareness of the consequences of maternal fasting. Granted, it will take time and delicate method of communication, but parents’ decision to fast is something which could be influenced, unlike more severe nutrition shocks such as the Dutch Famine. Besides midwives and doctors, husbands and religious leaders have a role to play here as they are also a main influence on why women decide to fast. Religious leaders could inform and promote delayed fasting, as is allowed by Islamic law. Already existing channels could be utilized to share this information, one example being WHO ‘Safe Motherhood Program’ in Indonesia. An increased access to highly skilled midwives could lead to more informed individuals and hopefully better health choices. A collectivist holistic

25

approach could make all parties better off and offset adverse health and socio-economic effects of in-utero nutritional shocks during Ramadan and beyond.

6. References

Almond, D. (2006). Is the 1918 Influenza Pandemic Over? Long‐Term Effects of In Utero Influenza Exposure in the Post‐1940 U.S. Population. Journal of Political Economy, 114(4), 672-712. doi:10.1086/507154

Almond, D., & Currie, J. (2011). Killing Me Softly: The Fetal Origins Hypothesis. The Journal of Economic Perspectives : A Journal of the American Economic Association, 25(3), 153–172. http://doi.org/10.1257/jep.25.3.153

Almond, D., & Mazumder, B. (2011). "Health Capital and the Prenatal Environment: The Effect of Ramadan Observance during Pregnancy." American Economic Journal: Applied Economics, 3 (4): 56-85. DOI: 10.1257/app.3.4.56

BARKER, D. J. (1995). The fetal and infant origins of disease. European Journal of Clinical Investigation, 25: 457-463. doi:10.1111/j.1365-2362.1995.tb01730.x

van Bilsen, L.A., Savitri, A.I., Dwirani, A., Baharuddin,M., Diederick, E. Grobbee, C., Uiterwaal, S.P.M.. (2016). Predictors of Ramadan fasting during pregnancy, Journal of

Epidemiology and Global Health. Volume 6, Issue 4, Pages 267-275,

https://doi.org/10.1016/j.jegh.2016.06.002.

Calender.sk. Historical Ramadan dates. Retrieved on 15.07.2018. Retrieved via: https://calendar.zoznam.sk/islamic_calendar-en.php?ly=2000

Chen X. (2014). Fetus fasting, and festival: the persistent effects of in utero social shocks. Int J Health Policy Manag 2014; 3: 165–169. doi: 10.15171/ijhpm.2014.92

Dikensoy, E., Balat, O., Cebesoy, B., Ozkur, A., Cicek, H., Can, G. (2009). The effect of Ramadan fasting on maternal serum lipids, cortisol levels and fetal development. Arch Gynecol Obstet. 279(2):119-23.

van Ewijk, R., (2009). Long-Term Health Effects on the Next Generation of Ramadan Fasting During Pregnancy. CEP Discussion Paper No. 926. http://dx.doi.org/10.2139/ssrn.1402632

26

Gaduh, Arya (2011). Religion and Cooperative Attitudes: Evidence From Indonesia. Working Paper. University of California.

Joosoph, J., Abu, J., Yu, S. (2005). A survey of fasting during pregnancy. Singapore medical journal. 45. 583-6. https://www.ncbi.nlm.nih.gov/pubmed/15568120

Jukic, A.M., Baird, D.D., Weinberg, C.R., Mc Connaughey, D.R., Wilcox, A.J. (2013). Length of human pregnancy and contributors to its natural variation. Hum Report. 28(10): 2848–2855. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3777570/

Lou, A., & Hammoud, M. (2016). Muslim patients’ expectations and attitudes about Ramadan fasting during pregnancy. International Journal of Gynecology & Obstetrics, 132: 321-324. doi:10.1016/j.ijgo.2015.07.028

Mirghani, H.M., Weerasinghe, S., Al-Awar, S., Abdulla, L., Ezimokhai, M.J. Perinatol. (2005). The effect of intermittent maternal fasting on computerized fetal heart tracing. 25(2):90-2. https://www.ncbi.nlm.nih.gov/pubmed/15526011/

Muhammad F.M. (2015). The persistent effects of in utero nutrition shocks over the life cycle: Evidence from Ramadan fasting, Journal of Development Economics, Volume 117, 2015, Pages 48-57, ISSN 0304-3878, https://doi.org/10.1016/j.jdeveco.2015.06.006.

Pradhan, M. P., Suryadarma, D., Beatty, A., Wong, M., Alishjabana, A., Gaduh, A., & Artha, R. (2011). Improving educational quality through enhancing community participation: Results from a randomized field experiment in Indonesia. (Policy Research Working Papaer Series; No. 5795). New York: The World Bank.

Ravelli, A.C., van der Meulen J.H., Michels, R.P., Osmond, C., Barker, D.J., Hales, C.N., Bleker, O.P. (1998). Glucose tolerance in adults after prenatal exposure to famine. 351(9097):173-7.

Raisler, J., & Kenned, H. (2005). Midwifery care of poor and vulnerable women, 1925–2003, Journal of Midwifery & Women's Health, Volume 50, Issue 2,

27

7. Appendices

7.1 Reasons for Fasting

It seems surprising that around 70-90% of pregnant women still observe the fast even though they are legally exempt from fasting doing so (van Ewijk, 2011). The need to disentangle the effects that the surroundings have on in utero nutrition choices is vital to understanding and designing effective policy measures (Majid, 2015). The most important paper’s views on in utero nutrition decision making are aggregated here. The main reasons can be roughly divided in subcategories: Religious, costs of delay, social, health and other.

1. Religion- misinformation around Ramadan Fatwa

Some Muslims might not be aware on the exemption of pregnant women fasting during Ramadan. Others might believe that fasting does not do harm, even during pregnancy, or can even be beneficial since it is Gods wish that they do so (Joosoph et al., 2004). Robinson and Raisler (2005) highlight that some Muslims consider it a sin to not fast during Ramadan. Van Ewijk (2015) discusses the ambiguous interpretation of the Fatwa: Some interpret it as all pregnant women have to fast unless they have significant health problems, and the other interpret is as all pregnant women are exempted from fasting. In van Bilsen’s et al. (2016) study none of the women who did not fast gave as a reason that ‘Because according to Islam, pregnant women do not have to fast’. When women did fast, 26.2% indicated that their reason to fast was because ‘According to Islam, pregnant women are obliged to fast’. Some women are aware that they do not have to fast but decide to fast anyway (61.5%). In conclusion, it seems that there is enough indication to think that lack of awareness and misinformation surrounding Ramadan fasting plays a role in deciding to fast or not.

2. Costs of delay

Usually, fasting cannot be skipped completely and has to be done later. This delay does come with some negatives: It results in the fact that these women now have to fast by themselves therefore being (fully) excluded from the communal activities surrounding Ramadan. A delay in fasting often requires a donation to charity or the poor to be made, these financial costs could be especially cumbersome for people with a lower income. Delaying the fast until after the pregnancy could hamper breastfeeding, if the woman choses to do so then. In Bilsen’s et al. (2016) study, 15.5% of women who fast indicated that ‘Because I do not want to have to make up the fasting later’ as a reason for them to fast, which does not look further into the reasons why exactly they do not want to make up later.

28

3. Social

As mentioned earlier, when women do not fast they will have to miss out of the communal activity and have to deal with the loss of feeling of participating in Ramadan. In van Bilsen’s et al. (2016) study 48.6% indicated that they continue fasting ‘Because I want to share

spiritual and social experiences with my family’. Another reason mentioned by van Ewijk

(2011) is that in an orthodox area women might be scrutinized if people decide to not participate, the fear of social scrutiny could be a reason to participate. Van Bilsen et al. (2016) emphasizes the importance of the husband’s opinion: When women were not fasting most men had no opinion on whether she should fast or not and of women that did fast most of their husbands said they should. Their study indicates that the husband’s opinion influences a woman’s decision to fast significantly, and they emphasize the gender roles that play an active role when making this decision. In their study from the women who were fasting 11.5% said that ‘Because my husband told me to fast.’ And when women did not fast 42% said that their reason was ‘Because my husband told me not to fast’ next to a 5.3% of participants indicating this reason: ‘Because my family and friends suggest me not to fast’. The education of the husband influences whether they are for- or against fasting, the higher the education the more likely to oppose fasting during pregnancy.

4. Health

Van Ewijk (2011) conducted interviews with doctors, midwives, health workers and others on Ramadan fasting. The observed regulation is that pregnant women are allowed to skip fasting if they believe that it may harm their own or their fetus’ health. In van Bilsen’s et al. (2016) study, some 42% expressed that they did not fast because they expressed concern on adverse health effects on their child and 15.8% did not fast ‘Because I am afraid that it will affect my health’. 13.2% did not fast due to ‘illness or pregnancy complications’. Around 32% did not fast because the doctor or midwife advised them not to. Conversely, of 16.9% of women who did fast said that ‘Because my midwife and doctor told me that Ramadan fasting during pregnancy is not harmful’ as a reason for them to decide to fast. It is possible that outside of this study this percentage is even higher since the study took place in a private hospital in Jakarta. It seems of vital importance to ensure that doctors and midwives have the correct information and discuss that with their patients. Fear that fasting is harmful is common in both fasting and non-fasting groups according to van Bilsen et al. (2016), this concern itself can be manifested in psychological stress which is associated with many complications during pregnancy and can even influence fetal programming and lead to lower mental and motor

29

skills of infants. This could be a possible influence that is captured in our research design which is not Ramadan fasting itself.

5. Other

Some pregnant women attempt to fast but gave up because it was too difficult. Van Bilsen et al. (2016) study showed that the main reasons women did not fast is that they indicated that it is too difficult to do so when pregnant (84.2%). Van Bilsen et al. (2016) has some other reasons people fast during Ramadan, 10.8% of women fasted ‘because I did not know I was pregnant during Ramadan’ which concerns Ramadan exposure in first trimester primarily. Also the reason of not fasting being too difficult is more prevalent amongst women who are in second or third trimester. Another reason is weight gain control, people who have a higher BMI at the start of their pregnancy are more likely to fast during Ramadan. Lastly, 18.2% of women did not fast ‘Because I also did Ramadan fasting on my previous pregnancy(ies) and there is no problem at all’. Also a reason given is that they are used to fasting, for 56.8% in van Bilsen’s et al. (2016) study.

7.2 Duration Pregnancy Tables

Please take into consideration that duration pregnancy here is expressed in weeks starting from the beginning of the last period and not from moment of conception.

Table A1: Duration Pregnancy Frequency Table- Muslims

Duration pregnancy Not exposed Exposed Total

21 Weeks 2 0% 0 0% 2 24 Weeks 10 0% 14 0% 24 25 Weeks 1 0% 2 0% 3 26 Weeks 2 0% 1 0% 3 27 Weeks 4 0% 2 0% 6 28 Weeks 81 2% 80 1% 161 29 Weeks 1 0% 1 0% 2 30 Weeks 6 0% 3 0% 9 31 Weeks 2 0% 1 0% 3 32 Weeks 186 5% 225 3% 411 33 Weeks 15 0% 17 0% 32 34 Weeks 18 1% 41 1% 59

30 35 Weeks 20 1% 23 0% 43 36 Weeks 2,565 72% 4,594 68% 7,159 37 Weeks 84 2% 155 2% 239 38 Weeks 69 2% 94 1% 163 39 Weeks 39 1% 105 2% 144 40 Weeks 383 11% 1082 16% 1,465 41 Weeks 30 1% 89 1% 119 42 Weeks 26 1% 97 1% 123 43 Weeks 9 0% 17 0% 26 44 Weeks 12 0% 48 1% 60 45 Weeks 0 0% 8 0% 8 48 Weeks 5 0% 29 0% 34 52 Weeks 0 0% 1 0% 1 Total 3,570 6,729 10,299

Table A2: Duration pregnancy Frequency Table- Non-Muslims

Duration pregnancy Not exposed Exposed Total

24 Weeks 3 1% 2 0% 5 26 Weeks 0 0% 1 0% 1 27 Weeks 0 0% 1 0% 1 28 Weeks 11 3% 11 1% 22 29 Weeks 2 0% 4 0% 6 30 Weeks 2 0% 1 0% 3 32 Weeks 25 6% 75 9% 100 33 Weeks 1 0% 3 0% 4 34 Weeks 6 1% 9 1% 15 35 Weeks 6 1% 9 1% 15 36 Weeks 322 74% 538 63% 860 37 Weeks 10 2% 27 3% 37 38 Weeks 7 2% 13 2% 20 39 Weeks 3 1% 9 1% 12 40 Weeks 30 7% 98 12% 128

31 41 Weeks 3 1% 17 2% 20 42 Weeks 2 0% 7 1% 9 43 Weeks 1 0% 2 0% 3 44 Weeks 1 0% 15 2% 16 45 Weeks 0 0% 3 0% 3 48 Weeks 0 0% 3 0% 3 52 Weeks 0 0% 1 0% 1 Total 435 849 1,284

7.3 Individuals born during Ramadan

Relative size Birth weight W-score Log hours Self-Employed

M Non M Non M Non M Non M Non

Born during Ramadan -0.06* -0.038 0.04 -0.08 -0.93 9.75** 0.02 0.03 0.004 -0.01

SD 0.4 0.1 0.035 0.14 1.53 3.95 0.036 0.089 0.012 0.03

Observations 5,693 660 5,542 559 28,138 3,150 5,674 621 21,976 2,523 Control variables used: Birth-month fixed effects, age and age2 for all regressions. For

Relative size and birth weight multiples was also added as a control. Robust standard errors are clustered at household level in parentheses. M is Muslims, and Non is Non-Muslims. Exposure dummy is 1 if an individual is exposed to full month of Ramadan.

*** p<0.01, ** p<0.05, * p<0.1