You are on page 1of 28

The Kids Are All Right: The Impact of Parental Migration in the Philippines

Pajaron, Marjorie
School of Economics
University of the Philippines

Abstract

This study contributes to the debate on the impact of parental migration on children in the
following ways: differentiating the welfare of left-behind children and children of non-
migrant parents in the Philippines; properly addressing identification issues related to
parental migration; and exploring heterogeneity in the impact of parental migration.
Using previously unexamined data on left-behind children in the Philippines, the results
are robust across six econometric specifications (OLS, probit, multinomial logit,
treatment effects, bivariate probit, and propensity score matching). Children with a
migrant parent have better outcomes in education (in terms of grade level, grades, and
study habits), health (psychological and physical), and labor (child labor force
participation) than children of non-migrant parents. The results also show heterogeneity
in the impact of parental migration conditional on the gender of the left-behind child,
gender of the household head, and gender bias of the household head.

Keywords: Migration, Children’s Welfare, Instrumental Variable, Bivariate Probit, PSM,


Left-Behind Children
JEL codes: F22, J13, O15

Page 1 of 28
1. Introduction

For the past four decades, there has been a growth in the number of Filipino

migrant workers who are leaving the country in search of better job opportunities and

higher income (International Organization for Migration [IOM] 2013). The IOM reports

that, as of 2011, the number of Filipino migrants had reached about 10.5 million people

working in about two hundred other countries—placing the Philippines among the top

countries in the world in the export of labor.

According to the IOM (2013), 53% of the total number are temporary or irregular

migrants, meaning they have families back home that depend on their income. These

temporary migrants, known as Overseas Filipino Workers (OFWs), have often been

credited for facilitating the growth of the Philippine economy over the past years. The

government has recently been making efforts to promote labor exports and increase the

length of overseas contract working periods, causing migrants to spend even more time

abroad. The results of such heavy labor migration likely include changes in household

roles and composition. The net effects of this on Filipino families, however, remain

ambiguous.

Recent studies have shown that household structure can impact the welfare of

children. For example, children in the United States living in single-parent households are

more likely to perform poorly in school, participate in the labor force, and exhibit issues

with their physical, emotional, and social well-being (Milne, Myers, Rosenthal, and

Ginsburg 1986; Moehling 2003; Robl, Jewell, and Kanotra 2012). The same is true in

Canada, where youth living with single parents are more likely to smoke (Razaz-

Rahmati, Nourian, and Okoli 2012). In addition, the longer a child spends in a single-

Page 2 of 28
parent home, the more adverse the effects are, as observed in the United States (Krein

and Beller 1988).

Households with migrant parents may not experience the same effects as other

single-parent households because members can allocate remittances to benefit children’s

welfare. In several cases, children with migrant parents have been shown to stay longer

and perform better in school, including in the Philippines, Mexico, China, and Nepal

(Arguillas and Williams 2010; Antman 2012; Asis and Ruiz-Marave 2013; Hu 2013;

Acharya and Leon-Gonzales 2014). In contrast, however, other studies have found that

children of migrants have higher dropout rates because of social problems and increased

household responsibilities, as observed among children in Mexico and Caribbean states

(McKenzie and Rapoport 2006; Bakker, Elings-Pels, and Reis 2009).

In terms of health, children of migrants in the Philippines, Romania, and Sri

Lanka have been shown to suffer nutritional deficiencies and emotional distress

(Smeekens, Stroebe, and Abakoumkin 2012; Botezat and Pfeiffer 2014; Wickramage et

al. 2015). However, other studies have reported that in Indonesia, the Philippines,

Thailand, and Vietnam, migration does not have significant negative effects on children’s

health (Battistella and Conaco 1998; Graham and Jordan 2011).

The contradictory results of these studies suggest two possible, opposing effects

of parental migration on the welfare of left-behind children: living in a migrant household

may be detrimental to a child’s welfare due to the lack of parental involvement; however,

the contribution of remittances might compensate for the parent’s absence to some extent

by increasing the household’s income.

Page 3 of 28
This paper contributes to the existing literature on parental migration and the

welfare of left-behind children in three different ways. First, the analysis includes

detailed measures of the welfare of left-behind children using new nationwide survey

data from the Philippines that, to the best of our knowledge, has not been used for such a

study before. We use eight different measures of welfare outcomes: four measures of

educational outcomes (current grade level as a continuous variable, current grade level as

a categorical variable, probability of having poor grades, and probability of having good

study habits), two measures for health outcomes (physical and psychological), and two

measures for labor outcomes (probability of the child having worked in the past week and

in the past year).

Second, the study addresses potential identification issues pertaining to parental

migration by using an instrumental variable (IV) analysis (estimating treatment effects

and bivariate probit models) and propensity score matching (PSM), and comparing the

results with those of ordinary least squares (OLS), multinomial logit, and probit

regressions.

Third, the study examines possible heterogeneity in the impact of parental

migration on children’s welfare conditional on the gender of the left-behind children and

the gender of the heads of the households in which the children live (91% of the children

live with their left-behind parent while 8% live with their left-behind grandparent). We

also consider the possibility of gender bias and test whether there is an interaction

between gender of the child and gender of the child’s household’s head. To put this

another way, we analyze whether differences in welfare outcomes exist between boys and

Page 4 of 28
girls conditional on the gender of the person responsible for them, whether that is a left-

behind parent or a grandparent.

The results show that the migration of parents can have a positive impact on the

welfare of their children. In particular, for the period studied, children of migrant parents

had higher current grade levels, lower probability of poor grades, higher probability of

studying regularly, less probability of being perceived as sickly and temperamental, and

less likelihood of having worked in the past week and past year compared to children of

non-migrant parents. The results are robust across the different econometric strategies

used and even after addressing identification issues such as endogeneity of parental

migration.

Heterogeneity in the impact of migration on children’s welfare also exists: left-

behind boys are more likely to have good study habits compared to left-behind girls, and

children of migrants in female-headed households are less likely to have poor grades

compared to those in male-headed households. In addition, in female-headed households,

left-behind girls are less likely to be sickly but more likely to be temperamental compared

to left-behind boys. The results of post-estimation Wald tests, on the other hand, suggest

that in male-headed households, left-behind male children are 4% more likely to have

good study habits, but they are also marginally more likely to have negative health

outcomes (more likely to be sickly and temperamental) compared to left-behind female

children.

The paper is organized as follows. Section 2 reviews existing literature related to

migration and its impact on child welfare. Section 3 contains a description of the data

used in this study along with a discussion of methodology and econometric strategies.

Page 5 of 28
Section 4 elaborates on the findings from the regressions and their implications. Section 5

discusses our conclusions and recommendations.

2. Review of Related Literature

2.1. Background on Migration

Labor migration is considered a human capital investment because workers

venture out to increase their productivity and wages despite the transportation and

relocation costs incurred (Ehrenberg and Smith 1988). Non-monetary factors such as

living environment, social services, political stability, and security may drive individuals

to migrate to a certain location (World Bank 2006). For the Philippines, migration has

resulted in improved living standards, which has triggered further migration (IOM 2013).

Certain trends can be observed in the individuals who choose to migrate. Based

on data gathered by the United Nations Department of Economic and Social Affairs (UN-

DESA) and the Organization for Economic Co-operation and Development (OECD) in

2013, about half of all international migrants residing in OECD countries come from

Asia, Latin America, or the Caribbean. Increasingly more of these migrants are low-

skilled workers who have not completed secondary school. However, high emigration

rates for the highly educated still persist; this is especially true in low and middle income

countries, and more so for women than men.

Given the increasing trend in migration outflows, it is timely to consider the

impacts—beyond the usual positive income effect—of parental migration on the children

they leave behind.

Page 6 of 28
2.2. Impact of Migration on Educational Outcomes

Migration may have adverse impacts on educational attainment and school

performance of left-behind children. In Mexico, boys had higher chances of migrating

and dropping out in junior high school while girls had higher chances of dropping out in

high school to do housework (McKenzie and Rapoport 2006). In the Caribbean, children

were at higher risk of having poor grades and not completing their schooling (Bakker et

al. 2009). Younger siblings were more likely to drop out due to coping difficulties and

increased fighting incidences in school, while older siblings were more likely to drop out

due to the new household responsibilities they had to assume in the absence of their

migrant parent.

There are recent studies, on the other hand, that have recognized the positive

income effect that stems from migration. In Nepal, migration of uneducated mothers and

those from poor households actually resulted in higher child enrollment rates and

educational investment (Acharya and Leon-Gonzales 2014). In northwest China, parental

absence resulted in poorer educational performance but remittances partially offset this

(Hu 2013). Girls are more affected, which may suggest a gender bias in the impact of

parental migration. In Mexico, the migration of fathers to the United States resulted in

increased decision-making by mothers and better educational outcomes for daughters

(Antman 2012). In the Philippines, parental migration likewise has a positive impact on

the education of children; the effect is more pronounced in households where the father

migrates while the mother stays at home (Battistella and Conaco 1998; Arguillas and

Williams 2010; Asis and Ruiz-Marave 2013).

Page 7 of 28
2.3. Impact of Migration on Psychological and Physical Health

Migration may affect the psychological and physical health of left-behind children

in two opposing ways: decreased time allotted by parents for maintaining their children’s

health, which may have an adverse effect, and augmented household income through

remittances, which may have a positive effect.

Recent studies have shown that parental migration often leads to the deterioration

of the physical and psychological health of left-behind children. For example, in Sri

Lanka, parental migration resulted in children’s increased vulnerability to malnutrition

and mental, emotional, psychiatric, and hyperactive disorders (Wickramage et al. 2015).

In Romania, left-behind children were more likely to get sick and suffer from depression,

especially in rural areas (Botezat and Pfeiffer 2014). In the Caribbean, children are more

likely to display a wide array of emotional and psychological problems, including low

self-esteem, increased anger and violence, lack of trust in relationships, and depression

(Bakker et al. 2009).

In the Philippines, however, different studies provide conflicting results. For

example, Battistella and Conaco (1998) found that parental migration, particularly

maternal migration, adversely affected the welfare of children, such as their emotional

health. Smeekens, Stroebe, and Abakoumkin (2012) also found that parental migration

was associated with a negative impact on the physical and emotional condition of left-

behind children. On the other hand, Graham and Jordan (2011) found, after examining

migration data from several Southeast Asian countries including the Philippines, that

there were no significant differences between the psychological well-being of children of

migrants and children of non-migrant parents.

Page 8 of 28
2.4. Impact of Migration on Child Labor Force Participation

To our knowledge, there is little existing literature on the direct effects of

migration on child labor force participation. A paper by Rickey (2009) discussed the

specific determinants of child labor in the Philippines and mentioned that, theoretically,

the presence or absence of a parent or primary caregiver can play a role in whether a

child will choose to work or not. This theory is supported by Moehling (2003) in a study

regarding child labor force participation in different family structures; she reported that in

America, children living in single-parent households had a greater labor market

participation rate when compared to children in two-parent households. The probability

of a child participating in the labor force also increased if his/her gender was the same as

that of the remaining parent. However, households with migrant parents may differ from

other single-parent households in that the former may have a relatively higher household

income due to remittances, which makes them interesting to analyze.

3. Data and Methodology

3.1. Data Description

This article primarily uses the 2011 Survey on Children (SOC), which is a joint

project of the International Labor Organization (ILO) and the Philippine National

Statistics Office (NSO), for the child of migrant indicator, child welfare outcomes, and

other control variables (socioeconomic characteristics of children and households). SOC

gathers information about children to better understand their activities, labor force

participation, and working conditions in the Philippines. The dataset is split into

household-level characteristics, such as location, income level, resources, and child-

Page 9 of 28
centered socioeconomic outcomes. A total of 27,348 households were included in this

survey with a sample of 70,707 children; about 3,234 of these children (4.6%) have

migrant parents (Table 1).

<Table 1 about here>

Table 1 depicts the different child welfare outcomes used in this article: (a) four

educational outcomes, as a continuous variable (current grade level), as binary variables

(indicator for poor grades and indicator for studying regularly), and as a categorical

variable (four categories of current grade level); (b) two health outcomes—physical and

psychological—as binary variables (indicator for whether the child is perceived to have

poor health and indicator for whether the child is perceived to have anger issues or tend

to be emotional); and (c) two labor outcomes as binary variables (indicators for whether

the child had worked in the past week and in the past year).

On average, migrant children had a higher current grade level (by about one level)

and more of them were in high school and college compared to children of non-migrants.

Left-behind children were also less likely (by about 1%) to have poor grades and more

likely (by about 9%) to have good study habits.

For the two health outcomes, Table 1 shows that out of the 70,707 children, only

about 354 (0.5%) were perceived to be sickly or temperamental, with the rates slightly

higher on average for children of non-migrant parents than for children of migrant

parents.

For labor outcomes, around 6,080 had worked in the past week while around

9,263 had worked in the past year. In addition, on average, children of non-migrant

Page 10 of 28
parents were more likely to have worked in the past week (5.6% more) and in the past

year (8.1% more) than children of migrant parents.

<Table 2.1 about here>

Table 2.1 shows the definition and descriptive statistics of the different

demographic and socioeconomic characteristics of the children and households in the

dataset that are predicted to impact children’s welfare outcomes. In terms of child

characteristics, there is not much difference between children of migrants and non-

migrants. On average, about half of the children in the dataset are male, the second to

youngest sibling, and about 11 years of age.

Differences between the two types of households can be seen in the household

characteristics. Households of left-behind children have younger household heads (about

two years younger), are mostly in urban areas (about 51% of them), have slightly smaller

household sizes, usually have access to water and light sources, and, although they own

less agricultural land, more of them belong to higher income brackets compared to the

households of children of non-migrant parents.

Regional data, needed to address the possible endogeneity of parental migration,

are taken from three sources, which are merged with the SOC dataset: the 2004 Survey

on Filipinos (SOF), the 2011 and 2014 Philippine Statistical Yearbooks (PSY), and the

2011 Annual Poverty Indicator Survey (APIS). The SOF, which contains data on the

amount of remittances and on the socioeconomic characteristics of workers who were

working or had worked abroad, is used to derive the regional migration rate. The APIS

provides different poverty indicators and is used to derive regional family income,

percentage of families who experienced hunger per region, and regional school

Page 11 of 28
attendance. The PSY, which compiles major economic and social information about the

Philippines, is used to derive regional population and number of schools.

<Table 2.2 about here>

Regional characteristics, depicted in Table 2.2, reveal regional differences. First,

the children of both migrants and non-migrants lived mostly in the National Capital

Region (NCR) and the region next to it (CALABARZON). Second, households living in

NCR had the highest average income while those living in the Autonomous Region of

Muslim Mindanao (ARMM) had the lowest.

Third, in terms of families who experienced hunger, the highest percentages were

in the Eastern Visayas, consisting of the three main islands of Samar, Leyte, and Biliran

(16.2%), followed by the three non-ARMM regions of Mindanao (all over 10%).

Fourth, regional differences in terms of number of schools (relative to regional

population) and school attendance are not that significant, with differences of about 8% at

most. School attendance is the regional percentage of individuals who attended school in

2011–2012, which encompassed those aged 3 to 24 for both public and private schools.

The Cordillera Administrative Region (CAR) had the highest school attendance (71%)

while ARMM had the lowest (63%). Fifth, the northernmost part of the Philippines

(Ilocos and Cagayan regions) had the highest migration rates, followed by NCR and

CALABARZON.

Page 12 of 28
3.2. Empirical Models

3.2.1. Impact of Parental Migration on Children’s Welfare

This article first estimates the following model to examine the impact of parental

migration on children’s welfare:

Yni = β0 + β1 childofmigranti + β2 Xi + ui (1)

where Yni is the nth child welfare outcome of parental migration on ith child;

childofmigrant is a binary variable for child of migrant (1 if the child has at least one

migrant parent); X is a vector of control variables that affect child welfare outcomes,

which include child’s characteristics (sex, child’s rank among his/her siblings, and age),

household’s characteristics (household head’s sex, age, and spouse’s age; location; water

and light source; ownership of agricultural land; and average monthly gross income); and

u is the error term.

Child welfare outcomes (Yni), as mentioned above, are measured in eight different

ways, which are described in Table 1. The first measure of educational outcomes, current

grade level as a continuous variable, is estimated using ordinary least squares (OLS)

while the second measure, current grade level as a categorical variable with four

categories (no grade completed, primary, secondary, and tertiary) is estimated using

multinomial logit.1 The rest of the welfare outcomes are binary variables and they are

1
For estimating the effect on current grade as a categorical variable, we use the following multinomial
logistic regression model:
� =�
Logit(y=m)=log = δ0 + δ1 childofmigrant + δ2 Xi + ei, m=1, 2, 3, 4 (2)
1− �=�
where y equals the four categories for the child’s current grade level variable, and the category being
tested is indicated by m. The base category used here is primary school, because the majority of our
sample is in this group. In addition, childofmigrant is a dummy variable indicating whether or not a child
has a migrant parent and X is the vector of controls.

Page 13 of 28
estimated using probit.2

Identification Issues (Endogeneity of Migration)

In the base model above, the assumption is that migrant and non-migrant

households are similar in all observable and unobservable characteristics. However, it is

likely that the decision to migrate is correlated with unobserved characteristics that affect

the household’s decision to invest in child welfare. For example, it is possible that

migrant parents’ preferences indicate that they highly value their children’s education or

health, making them naturally more inclined to invest in these matters. To avoid bias and

overestimation of the effects of migration, three more regression methods are used:

treatment effects, bivariate probit, and propensity score matching (PSM).

For both the treatment effects and bivariate probit, the probability that a child has

migrant parents is instrumented using historical regional migration rate, because this

variable reflects regional migration networks, which facilitate current parental migration

by helping to make it more convenient and less costly to migrate (McKenzie and

Rapoport 2006; Botezat and Pfeiffer 2014). Hence, there is a high correlation between

historical migration rates and current parental migration, which results in having left-

behind children. We assume that these past migration rates do not directly affect the

2
For our binary outcomes, a simple linear probability model (LPM) would violate the assumptions of OLS,
namely that error terms have equal variances for all X’s and that error terms are normally distributed.
Thus, in order to address this issue, a probit regression is included and is modeled as follows:
Pr (Yni =1|X) = G(ɣ0 + ɣ1childofmigranti + ɣn Xi) (3)
s.t. G(z) = Φ(z) = ∫−∞ � � ��
where Yni represents child welfare outcome n of individual i; and childofmigrant and X pertain to the
variables described for Equations (1) and (2). G(z) is the standard normal cumulative distribution function.

Page 14 of 28
present outcomes of children, except through current parental migration. The first-stage

instrumental regression equation is:

childofmigranti = q 0 + q 1M j + q 2 Xi + q 3R j + u i (4)

where childofmigranti is a discrete variable for whether the ith child has migrant parents,

Mj is the historical regional migration rate in 2003 computed as the ratio of the total

regional number of migrants relative to the total population per region j, Xi is the vector

of control variables similar to those described in Equation (1), Rj pertains to regional

variables such as average family income, the percentage of families who experienced

hunger, number of schools, and school attendance.3 For the continuous measure of

current grade level, a treatment-effects model is used, while bivariate probit is used for

the rest of the binary outcome variables.

PSM, another method that addresses possible endogeneity of parental migration,

is a quasi-experimental method that attempts to mimic randomization (Khandker,

Koolwal, and Samad 2010). PSM constructs a counterfactual by matching observations

based on their propensity scores or the probability that a child would have been a child of

a migrant based on a given a set of characteristics:

Pr (childofmigrant=1 | Y) = G(ω0 + ωnY) (5)


s.t. G(z) = Φ(z) = ∫−∞ � � ��

3
The year 2003 is chosen because it falls shortly after the boom of overseas Filipino workers that took
place at the end of the 20th century.

Page 15 of 28
PSM requires that covariates Y, which encompass all the independent variables

described in Equation (4), be not affected by the probability that a child is a left-behind

child.

Children are then paired based on their propensity scores, and average treatment

effects are calculated from the average of the differences between the outcomes of

matched children.

3.2.2. Heterogeneity in the Impact of Parental Migration across Gender of Child

To test whether there exists a differential in the impact of parental migration

conditional on the gender of the child, an interaction of indicator for child of a migrant

parent and indicator for gender of the child (childofmigranti * child’ssexi) is added to

Equation (1); the rest of the variables are similar to those in Equation (1):

Yni = �0 + � 1 childofmigranti + � 2 child’ssexi + � 3 childofmigranti *


child’ssexi + � 4 Xi + ei (6)

Equations (7–8) below depict the impact of parental migration on girls and on

boys, respectively, while Equation (9) describes the heterogeneity in the impact of

parental migration between these two groups:

¶ Yni
=p 1
¶ child _ Mig child 'ssex=0
(7)

¶ Yni
=p 1+p 3
¶ child _ Mig child 'ssex=1
(8)

Page 16 of 28
¶ Yin ¶ Yin
- =p 3
¶ child _ Mig child 'ssex=1 ¶ child _ Mig child 'ssex=0
(9)

3.2.3. Heterogeneity in the Impact of Parental Migration across Gender of Household


Head

To determine whether the impact of parental migration on children’s welfare

outcome depends on the gender of the household head, an interaction of indicator for

child of a migrant parent and indicator for gender of the household head (childofmigranti

* head’ssexi) is added to Equation (1); again the rest of the variables are as described in

Equation (1):

Yni = α0 + α1 childofmigranti + α2 head’ssexi + α3 childofmigranti *


head’ssexi + α 4 Xi + ei (10)

Equations (11–12) below show the effect of parental migration on children in

female-headed and male-headed households, respectively:

¶ Yni
=a 1
¶ child _ Mig head 'ssex=0
(11)

¶ Yni
=a 1 +a 3
¶ child _ Mig head 'ssex=1
(12)

Heterogeneity between female- and male-headed households in the effect of


parental migration on welfare outcomes of children is depicted in Equation (13) below:

¶ Yin ¶ Yin
- =a 3
¶ child _ Mig head 'ssex=1 ¶ child _ Mig head 'ssex=0
(13)

Page 17 of 28
3.2.4. Gender Bias toward Children Conditional on the Gender of Household Head

It is also possible that female heads and male heads have gender bias and may

allocate resources to children differently depending on the gender of the children. To

account for this, a triple interaction term is added (childofmigranti * child’ssexi *

head’ssexi):

Yni = 0 + 1 childofmigranti + 2 child’ssexi + 3 head’ssexi + 4 childofmigranti *


child’ssexi + 5 childofmigranti * head’ssexi + 6 child’ssexi * head’ssexi +
7 childofmigranti * child’ssexi * head’ssexi + 8 Xi + ei (14)

The hypothesis tests below aim to determine the impact of parental migration on

girls and boys in female-headed households (Equations 15–16) and in male-headed

households (Equations 17–18):

¶ Yin
=j 1
¶ child _ Mig child 'ssex=0; head 'ssex=0
(15)

¶ Yin
=j 1 +j 4
¶ child _ Mig child 'ssex=1; head 'ssex=0
(16)

¶ Yin
=j 1 +j 5
¶ child _ Mig child 'ssex=0; head 'ssex=1
(17)

¶ Yin
=j 1 +j 4 +j 5 +j 7
¶ child _ Mig child 'ssex=1; head 'ssex=1
(18)

Equation (19) tests the gender bias of female heads and Equation (20) tests that of

male heads of migrant households:

Page 18 of 28
¶ Yin ¶ Yin
- =j 4
¶ child _ Mig child 'ssex=1; head 'ssex=0 ¶ child _ Mig child 'ssex=0; head 'ssex=0
(19)

¶ Yin ¶ Yin
- =j 4 +j 7
¶ child _ Mig child 'ssex=1; head 'ssex=1 ¶ child _ Mig child 'ssex=0; head 'ssex=1
(20)

4. Results

4.1. Impact of Parental Migration on Children’s Welfare

4.1.1. Educational Outcomes

The first measure of educational outcome, current grade level as a continuous

variable, is estimated using ordinary least squares (OLS) and it can be gleaned from

Table 3, Column 1 that children of migrants studied half a year more than children of

non-migrants, keeping other factors constant. In the second measure of educational

outcome, the current grade level is considered as a categorical variable with four

categories (no grade completed, primary, secondary, and tertiary) and Equation (2) is

estimated using multinomial logit. Table 3 (Column 2) shows that parental migration is

associated with a 0.06-decrease in the relative log odds of a child having no grade

completed and a 0.02-increase in the relative log odds of a child having secondary

education compared to the child being enrolled in primary school (i.e., the base category).

<Table 3 about here>

The other two measures of educational outcomes are both treated as binary

variables; they pertain to the perception of the respondent with regard to the performance

of the child in school (whether the child has poor grades) and the study habits of the child

Page 19 of 28
(whether the child studies regularly). To estimate the impact of parental migration on

these two variables, the probit estimation method is used. Table 3 (Columns 5 and 6)

shows that left-behind children are less likely (by about 1%) to have poor grades and

more likely (by about 6%) to study regularly than children of non-migrant parents. In

addition, girls, regardless of the type of household, are 1% less likely to have poor grades

and 5% more likely to study than boys. This last result will be further explored in a later

section, when the differential impact of parental migration based on the gender of the

child is formally tested.

The results, thus far, indicate a positive impact of parental migration on the

educational outcomes of left-behind children, keeping everything else constant.

4.1.2. Health and Labor Outcomes

The probit regression results involving the two measures of health outcomes

(sickly and temperamental) show that there is no statistical difference between children of

migrant parents and children of non-migrant parents (Table 3, Columns 7 and 8). Girls

have better health outcomes compared to boys, albeit the differences are marginal.

Probit regressions reveal that children of migrant parents were 5% less likely to

have worked in the past week and 6% less likely to have worked in the past year

compared to children of non-migrant parents (Table 3, Columns 9 and 10, respectively).

The marginal effects for the other control variables suggest that boys were more likely

(4%) to participate in the labor market than girls. Meanwhile, rank among siblings and

age of children are positively correlated with the probability of working in both time

periods. In addition, for both labor outcome variables, children in urban areas, from

Page 20 of 28
larger families, with access to water and a light source, without agricultural land, and

whose household’s average gross income was higher were less likely to participate in the

labor market.

4.2. Addressing Endogeneity of Parental Migration

4.2.1. Two-Step Treatment-Effects Model and Bivariate Probit

Table 4 shows the regression results after considering endogeneity of parental

migration and using historical regional migration rate as a potential instrumental variable.

As discussed above, historical migration tends to create a regional migration network,

which may help facilitate more convenient and less costly current migration, suggesting

an intuitive association between the two.

<Table 4 about here>

For an instrument to be valid in this study, it must satisfy two requirements: it

must be correlated to parental migration, and it must be uncorrelated to the error term in

the children’s welfare outcome function (Equation 1). The first requirement is formally

tested by regressing parental migration on log of regional migration rate in 2003,

controlling for all other variables described in Equation (1) and the regional variables

depicted in Table 2.2. Table 4 (all first stage columns) shows that log of migration rate is

positively and statistically significant at the 1% level, suggesting a strong correlation

between the two. The second requirement can be tested using the post-estimation

likelihood ratio (LR) test or Wald test, which is an approximate of LR, to test for the

endogeneity of the instrument (Knapp and Seaks 1998). The instrumental variable passes

the endogeneity test only for the sickly, temperamental, and worked in the past year

Page 21 of 28
regressions (Wald test results in Columns 8, 10, and 14). Out of these three regressions

with the exogenous instrumental variable, it is only in the temper regression that the

coefficient for child of migrant is statistically significant at the 1% level, suggesting that

a child with a migrant parent is 3% less likely to be temperamental compared to a child of

non-migrant parents.

4.2.2. Propensity Score Matching (PSM)

The PSM approach assumes that after matching on all observable household,

child, and regional characteristics, assignment to the treatment (children of migrant

parents) or control group (children of non-migrant parents) is random. The average

outcomes for children in the treatment group are compared with those for the matched

controls. As suggested by Rosenbaum and Rubin (1983), instead of matching each child

of a parent migrant with a control child, which could be difficult given the sample size

and the need for simultaneous matching on every dimension, matching based on

“propensity score” or the probability that a child has migrant parents, given observable

characteristics, will suffice. The propensity score is derived from probit analysis to

determine the factors that impact the probability of parental migration (and hence, being a

child of a migrant).

The results of probit analysis, which has heteroskedastic-standard errors as in all

the other previous regressions, suggest that the log of historical migration rate is positive

and statistically significant, which suggests that as historical migration rate increases,

children are more likely to have migrant parents (Table 5.1). Among the children and

household characteristics, age of the child, water source, light source, and average gross

Page 22 of 28
income are all statistically significant and positively correlated with the probability of

parental migration.

<Tables 5.1 and 5.2 about here>

Table 5.2 displays the results of comparing the welfare outcomes of children of

migrant parents against the outcomes a matched control group (children of non-migrant

parents) that have similar observable characteristics using the propensity scores derived

in Table 5.1. The average treatment effect (ATE) from PSM reveals that for educational

outcomes, children of migrants were enrolled in grades 1.33 levels higher. Children of

migrants were also 1% less likely to have poor grades and 9% more likely to have good

study habits. For health outcomes, children of migrants were about 0.5% and 0.3% less

likely to be physically sick and temperamental, respectively. For labor outcomes, children

of migrant parents were about 1% less likely to have worked in the past week and in the

past year.

In summary, PSM results suggest that for all of the welfare outcomes, children of

migrant parents are better off compared to children of non-migrant parents. These results

are consistent with the OLS and probit results on educational and labor outcomes (Table

3) and the biprobit result on likelihood of a child being temperamental (Table 4, Column

10).

4.3. Heterogeneity in the Impact of Parental Migration

This section considers possible gender differences in the impact of parental

migration conditional on the gender of the child, the gender of the household head, and

the interaction of these two sources of gender differentials.

Page 23 of 28
4.3.1. Differences across the Gender of the Child

As mentioned above, the results presented in Table 3 indicate that girls, regardless

of parental migration, have better educational, health, and labor outcomes than boys. To

formally test whether the impact of parental migration on welfare outcomes varies across

the gender of the child, we include an interaction of child of a migrant and gender of the

child of a migrant in the regression analyses. The coefficients of child of a migrant from

Table 6 show that, following Equation (7), daughters of migrant parents have better

educational and labor outcomes compared to daughters of non-migrant parents (Columns

1, 2, 3, 6, and 7). Post-estimation Wald tests reveal similar results for sons of migrant

parents (after testing Equation 8 above). Heterogeneity between the two groups can only

be observed for the study habits regression through the interaction term coefficient (from

Equation 9). In particular, the impact of parental migration is 4% higher for boys than

girls in terms of good study habits (Table 6, Column 3).

<Table 6 about here>

4.3.2. Differences across the Gender of the Household Head

To explore the possibility that parental migration has a differential impact on the

welfare of children conditional on the gender of the household head, we add an

interaction for gender of the household head and being a child of a migrant. The

coefficients for child of a migrant in Table 7 reveal the welfare outcomes of children in

female-headed households, based on Equation (11). It can be gleaned that children of

migrants were in school at almost a grade level higher and 9% less likely to have poor

Page 24 of 28
grades relative to children of non-migrants (Columns 1 and 2, respectively). Testing

Equation (12), or testing the linear combination of coefficients for children of migrants

and the interaction term child of migrant and sex of household head, using post-

estimation Wald tests shows that in male-headed households, children of migrant parents

have better educational and labor outcomes than children of non-migrant parents. The

coefficient for the interaction term child of migrant and sex of household head reveals

heterogeneity in the impact of parental migration on welfare outcomes of children

conditional on the sex of the household head, following Equation (13). Column 2 shows

that the likelihood that the child of a migrant will have poor grades is 8% higher in male-

headed households than female-headed households.

<Table 7 about here>

4.3.3. Gender Bias toward Children Conditional on the Gender of the Household Head

Finally, in order to test whether gender bias or gender preference exists among

male heads and female heads of migrant households, Equations (15–20) above are tested;

the results are shown in Table 8.

<Table 8 about here>

The coefficient for child of migrant shows the impact of parental migration on

girls in female-headed households – they are less likely to have poor grades and less

likely to be sickly compared to girls in non-migrant households headed by female. On

the other hand, the results of testing Equation (16), displayed in Table 8, Panel A, show

that boys in female-headed households are enrolled in a higher grade level (about a year)

Page 25 of 28
and less likely to have poor grades (9%), more likely to be sickly (5%), and less likely to

be temperamental (5%).

In male-headed households, the second Wald postestimation test in Table 8, Panel

B tests for Equation (17) and the results suggest that girls have better educational, health,

and labor outcomes but more likely to be physically sick albeit a small probability.

Similarly, testing Equation (18) shows that sons have improved educational, health, and

labor outcomes (Table 8, Panel C).

The coefficient for the interaction term child of migrant and sex of child, which

tests for gender bias in female-headed households (Equation 19), reveals that in these

migrant households, male children are more likely to be sickly (9%) but less likely to be

temperamental (5%) than girl children.

The results of testing for the linear combination of the coefficients of the

interaction term (child of migrant and sex of child) and the triple interaction term, which

tests for gender bias in male-headed households (from Equation 20), suggest that in male-

headed households, parental migration results in male children being more likely to have

good study habits (4%) but also more likely to be sickly (less than 1%) and

temperamental (1%) than female children (Table 6, Panel D).

5. Conclusion

The goal of this paper is to examine the effects of parental migration on the

welfare of the children left behind. This topic is relevant because the Philippines is a

leading exporter of labor, with millions of migrant workers living in other countries,

many of whom have families at home in the Philippines. The impact of parental

Page 26 of 28
migration is interesting to analyze because it has been shown to have both positive effects

(through increased income to the household) and negative effects (due to parental

absence) on child welfare.

Our findings are supportive of the literature that claims that migration improves

children’s welfare—or at the very least, does not diminish it. The clearest positive effects

we observe are in education and labor, as children of migrants are more likely to reach a

higher level of educational attainment, more likely to have good study habits, less likely

to have poor grades, and less likely to have worked in the past week and in the past year.

These results, which are robust across three different econometric models—OLS, probit,

and PSM—may imply that the income effect from remittances compensates for lack of

parental presence and involvement.

Parental migration also improves the health of the left-behind children, albeit only

marginally. PSM results suggest that the children of migrant parents are less likely to be

sickly and temperamental compared to the children of non-migrant parents. The result for

psychological health (temper) is robust across PSM and biprobit.

Our findings also suggest that aside from the income effect of remittances, the

positive impact of parental migration on the welfare of left-behind children may also be

ascribed to the gender of the household head. For example, our results suggest that, when

they live in female-headed households, left-behind children are less likely to have poor

grades and more likely to reach a higher grade level. The gender of the child also matters;

at least for good study habits, where boys appear to perform better than girls.

Gender bias is also observed in three different outcomes. First, boys have better

study habits when they live in male-headed households. Second, girls are less likely than

Page 27 of 28
boys to be physically sick in both male-headed and female-headed households. Third,

girls are perceived as more temperamental when they live in female-headed households,

while the same is true for boys living in male-headed households.

Our research could be improved in the following ways, provided that the data

become available. First, inclusion of parents’ educational attainment could improve our

IV and PSM results, because this variable could potentially influence the probability of

migration. Second, the following variables could be measured in actual terms instead of

as perceptions: grades, sickliness, and temper—especially as these subjective outcomes

are only observed in less than 1% of the total number of children in our sample. Third,

considering the actual amount of remittances and the length of parental absence would

provide a deeper and broader analysis of the impact of parental migration on children’s

welfare. Fourth, it would be interesting to explore further the heterogeneity in the impact

of migration of parents on their children in terms of gender differences across the

household heads and across the children. Fifth, it would also be of interest to study

whether the development of new long-distance communication technology has helped

mitigate the adverse effects of parental absence from children’s lives.

Page 28 of 28

You might also like