impact of foreign maids on female labor force participation in malaysia

Impact of Foreign Maids on Female LaborForce Participation in Malaysia*

Peck-Leong Tan and John Gibson

Received 23 March 2011; accepted 10 December 2012

Demand for foreign maids by households in Malaysia has increased rapidly andexpected to affect female labor force participation. Simple comparisons of house-holds with maids and those without suggest that maids raise the participation rateof their female employers by 26 percentage points. However, such comparisonsare not directly comparable. Households employing maids differ in many waysfrom those that do not. When propensity score matching methods are used toestimate the treatment effect of having a foreign maid, the female labor forceparticipation rate is estimated to be only 18 percentage points higher in 1993/1994and 13 percentage points higher in 2004/2005. This decline over time in thetreatment effect is not apparent when simpler but potentially biased methods areused. The small and declining impact of hiring maids suggests financial losses tothe host households.

Keywords: female labor force participation, household production, foreign maids,maternal employment, migration, treatment effects.

.JEL classification codes: J21, J61.

doi: 10.1111/asej.12008

I. Introduction

One of the most important migration trends in Asia is the increasing number ofyoung women migrating to rapidly growing countries like Singapore, Hong Kongand Malaysia to work as domestic workers or maids (Wickramasekera, 2002).This international movement of young women also reflects deliberate policies ofhost countries wanting to raise the labor force participation of domestic educatedwomen, by importing domestic services. For example, in Malaysia, the officialdata recognize some 320 000 foreigners working as domestic workers or maids(henceforth, ‘maids’) under a policy that aims to enable more Malaysian mothersto work in the paid labor market (Ministry of Finance, 2007).

*Tan:Faculty of Business Management, Universiti Teknologi MARA, 40450 Shah Alam, Selangor,Malaysia. Email: [email protected]. Gibson (corresponding author): Department of Economics,University of Waikato, Private Bag 3105, Hamilton, New Zealand. Email: [email protected] are grateful for financial support from the Universiti Teknologi MARA and the Ministry of HigherEducation, Malaysia and for assistance from the Department of Statistics, Malaysia. We thank PhilipMcCann, Steven Lim and an anonymous referee for their valuable comments. The paper was com-pleted while Gibson was a Visiting Fellow at the Center for Global Development, whose hospitalityis much appreciated.

bs_bs_banner

Asian Economic Journal 2013, Vol. 27 No. 2, 163–183 163

© 2013 The AuthorsAsian Economic Journal © 2013 East Asian Economic Association and Wiley Publishing Pty Ltd

This growing reliance on importing maids to perform domestic services reflectstwo factors. First, women in countries such as Malaysia have become moreeducated and so they are more likely to join the labor force and pursue a careerrather than staying at home solely as a housewife. Second, working women inAsia also are expected to play household production roles, unlike men. In par-ticular, working women are still expected to care for their family and undertakehousehold chores. Therefore, foreign maids are employed as a way to releasewomen from childcare and household chores, leaving them more time to pursuetheir career. According to Cortés and Tessada (2011), temporary migration offoreign maids has lowered the cost of household services and, thus, enabled morewomen, especially highly-skilled women, to work longer hours and to becomesuccessful in their career.

However, the impact of foreign maids on labor force participation of women inthe host countries remains unclear. Past studies from the Asian region showmixed effects of maids on the female labor force participation of their employers.For example, one study in Hong Kong showed that their presence increasedmarried women’s labor force participation by 22 percentage points (Suen, 1994).However, further study in Hong Kong by Chan (2006) found that the effect ofmaids varies according to household income and women’s educational attain-ment, with increases in activity rates that range from 16 to 73 percent. Con-versely, in Singapore, a study by Guha (2007) shows that increasing the use ofmaids does not necessarily increase the labor force participation of women fromtheir employing households.1

These mixed results may be due to an imperfect treatment of sample selectionissues in the existing literature. Simple comparisons of women in households withand without maids may be misleading if the two groups differ systematically.Sample selectivity arises because it is a choice of the household whether toemploy a maid, and factors correlated with this choice may also be correlated withwomen’s labor force participation, leading to biased estimates if this correlationis not accounted for (Heckman, 1979). This selection problem may matter inparticular for Malaysia, where there are rules for hiring foreign maids that stipu-late that employers must not only be financially able to pay for but also mustprove a need for a maid, such as having young children or elderly residents athome. Hence, the set of households that do not have a maid are not all equallylikely to provide a counterfactual for the outcomes that would have occurred forthe households with maids, if those households hiring maids had in fact not hiredthem. Therefore, econometric estimators that use all non-maid households to forma counterfactual may not provide robust estimates of the impact of maids onfemale labor force participation.

The aim of this paper is to investigate the impact of foreign maids on the femalelabor force participation rate in Malaysia. In contrast to previous research in this

1 In addition to these positive studies, a normative study suggests that admitting 7 percent moreforeign maids into Singapore could potentially increase welfare among natives, by an amount equiva-lent to a 1.1 percent increase in GDP (Kremer and Watt, 2009).

ASIAN ECONOMIC JOURNAL 164


area, a matching technique is used because the econometric techniques of previ-ous studies may result in some biases. The present study uses Household Expen-diture Survey (HES) data collected by the Department of Statistics, Malaysiaapproximately every 10 years. We compare estimates using the 1993/1994 HESwith those from the 2004/2005 HES. The matching results suggest that theaverage treatment effect of foreign maids on the female labor force participationrate has fallen from 0.18 in 1993/1994 to 0.13 in 2004/2005. Thus, the probabilitythat women who have maids at home will join the labor force has decreased overthe decade studied. This decrease would not be observed if we were to use theprobit method, which is the (potentially biased) approach relied upon in theexisting literature.

These findings should be of broad interest, given that relatively few studiesexamine the impact of migrant maids on female labor force participation in thehost countries. These findings are especially relevant for Malaysia and Indonesiabecause there are at least 0.3 million Indonesian maids working in Malaysia.These Indonesian women are negatively perceived by the Malaysian public:maids have been accused of abusing children and elderly under their care. At thesame time, stories of maids being abused have been widely reported (Chin, 1997,2005; Ariffin, 2001). Consequently, both governments periodically restrict andrelax migration policies for Indonesian maids. Yet, despite such stories, there hasbeen little formal study of this important bilateral migration flow. The currentpaper aims to fill this gap.

The structure of this paper is as follows. In Section II, a brief background onforeign maids in Malaysia is provided and the data used in the study are described.The methodology is outlined in Section III. Section IV discusses the empiricalresults and Section V concludes.

II. Background

II.1 Foreign maids in Malaysia

Malaysia is one of the most developed countries in South-East Asia and is rankedthird in real GDP per capita among ASEAN members (International LabourOrganization, 2007). Over the past 25 years, annual GDP growth has averaged5–6 percent and Malaysia aims to be a developed country by the year 2020. Thisvision has spurred the country to raise productivity and the labor force participa-tion rate, especially for women. Moreover, the female adult literacy rate increasedfrom 64.7 percent in 1980 to 94.7 percent in 2004 and women’s gross enrolmentrate in public universities increased from 36 percent in 1980 to 60 percent in 2006(Ministry of Women Family and Community Development, 2007). Under theNinth Malaysian Plan, from 2006 to 2010, development of human capital is oneof the main thrusts to empower Malaysians with knowledge that will increasetheir productivity and output. Hence, there should be a continuation of the rapid

FOREIGN MAIDS AND LABOR FORCE PARTICIPATION 165


increase in the educational attainment of women, allowing more women to jointhe labor force in future.

However, Malaysian women must perform household production roles, such aschildcare and household duties, even if they choose to join the labor force. Thebalance between household production and contributing to economic growth is wors-ened by the reduction in the average family size over the past two decades, from 5.2in 1980 to 4.7 in 1994 due to the increase in the nuclear family structurein place of extended families (Ministry of Women Family and CommunityDevelopment, 2007). Hence, there is less scope for grandparents to relieve workingmothers of some of their household production tasks. One way to cope with thesecompeting demands is to have someone else take over the non-economic householdroles. Thus, over the last two decades, many Malaysians have hired foreign maids.

At the same time, the large gap in wages between Malaysia and its neighbors,such as Indonesia and Philippines, has attracted migrants from these countries.There were approximately 1.8 million foreign workers in Malaysia by the end of2006 (Ministry of Finance, 2007). These migrants are mostly unskilled and occupypositions that locals refuse to work, particularly in agriculture, construction andservices, such as domestic services (Wickramasekera, 2002; Kanapathy, 2006).Consequently, the employment of foreign maids by Malaysian households hasincreased rapidly, from 4000 employees in the early 1980s, to some 320 000 by theend of 2005 (Ministry of Finance, 2007). Many of these maids are from Indonesiabecause of the many similarities in culture, language and religion between the twocountries.2 Yet, despite these many similarities with the locals, foreign maids arebelieved to create new social problems (e.g. abuse of children and elders, and tocommitting petty crime). As such, foreign maids are viewed negatively by thepublic but are still considered a necessary to sustain a modern family life (Chin,1997). However, only a handful are highly valued by their employers (Ooi, 2007).Nevertheless, at least from an economic point of view, their presence in thehousehold should potentially allow up to 320 000 more married Malaysian womento participate in the workforce. This potential increment to the labor force followsfrom the criteria for hiring a maid, which is that there is some dependent person in

2 The supply of foreign maids is impacted by policy determined by both sending and receivingcountries (Guha, 2007). Receiving countries like Singapore and Malaysia can influence the number ofmaids hired, and their net price, by imposing a levy and also by limiting the hiring of maids to certaintypes of qualifying households. For example, the levy for foreign maids in Malaysia is approximatelyRM360 (USD100) per year. This levy has been held at the same rate since 1995 and so should havebecome less binding. On the supply side, sending countries may impose quotas and temporarymoratoria to halt the supply of new maids, as occurred in Indonesia in July 2009. The extent to whicha moratorium from a single country affects the market for maids in Malaysia depends on the degreeof substitutability between maids from various countries. With our data we cannot estimate thissubstitutability but future work could attempt to do so. Therefore, we assume that because Malaysianemployers can also hire maids from Vietnam, Cambodia, Laos, Thailand, India and Sri Lanka(Immigration Department of Malaysia, 2010), actions from a single supplying country do not mark-edly change the cost of maids in Malaysia and are not the main factor determining either theemployment rate of maids or their impact on Malaysian women’s labor force participation.



the household; previously a married woman may have cared for that dependent butonce a maid is hired that married woman is able to enter the paid labor force. Femaleeducation has risen rapidly in Malaysia, and the mothers released into theworkforce should raise the overall productivity, with less educated migrantworkers carrying out unskilled household duties, while more educated Malaysianwomen undertake skilled tasks in the labor market.

Despite the scope for increasing numbers of foreign migrants to enable therelease of more Malaysian women into the labor force, the female labor forceparticipation rate has remained just below 50 percent for the past 25 years.Therefore, the presence of foreign maids might not have had a positive impact onMalaysian women’s decision to work. In this case, it may not be economicallyefficient to allow foreign maids into Malaysia unless the objective of such a policyis to increase non-income dimensions of welfare, such as women’s leisure time,for Malaysian households. A thorough investigation is necessary to measure theimpact of foreign maids on female labor force participation in Malaysia.

II.2 Data description

This study uses HES data collected by the Department of Statistics, Malaysiaapproximately every 10 years. The first survey was carried out in 1993/1994 andthe second survey in 2004/2005. The HES captures not only the expenditure ofhouseholds but also their demography and the economic activities of eachmember. This feature makes HES data suitable for estimating the change in theprobability of women joining the labor force in response to their householdemploying a foreign maid.3

The 1993/1994 and 2004/2005 surveys collected information from 14 627 and11 227 households, respectively. However, for the purpose of this study, thesample is restricted to households with women of working age (16 to 64 yearsold) and currently not studying. As such, a total of 10 722 and 8869 householdsfor 1993/1994 and 2004/2005 were selected. In this study, we focus on femaleforeign maids who work and live in the household of their employer. Under thisdefinition, there are some 184 and 206 sampled households with foreign maids in1993/1994 and 2004/2005, respectively. These represented 1.3 and 1.8 percent ofall households in these years.

III. Methodology

Households with foreign maids may have different characteristics to other house-holds. This is especially true in Malaysia, where rules stipulate that, to hire a

3 Because Malaysia conducts a separate Household Income/Basic Amenities Survey (which is noteasily available to researchers), the HES has only limited income information. In particular, it does notallow the modeling of labor/leisure tradeoffs. Therefore, the full range of income and non-incomeimpacts from the choice of households to employ a maid cannot be studied with the available data andremains a topic for future research.



foreign maid, employers must not only have the financial wherewithal to pay amaid but also must prove a need for a maid, for example by having young childrenor elderly at home. Therefore, in order to have the most valid estimates of theimpact of maids on female labor force participation, it is necessary to restrictthe comparisons to households with similar characteristics that employ and thatdo not employ foreign maids.

Although regression models such as OLS and probit models could control forthe differences in average characteristics when studying outcomes for householdswith and without maids, these regression methods may be less successful indealing with the sample selection problem when subjects in non-experimentalstudies cannot be randomly assigned to ‘treatment’ and ‘control’ groups(Heckman, 1979). Such problems are relevant to attempts to measure the impacton labor force participation of women in households with maids because onlyfinancially-able households who meet the immigration requirements are meant toemploy foreign maids. Such a group are likely to be a non-random sample.

Obviously, it is impossible to observe both outcomes (employing a foreignmaid and not employing a foreign maid) for the same household at the sametime. Likewise, taking the mean outcome of households without maids as acounterfactual for what households with maids would be like without their maidis not satisfactory because the two groups of households may be quite different.However, the matching approach is one possible solution to the selectionproblem. Intuitively, this approach attempts to find a group of non-participants(households without maids) who are similar to the participants (households withmaids) in all relevant pre-treatment characteristics, under the assumption ofconditional independence (Rosenbaum and Rubin, 1983). This assumptionrequires that any differences in covariates between the control and treatmentgroups are independent of the treatment, and, hence, the differences in the out-comes are assumed to be only as a result of being given the treatment.4

Validation studies suggest that matching, and, in particular, propensity scorematching (PSM), can produce results close to experimental benchmarks if usedappropriately (Dehijia and Wahba, 2002). Using PSM to estimate the impact offoreign maids on female labor force participation requires first estimating a probitequation for the probability of a household having a foreign maid. The resultinghousehold propensity score is then attached to all working age women in thehousehold. Comparisons are then restricted to individuals from treatment andcontrol group households who have similar values of these propensity scores (thisis the ‘common support’ condition). A comparison of the two matched samples(households with maids and those without but having a similar probability ofhaving a maid) then gives an estimate of the ‘average treatment effect on thetreated’ (ATT). In this case, the ATT is an estimate of the likelihood that womenin households with foreign maids would have participated in the labor force if

4 In addition to this conditional independence assumption, which is also known as‘unconfoundedness’, a ‘balancing property’ is required, which is that, conditional on the propensityscore, the means of the covariates should be identical across the treatment and control groups.



they did not have a maid. Thus, PSM offers a way of structuring non-experimentaldata to look like experimental data, where for every subject in the ‘treated’ groupthe researcher finds comparable subjects in the ‘control’ group.

The available matching approaches include matching each treated observation,i, to the nearest neighbor(s) from the control group, kernel matching where aweighted average of the j control group neighbors is taken with weights propor-tional to the closeness of propensity scores for i and j, and radius matching wherecontrol group observations are used if their propensity scores are within a radiusr of the propensity score for treated observation i (Caliendo and Kopeinig, 2008).In this paper, both kernel matching and radius matching are used, and comparingthe two may indicate if these are robust estimates of treatment effects.

In the first step, a probit model is used to estimate the odds of each householdhiring a foreign maid. The covariates are the number of school children, toddlers,babies and elderly in the household, the number of working age women, and totalhousehold income. The propensity scores from this probit are used to restrict thecomparison of control and treatment groups to those households with similarprobabilities of being treated (i.e. having a maid). This is the imposition of the‘common support’ condition, and in the estimates reported below typically resultsin matched control group samples with only 90 percent of the original non-maidhouseholds (i.e. some non-maid households provide a poor counterfactual). In thenext step, kernel and radius matching is used to obtain the estimates of the averagetreatment effect. Kernel matching uses the weighted averages of all the individu-als in the control group to create the counterfactual outcome, while radius match-ing uses the weighted average of all individuals in the control group within theradii of 0.05.

Figure 1 displays the propensity scores (probabilities) from the probit model ofwhether households employ a foreign maid or not. The propensity scores forhouseholds with maids in 1993/1994 range from 0.00 to 0.56, with a mean of0.09. In contrast, the propensity scores for other households without maidshave a mean of 0.01, suggesting differences in the two groups, on average. Otherunreported density graphs for 2004/2005 and for different regions withinMalaysia show similar patterns of some households without maids having muchlower propensity scores, making members of these households less plausiblecounterfactuals for women in the households that do hire a foreign maid.

In addition to the propensity scores from the full sample, another set ofpropensity scores are estimated, restricting the sample to households withmonthly total income above RM3000.5 This follows from an institutional featurethat should create a discontinuity. The Malaysian immigration authorities stipu-late that only households with monthly income of more than RM3000 may hireforeign maids. Although there appears to be some violation of this requirement,with one-seventh of sampled households with maids having income below this

5 This is equivalent to USD1154 (USD1 = RM2.60) in 1993/1994 and USD790 (USD1 = RM3.80)in 2004/2005.



level, the requirement does lower the probability of households below the thresh-old having a maid so use of this extra information can help to form a moreconvincing counterfactual group.

In addition to the PSM estimates, we use probit regressions to estimate thetreatment effect of maids on women’s work status. In previous published studies,the probit regression is the main tool used to study the effect of maids on women’slikelihood of joining the labor force (Suen, 1994; Chan, 2006). Because the probitmodel does not provide a way to restrict the comparisons just to the commonsupport region, it may therefore suffer from selectivity bias, and so its resultsmight not be the same as those from PSM. Therefore, comparing the two sets ofresults might be informative in regards to possible biases in previous resultsreported in the literature.

IV. Results

IV.1 Descriptive statistics

Table 1 reports descriptive statistics for both 1993/1994 and 2004/2005, forMalaysia and the two main sub-areas, West Malaysia and East Malaysia. Descrip-tive statistics are also presented for the sample of households with income aboveRM3000 for reasons explained in the previous section. In the full sample, the

Figure 1 Propensity scores for households with and without maids: Malaysia1993/1994

0

5

10

15

0 0.2 0.4 0.6 0.8

Predicted probability

Household with maid Household without maid



percentage of women in the labor force has decreased from 48 percent in 1993/1994 to 43 percent in 2004/2005. The decrease is more substantial in EastMalaysia, down 8 percentage points from 45 percent in 1993/1994 to 37 percentin 2004/2005. Female labor force participation rates are higher among richer

Table 1 Households’ characteristics by region and year 1993/1994 and 2004/2005

Malaysia West Malaysia East Malaysia

1993/1994 2004/2005 1993/1994 2004/2005 1993/1994 2004/2005

Full samplea

Female labor forceparticipation rate

0.478 0.434 0.486 0.453 0.447 0.366(0.500) (0.496) (0.500) (0.498) (0.497) (0.482)

Average monthly nominalincome for workingwomen

703 1238 684 1217 794 1332

(1019) (1271) (1033) (1294) (941) (1159)

Average total monthlyhousehold income

2249 3051 2153 2988 2720 3313(3217) (3486) (2762) (3525) (4839) (3305)

Probability of householdhiring foreign maid

0.017 0.023 0.012 0.021 0.042 0.034(0.130) (0.150) (0.109) (0.142) (0.201) (0.182)

Average household size 5.114 4.921 5.073 4.86 5.33 5.172(2.203) (2.040) (2.167) (1.981) (2.363) (2.252)

Number of women in laborforce (16 to 64 years old)

1.389 1.328 1.370 1.299 1.484 1.451(0.717) (0.653) (0.697) (0.615) (0.800) (0.779)

Number of children aged 0to 2 years

0.402 0.302 0.393 0.29 0.445 0.35(0.623) (0.536) (0.614) (0.525) (0.663) (0.575)


0.543 0.467 0.537 0.46 0.567 0.498(0.752) (0.689) (0.746) (0.687) (0.781) (0.695)


0.791 0.743 0.792 0.746 0.787 0.729(0.992) (0.930) (0.992) (0.932) (0.995) (0.925)


0.573 0.618 0.58 0.624 0.543 0.592(0.876) (0.869) (0.876) (0.873) (0.877) (0.854)

Number of children aged 18years and above

0.547 0.495 0.539 0.496 0.585 0.494(0.988) (0.874) (0.969) (0.874) (1.074) (0.875)

Number of elderly aged 56to 64 years

0.222 0.244 0.231 0.258 0.186 0.185(0.495) (0.520) (0.500) (0.532) (0.464) (0.465)

Number of elderly aged 65years and above

0.167 0.166 0.165 0.177 0.177 0.121(0.430) (0.426) (0.426) (0.436) (0.448) (0.379)

Sample size (number ofhouseholds)

10722 8869 8902 7149 1820 1720

Restricted sampleb

Probability of womenworking

0.596 0.560 0.605 0.591 0.570 0.473(0.491) (0.496) (0.489) (0.492) (0.500) (0.499)

Average monthly nominalincome for working women

1261 1852 1265 1869 1252 1796(1661) (1595) (1784) (1668) (1195) (1305)

Average total monthlyhousehold income

5106 5483 5061 5461 5255 5555(2681) (2943) (2701) (2990) (2611) (2788)

Probability of householdhiring foreign maid

0.067 0.061 0.052 0.057 0.116 0.074(0.250) (0.240) (0.221) (0.232) (0.320) (0.262)

Sample size (number ofhouseholds)

2155 2909 1645 2218 510 691

Notes: Standard deviations are in parentheses. aSample is limited to households with women aged 16 to64 years (excluding students). bSample is further restricted to households with total monthlyincome of RM3000 and above.



households, but, again, there is a decline over time, from 60 percent in 1993/1994to 56 percent in 2004/2005. The trend is most obvious in East Malaysia, where thepercentage of women in households with monthly income above RM3000 whoare in the labor force has decreased from 57 percent in 1993/1994 to 47 percentin 2004/2005. Despite declining participation rates, nominal incomes for womenwith jobs have increased faster between the two surveys than have overall house-hold incomes.

Average total household income in 2004/2005 for all regions is just overRM3000, which is the stipulated minimum income required to employ foreignmaids. Hence, average Malaysian families can afford to employ foreign maids.Nevertheless, the probability of a household hiring a foreign maid in Malaysiawas just 0.02 for both periods. This stable average hides the fact that the prob-ability of hiring foreign maids in West Malaysia increased slightly from 0.01 in1993/1994 to 0.02 in 2004/2005, but the probability of hiring foreign maids inEast Malaysia decreased from 0.04 in 1993/1994 to 0.03 in 2004/2005. Thedecline in the probability of hiring foreign maids is especially apparent for highincome (�RM3000 per month) households in East Malaysia.

Table 1 shows other descriptive statistics for selected characteristics for bothperiods and both sub-areas. First, the average number of women of working age(15 to 64 years old) per household in Malaysia decreased from 1.4 in 1993/1994to 1.3 in 2004/2005. Second, the average household size in Malaysia decreasedfrom 5.1 in 1993/1994 to 4.9 in 2004/2005, and the number of children agedbelow 12 years decreased from 0.4 in 1993/1994 to approximately 0.3 in 2004/2005. Meanwhile, the average number of those aged 56-64 years has remainedaround 0.22 and the average number of the elderly (aged 64 years and above) ineach household remains at 0.17 in both 1993/1994 and 2004/2005. In summary,the households’ demography has changed significantly from 1993/1994 to 2004/2005 and also differs according to region and income level. Therefore, empiricalanalysis must account for these changes when estimating the impacts of foreignmaids.

IV.2 Raw comparisons

Simple comparisons of households with maids and other households suggest thatemploying foreign maids raised female labor force participation rates by 26percentage points in 1993/1994 and by 24 percentage points in 2004/2005(Table 2). However, women in households with maids differ significantly fromwomen in households without maids, in terms of demography and averageincome. Specifically, women in households with maids have more young chil-dren, fewer older children and fewer elderly people living in their homes. More-over, households with maids also have larger average household size but fewerwomen in the labor force as compared to other households.

The differences between households with maids and without maids are smallerwhen the sample is restricted to households with total income of more than



Tab

le2

Hou

seho

lds

and

indi

vidu

als

char

acte

rist

ics

bypr

esen

ceof

fore

ign

mai

din

Mal

aysi

afo

rye

ars

1993

/199

4an

d20

04/2

005

Ful

lsa

mpl

eR

estr

icte

dsa

mpl

e

1993

/199

420

04/2

005

1993

/199

420

04/2

005

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Cha

ract

eris

tics

ofin

divi

dual

fem

ale

aged

16to

64ye

arsa

Fem

ale

labo

rfo

rce

part

icip

atio

nra

te0.

735

0.47

40.

000

0.67

10.

429

0.00

00.

749

0.58

60.

000

0.68

60.

553

0.00

0A

vera

gem

onth

lyno

min

alin

com

efo

rw

orki

ngw

omen

1908

672

0.00

031

4211

730.

000

1801

1170

0.00

029

7617

410.

000

Age

34.5

836

.00

0.02

838

.87

37.7

60.

090

34.7

035

.67

0.16

438

.71

37.4

10.

072

Cha

ract

eris

tics

ofho

useh

olds

wit

hw

omen

aged

16to

64A

vera

geho

useh

old

size

5.94

05.

100

0.00

05.

684

4.90

30.

000

6.01

45.

424

0.00

05.

826

5.27

30.

004

Num

ber

ofw

omen

inw

orki

ngag

egr

oup

(16–

64ye

ars

old)

a1.

293

1.39

10.

039

1.22

31.

331

0.01

41.

299

1.57

30.

000

1.23

61.

436

0.00

0

Num

ber

ofch

ildr

enag

ed0

to2

year

s0.

652

0.39

70.

000

0.50

50.

297

0.00

00.

667

0.37

40.

000

0.47

80.

321

0.00

1N

umbe

rof

chil

dren

aged

3to

6ye

ars

0.65

20.

541

0.04

40.

748

0.46

10.

000

0.63

90.

483

0.01

40.

775

0.46

40.

000

Num

ber

ofch

ildr

enag

ed7

to12

year

s0.

745

0.79

20.

481

0.86

40.

740

0.09

10.

778

0.72

20.

486

0.91

00.

774

0.09

7N

umbe

rof

chil

dren

aged

13to

18ye

ars

0.33

00.

586

0.00

00.

351

0.62

80.

000

0.31

90.

566

0.00

00.

376

0.64

70.

000

Num

ber

ofch

ildr

enag

ed18

year

san

dab

ove

0.23

60.

563

0.00

00.

190

0.50

30.

000

0.28

50.

782

0.00

00.

197

0.60

40.

000

Num

ber

ofel

derl

yag

ed56

to64

year

s0.

065

0.22

60.

000

0.17

00.

246

0.01

80.

049

0.20

30.

000

0.16

30.

208

0.17

8N

umbe

rof

elde

rly

aged

64ye

ars

and

abov

e0.

141

0.16

80.

368

0.17

00.

166

0.89

60.

162

0.16

20.

902

0.17

40.

133

0.24

9S

ampl

esi

ze:

Num

ber

ofw

omen

wit

hm

aids

b23

814

656

252

1152

518

731

6233

4922

039

2241

42

Not

es:

a Exc

ludi

ngth

ose

who

are

curr

entl

yst

udyi

ng.

b The

rear

efe

wer

hous

ehol

dsth

anin

divi

dual

wom

en.



RM3000. First, there is a smaller apparent increase in female labor force partici-pation from having a foreign maid, of 16 and 13 percentage points for 1993/1994and 2004/2005. The average household size for households with maids is only 0.6more than for households without maids, compared to a difference of more than0.8 in the full sample. Households with maids and without maids are not signifi-cantly different in terms of the number of elderly and the number of children aged7 to 12 years old.

Tables 3 and 4 contain similar comparisons of the characteristics of householdswith and without foreign maids for the subsamples from the two different regionsin Malaysia. The impact of foreign maids appears larger in West Malaysia,especially in 2004/2005. However, there are also many differences in householdcharacteristics between those with and those without maids, so multivariateanalysis is required to estimate the effect of foreign maids. In particular, it isapparent that the average number of young children (0–2 years old) fell over the11-year period covered, reflecting the lower fertility rate and smaller family sizeas a result of Malaysia completing the demographic transition. It is notable alsothat the use of the restricted sample and focusing on the sub-regions may narrowthe differences between households with and without maids, leading to moreobservations in the common support (i.e. households with similar values of thepropensity score), and therefore potentially giving a more robust estimate of theaverage treatment effect.

IV.3 Propensity score matching estimates

The estimated ATT obtained via PSM is shown in Table 5. Five different groupsof covariates (number of members in various demographic groups, dummy vari-ables for household types, monthly incomes and their interactions) are used toobtain the propensity scores, to determine if the ATT estimates are robust.6

Results are reported for each year, for each of the two main regions of Malaysiaand also separately for the restricted sample with total income of RM3000 permonth or higher.

Foreign maids have positive impacts on female labor force participation,although this impact appears to have fallen over the period. In the full sample, andwhen using only demographic variables for matching, the estimated ATT isbetween 0.24 and 0.26 in 1993/1994 and between 0.18 and 0.24 in 2004/2005. Itis notable that the estimated ATT falls sharply, by approximately 0.06 in 2004/2005, once monthly household income is introduced as one of the matchingcovariates (Models 4 and 5).

In the regional results, the ATT for West Malaysia is between 0.24 and 0.27 in1993/1994 and between 0.22 and 0.29 in 2004/2005. There is a much smallerdecrease over time in the ATT in West Malaysia than in East Malaysia (using

6 Full details on the covariates are listed at the foot of Table 5. The selected sets of covariates foreach model all obey the balancing property required for credible PSM estimates.



Tab

le3

Hou

seho

ldan

din

divi

dual

char

acte

rist

ics

bypr

esen

ceof

fore

ign

mai

din

Wes

tM

alay

sia

for

year

1993

/199

4an

d20

04/2

005

Ful

lsa

mpl

eR

estr

icte

dsa

mpl

e

1993

/199

420

04/2

005

1993

/199

420

04/2

005

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Cha

ract

eris

tics

ofin

divi

dual

wom

enag

ed16

to64

year

sa

Fem

ale

labo

rfo

rce

part

icip

atio

nra

te0.

750

0.48

30.

000

0.73

90.

447

0.00

00.

755

0.59

60.

000

0.74

50.

582

0.00

0A

vera

gem

onth

lyno

min

alin

com

efo

rw

orki

ngw

omen

2098

670

0.00

034

3111

510.

000

1928

1170

0.00

031

5817

440.

000

Age

35.4

836

.35

0.31

639

.47

38.3

60.

170

35.0

635

.89

0.35

839

.45

37.6

10.

037

Cha

ract

eris

tics

ofho

useh

olds

wit

hw

omen

aged

16to

64ye

ars

Ave

rage

hous

ehol

dsi

ze5.

860

5.06

40.

000

5.46

24.

847

0.00

05.

976

5.35

30.

003

5.59

95.

208

0.02

0N

umbe

rof

wom

enin

wor

king

age

grou

p(1

6–64

year

sol

d)a

1.27

11.

371

0.10

21.

122

1.30

20.

000

1.29

41.

537

0.00

01.

142

1.39

40.

000

Num

ber

ofch

ildr

enag

ed0

to2

year

s0.

570

0.39

10.

011

0.49

00.

286

0.00

00.

552

0.36

60.

018

0.46

50.

317

0.00

8N

umbe

rof

chil

dren

aged

3to

6ye

ars

0.64

50.

536

0.11

50.

776

0.45

40.

000

0.64

70.

474

0.02

30.

811

0.47

00.

000

Num

ber

ofch

ildr

enag

ed7

to12

year

s0.

757

0.79

30.

670

0.78

90.

746

0.58

60.

812

0.72

40.

376

0.83

50.

787

0.59

3N

umbe

rof

chil

dren

aged

13to

18ye

ars

0.30

70.

592

0.00

00.

326

0.63

00.

000

0.36

50.

556

0.04

40.

339

0.65

30.

000

Num

ber

ofch

ildr

enag

ed18

year

san

dab

ove

0.20

10.

554

0.00

00.

181

0.50

50.

000

0.18

80.

749

0.00

00.

173

0.58

90.

000

Num

ber

ofel

derl

yag

ed56

to64

year

s0.

084

0.23

30.

000

0.17

70.

260

0.03

40.

047

0.20

30.

000

0.16

50.

203

0.34

2N

umbe

rof

elde

rly

aged

64ye

ars

and

abov

e0.

131

0.16

60.

295

0.16

30.

177

0.71

80.

141

0.15

40.

737

0.17

30.

132

0.33

7S

ampl

esi

ze:

Num

ber

ofw

omen

wit

hm

aids

b13

612

058

165

9118

110

2398

152

2940

Not

es:

a Exc

ludi

ngth

ose

who

are

curr

entl

yst

udyi

ng.

b The

rear

efe

wer

hous

ehol

dsth

anin

divi

dual

wom

en.



Tab

le4

Hou

seho

lds

and

indi

vidu

als

char

acte

rist

ics

bypr

esen

ceof

fore

ign

mai

din

Eas

tM

alay

sia

for

year

1993

/199

4an

d20

04/2

005

Ful

lsa

mpl

eR

estr

icte

dsa

mpl

e

1993

/199

420

04/2

005

1993

/199

420

04/2

005

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Wit

hm

aid

Wit

hout

mai

dp-

valu

efo

run

equa

lm

eans

Cha

ract

eris

tics

ofin

divi

dual

wom

enag

ed16

to64

year

sa

Fem

ale

labo

rfo

rce

part

icip

atio

nra

te0.

716

0.43

60.

000

0.54

00.

360

0.00

10.

740

0.55

50.

001

0.57

30.

467

0.07

7A

vera

gem

onth

lyno

min

alin

com

efo

rw

orki

ngw

omen

1643

740

0.00

023

9112

740.

000

1614

1167

0.00

425

1917

320.

001

Age

33.3

734

.38

0.30

037

.72

35.4

70.

047

34.1

734

.97

0.47

837

.28

36.8

50.

730

Cha

ract

eris

tics

ofho

useh

olds

wit

hw

omen

aged

16to

64ye

ars

Ave

rage

hous

ehol

dsi

ze6.

052

5.28

30.

001

6.23

75.

134

0.01

76.

068

5.67

20.

172

6.39

25.

484

0.08

5N

umbe

rof

wom

enin

wor

king

age

grou

p(1

6-64

year

sol

d)a

1.32

51.

491

0.03

01.

475

1.45

00.

848

1.30

51.

694

0.00

01.

471

1.57

30.

473

Num

ber

ofch

ildr

enag

ed0

to2

year

s0.

766

0.43

10.

000

0.54

20.

343

0.02

40.

831

0.40

00.

000

0.51

00.

338

0.07

0N

umbe

rof

chil

dren

aged

3to

6ye

ars

0.66

20.

565

0.29

00.

678

0.49

20.

058

0.62

70.

514

0.31

20.

686

0.44

50.

032

Num

ber

ofch

ildr

enag

ed7

to12

year

s0.

727

0.78

90.

583

1.05

10.

718

0.03

80.

729

0.71

20.

903

1.09

80.

731

0.04

3N

umbe

rof

chil

dren

aged

13to

18ye

ars

0.33

00.

586

0.00

00.

351

0.62

80.

000

0.25

40.

599

0.00

00.

376

0.64

70.

000

Num

ber

ofch

ildr

enag

ed18

year

san

dab

ove

0.23

60.

563

0.00

00.

190

0.50

30.

000

0.42

40.

900

0.00

10.

197

0.60

40.

000

Num

ber

ofel

derl

yag

ed56

to64

year

s0.

039

0.19

30.

000

0.15

30.

187

0.53

20.

051

0.20

40.

000

0.15

70.

223

0.28

7N

umbe

rof

elde

rly

aged

64ye

ars

and

abov

e0.

156

0.17

80.

675

0.18

60.

119

0.24

10.

204

0.18

80.

834

0.17

60.

138

0.53

8S

ampl

esi

ze:

Num

ber

ofw

omen

b10

225

9887

2407

7776

475

1007

Not

es:

a Exc

ludi

ngth

ose

who

are

curr

entl

yst

udyi

ng.

b The

rear

efe

wer

hous

ehol

dsth

anin

divi

dual

wom

en.



Tab

le5

Ave

rage

trea

tmen

tef

fect

offo

reig

nm

aids

onfe

mal

ela

bor

forc

epa

rtic

ipat

ion

inM

alay

sia,

Wes

tM

alay

sia

and

Eas

tM

alay

sia

for

1993

/199

4an

d20

04/2

005

usin

gfu

llan

dre

stri

cted

sam

ple

Ful

lsa

mpl

eR

estr

icte

dsa

mpl

e

Mal

aysi

aW

est

Mal

aysi

aE

ast

Mal

aysi

aM

alay

sia

Wes

tM

alay

sia

Eas

tM

alay

sia

1993

/199

420

04/2

005

1993

/199

420

04/2

005

1993

/199

420

04/2

005

1993

/199

420

04/2

005

1993

/199

420

04/2

005

1993

/199

420

04/2

005

Mod

el1

0.26

00.

241

0.26

60.

292

0.28

00.

181

0.16

40.

133

0.15

90.

163

0.19

00.

107

(0.0

31)*

**(0

.030

)***

(0.0

39)*

**(0

.035

)***

(0.0

46)*

**(0

.062

)***

(0.0

33)*

**(0

.030

)***

(0.0

45)*

**(0

.037

)***

(0.0

56)*

**(0

.057

)*M

odel

20.

260

0.24

10.

266

0.29

00.

278

0.18

10.

156

0.13

30.

148

0.16

20.

174

0.10

6(0

.032

)***

(0.0

28)*

**(0

.037

)***

(0.0

38)*

**(0

.039

)***

(0.0

54)*

**(0

.034

)***

(0.0

30)*

**(0

.041

)***

(0.0

37)*

**(0

.055

)***

(0.0

62)*

Mod

el3

0.26

10.

241

0.26

60.

290

0.27

00.

183

0.15

60.

130

0.15

00.

161

0.18

20.

103

(0.0

34)*

**(0

.031

)***

(0.0

38)

(0.0

35)*

**(0

.046

)***

(0.0

52)*

**(0

.035

)***

(0.0

36)*

**(0

.043

)***

(0.0

34)*

**(0

.057

)***

(0.0

53)*

*M

odel

40.

239

0.17

60.

247

0.22

10.

253

0.09

10.

175

0.12

70.

174

0.16

10.

180

0.07

4(0

.029

)***

(0.0

29)*

**(0

.042

)***

(0.0

40)*

**(0

.040

)***

(0.0

57)*

(0.0

31)*

**(0

.032

)***

(0.0

45)*

**(0

.038

)(0

.054

)***

(0.0

69)

Mod

el5

0.24

40.

175

0.24

10.

225

0.25

80.

043

0.18

20.

129

0.18

40.

163

0.20

90.

058

(0.0

29)*

**(0

.029

)***

(0.0

39)*

**(0

.032

)***

(0.0

52)*

**(0

.065

)(0

.031

)***

(0.0

33)*

**(0

.041

)***

(0.0

38)*

**(0

.058

)***

(0.0

71)

Not

es:

***,

**an

d*

repr

esen

tsi

gnifi

canc

eat

the

1,5

and

10%

leve

l,re

spec

tive

ly.

For

full

sam

ple:

N=

238

trea

ted

in19

93/1

994

and

from

N=

1344

7to

1465

6co

ntro

lgr

oup

mem

bers

(var

ying

acro

ssth

em

odel

sbe

caus

eth

eco

mm

onsu

ppor

tch

ange

sw

ith

diff

eren

tco

vari

ates

).N

=25

2tr

eate

din

2004

/200

5an

dfr

omN

=75

26to

1152

5co

ntro

lgro

upm

embe

rs.C

ovar

iate

sfo

rM

odel

1ar

enu

mbe

rof

chil

dren

aged

0–2

year

s,nu

mbe

rof

chil

dren

aged

3–6

year

san

ddu

mm

yfo

rhou

seho

ldw

ith

prim

ary

scho

olch

ildr

en.C

ovar

iate

sfo

rMod

el2

are

Mod

el1

plus

dum

mie

sfo

rhou

seho

lds

wit

hel

derl

yag

ed56

to64

year

san

dho

useh

olds

wit

hel

derl

yag

ed65

year

san

dab

ove.

Cov

aria

tes

for

Mod

el3

are

Mod

el2

plus

dum

mie

sfo

rho

useh

olds

wit

hse

cond

ary

chil

dren

and

hous

ehol

dsw

ith

mar

ried

chil

dren

.Cov

aria

tes

for

Mod

el4

are

Mod

el3

plus

num

ber

ofw

omen

inw

orki

ngag

egr

oup

and

log

tota

lm

onth

lyho

useh

old

inco

me.

For

rest

rict

edsa

mpl

e:N

=18

7tr

eate

din

1993

/199

4an

dfr

om27

84to

3164

cont

rol

grou

pm

embe

rs,w

hile

in20

04/2

005

ther

ear

eN

=22

0tr

eate

dan

dfr

om37

37to

3922

cont

rol

grou

pm

embe

rs.C

ovar

iate

sfo

rM

odel

1ar

eho

useh

olds

wit

hch

ildr

enag

ed0–

2,3–

6an

d7–

12ye

ars.

Cov

aria

tes

for

Mod

el2

are

hous

ehol

dsw

ith

elde

rly

aged

56to

64ye

ars

and

hous

ehol

dsw

ith

elde

rly

aged

65ye

ars

and

abov

e.C

ovar

iate

sfo

rM

odel

3ar

eco

vari

ates

from

Mod

els

1an

d2.

Cov

aria

tes

for

Mod

el4

are

cova

riat

esfr

omM

odel

3an

dlo

gto

tal

mon

thly

hous

ehol

din

com

e.M

odel

5:C

ovar

iate

sfo

rM

odel

5in

both

full

and

rest

rict

edsa

mpl

esar

eto

tal

num

ber

ofch

ildr

enbe

low

12ye

ars,

log

tota

lmon

thly

hous

ehol

din

com

ean

din

tera

ctio

nbe

twee

nth

ese

two

cova

riat

es.T

heco

vari

ates

for

allfi

vem

odel

s,ap

plie

dto

all

sam

ples

and

tim

epe

riod

s,ar

eba

lanc

edbe

twee

nth

etr

eatm

enta

ndco

ntro

lgro

ups

(i.e

.wit

hin

bloc

ksde

fine

dby

the

prop

ensi

tysc

ores

,the

cova

riat

esha

veeq

ual

mea

nsfo

rtr

eatm

ent

and

cont

rol

grou

ps).



Model 5, the ATT declines by 0.01 in West Malaysia and by 0.22 in EastMalaysia).7 All these declines in the ATT, for Malaysia and for each region, aresignificant at the 1-percent level (columns 2 and 6, Table 6). The falling treatmenteffect of foreign maids on the female labor force participation rate in Malaysiasuggests that maids may, increasingly, be augmenting Malaysian women’s leisuretime rather than enabling them to either join or rejoin the labor force.

The results in the last panel of Table 5 come from the restricted sample, withtotal monthly income exceeding RM3000 (the stipulated requirement for hiringmaids). The fulfillment of the common support assumption is more likely for thisrestricted sample because the sample restriction rules out lower income house-holds who, legally, should have no chance of obtaining the treatment (Caliendoand Kopeinig, 2008). Consequently, the estimated ATT on this sample is morerobust, in the sense of varying less with the choice of covariates, and is lower thanthe ATT from the full sample. Foreign maids are estimated to have increased thelikelihood of labor force participation by their female employers in these richerhouseholds by approximately 0.18 in 1993/1994 and by only 0.13 in 2004/2005.This decline in the average treatment effect is most apparent in East Malaysia.

IV.4 Probit model

Table 7 reports marginal effects from probit models of women’s labor forceparticipation. According to the estimates, having a maid increased the probabilityof women participating by 0.18 in 1993/1994 and by 0.14 in 2004/2005. When theanalysis is restricted to women in households with total monthly income exceed-ing RM3000, the impact of maids on the probability of participating is 0.16 inboth years. This result contrasts with the finding of a declining treatment effect inthe PSM estimates, with the comparison between the two sets of results reportedin Table 6. The different trends with the two approaches may occur because,unlike the PSM results, the probit estimates use all non-maid households (whethergood counterfactuals or not) and have no mechanism for dealing with sampleselection issues.

The other covariates in the probit models show some interesting patterns thatare not highlighted in the PSM results (because the covariates are just used toform the propensity scores prior to the matching and for reasons of brevity are notreported here). There is a higher female labor force participation rate in WestMalaysia, with this pattern becoming stronger in the second survey. Women withhigher education levels are more likely to participate, as are those who areunmarried, and (in the full sample) come from richer households. Compared withthe excluded category of households with no children and no elderly, the femalelabor force participation rate is lower if there are children, especially in theyounger age groups.

7 Two differences between West Malaysia and East Malaysia that may account for this pattern arethat income has risen more rapidly in West Malaysia (rising from 79% of the average in East Malaysiain 1993/1994 to 90% of the average by 2004/2005) and stricter law enforcement is likely to havedecreased the number of maids in East Malaysia informally migrating from neighboring Kalimantan.



Tab

le6

Cha

nges

over

tim

ein

aver

age

trea

tmen

tef

fect

sof

fore

ign

mai

dson

the

fem

ale

labo

rfo

rce

part

icip

atio

nra

tein

Mal

aysi

a

Ful

lSa

mpl

eR

estr

icte

dSa

mpl

e

Mea

ndi

ffere

nce

Pro

bit

mod

elP

SM:

Ker

nel

mat

chin

g(M

odel

5)M

ean

diffe

renc

eP

robi

tm

odel

PSM

:K

erne

lm

atch

ing

(Mod

el5)

PSM

:R

adiu

sm

atch

ing

(0.0

5ra

dii)

Mal

aysi

a19

93/1

994

0.26

10.

181

0.24

40.

163

0.15

50.

182

0.15

7(0

.037

)***

(0.0

29)*

**(0

.036

)***

(0.0

31)*

**(0

.033

)***

2004

/200

50.

242

0.14

40.

176

0.13

30.

158

0.12

90.

128

(0.0

43)*

**(0

.029

)***

(0.0

39)*

**(0

.033

)***

(0.0

28)*

**D

iffe

renc

ea-0

.019

-0.0

37-0

.068

-0.0

30.

003

-0.0

53(0

.009

)**

(0.0

03)*

**(0

.002

)***

(0.0

04)*

**(0

.002

)(0

.002

)***

Wes

tM

alay

sia

1993

/199

40.

267

0.18

00.

241

0.15

90.

164

0.18

40.

146

(0.0

37)*

**(0

.039

)***

(0.0

40)*

**(0

.041

)***

(0.0

46)*

**20

04/2

005

0.29

20.

144

0.22

50.

163

0.16

00.

163

0.15

7(0

.043

)***

(0.0

32)*

**(0

.040

)***

(0.0

38)*

**(0

.038

)***

Dif

fere

ncea

0.02

5-0

.036

-0.0

160.

004

-0.0

04-0

.021

(0.0

11)*

*(0

.003

)***

(0.0

02)*

**(0

.005

)(0

.002

)**

(0.0

03)*

**E

ast

Mal

aysi

a19

93/1

994

0.28

00.

169

0.25

80.

185

0.16

80.

209

0.18

3(0

.033

)***

(0.0

52)*

**(0

.041

)***

(0.0

58)*

**(0

.055

)***

2004

/200

50.

180

0.14

30.

043

0.10

60.

169

0.05

80.

105

(0.0

44)*

**(0

.065

)(0

.043

)***

(0.0

71)

(0.0

61)*

Dif

fere

ncea

-0.1

00-0

.026

-0.2

15-0

.079

0.00

1-0

.151

(0.0

16)*

**(0

.003

)***

(0.0

04)*

**(0

.007

)***

(0.0

03)

(0.0

04)*

**

Not

es:

***,

**an

d*

repr

esen

tsi

gnifi

canc

eat

the

1,5

and

10%

leve

l,re

spec

tive

ly.

Rob

ust

stan

dard

erro

rsar

ein

pare

nthe

ses

for

the

prob

itm

odel

.B

oots

trap

ped

stan

dard

erro

rsba

sed

on10

0re

plic

atio

nsap

pear

inpa

rent

hese

sfo

rth

epr

open

sity

scor

em

atch

ing

(PS

M)

mod

el.A

vera

getr

eatm

ente

ffec

tsfo

rP

SM

mod

elar

eba

sed

onM

odel

5in

Tab

le5.

a Dif

fere

nce

=20

04/2

005–

1993

/199

4an

dbo

otst

rapp

edst

anda

rder

rors

base

don

500

repl

icat

ions

.



Table 7 Marginal effect of foreign maids on female labor force participation rate inMalaysia for year 1993/1994 and 2004/2005 using probit model

Full sample Restricted sample

1993/1994 2004/2005 1993/1994 2004/2005

Maid 0.181 0.144 0.155 0.158(0.037)*** (0.043)*** (0.036)*** (0.039)***

Region (West Malaysia = 1) 0.052 0.082 0.028 0.066(0.012)*** (0.016)*** (0.022) (0.028)**

Household characteristicsNumber of children aged 0

to 2 years-0.111 -0.108 -0.058 -0.096(0.008)*** (0.015)*** (0.015)*** (0.025)***


-0.049 -0.058 -0.035 -0.041(0.006)*** (0.012)*** (0.014)** (0.019)**


-0.003 -0.026 -0.012 -0.035(0.005) (0.009)*** (0.011) (0.016)**


-0.005 -0.000 -0.002 -0.022(0.005) (0.008) (0.011) (0.015)

Number of children agedabove 18 years

-0.018 -0.017 -0.006 0.013(0.005)*** (0.009)** (0.008) (0.013)

Number of elderly aged 56to 64 years

0.010 0.016 0.009 0.008(0.010) (0.016) (0.021) (0.027)

Number of elderly agedabove 64 years

0.025 0.042 0.041 0.026(0.011)** (0.017)** (0.022)* (0.031)

Ethnic group for head ofhousehold (Malay)

0.050 0.047 0.118 0.095(0.016)*** (0.025)* (0.032)*** (0.042)**

Ethnic group for head ofhousehold (Chinese)

0.007 -0.087 0.132 -0.083(0.016) (0.041)** (0.032)*** (0.060)

Number of women inworking age group(16–64 years old)a

0.022 -0.020 0.020 -0.029

(0.007)*** (0.012)* (0.011)* (0.017)*

Log total household income 0.060 0.133 -0.105 -0.027(0.007)*** (0.013)*** (0.026)*** (0.035)

Individual characteristicsMarital status (never

married)0.145 -0.031 0.132 0.073

(0.023)*** (0.040) (0.053)** (0.081)Marital status (currently

married)-0.155 -0.234 -0.103 -0.098(0.018)*** (0.031)*** (0.048)** (0.069)

Level of education: Tertiary 0.299 0.235 0.333 -0.349(0.018)*** (0.027)*** (0.021)*** (0.044)***

Level of education:Secondary

0.087 0.016 0.161 -0.233(0.011)*** (0.019) (0.022)*** (0.032)***

Age 0.043 0.048 0.049 0.067(0.003)*** (0.005)*** (0.007)*** (0.010)***

Age2 -0.001 -0.001 -0.001 -0.001(0.000)*** (0.000)*** (0.000)*** (0.000)***

Observations 14894 11781 3352 4145Wald c2 1815.37 751.18 512.9 302.6Probability > c2 0.000 0.000 0.000 0.000Predicted probability 0.484 0.451 0.599 0.604Pseudo R2 0.115 0.132 0.160 0.163

Notes: ***, ** and * represent significance at the 1, 5 and 10% level, respectively. Robust standarderrors are in parentheses. The control group for marital status is divorced/widow/widower/separated; the control group for education level is those with primary education and below; andthe control group for ethnic group is other ethnic group. aExcluding those who are currentlystudying.



V. Discussion and Conclusions

The most plausible estimates of the average treatment effect coming from thepropensity score matching method is that foreign maids increased the likelihoodof labor force participation by working age Malaysian females in their employinghouseholds, by 0.18 in 1993/1994 and 0.13 in 2004/2005. These estimates comefrom matching family demographics and household income, and from using thesample legally able to hire foreign maids. The estimated treatment effects aresomewhat higher if households whose income is too low to legally hire a maid areincluded: a finding that makes sense because the incentive to ignore the immi-gration rules would most likely be greatest for households where the impact of themaid was largest.

The ostensible aim of allowing migrant maids to work in Malaysia is to raisethe participation rate of Malaysian women, by effectively outsourcing householdproduction activities to these lower-cost foreigners (Ministry of Finance, 2007).The treatment effects on female labor force participation that are reported aboveare a key input into an assessment of the net financial benefits of this policy. Thesefinancial benefits depend on whether the value of additional production generatedby Malaysian women joining or rejoining the labor force exceeds the cost ofhiring foreign maids. Under the assumption that labor markets are competitive,wages should equal the value of extra production, so the net financial benefit toMalaysia from using foreign maids, B, should be:

B a w m= × − ,

where: a = ATT of foreign maids on likelihood of female labor force participa-tion; w = average monthly labor market income of Malaysian women with maids;and m = average monthly cost of hiring a foreign maid.

The cost of hiring a maid, m, is estimated to be RM750 per month. The threecomponents of this cost are the monthly salary, which averages RM400 in asurvey of over 100 Indonesian maids carried out by the first author (Tan andGibson, 2010), costs for food and lodging of RM300 (estimated by the authors)and average monthly administrative fees of RM53 (as reported by Chin, 2005).The Malaysian consumer price index increased only 13 percent from 2005 to2010, so deflating these costs back to the timing of the last HES would reducethem only modestly, to approximately RM650 per month.

According to the HES estimates in Table 1, the average monthly income forMalaysian women in the labor force and with a maid in their household wasapproximately RM1850 in 2004/2005. The PSM results for 2004/2005 suggestthat having a maid raises the likelihood of a Malaysian women being in the laborforce by just 0.13. Therefore, the expected gross financial benefit induced by

8 Similarly, the welfare calculations by Cortés and Pan (2011) suggest that mothers of youngchildren in Hong Kong may reap a consumer surplus of US$200 per month from the availability offoreign maids.



having the maid is RM240 per month (i.e. 0.13 ¥ 1850) for the average householdthat employs a foreign maid (and meets the requirements, specifically havingmonthly income greater than RM3000).

Subtracting the costs, it appears that the net financial benefit is –RM400 permonth (specifically, B = 240 - 650 ª -400). In other words, for the averagehousehold in Malaysia employing a foreign maid in 2004/2005, there appears tohave been a net financial cost of RM400 per month. This follows from the impactof maids on female labor force participation being rather low, so that the expectedincrease in family labor earnings is much less than the family outlays on the maid.Obviously, because these households choose to employ a foreign maid, it mustmake them better off in other ways, so an alternative way to interpret thiscalculation is that when aggregating across all 300 000 foreign maids in Malaysia,the welfare benefit that they provide must be equivalent to RM1 billion per yearbecause that is the total value of the net financial cost of hiring foreign maids.8 Itis possible that by reducing the need for Malaysian housewives to engage inhousework, such as cleaning and cooking, maids allow mothers more time tospecialize in caring for their children (Chan, 2006). This extra parental inputcould perhaps create better future human capital, which is a subject for futureresearch.

Moreover, it is notable that the likelihood of women going to work as a resultof having maid declined between 1993/1994 and 2004/2005. This might reflectthe ongoing increase in average incomes in Malaysia, which should increase thedemand for leisure (as a normal good), which would then see more of the impactof maids on women’s leisure rather than on labor force participation. Such aneffect is at odds with the publicized aim of the policy allowing foreign maids tobe hired. It would take further analysis, using time use surveys, to measure thisimpact on leisure and to fully account for the welfare effects of foreign maids inMalaysia. That remains a topic for future research.

References

Ariffin, R., 2001, Domestic work and servitude in Malaysia. Hawke Institute Working Paper SeriesNo. 14, University of South Australia, Adelaide.

Caliendo, M. and S. Kopeinig, 2008, Some practical guidance for the implementation of propensityscore matching. IZA Discussion Paper No. 1588. Institute for the Study of Labor, Bonn.

Chan, A. H., 2006, The effects of full-time domestic workers on married women’s economic activitystatus in Hong Kong, 1981–2001. International Sociology, 21, 133–57.

Chin, C. B. N., 1997, Walls of silence and late twentieth century representations of the foreign femaledomestic worker: The case of Filipina and Indonesian female servants in Malaysia. InternationalMigration Review, 31, 353–85.

Chin, C. B. N., 2005, Neither at work nor at home: Asian transnational domestic workers in Malaysia.In: Asian Women as Transnational Domestic Workers (eds Huang S., Yeoh B. S. A. and RahmanN. A.), pp. 262–87). Marshall Cavendish Academic, Singapore.

Cortés, P. and J. Pan, 2011, Outsourcing household production: The demand for foreign domestichelpers and native labor supply in Hong Kong. Mimeo. Boston University School of Management,Boston.



Cortés, P. and J. Tessada, 2011, Low-skilled immigration and the labor supply of highly educatedwomen. American Economic Journal: Applied Economics 3, 88–123.

Dehijia, R. H., and S. Wahba, 2002, Propensity score matching methods for nonexperimental causalstudies. Review of Economics and Statistics 84, 151–61.

Guha, B., 2007, Maids and mistresses: Migrating maids and female labor force participation. Eco-nomics Bulletin 10, 1–9.

Heckman, J., 1979, Sample selection bias as a specification error. Econometrica: Journal of theEconometric Society, 153–61.

Immigration Department of Malaysia, 2010, Foreign domestic helper. Immigration Departmentof Malaysia, Putrajaya [online; cited 6 December 2010]. Available from URL: http://www.imi.gov.my/index.php/en/main-services/foreign-domestic-helper.

International Labour Organization, 2007, Labour and Social Trends in ASEAN 2007. InternationalLabour Organization, Bangkok.

Kanapathy, V., 2006, Relevance, risk and benefits of labour migration in Malaysia. Paper presentedat the Challenges of Global Migration and Forced Displacement, Kuala Lumpur (1–2 August2006).

Kremer, M. and S. Watt, 2009, The Globalization of Household Production. Paper presented at theThe Second Conference on International Migration and Development, 10–11 September 2009.

Ministry of Finance, 2007, Economic Report 2007/2008. Ministry of Finance Malaysia, KualaLumpur.

Ministry of Women Family and Community Development, 2007, Measuring and Monitoring GenderEquality, Malaysia’s Gender Gap Index. Ministry of Women Family and Community Develop-ment Malaysia, Kuala Lumpur.

Ooi, X., 2007, Maids on my mind. The Star, 26 September.Rosenbaum, P. R. and D. B. Rubin, 1983, The central role of the propensity score in observational

studies for causal effects. Biometrika 70, 41–55.Suen, W., 1994, Market-procured housework: The demand for domestic servants and female labor

supply. Labour Economics 1, 289–302.Tan, P-L. and J. Gibson, 2010, The impacts of temporary emigration of lower-skilled females on

sending households in Indonesia. Paper presented at the Australian Development EconomicsWorkshop, University of Western Sydney, Sydney.

Wickramasekera, P., 2002, Asian labour migration: Issues and challenges in an era of globalization.International Migration Papers 57. International Labour Office, Geneva.



impact of foreign maids on female labor force participation in malaysia

Documents