Employment Choices of PWDs
Employment Choices of PWDs
Employment Choices of PWDs
Working Paper
Employment Choices of Persons with Disability
(PWDs) in Metro Manila
Suggested Citation: Mina, Christian D. (2010) : Employment Choices of Persons with Disability
(PWDs) in Metro Manila, PIDS Discussion Paper Series, No. 2010-29, Philippine Institute for
Development Studies (PIDS), Makati City
Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your
Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.
Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial
Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them
machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise
use the documents in public.
Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen
(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open
gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you
genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated
licence.
Philippine Institute for Development Studies
Surian sa mga Pag-aaral Pangkaunlaran ng Pilipinas
November 2010
Christian D. Mina†
†
Supervising Research Specialist, Philippine Institute for Development Studies (PIDS)
NEDA sa Makati Building, 106 Amorsolo Street, Legaspi Village, Makati City 1229, Philippines
This paper examined how PWDs in the Philippines make a living and determined the factors that
influenced them to select a particular labor market status and/or a source of personal income. It
also analyzed the different types of work these PWDs engage in and their association with levels
of educational. Using dataset from the pioneering survey on socioeconomic conditions of PWDs
in selected cities in Metro Manila, Philippines conducted in 2008, econometric analyses were
conducted. The paper then presented some insights that could help the government craft better
policy strategies in the provision of livelihood to, as well as empowerment of, PWDs.
I. Introduction
In 2008, employment had already been included in the set of targets under the Millennium
Development Goals (MDGs). The member-countries are now committed to monitor progress
towards achievement of full and productive employment as well as decent work for all segments
of the population, including persons with disabilities (PWDs). In the Philippines, various efforts
have long been undertaken by the government in providing employment and livelihood supports
to PWDs. One of those is the Magna Carta for Disabled Persons (MCDP), which ensures equal
opportunities for suitable employment to PWDs as their able-bodied counterparts. Among the
remarkable employment and livelihood programs of the government are the Tulong Alalay sa
Taong May Kapansanan (TULAY) program, or Support services to persons with disabilities, of
the Department of Labor and Employment (DOLE); Assistance package for PWDs of the
Department of Trade and Industry (DTI); Philippine National Skills Competition for PWDs of
the Technical, Education and Skills Development Authority (TESDA); and, Science and
Technology Intervention for the Poor, the Vulnerable and PWDs of the Department of Science
and Technology (DOST) (Mori et al. 2009; Purcil 2009; DRPI and KAMPI 2008).
In spite of the various efforts carried out by the government in providing full and productive
employment to PWDs, it seems that employment opportunities for this segment of the population
are still limited. Schelzig (2005) noted that out of more than 100,000 employable PWDs that are
registered with the DOLE, only less than 10 percent are wage employed. A pioneering survey on
the socioeconomic conditions of PWDs in selected cities in Metro Manila, which was conducted
in August 2008 jointly by the Institute of Developing Economies (IDE) and the Philippine
Institute for Development Studies (PIDS), revealed that slightly more than half of the PWD
2
respondents have income-generating job/business. Out of those with job/business, almost 50
percent engage themselves (either full-time or part-time) into other small-scale income-
generating activities, which are usually under informal arrangements. Meanwhile, around 25
percent of the PWD respondents rely merely on supports from family members/friends, benefits
This paper thus aimed at examining how PWDs in the Philippines make a living and determine
the factors, mainly their individual and household characteristics, which influenced them to
select a particular labor market status and/or a source of personal income. It also analyzed the
different types of work these PWDs engage in and their association with levels of education.
Based on the insights drawn from econometric analyses, it presented some recommendations that
can be of help to the government in crafting better policy strategies in the provision of livelihood
3
II. Review of Literature
A number of studies have been made on the factors affecting labor force participation and
employment. Using census data on 2,825 households, Khan (2007) estimated a multinomial logit
model and found significant individual, household and community-related factors affecting
employment choices in rural northwest Pakistan. Some of these include household per capita
income, age and education of household head, household size, number of working members in a
In the Philippines, Alba and Esguerra (1998) estimated a mixed logit regression model using the
matched files of the Family Income and Expenditure Survey and Annual Poverty Indicators
Survey to find a set of determinants of labor force participation. Using around 20,876 working-
age individuals, the following variables appeared significant: personal income; education;
regional location; number of elderly in the household; interaction between region and urban/rural
classification; interaction between age and education; and, interaction among sex, marital status
There had also been studies particularly focused on labor force participation among the disabled.
Kidd et al. (2000) found that education, marital status, race and location have significant effect
on labor force participation of both able-bodied and disabled British males. Among the disabled,
type of disability was noted to have significant effect on the probability of employment. The
study employed probit analysis using the 1996 British Labour Force Survey.
4
A study by Scott and Mete (2008) also revealed that individual characteristics of PWDs in
Uzbekistan such as age, sex and education were also found to have significant influence on the
probability of being economically. As age of the disabled increases, he/she is more likely to
participate in the labor market. Similarly, years of education increases the odds of being
employed. On the other hand, being female decreases the probability of participating in the labor
force. In addition to individual characteristics, household and some other characteristics like
being head of a household, number of children aged 15 and below, number of pension-aged
adults, and household size have significant relationship with employment probability. The first
three factors were found to have significant positive relationship with labor force participation
while the last one negatively influences the probability of being economically active. Location
was also noted to be a significant factor of employment choice. Meanwhile, disability status was
Park et al. (2007) also scouted for significant correlates of labor force participation among
mobility-impaired in South Korea. In this study, gender came out be one of the most significant
factors influencing employment. It also noted that having a less severe disability tends to
Using a pooled data from household and labor force surveys in Great Britain, Blackaby et al.
(1999) analyzed the occupational and labor market statuses of disabled and non-disabled. One of
the findings of the study is the U-shaped relationship between age and unemployment, implying
that the unemployed are more likely to be either too young or too old relative to those in work.
The study also found that ethnic background matters in the labor market status of the disabled.
5
Those who are white tend to have higher probability of being employed relative to those
classified as an ethnic minority. Similar to findings in other studies, education also plays a
significant role in increasing one’s chance of being employed. Another interesting finding is that
unemployment is highly probable among individuals who have not married yet. The presence of
young children, on the other hand, was found to have significant positive relationship with being
the possible social deprivation and reduced geographical mobility associated with council
housing.
6
III. Methodology
In order to answer the objectives of this study, various statistical analyses were conducted. To
determine the factors affecting PWDs to participate in the labor market as well as those affecting
them to depend on certain sources of income, econometric models were developed. A descriptive
technique was also employed to examine the relationships among different types of occupation
Manila was conducted in August 2008 by a team of PWD enumerators (representing the
team from the PIDS. The selected cities were Makati, Quezon, Pasay, and Valenzuela,
representing local government units (LGUs) with different income. The NSO list, which was
based on the 2000 Census of Population, was supposed to be the sampling frame of the study.
However, because the list contains many incorrect information, samples were also drawn from
other sources: LGU list of program beneficiaries, member lists from different Disabled People
Organizations (DPOs), and PWDs on-site. LGU list covers more or less lower-income PWDs
because they are usually the beneficiaries of programs at the local level. DPO member lists, on
the other hand, usually include PWDs at higher income deciles. Although the sample PWDs are
in some sense representative of PWDs in the sampled areas, in terms of socioeconomic status,
weighting of the sample respondents cannot be applied to yield unbiased estimates of parameters.
Data analysis can thus only be carried out as though the sample generated is purposive (Mori et
al. 2009).
7
A. Factors affecting PWDs’ mode of labor force participation and major source of
personal income
Variables
(2) Mode of labor force participation: 1 if a PWD was ‘pseudo’ fully employed; 2 if ‘pseudo’
(4) Dummy variables of mode of labor force participation: (a) ‘pseudo’ fully employed; (b)
inactive;
(5) Dummy variables of major source of personal income: (a) wage income; (b)
Job/business indicator provides the working status of a PWD; whether he/she was engaged in
The mode of labor force participation further detail the working status of a PWD. Labor force
concepts used in this study, however, do not fully conform to the official definitions of the ILO.
Some items that are required to satisfy the official definition of modes of labor force
participation are not present in the IDE-PIDS PWD survey. For instance, based on the official
8
definition of International Labour Organization (ILO), if a person worked for pay of profit for at
least an hour during the reference period, provided that he/she is of working age (15-70), he/she
is considered employed. However, because there was no question in the survey that asked the
respondent if he/she wanted additional or longer working hours, but not necessarily additional
job, he/she may not be considered as fully employed. Thus, those respondents who reported that
they had a job/business but were not looking for work are tagged as ‘pseudo’ fully employed.
Similarly, underemployed in this study is referred to as ‘pseudo’ underemployed. These are the
respondents who reported having a job/business and were looking for work. It excludes, for
instance, those who wanted only additional or longer working hours at their present job but not
necessarily additional job. Unemployed and not in the labor force in this study are also termed as
as those who had no job/business but was looking for work while economically inactive is
defined as those who had no job/business and not looking for work. The ILO definition requires
additional ‘screening’ questions on the person’s availability for work (whether he/she, although
currently not looking for work, is available for work for the next two weeks from the reference
A PWD respondent may have more than one job or business. In order to determine his/her major
source of income, all incomes from various sources were aggregated and then the share of each
source to total was estimated. The source with the highest percentage share was then picked as
the major source of personal income of the respondent. These sources of income were then
grouped into major categories, namely: wage income, entrepreneurial income and transfer
income. Paid employment income such as wages and salaries, including cash gifts and bonuses
9
received, is defined here as wage income. Entrepreneurial income here combines profits from
businesses and rent for buildings/rooms/lands. A few respondents who are renting their
rooms/houses to others consider this as business and income from it as profits. Transfer income
here includes receipts from family members (who are usually overseas Filipino workers or
OFWs) and friends. It also includes pension income and benefits/allowances from the
government. For the purpose of this study, receipts from institutions such as church or
federations, and even from begging, were also included under this category.
Because the dependent variables are somehow related to each other, only one set of explanatory
variables were used in all of the models. Most of these explanatory variables were selected based
on what were identified in the literature. These include the following: dummy variable for
location (Makati City and others); age, sex, marital status, and education of the PWD respondent;
household head indicator (whether the respondent is a household head or not); household size;
ownership of household assets (proxy for household income1); presence of OFWs in the
necessity of a personal assistant (PA) at home and when going out of the house as well as the
number of assistive devices (proxy for physical functioning2); and, dummy variables for the three
major types of impairment (mobility, visual, hearing) and for multiple impairments.
Note that two of the explanatory variables were indices generated using Principal Component
Analysis (PCA), namely: asset index and functioning index. Asset index was generated using the
variables on ownership of household assets (tenure status of house and lot, housing type and
ownership of durable assets). Functioning index was generated using the variables on the
availability and perceived necessity of PA at home and when going out as well as possession of assistive
10
devices. Because there are only a limited number of sample observations in this study,
explanatory variables entered in the model should be maintained at a lower number. Instead of
including all of the relevant variables in the model, PCA performs some statistical computations
to come up with a composite index that would represent all of these variables as one explanatory
variable in the model. This is also one of the statistical techniques used to address the
multicollinearity problem in model estimation. (Refer to Table 1 for the complete list and
Meanwhile, it would have been better if the numbers of children and elderly in the household
were included as explanatory variables but the survey data did not permit the generation of these
potential variables. Age of each of the household members was not asked in the survey.
Models to be fitted
Two classes of models were estimated in this study, namely: logistic regression (LR; both
multinomial, or MLR, and binary, or BLR) and multivariate adaptive regression splines
(MARS). The general specification of the models used in this study can be written as follows:
= explanatory variable, or the potential factor affecting the choice of the PWD
= stochastic error
11
If the model to be estimated is LR, ln , where: 1 is
the probability that the ith individual PWD will favor the jth choice. On the other hand, if the
M
model to be fitted is MARS, the following is the function approximation: Y = a 0 + ∑ a m Bm( q ) (x ) ,
m =1
such that: a0 is the coefficient of the constant basis function (or the constant term); {a m }1 is the
M
(x ) = ∏ [s km ⋅ (xv ( k ,m) − t km )]
Km
(q)
B m is the vector of non-constant (truncated) basis functions (or the
k =1 +
tensor product spline basis); where: m is the number of non-constant basis functions (1, 2, …,
M); q is the power to which the spline is raised in order to control for the degree of smoothness
of the resultant function estimate (which in this case is equal to 1); (+) denotes that only positive
results of the right-hand side of the equation are considered (otherwise, the functions evaluate to
0; thus, the term truncated); s km indicates the (left/right) sense of truncation that assumes only 2
values (i.e., ± 1, representing the standard basis function and its mirror image3); x v ( k ,m ) refers to
the value of the predictor; v(k,m) refers to the label of the predictor (1 ≤ v(k,m) ≤ n); n is the
number of predictors; t km is the “knot” location on the corresponding predictor space or region
(or the value that defines an inflection point along the range of the predictor); and, k is the
maximum level or order of interaction (or the number of factors) in the mth basis function (1, 2,
…, Km).
LR is a better alternative to multiple linear regression under the following conditions: (i) the
dependent variable is categorical; (ii) the independent variables are a mixture of continuous and
categorical variables, with the latter dominating the model, and; (iii) the independent variables
12
do not satisfy the multivariate normality assumption, especially when the total number of
observations is very few. On the other hand, MARS4 is a nonparametric method for fitting
adaptive regression that uses piecewise basis functions to define relationships between a
dependent variable and a set of predictors. Basis functions are a set of functions used to represent
the information contained in one or more variables. Like principal components, basis functions
re-express the relationship of the predictors with the dependent variable. Unlike any other spline-
based techniques, MARS uses the so-called truncated power (multivariate) spline basis functions
with q=1 to estimate a simple linear function, similar to that of the linear regression.
Variables
Current occupations reported by employed PWD respondents were used in this analysis. Major
Classification (PSOC), were also used. Meanwhile, the original categories of the variable on
educational attainment (i.e., from no grade completed to postgraduate) were also used in this
analysis.
Instead of the simple two-way tabulations, a Correspondence Analysis (CA) was carried out to
be able to highlight all of the important patterns in the two-way tabulations of occupations and
educational attainment of PWDs. A biplot was then displayed to graphically show the statistical
association of certain types of occupations with certain levels of education attained by PWDs.
13
IV. Empirical Results
Diagnostic Checking
Before proceeding to model estimation, some diagnostic checks were performed. Pairwise
correlation matrix revealed that the strongest correlation, which exists between age and marital
status at -0.5167, did not reach the 0.70 threshold. This implies that multicollinearity should not
be a problem in model estimation. Both correlation tests and univariate regression analyses, on
the other hand, provided idea on the hypothesized relationships of the dependent variables with
All LR models satisfied the Hosmer-Lemeshow’s goodness-of-fit and specification link tests.5.
Interestingly, all of them have relatively good predictive ability since they were able to predict
A. Factors affecting PWDs’ mode of labor force participation and major source of
personal income
Results of LR analyses
Based on the findings from the multinomial and binary LR analyses, the three most significant
factors that differentiate the PWDs across mode of labor force participation and major source of
personal income are household head indicator, membership in a DSHO and functioning index
(Refer to Tables 4-8). Being a household head increases the probability of being employed and
of sourcing most of his/her personal income from wage employment. Similarly, being a member
14
of a DSHO also increases the odds of being wage employed. Members of DSHO are also less
likely to depend on transfer income and, although some of them are unemployed, they tend to
look for work. Meanwhile, having access to PA and assistive devices and viewing PA’s help as
necessary increases the probability of being economically inactive. This segment of PWD
samples also tends to have no personal income and if they have, majority of their income are
Some other factors were also found to have significant effect on labor market status and income
source of PWDs. Most of the visually-impaired in the sample were found to be wage employed,
although some of them still look for other jobs. There is thus a very low probability of finding a
Also, many of the PWDs with more assets are employed and source most of their income from
transfers from other people. Those who are at least high school graduate are economically active,
although not usually under full employment, and primarily depend on wage income. On the other
hand, PWDs living in a household with at least one OFW are economically inactive and are
dependent on transfer income. Moreover, female PWDs and those residing in Makati City are
more likely to be economically inactive and either have no reported income or dependent on
transfer income.
Examining interactions between and among the explanatory variables might uncover some more
interesting patterns in the data. Through the estimated MARS models, factors that were not
found significant in LR models became significant when they were allowed to interact with other
15
factors (Tables 9-17). One of the findings from the MARS analyses is that male PWDs who are
heads of their households have higher probability of being employed. In addition, heads that are
aged 53 and below and are visually-impaired tend to be in wage employment. Meanwhile, heads
that are not single and have lesser access to PA and/or assistive devices are more likely to be
On the other hand, PWDs who are not heads in relatively smaller households (below 5 members)
that have at least one OFW member are less likely to have an income-generating job/business.
PWDs who are not household heads, below 32 years old and not high school graduate tend to be
underemployed. However, those who are in the mid-30’s and above were found to have no
job/business. PWDs who are just members of relatively poorer households (owning fewer assets)
have higher probability of not participating in the labor force. From this group, those who have
never been married did not even report any personal income.
Membership in a DSHO increases a PWD’s chance of being employed. In fact, although many of
the visually-impaired are employed, if he/she is not a member of a DSHO, there is a high
probability that he/she is not fully employed. Among the visually-impaired PWDs who are
DSHO members, many are in wage employment, particularly those who belong to non-OFW
households. However, not all members of a DSHO for the blind have wage employment as their
major source of personal income. If he/she did not even finish secondary education, he/she has
lower probability of being wage employed. Meanwhile, those who are visually-impaired and are
not DSHO members but do not depend much on PAs and/or assistive devices have higher
16
Moreover, PWDs who are not members of DSHO and are aged 21 and above are more likely to
be economically inactive. Those who are not members of any DSHO and are highly dependent
on PAs and/or assistive devices have relatively lower probability of being employed.
Higher functioning index was generally found to have negative effect on the probability of being
employed. This is particularly true among visually-impaired PWDs who belong to a two-member
household. Also, PWDs who are either mobility- or hearing-impaired with higher functioning
index were not in the labor force. Those who are either visually- or mobility-impaired with
higher functioning index and are aged 46 and below have lower probability of engaging in
entrepreneurial activities. On the other hand, PWDs who have lower functioning index and
Interestingly, visually-impaired PWDs aged 44 and above and members of households with
relatively more assets are usually fully employed. Female PWDs belonging to a relatively
wealthier household and are either mobility- or hearing-impaired often have personal income,
PWDs who are at least high school graduate usually participate in the labor force. It is thus not
surprising to find out that those who did not finish secondary education and are either mobility-
or hearing-impaired are more likely to have no job/business, especially those who are 30 years
old and older. They usually depend on transfer income, rather than income from wage
employment or entrepreneurial activities. However, not all PWDs who have lower educational
17
attainment have no job/business. The survey data also revealed that PWDs who are not visually-
impaired, did not finish high school and members of non-OFW households either engage in
The results of the CA suggest that some types of occupation are significantly associated with
some levels of education. Using the original categories of the occupation variable, it can be
deduced from the results of the CA and its biplot (shown in Figure 1) that having a master’s or
any post-graduate degree is closely associated with teaching as the primary occupation. Being an
office clerk seems to be the most common occupation of PWDs with post-secondary degrees. A
lot of the sample PWDs which took up diploma and certificate courses might have acquired jobs
related to office or clerical works. Having finished a vocational degree is closely associated with
getting jobs such as store keeper/manager as well as artist/musician. ICT-related workers have
been significantly associated with PWDs who have vocational degrees and college/university
degrees. PWDs who are owners of small-scale businesses, electricians, launderers/pressers, and
drivers usually have vocational and/or tertiary education. Moreover, many of those who had no
formal education or have only reached elementary education end up with occupations such as
masseurs, street and market vendors, buy and sell agents, and helpers (either in private
households or establishments).
Associating the major groups of occupations (based on PSOC) with educational attainment,
Figure 2 suggests that PWDs who have pursued post-graduate degrees are more likely to be
18
among the pool of professionals. PWDs with college, vocational or other post-secondary degrees
tend to become clerks. Those who have pursued vocational education end up being among those
trades and related workers. Lower level of education and no formal education are more likely
associated with being an agricultural worker and having no job/business. Ironically, many of the
laborers and unskilled workers reached tertiary level education. Meanwhile, other groups of
occupations such as officials, technicians and associate professionals, service and shop and
market sales workers, and plant and machine operators seem to have finished college or
vocational education.
19
V. Concluding Remarks
Notwithstanding the limited sample size in this study, some insights can be drawn from
econometric and descriptive analyses. The findings suggest that certain groups of PWDs
possessing different individual and household characteristics need specific types of interventions
from the government. For instance, being a household head seems to put additional pressure for a
PWD to seek employment. This is particularly true among those who are male, have their own
families, have relatively small number of assets, and cannot afford and/or does not urgently need
assistance from PAs and/or any special devices. Apparently, these PWDs belong to relatively
poorer households as evidenced by ownership of fewer assets and lesser access to PAs and/or
assistive devices. Thus, the government might want to provide special assistance to this
particular group of PWDs through provision of additional source of livelihood to their families.
Another important factor is membership in a DSHO. It seems that DSHOs play a significant role
in providing livelihood to and empowering their members. The government, through the LGUs,
might want to implement programs and activities that are similar to those conducted by DSHOs
for the benefit of their members. Alternatively, the government might want to work in
Higher access to PAs and/or assistive devices may imply higher degree of disability (or limited
physical functioning) or higher household wealth. Many PWDs who belong to any of these
groups tend to have no job/business. Rehabilitation and livelihood assistance may be provided by
the government to these PWDs, not only to make earnings for themselves, but also to let them
20
Moreover, education can be considered as a critical factor in gaining employment. The
government might offer scholarships to school-aged PWDs who cannot attend school because of
financial constraints. Alternative learning sessions might also be conducted (on a regular basis
and free of charge) to PWDs who are already beyond the school-age but do not have high school
diploma (which is usually the minimum qualification set by employers). Special sessions might
also be catered to hearing-impaired who do not know sign language (especially those in far-flung
areas where deaf schools are inaccessible) and are not knowledgeable with Filipino (since
English is the medium of instruction in deaf schools). The government might also allocate
additional funds for programs and activities related to skills training and livelihood assistance to
Meanwhile, this study may be further improved by employing some sophisticated statistical
techniques such as the nonparametric counterpart of the Heckman correction to correct for the
possible problem of endogeneity. In addition, better insights might be drawn if the labor supply
equation would be correctly identified by taking into account the income and substitution effects
21
References
Alba, M.M. and E.F. Esguerra, 1998, A mixed logit model of modes of labor force participation.
MIMAP Research Paper No. 32. Makati City, Philippines: MIMAP Project, Policy and
Development Foundation, Inc.
Blackaby, D., K. Clark, S. Drinkwater, D. Leslie, P. Murphy, and N. O’Leary, 1999, Earnings
and employment opportunities of disabled people: secondary analysis of the General
Household Survey and the Labour Force Survey. Research Report No. 133. Norwich:
The Crown Copyright.
Friedman, J.H., 1991, Multivariate additive regression splines. Ann. of Stat., 19(1), pp. 1-67.
Friedman, J.H. and C. Roosen, 1995, An introduction to MARS. Statistical Methods in Medical
Research, 4, pp. 197-217.
Khan, M.A., 2007, Factors affecting employment choices in rural northwest Pakistan. Paper
presented during the Conference on International Agricultural Research for Development,
University of Göttingen, 9-11 October 2007.
Kidd, M.P., P.J. Sloane and I. Ferko, 2000, Disability and the labour market: an analysis of
British males. Journal of Health Economics, 19, pp. 961-981.
Leathwick, J.R., J. Elith and T. Hastie, 2006, Comparative performance of generalized additive
models and multivariate adaptive regression splines for statistical modeling of species
distributions. Ecological Modeling, 199, pp. 188-196.
Mori, S., C. Reyes and T. Yamagata, 2009, Poverty reduction for the disabled in the Philippines:
Livelihood analysis from the data of PWDs in Metro Manila. Joint Research Program
Series No. 151. Chiba, Japan: Institute of Developing Economies.
Purcil, L., 2009, Monitoring the human rights of persons with disabilities: laws, policies and
programs in the Philippines. Ontario, Canada: Disability Rights Promotion International.
Schelzig, K., 2005, Poverty in the Philippines: income, assets, and access. Mandaluyong City,
Philippines: Asian Development Bank.
Scott, K. and C. Mete, 2008, Measurement of disability and linkages with welfare, employment,
and schooling: the case of Uzbekistan. In: Economic implications of chronic illness and
disability in Eastern Europe and the former Soviet Union. Washington, D.C.: The World
Bank.
22
Table 1. Variable definition
Variable Definition
Dependent variables
Model 1: Job/business indicator Job/business indicator: 1 - has a job/business; 0 - otherwise
Model 2: Mode of labor force Mode of labor force participation: 1 - 'pseudo' fully employed; 2 - 'pseudo' underemployed; 3 - 'pseudo'
participation unemployed; 4 - 'pseudo' economically inactive
Model 3: Major source of personal Major source of personal income: 0 - no reported income; 1 - wage income; 2 - entrepreneurial income; 3 -
income transfer income
Model 4: 'Pseudo' fully employed Dummy for 'pseudo' fully employed: 1 - 'pseudo' fully employed; 0 - otherwise
Model 5: 'Pseudo' underemployed Dummy for 'pseudo' underemployed: 1 - 'pseudo' underemployed; 0 - otherwise
Model 6: 'Pseudo' unemployed Dummy for 'pseudo' unemployed: 1 - 'pseudo' unemployed; 0 - otherwise
Model 7: 'Pseudo' economically Dummy for 'pseudo' economically inactive: 1 - 'pseudo' economically inactive; 0 - otherwise
inactive
Model 8: Wage income Dummy for wage income: 1 - major source of personal income is wage employment; 0 - otherwise
Model 9: Entrepreneurial income Dummy for entrepreneurial income: 1 - major source of personal income is entrepreneurial activities;
0 - otherwise
Model 10: Transfer income Dummy for transfer income: 1 - major source of personal income is transfers or receipts from family
members/friends/others; 0 - otherwise
Model 11: No income Dummy for no income: 1 - no reported income; 0 - otherwise
Independent variables
All models:
Makati Dummy for Makati City: 1 - resides in Makati City; 0 - otherwise
Age (squared) Squared age of the respondent (standardized)
Female Sex of the respondent: 1 - female; 0 - male
Single Marital status of the respondent: 1 - never been married; 0 - otherwise
At least high school graduate Highest educational attainment of the respondent: 1 - at least high school graduate; 0 - otherwise
Household head Household head indicator: 1 - household head; 0 - otherwise
Household size Total number of members in a household
Asset index Index for household assets (tenure status of house and lot, housing type and ownership of durable assets);
generated using Principal Component Analysis
OFW OFW indicator: 1 - if there is an OFW in a household; 0 - otherwise
DSHO member Membership in a Disability Self-Help Organization: 1 - member; 0 - otherwise
Functioning index Index for the availability and necessity of PA at home and when going out, and possession of assistive
devices; generated using Principal Component Analysis
Impairment dummies:
Mobility Dummy for Mobility: 1 - mobility-impaired only; 0 - otherwise
Visual Dummy for Visual: 1 - visually-impaired only; 0 - otherwise
Hearing Dummy for Hearing: 1 - hearing-impaired only; 0 - otherwise
Multiple (base category ) Dummy for Multiple: 1 - has more than one impairment; 0 - either mobility-, visually- or hearing-impaired
23
Table 2. Results of correlation tests
Dependent / Job/business 'Pseudo' fully 'Pseudo' under- 'Pseudo' 'Pseudo' econ'ly Wage Entrep'l Transfer
No income
Independent indicator employed employed unemployed inactive income income income
Makati -0.1062 ** -0.1253 ** 0.0174 0.0163 0.1057 ** -0.0711 -0.0364 0.0218 0.0891 *
Age (squared) 0.0105 0.0901 * -0.0760 -0.0146 0.0028 -0.0466 0.0592 0.0122 -0.0321
Female -0.1108 ** -0.0002 -0.1067 ** 0.0296 0.0975 * -0.1087 ** 0.0192 0.0258 0.0616
Single -0.0331 0.0552 -0.0844 * 0.0371 0.0002 0.0234 -0.0281 -0.0415 0.0522
At least high school 0.0349 -0.0450 0.0765 0.1054 ** -0.1466 *** 0.0801 -0.0461 -0.0012 -0.0286
graduate
Household head 0.1944 *** 0.0993 ** 0.0957 * -0.0917 * -0.1329 *** 0.1031 ** 0.0411 0.0235 -0.1735 ***
Household size 0.0298 0.0506 -0.0195 -0.0677 0.0349 -0.0554 0.0432 -0.0036 0.0118
Asset index 0.0997 ** 0.0652 0.0344 -0.0151 -0.0994 ** -0.0570 0.0425 0.1632 *** -0.1650 ***
OFW -0.1264 ** -0.0630 -0.0626 0.0505 0.0945 * -0.0711 -0.0503 0.1397 *** -0.0232
DSHO member 0.1967 *** 0.1454 *** 0.0533 -0.0327 -0.1936 *** 0.1351 *** 0.0554 -0.1252 ** -0.0645
Functioning index -0.1559 *** -0.1375 *** -0.0198 -0.0264 0.2040 *** -0.0493 -0.1292 ** 0.0771 0.1118 **
Mobility-impaired 0.0268 0.0065 0.0197 0.0437 -0.0752 -0.0537 0.0523 -0.0221 0.0198
Visually-impaired 0.1313 *** 0.0654 0.0652 -0.0145 -0.1363 *** 0.1071 ** 0.0471 -0.1174 ** -0.0348
Hearing-impaired -0.0222 -0.0293 0.0065 0.0463 -0.0218 0.0299 -0.0268 0.0057 -0.0064
Note: Reported figures are partial correlation coefficients and estimated coefficients
from the correlation tests and univariate regression analyses, respectively.
Coefficients with ***, ** and * are significant at 1%, 5% and 10% level, respectively.
24
Table 3. Results of univariate logistic regression
Dependent / Job/business 'Pseudo' fully 'Pseudo' under- 'Pseudo' 'Pseudo' econ'ly Wage Entrep'l Transfer
No income
Independent indicator employed employed unemployed inactive income income income
Makati -0.1847 -0.3538 0.1028 0.2777 0.0100 -0.3537 -0.0112 0.0724 0.2905
Age (squared) 0.1405 0.1778 -0.0071 -0.1853 -0.0371 -0.1240 0.2307 ** 0.0632 -0.2193
Female -0.7257 *** -0.1969 -0.6720 *** 0.2958 0.7192 *** -0.6292 ** -0.0480 0.0627 0.6183 **
Single -0.6620 *** -0.2403 -0.5372 ** 0.4945 * 0.4762 ** -0.1520 -0.4613 ** -0.0012 0.6921 ***
At least high school 0.4717 ** -0.0350 0.5750 *** 0.4447 * -1.0372 *** 0.3418 -0.0315 0.1343 -0.4937 **
graduate
Household head 1.5870 *** 0.7546 *** 0.8592 *** -0.8624 *** -1.6302 *** 0.9594 *** 0.4759 ** -0.3445 -1.6253 ***
Household size -0.0398 -0.0102 -0.0374 -0.0099 0.0594 * -0.0936 * 0.0021 0.0490 0.0258
Asset index -0.0300 -0.0325 -0.0044 0.0779 -0.0270 -0.1483 ** -0.0276 0.2622 *** -0.1346 **
OFW -0.6463 ** -0.4743 -0.3556 0.4064 0.4887 * -0.7421 ** -0.3477 1.0481 *** -0.3114
DSHO member 0.8965 *** 0.4955 ** 0.5417 ** -0.1262 -1.1688 *** 0.8275 *** 0.1914 -0.6910 *** -0.3096
Functioning index -0.1127 * -0.1423 * -0.0115 -0.0639 0.1868 *** -0.0258 -0.1708 ** 0.0781 0.0967
Mobility-impaired -0.1784 -0.1975 -0.0249 0.3792 -0.0922 -1.3052 *** 0.3698 0.3865 0.2549
Visually-impaired 1.4494 *** 0.8689 *** 0.6558 *** -0.8458 *** -1.3917 *** 1.4406 *** 0.3828 * -1.4510 *** -0.6609 **
Hearing-impaired -1.0395 *** -0.7699 *** -0.5806 ** 0.4977 * 0.8997 *** -0.3483 -0.8225 *** 0.6970 *** 0.3211
Note: Reported figures are partial correlation coefficients and estimated coefficients
from the correlation tests and univariate regression analyses, respectively.
Coefficients with ***, ** and * are significant at 1%, 5% and 10% level, respectively.
25
Table 4. Estimated logistic regression model for job/business indicator
Dependent variable: Job/business indicator
Independent variables Coefficient
Intercept -0.6002
(0.6204)
Makati -0.8062 **
(0.3159)
Age (squared) 0.0185
(0.1546)
Female -0.6155 **
(0.2701)
Single -0.1863
(0.2938)
At least high school graduate 0.1571
(0.2631)
Household head 1.3292 ***
(0.3324)
Household size 0.0327
(0.0357)
Asset index 0.1370 **
(0.0664)
OFW -0.8011 **
(0.3206)
DSHO member 1.2889 ***
(0.3094)
Functioning index -0.2255 ***
(0.0726)
Impairment dummies
(base category : Multiple)
Mobility 0.1803
(0.4527)
Visual 1.2977 ***
(0.4661)
Hearing -0.2160
(0.4839)
Note: Reported figures are the estimated coefficients. Those in parentheses are robust standard errors.
Coefficients with ***, ** and * are significant at 1%, 5% and 10% level, respectively.
26
Table 5. Estimated multinomial logistic regression model for mode of labor force
participation
Dependent variable: Mode of labor force participation
(base category : 'Pseudo' economically inactive)
'Pseudo' fully 'Pseudo' 'Pseudo'
Independent variables
employed underemployed unemployed
Intercept -1.5112 -0.4740 -0.8925
(0.9163) (0.8640) (0.9941)
Makati -1.5189 *** -0.8863 ** -0.6341
(0.4888) (0.4370) (0.4759)
Age (squared) 0.1478 -0.1837 -0.0911
(0.2178) (0.2125) (0.2225)
Female -0.5517 -0.9262 *** -0.2852
(0.3824) (0.3580) (0.3653)
Single 0.1905 -0.3232 0.1713
(0.4199) (0.3758) (0.3878)
At least high school graduate 0.5112 0.9169 *** 1.0937 ***
(0.3620) (0.3490) (0.3776)
Household head 1.7681 *** 1.6877 *** 0.6841
(0.5294) (0.4899) (0.5174)
Household size 0.0300 -0.0144 -0.0546
(0.0488) (0.0523) (0.0495)
Asset index 0.2192 ** 0.1888 ** 0.1127
(0.0980) (0.0915) (0.1008)
OFW -0.9418 ** -0.9205 ** -0.2360
(0.4533) (0.4297) (0.3956)
DSHO member 2.1030 *** 1.6783 *** 1.0453 **
(0.4516) (0.4227) (0.4671)
Functioning index -0.4569 *** -0.3011 *** -0.2756 ***
(0.1281) (0.0869) (0.0977)
Impairment dummies
(base category : Multiple)
Mobility 0.3610 0.4724 0.6021
(0.6734) (0.6208) (0.6994)
Visual 1.6718 ** 1.7029 *** 0.8665
(0.6954) (0.6391) (0.7541)
Hearing -0.3236 0.0420 0.3409
(0.7287) (0.6561) (0.7155)
Note: Reported figures are the estimated coefficients. Those in parentheses are robust standard errors
Coefficients with ***, ** and * are significant at 1%, 5% and 10% level, respectively.
27
Table 6. Estimated multinomial logistic regression model for major source of personal
income
Dependent variable: Major source of personal income
(base category : Wage income)
Entrepreneurial
Independent variables Transfer income No income
income
Intercept -0.1455 0.7397 0.4687
(0.9263) (0.9875) (0.9797)
Makati -0.0274 0.4113 0.8439 *
(0.3753) (0.4186) (0.4751)
Age (squared) 0.2506 0.1426 0.0612
(0.1908) (0.2228) (0.2216)
Female 0.5166 0.6317 * 0.7734 **
(0.3403) (0.3690) (0.3817)
Single -0.1880 -0.2705 0.2312
(0.3710) (0.4249) (0.4160)
At least high school graduate -0.5088 -0.3373 -0.4879
(0.3142) (0.3492) (0.3666)
Household head -0.2597 -0.4173 -1.7260 ***
(0.3683) (0.4362) (0.4774)
Household size 0.0636 0.0415 0.0347
(0.0557) (0.0569) (0.0587)
Asset index 0.0894 0.1932 ** -0.1621
(0.0806) (0.0816) (0.1027)
OFW 0.2413 1.0123 ** 0.3257
(0.4959) (0.4816) (0.5391)
DSHO member -0.2827 -1.2395 *** -1.0733 **
(0.3428) (0.3973) (0.4323)
Functioning index -0.0788 0.1832 * 0.2387 **
(0.0982) (0.1066) (0.1104)
Impairment dummies
(base category : Multiple)
Mobility 0.9608 0.4888 0.7310
(0.7580) (0.7539) (0.7674)
Visual -0.3801 -1.7484 ** -1.2006 *
(0.6874) (0.6838) (0.7032)
Hearing -0.5503 -0.2631 -0.3476
(0.7657) (0.7050) (0.7488)
Note: Reported figures are the estimated coefficients. Those in parentheses are robust standard errors.
Coefficients with ***, ** and * are significant at 1%, 5% and 10% level, respectively.
28
Table 7. Estimated logistic regression models for each mode of labor force participation
Model 4 Model 5 Model 7
Model 6
(Dependent variable: (Dependent variable: (Dependent variable:
Independent variables (Dependent variable:
'Pseudo' fully 'Pseudo' 'Pseudo' economically
'Pseudo' unemployed)
employed) underemployed) inactive)
Intercept -2.3020 -0.8040 -1.6719 -0.2370
(0.6768) (0.6674) (0.7871) (0.7775)
Makati -0.8062 ** 0.1068 0.1752 0.9946 **
(0.3325) (0.2812) (0.3387) (0.4062)
Age (squared) 0.2854 * -0.2157 -0.0448 0.0572
(0.1526) (0.1522) (0.1734) (0.1840)
Female 0.0131 -0.5588 ** 0.2082 0.5978 *
(0.2768) (0.2580) (0.2944) (0.3108)
Single 0.3258 -0.4707 * 0.2617 0.0155
(0.3203) (0.2821) (0.3180) (0.3282)
At least high school graduate -0.2294 0.3919 0.6430 ** -0.8563 ***
(0.2561) (0.2464) (0.3137) (0.3068)
Household head 0.5482 * 0.4887 * -0.7198 ** -1.4456 ***
(0.3228) (0.2839) (0.3563) (0.4533)
Household size 0.0418 -0.0181 -0.0582 0.0171
(0.0375) (0.0438) (0.0395) (0.0411)
Asset index 0.0836 0.0412 -0.0283 -0.1735 **
(0.0649) (0.0594) (0.0736) (0.0843)
OFW -0.4291 -0.4111 0.3465 0.6635 *
(0.3467) (0.3216) (0.3260) (0.3492)
DSHO member 0.8578 *** 0.2814 -0.2692 -1.5942 ***
(0.2902) (0.2719) (0.3411) (0.3838)
Functioning index -0.2379 ** -0.0299 -0.0512 0.3315 ***
(0.1057) (0.0651) (0.0806) (0.0801)
Impairment dummies
(base category : Multiple)
Mobility 0.1256 0.2609 0.4619 -0.4794
(0.5266) (0.5089) (0.5716) (0.5418)
Visual 0.6317 0.6624 -0.2691 -1.4748 ***
(0.5294) (0.5069) (0.6095) (0.5731)
Hearing -0.3328 0.1208 0.4955 -0.0684
(0.5729) (0.5395) (0.5832) (0.5648)
Note: Reported figures are the estimated coefficients. Those in parentheses are robust standard errors.
Coefficients with ***, ** and * are significant at 1%, 5% and 10% level, respectively.
29
Table 8. Estimated logistic regression models for each major source of personal income
Model 8 Model 9 Model 10 Model 11
Independent variables (Dependent variable: (Dependent variable: (Dependent variable: (Dependent variable:
Wage income) Entrepreneurial income) Transfer income) No income)
30
Table 9. Estimated MARS model for job/business indicator
Dependent variable: Job/business indicator
Parameter Est. coeff. Std. error p-value Basis Functions
Constant 0.711 0.034 0.0000
BF5 -0.181 0.035 0.0000 BF1 = (HHEAD = 0)
BF6 -0.961 0.144 0.0000 BF2 = (HHEAD = 1)
BF7 0.032 0.012 0.0060 BF3 = (VISUAL = 0)
BF11 -0.085 0.017 0.0000 BF5 = max(0, AGE_SQ - 0.841) * BF3
BF13 -0.474 0.171 0.0060 BF6 = max(0, 0.841 - AGE_SQ) * BF3
BF20 0.250 0.049 0.0000 BF7 = max(0, ASSET_INDEX + 1.842) * BF5
BF27 -0.211 0.065 0.0010 BF9 = (DSHO = 0)
BF11 = max(0, FUNCTION_INDEX + 1.176) * BF9
Final Model BF13 = max(0, - 2.310 - ASSET_INDEX) * BF1
Y = 0.711 - 0.181 * BF5 - 0.961 * BF6 + 0.032 * BF7 - 0.085 * BF11 BF15 = (OFW = 1) * BF1
- 0.474 * BF13 + 0.250 * BF20 - 0.211 * BF27 BF20 = (FEMALE = 0) * BF2
BF27 = max(0, 5.000 - HSIZE) * BF15
31
Table 11. Estimated MARS model for ‘pseudo’ underemployed
Dependent variable: 'Pseudo' underemployed
Parameter Est. coeff. Std. error p-value Basis Functions
Constant 0.650 0.054 0.0000
BF4 -0.145 0.035 0.0000 BF1 = (HHEAD = 0)
BF7 0.111 0.035 0.0010 BF2 = (HHEAD = 1)
BF9 0.066 0.023 0.0040 BF4 = max(0, 2.040 - FUNCTION_INDEX) * BF2
BF13 -0.507 0.071 0.0000 BF5 = max(0, AGE_SQ - 0.784)
BF15 -0.344 0.068 0.0000 BF7 = (SINGLE = 0) * BF4
BF17 0.170 0.054 0.0020 BF9 = max(0, HSIZE - 12.000)
BF21 -0.023 0.004 0.0000 BF13 = (HIGHSCH = 0) * BF1
BF14 = (HIGHSCH = 1) * BF1
Final Model BF15 = (MAKATI = 0) * BF14
Y = 0.650 - 0.145 * BF4 + 0.111 * BF7 + 0.066 * BF9 - 0.507 * BF13 BF17 = max(0, AGE_SQ - 1.024) * BF13
- 0.344 * BF15 + 0.170 * BF17 - 0.023 * BF21 BF21 = max(0, HSIZE - 1.000) * BF5
32
Table 13. Estimated MARS model for ‘pseudo’ economically inactive
Dependent variable: 'Pseudo' economically inactive
Parameter Est. coeff. Std. error p-value Basis Functions
Constant -0.066 0.034 0.0530
BF2 2.242 0.495 0.0000 BF1 = max(0, AGE_SQ - 0.441)
BF3 0.126 0.038 0.0010 BF2 = max(0, 0.441 - AGE_SQ)
BF5 0.128 0.020 0.0000 BF3 = (HHEAD = 0)
BF12 0.539 0.092 0.0000 BF5 = (DSHO = 0) * BF1
BF14 -0.138 0.038 0.0004 BF10 = (VISUAL = 0)
BF15 -0.276 0.075 0.0003 BF12 = (HIGHSCH = 0) * BF10
BF17 0.064 0.012 0.0000 BF14 = max(0, AGE_SQ - 0.225) * BF12
BF19 0.391 0.142 0.0060 BF15 = (OFW = 0) * BF12
BF17 = max(0, FUNCTION_INDEX + 1.176) * BF10
Final Model BF19 = max(0, - 2.310 - ASSET_INDEX) * BF3
Y = -0.066 + 2.242 * BF2 + 0.126 * BF3 + 0.128 * BF5 + 0.539 * BF12
- 0.138 * BF14 - 0.276 * BF15 + 0.064 * BF17 + 0.391 * BF19
33
Table 15. Estimated MARS model for entrepreneurial income
Dependent variable: Entrepreneurial income
Parameter Est. coeff. Std. error p-value Basis Functions
Constant 0.320 0.031 0.0000
BF2 -0.156 0.039 0.0001 BF2 = max(0, 2.116 - AGE_SQ)
BF3 0.167 0.047 0.0005 BF3 = (HEARING = 0) * BF2
BF5 -0.066 0.024 0.0070 BF5 = max(0, FUNCTION_INDEX - 0.315) * BF3
Final Model
Y = 0.320 - 0.156 * BF2 + 0.167 * BF3 - 0.066 * BF5
34
Table 17. Estimated MARS model for no income
Dependent variable: No income
Parameter Est. coeff. Std. error p-value Basis Functions
Constant 0.049 0.037 0.1890
BF7 0.354 0.089 0.0001 BF1 = (HHEAD = 0)
BF8 0.381 0.090 0.0000 BF5 = (SINGLE = 1) * BF1
BF10 -0.361 0.087 0.0000 BF7 = max(0, - 1.484 - ASSET_INDEX) * BF5
BF12 -0.058 0.02 0.0040 BF8 = (VISUAL = 0)
BF21 0.486 0.114 0.0000 BF10 = (FEMALE = 0) * BF8
BF23 0.158 0.039 0.0001 BF11 = (FEMALE = 1) * BF8
BF12 = max(0, ASSET_INDEX + 3.431) * BF11
Final Model BF18 = max(0, 0.961 - AGE_SQ) * BF8
Y = 0.049 + 0.354 * BF7 + 0.381 * BF8 - 0.361 * BF10 - 0.058 * BF12 BF21 = (OFW = 0) * BF18
+ 0.486 * BF21 + 0.158 * BF23 BF23 = (HEARING = 0) * BF1
Figure 1. Correspondence analysis biplot between types of Figure 2. Correspondence analysis biplot between major
occupation and levels of education groups of occupation and levels of education
2
6
10 333
1
7
4
2 00
Dimension 2 (21.4%)
Dimension 2 (23.4%)
9 425777
6
8 3
0
4 3
5 8
2
2 1
691 7
-1
8 697 9 9 6
96 95
0
3393
2
5 8
54 98
777
094 92
-2
4
1
333
-2
7
-3
10
7
-4
-4
-2 0 2 4 6 8 10 -8 -6 -4 -2 0 2
Dimension 1 (37.0%) Dimension 1 (46.2%)
35
End notes
1
Since most of the PWD respondents had difficulty estimating their total household income, variable on
household income was deemed unreliable. Thus, ownership of assets was used as a proxy variable to be
able to measure the welfare of the household.
2
In this study, availability and perceived necessity of a personal assistant, both at home and when going
outside of the house, as well as presence of at least one assistive device were used to generate a composite
index that will serve as proxy for physical functioning.
3
For s km equal to +1, the basis function will have a value x-t if x>t and 0 if x≤ t. If it is -1, the basis
function will have a value t-x when x<t, while 0 if x≥t.
4
Theoretical discussions presented here are based mostly on Friedman (1991), Friedman and Roosen
(1995), and Leathwick et al. (2006).
5
Hosmer-Lemeshow’s goodness-of-fit test checks whether the model fits reasonably well by testing the
null hypothesis that the dependent variable follows binomial distributions with probabilities specified by
the logistic regression model. The Specification link test for single-equation models, on the other hand,
tests the null hypothesis that the model is specified correctly.
36