adb economics working paper series no

8/3/2019 ADB Economics Working Paper Series No

1/17

ADB Economics Working Paper Series No. 199

Education Outcomes in

the Philippines

Dalisay S. Maligalig, Rhona B. Caoli-Rodriguez,

Arturo Martinez, Jr., and Sining Cuevas

May 2010

(Revised: 17 January 2011)


2/17

The ADB Economics Working Paper Series is a forum for stimulating discussionand eliciting feedback on ongoing and recently completed research and policy studiesundertaken by the Asian Development Bank (ADB) staff, consultants, or resourcepersons. The series deals with key economic and development problems, particularlythose facing the Asia and Pacific region; as well as conceptual, analytical, or

methodological issues relating to project/program economic analysis, and statisticaldata and measurement. The series aims to enhance the knowledge on Asiasdevelopment and policy challenges; strengthen analytical rigor and quality of ADBscountry partnership strategies, and its subregional and country operations; andimprove the quality and availability of statistical data and development indicators formonitoring development effectiveness.

The ADB Economics Working Paper Series is a quick-disseminating, informalpublication whose titles could subsequently be revised for publication as articles inprofessional journals or chapters in books. The series is maintained by the Economicsand Research Department.

AbstractThis paper identifies key determinants of individual, school, and quality of

education outcomes and examines related policies, strategies, and project interventionsto recommend reforms or possible reorientation. Two sets of data were used: (i) data onschool resources and outputs from the administrative reporting systems of theDepartment of Education; and (ii) the 2002, 2004, and 2007 Annual Poverty IndicatorSurveys. Analysis of individual, school, and quality of education outcomes showedthat although school resources such as pupilteacher ratio is a key determinant for bothindividual and school outcomes, and that per capita miscellaneous operating and other

expenses are significant factors in determining quality of education outcome,socioeconomic characteristics are stronger determinants. Children of families in thelower-income deciles and with less educated household heads are vulnerable and lesslikely to attend school. Girls have better odds of attending school than boys. Workingchildren, especially males, are less likely to attend secondary school. On the basis ofthese results, recommendations in the areas of policy and programs are discussed tohelp address further deterioration, reverse the declining trend, and/or sustain gains sofar in improving basic education system performance outcomes.

I. IntroductionFilipino parents value education as one of the most important legacies they can

impart to their children. They believe that having a better education opensopportunities that would ensure a good future and eventually lift them out of poverty.Thus, they are willing to make enormous sacrifices to send their children to school(Dolan 1991, De Dios 1995, LaRocque 2004). However, with a poor familys severelylimited resources, education tends to be less prioritized over more basic needs such asfood and shelter. Hence, the chances of the family to move out of poverty are unlikely.It is therefore, important that the poor be given equitable access to education.


3/17

The 1987 Philippine Constitution declares that education, particularly basiceducation, is a right of every Filipino. On this basis, government education policies andprograms have been primarily geared toward providing access to education for all. ThePhilippines is committed to the World Declaration on Education for All (EFA) and thesecond goal of the Millennium Development Goals (MDG) to achieve universal

primary education by 2015.EFAs framework of action has six specific goals in the areas of: (i) early

childhood care and education (ECCE); (ii) universal primary/basic education; (iii) lifeskills and lifelong learning; (iv) adult literacy; (v) gender equality; and (vi) quality. Inline with this framework of action, the Philippine EFA 2015 National Action Plan(UNESCO 2010) adopted in 2006 was formulated as the countrys master plan for basiceducation.

In 2000, the Philippines reported that it has achieved substantial improvement interms of access to basic education, but still faces challenges in the areas of earlychildhood care and development, internal efficiency, and learning outcomes (NCEFA

1999). Through the governments efforts to achieve the 2015 MDG targets, recent studiessuch as the Philippines Midterm Progress Report on the MDGs (NEDA and UnitedNations Country Team 2007, Table 1) assess that the probability of achieving universalprimary education (MDG 2) in the country is low (based on net enrollment rate, cohortsurvival rate, and completion rate). Similarly, the 2009 EFA Global Monitoring Report(UNESCO 2008) identified the Philippines to be among the countries with decreased netenrollment rate from 1999 to 2006, and with the greatest number of out-of-schoolchildren (more than 500,000). The Philippiness current performance in education basedon the trends identified by the EFA and MDG indicators as shown in Appendix Table 1is not also promising. It is quite likely that the EFA and MDG targets will not be met by

2015. Overall, the Philippines has suffered a setback in most education outcomeindicators. Although signs of recovery have been registered by some indicators,national targets for key EFA indicators such as intake and enrollment rates will stilllikely be missed in 2015.

How can the decline in the performance of EFA indicators of education outcomesbe averted and improvements in those that registered recovery be sustained? This paperaims to address this question by identifying key determinants of selected majoreducation outcomes, and on this basis, examine concomitant or related policies,strategies, and project interventions for purposes of recommending reforms or possiblereorientation.

Previous studies have suggested that poverty incidence (socioeconomic status),government expenditure on education (as a percentage of gross domestic product[GDP]) and pupilteacher ratio (PTR) are key determinants of school attendance or netenrollment rate. Except for a few studies covering a specific area in the country, mostrelated studies in the Philippines examine the relationships of education outcomes andinputs using exploratory correlations and regressions of inputs and factors that mayaffect education outcomes. These studies do not have an explicit theoretical model toguide the analysis, and hence could be considered to have been done on a piecemeal


4/17

basis, without being able to explore the relationships of all the major factors in onecomprehensive analysis. For example, Maligalig and Albert (2008) concluded that thereis evidence that government expenditure on education and poverty incidence aredirectly related to net enrollment ratio, but failed to ascertain the degree of therelationships as well as the efficacy of other factors that may affect school enrollment.

There are many other methods that could be employed in identifying keydeterminants of education outcomes, such as the education production function, whichhas been used by many studies cited throughout this paper. Another method is therandomized evaluations that have already been done in other countries like Kenya,Nicaragua, and United States; or the natural experiments study conducted in Indonesiaby Duflo (2001); or the qualitative methods that are being conducted as part of theTrends in International Mathematics and Science Study. The education productionfunction approach usually refers to a mathematical equation between outcomes andinputs and a statistical method for estimating those relationships. The success of thisapproach is contingent upon available data and the application of suitable statistical

methods in estimating the production function. Both randomized evaluation andnatural experiments render controlled comparisons. However, both require extensiveplanning prior to the implementation of the study.

For the purposes of this study, as randomized evaluations and naturalexperiment were not possible, key determinants of education outcomes were identifiedby estimating an education production function based on the combination of data fromthe Department of Education (DepEd) administrative reporting systems, and theAnnual Poverty Indicator Survey (APIS) conducted by the National Statistics Office(NSO) in between the Family 2 | ADB Economics Working Paper Series No. 199Incomeand Expenditure Survey (FIES). Section II of this paper identifies the conceptual

framework that was used; Section III presents the results; while Section IV discusses thepolicy implications. The last section presents the conclusions and recommendations ofthe study.

II. Conceptual FrameworkMany studies on the determinants of education outcomes are based on an

education production function that defines a mathematical relationship between inputsand education outcomeY such as

Y = Y (I, F,R) +e (1)where Y is a function of I and F, which are individual characteristics and familysocioeconomic factors, respectively, R is school resources, and e represents unmeasuredfactors influencing schooling quality. Depending on the availability of data, thismathematical relationship is estimated using suitable statistical models, of which thebest is identified through evaluation of the models goodness of fit and adherence toassumptions.

The output of an education production function is usually some achievementthat can be measured through indicators. Among these are intake and enrollment rates,cohort survival rate, dropout rate, and repetition rate, which are all EFA indicators.


5/17

Another key education outcome indicator is the learning achievement rate or learningoutcomes usually measured through national standardized tests.

The education production function described in equation (1) requires bothmeasures of individual and family socioeconomic characteristics as well as schoolresources. Previous studies in the Philippines as well as in other countries indicate that

there are individual and household characteristics that influence childrensparticipation and performance in basic education (Bacolod and Tobias 2005, DeGraffand Bilsborrow 2003, UIS 2005). These studies suggest that family background andsocioeconomic factors are as important as school resources in determining whether achild will attend school, survive, and complete an education level, and achieve anacceptable level of learning outcome. In fact, Hanushek (1986) concluded thatsocioeconomic factors are stronger determinants compared to school resources.

Individual characteristics such as age, sex, and parents educational attainmentare important factors in achieving better education outcomes. For example, based on the2004 APIS, Maligalig and Albert (2008) concluded that, assuming all other factors stay

the same (ceteris paribus), boys are 1.39 times more likely not to attend school thangirls. Similarly, in examining Indonesias 1987 National Socioeconomic Survey,Deolalikar (1993) found that males have significantly lower returns to schooling thanfemales at the secondary and tertiary levels. The returns to university education are 25%higher for females than males. Deolalikar also cited some evidence that older householdheads and better-schooled female household heads provide relatively more schoolingopportunities for their female relatives. Furthermore, community characteristics such asproportion of villages in the district of residence having access to all-weather roads,access by water, lower secondary school, etc. have relatively few significant effects onschool enrollment.

School resources, on the other hand, are typically the basic inputs in education,the most fundamental being the classrooms and teachers. Other important inputs arethe curriculum, textbooks and other instructional materials, water and sanitationfacilities such as toilets, libraries, and science laboratories. Bacolod and Tobias (2005)find that the presence of electricity is an important school input positively affectinglearning outcome in Cebu. As measure of school quality, school resources are expressedas PTR and pupilclassroom ratio, among others.

Previous studies have mixed observations on the effects of school resources oneducation outcomes. Case and Deaton (1999) found that prior to the democraticelections in South Africa in 1999 and conditional on age, lower test scores, and lowerprobabilities of being enrolled in education, schools with high PTRs discourageeducational attainment. In their study of time series data from 58 countries, Lee andBarro (2001) found strong relationships between measures of school resources andmeasures of outcomes such as subject test scores, dropout rate, and repetition rate. Onthe other hand, Hanushek and Kimko (2000) concluded, based on data from 39countries, that traditional measures of school resources such as PTR and per capitaeducation expenditures do not have strong effects on test performance. Also, Hoxby(2000) on her study of 649 elementary schools in the United States concluded that


6/17

reduction in class size has no effect on students achievement. Hanushek (2003)compiled 376 production functions from 89 individual publications on educationoutcomes across the United States and concluded that the evidence on the PTR as animportant determinant of education outcomes is not conclusive. These studies,however, differ on the statistical methods and data used. The suitability of the

econometric methods was not considered nor was data quality examined. As Case andDeaton (1999) have pointed out, many of these studies were concerned with theestimation of detailed educational production functions that try to sort out effects ofdifferent resources on education such as PTR, textbook-to-student ratio, pupilclassroom ratio, school buildings, presence of library, per capita expenditure oneducation, among others.

A.Data SourcesEducation production functions will be modeled using two major sources:

(i) the 2002, 2004, and 2007 APIS conducted by the NSO; and (ii) administrative

data obtained from the Basic Education Information System (BEIS) and theNational Educational Testing and Research Center (NETRC) of DepEd as well asfrom its budget appropriations.

The first source of data consists of three rounds of APIS that used almostthe same questionnaire. These surveys are of national coverage with regions asdomains, barangays or enumeration areas as primary sampling units, andhousing dwellings as the ultimate sampling units. Households in the selectedhousing dwellings are enumerated on the households income and expendituresand the socioeconomic characteristics of each member of the household. Aresponsible adult in the household was asked about each members age, sex,

educational attainment, school attendance, reason for not attending school, aswell as household income and expenditures, among others. More than 50,000households were surveyed covering the 85 provinces in the Philippines.

The APIS is undertaken during the intervening years of the FIES.Beginning 2004, the 2003 master sample design was used for all householdsurveys of national coverage including APIS. The basis of the sampling frame forthe 2003 master sample is the 2000 Census of Population and Housing as well asresults of past national surveys, such as the 2000 FIES, the 2001 Labor ForceSurvey, and the 1997 Family Planning Survey.

Administrative data from DepEds reporting systems stored at thedivision level could either be from a province or an independent city. Forpurposes of consistency with APIS, the province was set as the unit of analysis.Data were on the most recent five years (20022007).

The APIS gathers information on the demographic, economic, and socialcharacteristics of households, which include health and education data on eachfamily member. Data on education include school attendance, highesteducational attainment, and reasons for not attending school. Among the citedreasons for absence from school are cost of education, distance between home


7/17

and school, availability of transportation, existence of illness or disability, andwhether the member is working or looking for work (Appendix 4).

BEIS was established in 2002 to improve the monitoring and evaluation ofbasic education performance. Prior to BEIS, the basic education data system wasladen with an almost 3-year backlog. The BEIS significantly reduced data backlog

with its quicker consolidation and validation process. It includes data on schoolinputs (number of teachers, classrooms, other school facilities) and outcomeindicators crucial in assessing basic education performance in terms of access,internal efficiency, and quality. For school resources, the BEIS uses a color codingsystem that indicates the status of divisions and even schools with respect tothese resources.

The BEIS uses three modules. Module I is the Quick Count Module, whichgets total data from the schools (e.g., total enrollment, total number of teachersetc.) by the end of December every year. The information is used for planningand budgeting for the next school year. Module II is the School Statistics Module,

which collects school data in detail (e.g., enrollment by grade/year, age profilesof enrollees, etc.). This module is designed to collect information from bothpublic and private schools. Module III is the Performance Indicators Module,which processes the data and presents the outcome indicators.

Figure 1 describes the BEIS data collection process. Annual data collectionstarts upon the issuance of a DepEd order to collect public school profiles. Theorder is disseminated down to the schools where base data on enrollment,dropouts, repeaters, number of classrooms, teachers, etc. are manually recordedusing annual data gathering forms (government school profile forms forelementary and secondary levels) under Module II. These forms are submitted to

the division offices where they are encoded and consolidated in MS Excel files.The division offices are also responsible for validating the accuracy ofinformation with the schools before they are submitted to the regional offices forfurther consolidation. The regional offices then submit the data to the centraloffices Research and Statistics Division, which maintains and updates the BEISannually, processes the data, and presents the outcome indicators under ModuleIII. The data remains in MS Excel files that because of their bulk cannot beuploaded on the DepEds website. Researchers and other users can only accessfrom the internet a one-page fact sheet on basic education statistics showing thenational aggregates of major indicators for the last 5 years. The researchers mayobtain more information from the BEIS through a written request addressed tothe Research and Statistics Division, which provides the information in soft copy.The BEIS is also internally accessible among DepEds various offices and unitsthrough its local area network.


8/17

The DepEd intends to continuously improve BEIS. Under the BESRA, a proposal

for Enhanced BEIS is being explored. This involves developing an automated databasesystem where even data down the schools (School Information System) can be accessedfrom the web. Moreover, DepEd is currently in the process of adopting an ICT-baseddata collection scheme that will put in place effective quantitative and qualitative datacollection as well as student tracking systems.

Gross and net intake rates, gross and net enrollment rates, dropout rate,

repetition rate, and cohort survival rate are the key outcome indicators estimated andcompiled by BEIS. These indicators gauge the level of the childrens access to formalbasic education and the school effectiveness in keeping the children.

Indicators such as repetition rate, dropout rate, cohort survival rate, PTR, etc. arecomputed based on actual intake and year-to-year enrollment. As such they can beestimated at the school level and aggregated upward to district, division, regional, andnational levels. Intake and enrollment rates, however, can only be computed at thedivision level based on the consolidated actual enrollment data, because thedisaggregation of population estimate from the NSO are available down to the divisionlevel only.

The gross intake rate is the total number of enrollees in Grade 1, regardless ofage, expressed as a percentage of the population in the official primary education entryage, which is currently 6 years old. On the other hand, net intake rate accounts forGrade 1 enrollees expressed as a percentage of the 6-year-old population. The grossenrollment rate is defined as the total number of children, regardless of age, enrolled ina particular education level, measured as a proportion of the age group correspondingto that level. Meanwhile net enrollment rate (NER) accounts for the participation of


9/17

children who fall within a defined official school-age group. While the gross enrollmentrate reflects total participation and, to some extent, the capacity of the education system,the net enrollment rate is indicative of both the quantity and quality of educationsystem performance and effectiveness with respect to the target age group.

Box 1: Investigating the Accuracy of the Philippiness Net Enrollment RateOne of the key education indicators is the net enrollment rate (NER), which is

chiefly used to measure developments in primary education. In fact, both the EFAand MDG programs utilize this to evaluate the progress in their respective Goal 2objectives. On the basis of the NER current trends (Box Figure 1), it is projected thatthe Philippines will not likely attain universal primary education by 2015.

The NER is the ratio of the enrollment for the age group corresponding to theofficial school age in the elementary/secondary level to the population of the sameage group in a given year. The official school-age population for the primary level inthe Philippines is 611 years; thus, in order to estimate for the NER, the totalenrolled students aged 6-11 must be divided by the total population of the same agegroup. In theory, NER should range from 0 to 100%. However, in practice, as shownin Box Figure 2 where the box plots of NERs of provinces and independent cities areshown, there are many data points with more than 100% NERs.

This situation merits a closer look at how the data are compiled. There arethree possible sources of errors: (i) the population projections in the 611 age groupin provinces and cities are not accurate; (ii) the total enrollment of ages 611 is notproperly captured; or (iii) there are many cross-provincial enrollees for someprovinces and these are not captured at all in the DepEd administrative reporting

system (BEIS).Box Table 1 shows the comparison between APIS and DepEd data. The

figures for total population in the 611 age group that DepEd used to compute NERgrew at a steady 2.34% annually from 2002 to 2006 and dropped by 0.14% in 2007.The constant growth rate for 2002 to 2006 is equal to the national annual average

continued.


10/17

Box 1. continued.population growth rate that the NSO computed on the basis of the 1995 and 2000Census of Population and Housing. To derive the 611 population in 2007, DepEdthen adjusted the growth rate used and applied the average annual growth ratesfrom 2000 to 2007 on the 2000 Census 611 population. With a lower growth ratebasis of 2.04%, the 2007 population consequently exhibited a declining trend sincethe adjustment was not back-tracked. Usually, when new census figures becomeavailable, the population projections are also updated. This is not yet the case in thecurrent NER.

Therefore, the use of 2007 Census of Population and Housing estimateswithout back tracking the series may have caused an artificial increase in the 2007NER.

Another point investigated is the use of national population growth estimatesinstead of agespecific population growth rates. The 2.34% growth rate applied byDepEd to the 20022006 population is the 19902000 average annual growth rate ofthe Philippines. Similarly, the 2.04% growth used for the 2007 estimate is the also therate at the national level for the years 20002007. However, if the national averageannual population growth rate projections for 20012005 is to be computed, it is onlyabout 2.1%. And if the estimation is to be agespecific, the average annualpopulation growth rate for the 611 age group is only about 1.04%. These two

figures are lower than the 2.34% that DepEd employed to project total population ofages 611. Box Figure 3 shows the various NER trends based on (i) the 2.34%population growth rate used by DepEd for 20022006; (ii) the 2.04% rate if thepopulation adjustment will be back tracked; and (iii) the 1.04% rate, if the age-specific 611 growth rate is to be applied. Thus, the type of population estimatorused by DepEd has contributed to the rate of decline in NER from 2002 to 2006.

continued.


11/17

Box 1: continued.

To validate the total enrollment as compiled by BEIS, similar estimates fromthe Annual Poverty Indicator Survey were derived. The APIS is a survey of nationalcoverage that the NSO conducts in the intervening years of the Family Income andExpenditure Survey. All family members are asked about his/her age, whetherhe/she is attending school and if not, the reason for not doing so, among others.Hence, APIS could also provide estimates of the population in the primary agegroup as well as the population in the same age group who are in school. The totalenrollment estimates from APIS are within acceptable error margin (one standarderror) compared to the DepEds total enrollment and hence, there is no strong

evidence that DepEds total enrollment data is not accurate.It should be noted, however, that based on APIS data, a substantial number of6-year-olds are not yet in primary school even though by DepEds guidelines, theofficial age of entry to primary school is at 6 years old. About 830,900 6-year-oldchildren were not in primary school in 2007; 37.5% have not started school yet; while62.5% were still in preschool. This is equivalent to about 6.4% of the total populationin the 611 age group. On the other hand, examination of the composition ofenrolled 7-year old students showed that, although by DepEd guidelines, theyshould be in the Grade 2 level, most of them are still in Grade 1. In 2002, half of the7-year olds who are enrolled are in Grade 1. And although this proportion steeply

declined in 2004, it rose again in 2007 resulting to a nearly equal number of 7-year-old students in Grade 1 and Grade 2. This is an unexpected occurrence since it isanticipated that because DepEd has implemented its guidelines on the official age ofentry to primary school in 1995, the number of enrolled 7 year-olds in Grade 1should have been declining since then. These findings suggest that though theofficial school age starts at 6 years, there is still a significant percentage of families

continued.


12/17

The four indicators discussed aboveNER, gross enrollment rate, net intake rate,gross intake rateare compiled in BEIS at the division level using data from schools asnumerator and as denominator, the population projections for the corresponding agegroups from the NSO. A closer examination (see Box 1) of the net enrollment rate,which is the main indicator for universal primary or universal basic education goals ofboth EFA and MDG, reveals that there are flaws in the estimation process. For example,the fast decline of NER as reflected in the BEIS data series seems to be caused by thehigher population projections from NSO.

Once the children are in school, the next order of business is how to keep themengaged so that they are able to acquire the identified skills and levels of competenciesdefined in the curriculum. How well the schools can keep the children from leavingbefore completing a particular education level gauges the schools internal efficiency.Indicators of internal efficiency include cohort survival rate, dropout rate, and

Box 1: continued.

sending their children to primary school at a later year, thus contributing to theartificial decline of the NER.

Box Figure 4 shows the APIS and DepEd estimates of NER, which is another

form of validation that was used. While DepEds NER is steadily declining, theequivalent APIS indicator remained steady between 2002 and 2004, and showed aslight increase by 2007.


13/17

repetition rate. The cohort survival rate in a certain education level is the percentage ofa cohort of pupils/students enrolled in the first year of that level who reach the lastgrade/year of that particular education level. It indicates the holding power of theschool. A desirable pattern is that it should approach 100% and that its movementshould have a negative relation with the dropout rate.

Distortions in cohort survival rate are mainly the result of high dropout andrepetition rates. Dropout rate accounts for those pupils/students who leave schoolduring the year and those who complete the previous grade level but do not enroll inthe next grade/year level the following school year. It is expressed as a percentage ofthe total number of pupils/students enrolled during the previous school year.Repetition rate serves to measure the occurrence of pupils/students repeating a grade.It is technically defined as the percentage of a cohort of pupils enrolled in a grade at agiven school year who study in the same grade the following school year.

The National Achievement Test (NAT) is the primary indicator of school

effectiveness based on pupil/student scores in subjects like language, science, andmath. The NAT is administered by DepEd through its National Educational Testing andResearch Center, whose functions include analysis and interpretation of data for policyformulation and recommendation. Making a time-series comparison of NAT resultsfrom 2002 to 2007 is problematic since the tests are administered at different grade oryear levels annually. The NAT was first administered in 2002 to Grade 4 and 1st yearhigh school students. It included a diagnostic component conducted at the start ofschool year to determine the academic weaknesses or learning gaps of thepupil/students based on the curriculum- prescribed learning competencies at aparticular level. The results of this diagnostic test are compared with the achievementtests administered to the same group of pupils at the end of the school year todetermine learning progress. In the following school years, however, the NAT wasadministered in different grades and years.

Two indicators of school resources that will be used in the models are themiscellaneous operating and other expenses budget (MOOE) and the personnel salary(PS) budget. The budgeting division, working closely with Office of Planning Services,computes for the MOOE based on a formula (per capita student cost and school-based).They use the quick count data from BEIS to estimate the next schoolyears enrollmentand the MOOE. However, they also request the regional offices to submit MOOEproposals that they only use for validation purposes. The budget for PS is computed

based on current staff complement and increases only for new hires and promotions.Data on PS and MOOE used in this study were taken from various Congress-approvedGovernment Appropriations Acts based on the National Expenditure Programproposed by the government. Using the DepEd budget, however, does not present thecomplete basic education financing because it does not account for the contributions ofprivate schools, which comprise 8% of total elementary school enrollment and 21% ofsecondary school enrollment.


14/17

These data also do not include the contributions of the private sector and localgovernment units. DepEd has forged partnerships with private and business sectors inprojects such as Adopt-a-School and is implementing other private sector initiatives thathave resulted in valuable contributions that are also quantifiable but are not beingcaptured in the BEIS or by any DepEd unit. Local government units also contribute

significantly to basic resources needed by the schools. Among these local sources is theSpecial Education Funds (SEF) coming from the 1% real property tax earned by localgovernments and earmarked for basic education as provided for in the LocalGovernment Code. The SEF is used for construction and rehabilitation of classrooms aswell as for funding salaries of locally hired teachers.

The available administrative data do not include individual and householdcharacteristics of the pupils/students (e.g., socioeconomic status and ethnic or linguisticvariation). Moreover, accuracy is often an issue with administrative data, especiallysince the collector and processor of information are also its main users. As a result,over-reporting or under-reporting to influence decisions on funding and other

incentives can happen (UIS 2008).A more rigorous study that is also the approach taken by this research is to

combine education administrative data with census or household surveys. Althoughoften conducted less regularly, household surveys provide more information on thecharacteristics of individuals and households that often influence decisions related toeducation services made available by the government. Corresponding to the two majordata sources described above, two datasets were constructed: (i) thehousehold/individual data that combines APIS and the provincial-level PTR; and (ii)provincial-level data that consists of data from BEIS, NETRC, and the FinancialManagement System but which also includes provincial-level indicators from APISsuch as the proportion of females, median educational attainment of the householdhead, and median household per capita income.

B. Statistical Models

On the basis of the available data described above, a modeling frameworkwas developed (see Figure 2). In this framework, the decision to attend school isconsidered as an investment that promises future returns. First, it ishypothesized that the decision whether to attend school or not is mainlyinfluenced by personal circumstances. The process of deciding whether to attendschool or not usually starts at the household level and is depicted by the dotted

arrows pointing directly from household, personal resources, to the decision ofattending school. Once the household decides to send the child to school, thereare different possible education outcomes that are measured, such as dropoutrate, survival rate, repetition rate, and NAT score, among others. Theseeducation outcomes are directly influenced by education inputs, but householdand personal resources are also contributing factors.


15/17

Individual outcome (decision to attend school) is modeled using a combinationof the household/individual data from APIS and the provincial PTR from BEIS. Allschool outcomes including the quality of education outcome are modeled using thecombined administrative data and provincial estimates of key individual andhousehold variables from APIS.

In the case of the APIS dataset, for each year (2002, 2004, and 2007), a probabilitysample is drawn and hence, the set of households and individuals in the data set wereselected randomly. Because of this, a random effects model is explored, such thatsubject specific parameters {i} are treated as draws from an unknown population (andthus may be considered random). Moreover, the outcome that will be modeled for thisdata set is school attendance, a binary variable that can be modeled suitably by alogistic regression using random effects likelihood estimation. Unlike the administrativedataset, individuals, which are the unit of analysis, are only measured once; therefore, ifindividuals are considered the subject in the model, a longitudinal analysis approach isnot possible. However, since the regions are the domains of the APIS and housingdwellings are drawn from clusters or primary sampling units from strata defined

within regions (but are not similar across regions), the random effects that can beaccounted for clustering of responses are within the domains (region) and across years,such that


16/17

where ytdi is the education outcome of the ithindividual in region d and year t, xtdiisthe corresponding vector of explanatory variables, and td is the domain-specific nestedin time parameter representing heterogeneity across time and regions. The results of therandom effects model are also compared with that of the more commonly usedordinary logistic model.

Three types of explanatory variables are considered in the models: (i) individualcharacteristics such as sex and age; (ii) household characteristics such as household percapita expenditure, and age and educational attainment of the household head; and (iii)PTR at the provincial level representing school resources. The factor other thanhousehold characteristics that could affect the parents decision to send their children toschool is their perception on the capacity of the school. A measure of this perceptionthat is available is PTR because in general, parents believe that their children would getbetter education if the classrooms are not crowded. Other indicators of school resourceswere considered but dropped from the model because they were not used by parents orindividuals in their decision to attend school or not. These are the proxy for the average

teachers salary and the per capita MOOE. Moreover, these two indicators cover onlythe public school system and there are no corresponding data from the private schools.

For school education outcomes such as the NAT overall rating, NAT average testscores in Science, Math, English, and Filipino; dropout rate; cohort survival rate; andrepetition rates were considered. Since the BEIS dataset is the major data source formodeling these education outcomes, the unit of analysis was the province, since this isthe lowest disaggregation level at which the full set of data across the most recent 5years is available. Also, for most of the provinces, data have been recorded for the mostrecent 5 years. Thus, longitudinal analysis was conducted instead of cross sectionalanalysis. Longitudinal analysis is more complex than regression or time series analysisbut it has the ability to study dynamic relationships and to model differences amongsubjects. It can be shown that the educational outcomes significantly vary acrossprovinces. Hence, provincial-specific parameters will be included in the model suchthat

E (yit) = i + xit

where i is the ith province-specific parameters, yit is the educational outcome at year tand province I, while xit is the vector of explanatory variables. These variables arefurther described herein. There are two distinct approaches for modeling the quantitiesthat represent heterogeneity among the subjects (in this case, provinces) { i}: (i) fixed

effects model in which {i} are treated as fixed yet unknown parameters that need to beestimated and (ii) random effects model in which {i} are treated as draws from anunknown population and thus are random variables such that

E (yiti) = i + xit

Considering that measures from all provinces that are the subjects or units ofanalysis are included in the datasets, and that provincial-level measures were derivedfrom data of all schools in the province, the possibility of a provincial measure to vary


17/17

because of a random draw (sample) can be eliminated and hence, fixed effects model isdeemed appropriate.

Since the education production function is not complete without socioeconomiccharacteristics that are not found in BEIS or any other government administrative

reporting system, some provincial-level indicators from the APIS such as the proportionof females, median education attainment of the household head, and median householdincome were combined with the dataset. As a consequence, only 2002, 2004, and 2007data were included in the final data set.

There are many situations in educational and behavioral research in whichmultiple dependent variables are of interest. Usually, separate analyses are conductedfor each of these variables even though they are likely to be correlated and have similaralthough not identical set of predictor variables. In this research, a good example wouldbe the average NAT scores for English, Science, and Math that are also available formost of the provinces. These subject NAT scores are highly correlated and hence, to

accurately capture this situation, an alternative modeling approach, the seeminglyunrelated regression (SUR) was used. SUR is a technique for analyzing a system ofmultiple equations with cross-equation parameter restrictions and correlated errorterms.

The SUR technique estimates separate error variances for each equation; henceseparate R2s can be computed. Numerous parameter restrictions employed in SUR,however, may lead to negative R2. A potential advantage of its application in panel dataanalysis is to allow for same parameter estimates of the fixed effects using differentcorrelated dependent variables. Further, it moves away from the potential problem thatunbalanced data may cause under fixed or random effects framework.

Since separate data series for primary and secondary schools are provided in theadministrative dataset, separate models for primary and secondary age groups werederived and examined. To apply these models in the APIS dataset, the primary andsecondary age groups have to be designated. The issue of the official age of entry toprimary education arose in the process. Per DepEds policy, the official entry age toformal primary education is 6 years old. However, preliminary analysis of APISrevealed that a substantial numbers of 6-year-olds were not yet in school (21.5% for2002, 17.5% for 2004, and 15.2% in 2007) and a significant proportion is still in preschool(27.2% for 2002, 26% for 2004, and 25.3% for 2007).

adb economics working paper series no

Documents