DETERMINANTS OF HOUSEHOLDS EXPENDITURE IN BASIC EDUCATION IN COLOMBIA

DETERMINANTS OF HOUSEHOLDS EXPENDITURE IN BASIC EDUCATION IN COLOMBIA A Thess submtted to the Faculty of the Graduate School of Arts and Scences of Georgetown Unversty n partal fulfllment of the requrements for the degree of Master of Publc Polcy n Publc Polcy By Ana Mara Rojas Vllaml, B.A Washngton, DC Aprl 13, 2012

Copyrght 2012 by Ana Mara Rojas Vllaml All Rghts Reserved

DETERMINANTS OF HOUSEHOLDS EXPENDITURE IN BASIC EDUCATION IN COLOMBIA Ana Mara Rojas, B.A. Thess Advsor: Omar Robles, PHD ABSTRACT The mportance of early chldhood educaton s ndsputable. In the last decades developng countres have prortzed ther budget towards educaton but despte those efforts, nsttutonal nvestment n educaton stll does not translate n hgher enrollment rates n early educaton. Publc nvestment can provde educatonal facltes, qualty, and other nsttutonal measures but only household s nvestment wll enable ts utlzaton. Ths study pursued to dentfy poor household s constrants to nvestment n basc educaton n Colomba, to understand how educatonal polcy actons should be focalzed. The OLS and fxed effects analyss for 2008 and 2003 estmates evdence the gap between urban and rural households, suggestng that the educatonal attanment of the parents s the most robust determnant of expendture n early educaton and that data for sngle-mothers as the most vulnerable group s not conclusve enough. Polcy targetng ths populaton should be revsed and resources prortzed towards specfc programs to mprove educatonal attanment of parents and decentralzaton measures. Ths knd of research provdes emprcal evdence for Colomba and would be useful for polcy makers who seek a general understandng of educatonal regulatory polcy and ts mplcatons to natonal development.

The research and wrtng of ths thess s dedcated to my beautful mom and my lovng father. Con mucho amor, Ana Mara Rojas v

Table of Contents 1. INTRODUCTION... 1 2. LITERATURE REVIEW... 4 3. CONCEPTUAL FRAMEWORK... 6 3.1 Determnants... 6 3.1.1 Household s characterstcs... 6 3.1.2 Household s head characterstcs... 7 3.1.3 Regon Fxed Effects... 7 4. DATA AND METHODS... 8 4.1 Data Source... 8 4.2 Methodology... 9 4.2.1 Household s characterstcs... 10 4.2.2 Household s head characterstcs... 10 4.2.3 Regon Fxed Effects... 10 5. RESULTS... 11 5.1 Descrptve Results... 11 5.2 Regresson Results... 14 5.2.1 Household Sze... 14 5.2.2 Urban Households... 15 5.2.3 Female households... 15 5.2.4 Martal Status... 16 5.2.5 Household's Head Level of Educaton... 17 5.2.6 Household's Head Age... 17 6. DISCUSSION... 18 6.1 Valdty Threats... 18 6.2 Statstcal Lmtatons... 19 6.3 Polcy mplcatons... 20 7. CONCLUSION... 22 8. APPENDICES... 24 v

8.1 APPENDIX 1 Summary Statstcs... 24 8.2 APPENDIX 2 Regressons Specfcatons... 25 8.3 APPENDIX 3 Statstcal Lmtatons... 26 9. REFERENCES... 27 v

1. INTRODUCTION In the last decade early educaton ssues have partcularly captures the attenton of polcymakers after several studes 1 have been showng a consstent and close relatonshp between school attanment and economc development. Early educaton has become one of the global polcy prortes and the way of establshng equal educatonal opportuntes for chldren around the world 2. Internatonal ntatves such as Educaton for All (EFA) and the educaton Mllennum Development Goals, have promote developng countres to prortze ther budget towards educaton nvestment to mprove educatonal nfrastructure, qualty of educaton, and access. Despte all these efforts, these countres report that there are no sgnfcant changes n economc development and stll 77 mllon chldren all over the world are not enrolled n preschool and prmary educaton 3. Colomba s an nterestng case, where the prortzaton of educaton expendtures doesn t match wth hgher levels of prmary educaton enrollment. Followng nternatonal trends polcymakers n the country have focused on the allocaton of resources n nsttutonal efforts of provson of educaton (nfrastructure, qualty and access). In 2006 educaton expendture was arround11% of the total of government expendture, 4.8% of the country s GDP, a very mportant share consderng Colomba s a mddle-ncome country (Graph 1). 1 The Program for Internatonal Student Assessment (PISA, 2001, 2004, 2007), (HENDEY & STEUERLE, 2010). 2 UNESCO, 2008 3 Department of Educaton and Chld Development, Fact Sheet - Educaton and Tranng Reform Act 2006. 1

Graph 1 Educaton Expendtures of the Government n Colomba Source: UNICEF, Dvson of Polcy and Practce, Statstcs and Montorng Secton, www.chldnfo.org, May 2008 The publc expendture per capta as a percentage of the GDP n 2006 was estmated n 19.2%. But partcpaton rates stll remaned very low for ths same year. In the case of preschool educaton net enrollment rato was estmated n 34.6% for a preschool age populaton of 2,685 chldren and for prmary educaton 88.5% for a prmary school age populaton of 4,568 chldren 4. Why ths nconsstency between government s acton and polcy outcomes? Why so many chldren reman outsde the educaton system? Are resources and polcy actons addressng the demand sde of the early educaton ssue n Colomba? Publc nvestment can provde 4 UNESCO Insttute for Statstcs, Data Centre, January 2008. http://stats.us.unesco.org/unesco/reportfolders/reportfolders.aspx 2

educatonal facltes, qualty, and other nsttutonal measures but only household s nvestment wll enable ts utlzaton. Ignorng household s expendture n educaton restrctons or enhancers can be costly for educatonal polcy n the long run. Some affrmatve actons have been part of socal polcy n Colomba targetng vulnerable groups for resource allocaton and polcy actons, but how s that vulnerable populaton dentfed when addressng the early educaton ssue? Several household s characterstcs, whch could be called socal, cultural, educatonal, occupatonal and other factors, mght nfluence the nature and quantum of nvestments that households make n the educaton of ther chldren, and defne where polcy actons should be prortzed to. Unfortunately, there s not much research on the extent of household expendture on educaton n Colomba or on the determnants of household s expendture decsons. The purpose of the present study s to examne varous dmensons under whch Colomban household s take decsons about nvestng n early educaton. Identfyng poor household s constrants to nvestment n basc educaton n Colomba could help understandng how educatonal polcy actons should be focalzed. Ths analyss s worth pursung as emprcal evdence of the low level of Colomban government engagement wth educatonal polcy n a developng country where educatonal reforms have faled throughout the years. 3

2. LITERATURE REVIEW Becker (1964) developed the theory of human captal development, argung that when nvestment n educaton s undertaken t wll provde sklls to the populaton who wll be capable of workng, and wll eventually lead to economc growth. The World Development Report (1984) covered much of the earler lterature concernng the mpact of educaton on lowerng fertlty and mprovng maternal health. Behrman (1990), n a revew of human resources and poverty found strong evdence for the mpact of maternal years of schoolng on chld health. Snce 1990 ths lterature has contnued to confrm n more detaled ways the postve mpact of prmary educaton on the health of households across the generatons. Also, the mportance of early chldhood educaton and care programs on the development of chldren has been studed n many emprcal studes around the world (Hendey and Steuerle, 2010). These studes show evdence of the effect of educatonal polcy n mprovng economc welfare and health, reducng nequaltes and promotng more democratc poltcal systems. On the other hand, research on household expendtures on educaton s very lmted. The ssue has not attracted wde attenton of researchers so far. However, broadly wthn the framework of household s decson-makng behavor models, understandng how parents allocate resources across chldren has been researched n economcs. Becker (1967, 1981) examned an ndvdual maxmzaton model, where decsons regardng nvestment n educaton are manly made on the bass of effcency consderatons; whle Behrman, Pollak and Taubman (1982) ntroduced a famly model wth several other consderatons, ncludng equty between several chldren of 4

the famly, sons and daughters, younger and older chldren, etc., (or prejudces and bases, e.g., say dscrmnaton aganst grls), concludng that decsons regardng nvestment n educaton n general and more partcularly at lower levels of educaton are made by famles and rarely by the ndvdual concerned. Hence, famly/household expendture functon s consdered approprate. McMahon (1984) developed a future-orented famly utlty functon to explan why famles nvest n educaton n USA. Hs nvestment demand and supply functons ncluded varables on expected nonmonetary returns, famly dsposable ncome, tax subsdes, student loans, famly sze (number of brothers and ssters), order of brth, and the demand functon was estmated wth the help of academc scores, and schoolng level of parents. Ablty of the chldren n studes and mother s educaton were found to be very mportant. Wllams (1983) tred to explan the trends n prvate expendtures on educaton n Australa wth the help of government expendtures, real prce ndex of the cost of educaton, real personal dsposable ncome and the demographc term. More recently and closely related to ths thess study, Tunal, I. (2000) studed the determnants of school attanment of boys and grls n Turkey ncludng ndvdual, household and communty factors, ntroducng fxed effects for regons, and Tlak (2000) made a bref analyss of determnants of household expendture on educaton n Inda, to explan elastcty between nsttutonal and prvate expendtures, also usng households demographc characterstcs. 5

3. CONCEPTUAL FRAMEWORK Ths study pursues an statstcal analyss for Colomban household s expendture n educaton, ntroducng household s cultural and soco-economc characterstcs as determnants of famles behavor when nvestng n educaton of chldren (preschool and prmary educaton), wth a partcular emphass on the household s head characterstcs, and ncludng fxed effects for regons, to control for regonal dfferences that mght affect nvestment behavor. Decsons regardng nvestment n educaton n general and more partcularly at lower levels of educaton are made by famles and rarely by the ndvdual concerned. Hence, famly-household expendture functon s consdered approprate n the present context. 3.1 Determnants Household s expendture decsons depend on varous factors related to the household characterstcs, but n the case of expendture decsons on the educaton for the smallest chldren, partcularly factors related wth the household famly structure characterstcs. 3.1.1 Household s characterstcs: One of the man characterstcs nherent to a household s the number of members of the household. The expected relatonshp between educaton expendture and the sze of the household s ambguous. On one hand, resources have to be dstrbuted between more members, reducng the avalablty for educaton expendtures, but on the other hand, probably more people contrbute to the household budget, whch mght postvely affect allocaton of extra resources n basc educaton. 6

Other exogenous varables that are nherent to the household mght affect the expendtures n educaton: whether a household s located n an urban area for nstance, snce urban centers concentrate more educatonal nsttutons, ncreasng competton n the provson and facltatng access. Also benefts related wth low transportaton costs, and better qualty of the educaton. 3.1.2 Household s head characterstcs: In general, polcy tend to address affrmatve actons to sngle-mothers n many countres, because they are consdered a very vulnerable group n the populaton, and many resources are focalzed to help ther chldren. Colomba s not and excepton, but do the numbers support ths general assumpton? Soco-demographc characterstcs of the household s head lke gender, martal status, level of educaton, and age, should be examned. Do marred parents nvest more n educaton than sngle parents? Do sngle women nvest less n the educaton of ther chldren than sngle men? Do older parents nvest more n educaton? How does the level of educaton of the household head affect ther decson to nvest n educaton? 3.1.3 Regon Fxed Effects: Although hgh levels of enrollments have been acheved at the prmary school level for both boys and grls n much of Colomba (APENDIX 1), substantal regonal dfferences mght reman 5. The poltcal and admnstratve dvson of the country mples a decentralzaton of regulatons and resources, whch mght affect household s nvestment decsons. 5 Tunal, I. (2000). Determnants of school attanment of boys and grls n Turkey: ndvdual, household and communty factors Economcs of Educaton Revew 21 (2002) 455 470, Department of Economcs, Mddle East Techncal Unversty, 06531 Ankara, Turkey Receved 1 May 1999; accepted 15 February 2000 7

4. DATA AND METHODS 4.1 Data Source The analyss uses the Natonal Households Survey, conducted by the Colomban Natonal Department of Statstcs (DANE) n 2008. Ths data covered 13,600 homes n a geographcal area for Natonal Total, Ctes, Captal and 13 Metropoltan areas. Ths study takes a subsample of 3,013 households, to smplfy the analyss and focus on basc educaton expendture varables n poor households. The subsample meets the followng crtera: One famly households: Because most of the analyss focused on the household s head characterstcs, household s wth more than one head were excluded to smplfy the analyss. Households wth chldren under 14 years: Because the study s only seekng to draw conclusons about expendture on basc educaton, ths s the approprate group that generally would be attendng chld-care, preschool and prmary school. Households that have expendtures n educaton hgher than 0, and below 3 mllon pesos (1,500 dollars) per chld: Ths allows usng the logarthm functonal form of expendtures, and establshes a lmt that focused on the poor populaton. Unfortunately the 2008 Survey dd not follow the same households that the prevous Natonal Households Survey n 2003. Despte ths fact, for aggregate comparatve purposes, ths analyss also ncludes data from the 2003 survey whch has detaled observatons for chldren under 5 years old. Ths data covered 24,000 homes n a geographcal area for of nne regons. Ths study takes a subsample of 2,059 households by the same crtera that 2008 data. 8

4.2 Methodology Multvarable regresson analyss wth regonal fxed effects: Ths study estmates an expendture functon to dentfy the effect of the household s specfed characterstcs, ncludng fxed effects analyss to control for regonal dfferences. The conceptual model s expressed as a functonal relatonshp that relates expendtures to ts determnants: Equaton 1 l ) ( exed / chldren = ƒ ( x ) where l(exed/chlden) refers to the logarthm of the household expendture on basc educaton (preschool and prmary school) dvded by the number of chldren under 14 years old that belong to the household. The X denotes the set of ndependent varables specfed as determnants n the model. The equaton takes the followng functonal form: Equaton 2 l ( exduc/ chldren) = β + βsze + βurb + β female + βhmarr + βhed + βhage + βhage + µ 0 1 2 3 4 5 6 7 2 When ntroducng fxed effects the equaton takes the followng form: Equaton 3 l( exed / chldren ) + β reg 1 + β reg 2 + β 8 9 = β + β sze 0 10 1 reg 3 + β + β urb 11 2 reg 4 + β + β 12 3 fem reg 5 + β + β marr 13 4 reg 6 + β + β age 14 5 reg 7 + β + β age 15 6 2 reg 8 + µ + β ed 7 9

where ndependent varables are defned as: 4.2.1 Household s characterstcs: urb = Categorcal varable for whether the household s n a urban area sze = Number of chldren n the household 4.2.2 Household s head characterstcs: female= 1 f the household s head s a female, 0 f not, for household marred= 1 f the household head s marred, 0 otherwse, for household age = Age of the household s head, for household educ= Level of educaton of the household s head for household 0 = none, 1 = preknder, 2 = knder, 3 = preschool, 4 = prmary, 5 = Hgh School, 6 = Techncan, 7 = Technologc, 8 = Undergraduate, 9 = Graduate, 10 = hgher 4.2.3 Regon Fxed Effects: 9 regons: Atlantc (reference group), 1 = Orental, 2 = Central, 3 = Pacfca, 4 = Bogotá, 5 = San Andrés, 6 = Amazona Ornoqua, 7 = Antoqua, 8 = Valle 10

5. RESULTS 5.1 Descrptve Results Column (1) on Table 1 shows the multvarable regresson of all the household characterstcs and famly structure varables descrbed n ths study for 2008 on the log of expendture/ number of chldren under 14; column (2) ntroduces the regonal fxed effects on the same equaton. On the other hand columns (3) and (5) show the separate effect for females and males on the analyss, whle (4) and (6) show fxed effect for each gender respectvely. Table 1 Table 1: Household's Expendture n Educaton n Colomba 2008 Dependent varable: log (Expendture Edcuaton chldren under 14/Number of chldren under14) Fxed Effects Male Fxed Effects Female Fxed Effects Varable (1) (2) (3) (4) (5) (6) Household's Characterstcs Sze -0.1-0.1-0.0943-0.0962-0.1051-0.1047 [0.010]** [0.010]** [0.016]** [0.016]** [0.014]** [0.014]** Urban 0.1654 0.1408 0.2233 0.1802 0.1172 0.103 [0.046]** [0.048]** [0.070]** [0.074]** [0.083]** [0.064]* Household's Head Characterstcs Female 0.0483 0.0447 [0.042] [0.042] Marred 0.0047 0.0022 0.0364 0.0332-0.0236-0.0268 [0.042] [0.042] [0.064] [0.064] [0.057] [0.057] Educatonal Attanment 0.0363 0.0367 0.0425 0.048 0.0307 0.0294 [0.012]** [0.013]** [0.019]** [0.020]** [0.017]* [0.017]* Age 0.0038 0.0036 0.0017 0.0012 0.0055 0.005 [0.008] [0.008] [0.012] [0.012] [0.010] [0.010] R 2 0.035 0.035 0.033 0.033 0.038 0.038 Observatons 3013 5669 1407 1407 1606 1606 Notes: The table shows OLS estmates n (1), (3) and (5) and fxed effects n (2), (4) and (6). Standard Errors are n brackets. A sngle astersk denotes sgnfcance at the 10% level, two asterks denotes sgnfcance at the 5%. 11

Table 2 shows the separate analyss for expendtures made by households n chldcare and preschool vs. expendtures addressed to prmary school. In ths case, the OLS regresson of the famly structure varables for 2008 s on the log of expendture on preschool educaton/chldren under fve n column (1), and the log of expendture on prmary school/ chldren under fourteen (3), ncludng fxed effects columns (2) and (4) n each case. Table 1 Table 2: Household's Expendture n Educaton n Colomba 2008 (Preschool vs. Prmary) Dependent varable: log (Expendture Pre-school/ch <5) log (Expendture Prmary/ch<14) Fxed Effects Fxed Effects Varable (1) (2) (3) (4) Household's Characterstcs Sze -0.1445-0.1445-0.1129-0.1141 [0.030]** [0.029]** [0.010]** [0.010]** Urban 0.1532 0.1626 0.145 0.1166 [0.147] [0.152] [0.046]** [0.048]** Household's Head Characterstcs Female 0.0492 0.0333 0.0481 0.0457 [0.133] [0.133] [0.042] [0.042] Marred -0.251-0.2631 0.0253 0.0247 [0.133]** [0.134]** [0.042] [0.042] Educatonal Attanment 0.0134 0.0141 0.0345 0.0352 [0.037] [0.037] [0.013]** [0.013]** Age 0.0207 0.0268 0.0051 0.0058 [0.023] [0.023] [0.008] [0.008] Age Squared -0.0001-0.0002 0.00003-0.00003 [0.000] [0.000] [0.000] [0.000] R 2 0.066 0.066 0.044 0.044 Observatons 407 407 2873 2873 Notes: The table shows OLS estmates n (1) and (3) and fxed effects n (2) and (4). Standard Errors are n brackets. A sngle astersk denotes sgnfcance at the 10% level, two asterks denotes sgnfcance at the 5%. 12

Table 3 shows the OLS analyss for the 2003 dataset, whch only ncludes nformaton of expendtures on educaton of chldren under 5 years old. The multvarable regresson of the famly structure varables on the log of expendture/chldren under 5 n column (1); the lnear correlaton between these varables and the quadratc functon for the age of the household s head, column (2) ntroduces the regonal fxed effects on the same equaton. On the other hand columns (3) and (4) show the separate effect for females and males on the analyss, whle (4) and (6) show fxed effect for each gender respectvely. Table 2 Table 3: Household's Expendture n Educaton n Colomba 2003 Dependent varable: log (Expendture educaton chldren under 5/hldren under 5) Fxed Effects Male Fxed Effects Female Fxed Effects Varable (1) (2) (3) (4) (5) (6) Household's Characterstcs Sze -0.0617-0.0106-0.0763-0.0072-0.0453-0.0108 [0.016]** [0.014] [0.022]** [0.020] [0.024]** [0.021] Urban -0.8139-0.1313-0.7695-0.0954-0.9243-0.2192 [0.059]** [0.061]** [0.068]** [0.072]* [0.121]** [0.0610]** Household's Head Characterstcs Female -0.1575-0.129 [0.123] [0.109] Marred 0.1384 0.1699 0.2333 0.2082 0.1033 0.149 [0.130] [0.115]* [0.223] [0.199] [0.163] [0.143] Educatonal Attanment 0.1184 0.0765 0.1074 0.0635 0.1441 0.0984 [0.025]** [0.022]** [0.029]** [0.026]** [0.049]** [0.043]** Age 0.0128-0.0014 0.0371 0.0294-0.0167-0.0359 [0.015] [0.013] [0.019]** [0.017]* [0.045] [0.021]* Age Squared -0.0001 0.00001-0.0003-0.0003 0.0001 0.00001 [0.0001] [0.0001] [0.0002]** [0.0001]** [0.0002] [0.0002]* R 2 0.131 0.087 0.078 0.087 0.109 0.075 Observatons 2059 2059 1427 2059 632 632 Notes: The table shows OLS estmates n (1), (3) and (5) and fxed effects n (2), (4) and (6). Standard Errors are n brackets. A sngle astersk denotes sgnfcance at the 10% level, two asterks denotes sgnfcance at the 5%. 13

5.2 Regresson Results The analyss of household level data provdes mportant nsghts n understandng the determnants of household expendtures on educaton. Despte the households n the 2008 survey are not the same ones n the 2003 survey, the regresson analyss for both years adds consstency to some of the results. The number of observatons s consderable, and the fndngs show that that famly structure factors nfluences the levels of early educaton expendtures. Also, some strong regonal effects are evdenced, revelng dfferences n the socal structures across regons n Colomba, and some dfferences between subsamples of preschool vs. prmary school expendtures. Lastly some gender gaps are found that persst over tme. 5.2.1 Household Sze: For the 2008 households, controllng for the number of members of the famly show an on average decrease of approxmately 10% on the expendture on educaton. Ths outcome s sgnfcant at the 5% level and doesn t vary much when ntroducng the regonal fxed effects. For the preschool sample t shows an effect f 14%, whle for prmary school s 11%. These outcomes support the argument that fewer resources are avalable for educatonal nvestment n the case of Colomban households. For the 2003 households, the negatve effect s much smaller though. An effect of 6% decrease on average, and when ncludng the fxed effect t drops to 1% decrease. Ths effect s manly explaned by female-headed household s behavor. Asde of the evdenced dfferences between the 2003 and the 2008 surveys, these results mght be showng a very strong regonal heterogenety present n 2003 that mght have been moderated through the years through educatonal homogenzaton and standardzaton reforms related to enrollment and cost of basc educaton at a Natonal Level 6. 6 Plan Naconal Decenal de Educacón 2006-2016 (PNDE): The Ten-Year Natonal Plan for Educaton ncluded n the Regonal Development Plans, Sectoral plans and other ntatves of educatonal plannng and socal development n Colomba, mplemented after 2006. 14

5.2.2 Urban Households The regresson analyss for 2008 shows that urban households ncrease expendtures on educaton n 16% more compared to rural households. Ths outcome s expected snce urban centers are more lkely to provde more chld educaton optons and facltate access. When controllng by regons the effect s moderated to 14% probably because n most regons rural areas are less nsulated from urban centers, facltatng access to educaton centers and closng the gap. The gap between male and female households though s consderable. Whle males n urban households are estmated to ncrease the expendture n almost 22%, females only do t n 11%. Ths mght be explaned n part by an ncome gender gap. Also, a regonal effect s present n ths case: the fgures drop to 18 and 10% respectvely. On the other hand, for 2003 preschool educaton expendtures show a statstcally sgnfcant postve 83% varaton for urban households. But, ths effect drops dramatcally to 13% when controllng by regon characterstcs, whch suggest a lot of heterogenety among the dfferent regons access to basc educaton centers. In the separate gender analyss the strong regonal effect persst: a 76% effect for males drops to 9.5%, whle for females t changes from 92% to 21%. The dfference between the numbers of 2008 can be partly explaned agan by the homogenzaton and standardzaton reforms n 2006, whch moderates the fxed effects for the 2008 data. 5.2.3 Female households Female-headed households n 2008 have a postve but not sgnfcant effect n the overall sample of around 5% ncrease n expendtures on educaton. When breakng the sample between preschool educaton expendtures and prmary school expendtures, ntroducng fxed effects reduce the effect to 3% only for preschool expendtures, whch mght translate n the restrctons for mothers n some regons when decdng to nvest n the very frst years of educaton of ther chldren. 15

On the other hand the 2003 data, though wth no statstcal sgnfcance, show a negatve effect predctng a decrease of 15% on expendtures for chldren under 5, a consstent result wth current polcy actons that often address female-headed households 7. Probably the stuaton have change for female household heads en the latter years, through government support and educatonal reforms. The ncreasng employment rates for women n Colomba, mght also explan ths transton. 5.2.4 Martal Status Marred household s heads show a small postve effect of crca 0.4% on the expendture n the educaton of ther chldren n 2008. Ths effect s statstcally not sgnfcant for the complete sample, but when breakng the sample, preschool expendtures show a sgnfcant negatve effect of 25%. It could be argued that marred parents wth preschoolers behave dfferently snce one of the parents can stay wth the small chldren at home, and provde some knd of basc cogntve sklls to ther chldren, whle the other parent works, whch s not lkely for sngle parent households. On the other hand, n the 2003 sample the effect for chldren under 5 years old has a dfferent drecton. The fxed effect shows a postve effect of 17% on expendtures n educaton and a bg gap between male and female headed households, the were not present n the 2008 dataset. The analyss then shows nconsstent outcomes, from one year to the other. Lterature supports a postve ncdence of stable famly structures n educatonal outcomes. On the other hand, the probablty of two ncomes contrbutng to the household expendture decsons should ncrease expendtures. But, some other consderatons can be made at ths pont: the way of reportng martal status n a publc survey mght have msleadng outcomes. In the frst place, n a strongly catholc country lke Colomba, martal status has a number of socal 7 Famlas en Accón s a program run by the Colomban government as part of the RAS (Red de Apoyo Socal or Socal Support Network) and s fnanced by a loan from the World Bank and the IADB. It s a Condtonal Cash Transfer program, consstng of cash transfers to poor famles condtonal on chldren attendng school and meetng basc preventve health care requrements. The program targets manly female-headed households. 16

connotatons, whch mght affect the answer of ts respondents. Second, n poor households partcularly, people reman legally marred (hgh cost of dvorce), but n realty ther couples don t lve n the household or contrbute at all n the households budget. Thrd, the change n the methodology and data collecton strategy of the statstcs Colomban Natonal Department of Statstcs mght be reflectng the ntroducton of measurement errors, and other statstcal problems to the 2008 dataset. 5.2.5 Household's Head Level of Educaton: The level of educaton of the household s head s measured n terms of years of schoolng, and has a sgnfcant postve effect of 3.5% on the expendtures n basc educaton n 2008. When controllng by regonal characterstcs the effect persst, whle the data show that women wth hgher educatonal attanment nvest n basc educaton one percent pont less than more educated men. When dvdng the sample between preschool and prmary educaton expendtures, a postve effect of around 1.5% for preschool and a postve sgnfcant effect of 3.5% for prmary expendtures. The effect s also postve and sgnfcant for the 2003 dataset, 11%, whch agan shows a large regonal effect, droppng to 7% and a gender gap of around 4%, where more educated females spent more than ther male counterparts. Lterature supports ths outcome wdely argung that educaton ncreases the awareness of the benefts of educaton. The results here are consstent to show that the educatonal attanment of the parents has a postve effect on educatonal outcomes of ther chldren under 14. 5.2.6 Household's Head Age: The household s head age has no statstcally sgnfcance for any of the 2008 subsamples consdered n ths analyss. For the 2003 data though, a sgnfcant postve outcome s found for male headed households, of around 4% ncrease for chldren under 5 years old and a moderated regonal effect. These results though don t really gve a conclusve outcome as a determnant of expendture n educaton. 17

6. DISCUSSION 6.1 Valdty Threats Some valdty consderatons should be addressed before dscussng the results. In the frst place, usng two dfferent years that don t follow the same households mght have msleadng nterpretatons for external valdty. The most mportant dfferences dentfed between the 2003 and the 2008 surveys and should be remnded throughout the results analyss. For nstance, the 2003 survey ncludes only 30% female-headed households, compared to a 53% of the 2008 sample. On the other hand the 2003 survey ncludes 73% marred household heads, compared to a 50% n 2008, and a much less educated sample, 1,27 levels of educaton on average, vs. 3.5 levels of the 2008 sample. Lastly, the 2003 survey reports on average a slghtly younger sample of household heads, 41 years old, compared to 48 years n the 2008 case (APPENDIX 2). On the other hand, for nternal valdty concerns, usng a sub-sample that only ncludes households that have reported expendtures of at least $1 n basc educaton mght be ntroducng a bas n the estmatons, because every household wth chldren s not beng ncluded n the sample, probably the most vulnerable households, those unable to nvest at all n ths type of expendtures are beng left out. Also the fact that only one famly households are ncluded n the analyss, mght threat nternal valdty, those households mght have bgger problems when nvestng n early educaton, but because most of the analyss focused on the household s head characterstcs, household s wth more than one head would have lmted the analyss. 18

6.2 Statstcal Lmtatons Heteroscedastcty From a cross-sectonal analyss, heteroscedastcty could be expected, as vared types of households lvng n varous parts of the country are sampled together. Usng the sem-log regresson equaton (1) of the expendture functon, a post-examnaton of the estmated resdual squares ddn t exhbt any systematc pattern, meanng no heteroscedastcty was present n the model. Multcolnearty On the other hand, the results show very low values. Multcolnearty can be a problem. Ths means there sn t a lot of varaton that s ndependent from the other determnants consdered n the equaton. Some authors have suggested a formal detecton-tolerance or the varance nflaton factor (VIF) for multcollnearty: Equaton 4 where s the coeffcent of determnaton of a regresson of j on all the other varables. A tolerance of less than 0.20 or 0.10 and/or a VIF of 5 or 10 and above ndcates a multcollnearty problem 8. (APPENDIX 3) 8 O'Bren 2007 19

6.3 Polcy mplcatons Ths analyss pursued to dentfy poor household s constrants to nvestment n basc educaton n Colomba, to understand how educatonal polcy actons should be focalzed. In general, polcy should address populaton that s strugglng wth allocatng resources toward chld educaton. The OLS estmates of the regresson analyss and the fxed effects analyss helped dentfyng major determnants of household expendtures on educaton. Rural-Urban balance of educaton opportuntes Concentraton of economc actvtes, schools and tranng nsttutons n urban centers s a perennal problem n Colomba. Despte dfferences between 2003 and 2008 outcomes, n general data shows a persstng gap remans of 15% on average, ths suggest that decentralzaton polces could have an mportant effect n household expendtures on educaton. If qualty s mproved n rural schools, and more prepared teachers are ncentvzed to move to rural areas, parents mght also be ncentvzed to spend on the educaton of ther chldren. Also, employment opportuntes n rural areas mght ncrease household s budget, and more resources become avalable to allocate for chldren educaton. Other mportant tems of household expendtures on educaton have to do wth transacton costs of lvng n rural areas lke transportaton fees. Gender orented polcy Exstng socal polcy n Colomba targets manly sngle-mother households for elgblty n ncome-transfers and n-knd benefts, but n many households, as suggested before, sngle females can t/won t probe ther real martal status, or are not necessarly the household s chldren mothers. Some part of the data for 2003 n ths analyss supports ths polcy efforts showng that sgnfcant gender dfferences exst and they are normally aganst female-headed households. But on the other hand, results are not really conclusve snce they lack statstcal sgnfcance and consstency for the 2008 sample. Further analyss of ntra-household nequalty and gender gap consderatons should be studed by the government to reallocate polcy programs, f the stuaton have change for female household heads en the latter years. 20

Parents Educatonal Attanment The most consstent and statstcally sgnfcant outcome n ths analyss s the effect of educatonal attanment of households head educatonal attanment n expendtures on educaton. The effect though appears to be moderate. Further polcy acton should be focalzed on ncreasng adult s educatonal attanment, not only from the early educaton nterventons, but also through ntatves targetng youth and adult educaton attanment. Educatonal attanment mght also have employablty postve outcomes for parents. Other consderatons From the statstcal pont of vew, the mportance of the ncdence of famly structure n many socal outcomes, further efforts should be pursue to collect nformaton about martal status, one of the key varables n for ths knd of polcy analyss. 21

7. CONCLUSION Many developng countres have prortzed ther budget towards educaton nvestment to mprove educatonal nfrastructure and qualty of educaton. In Colomba, despte several nsttutonal efforts of publc nvestng n early educaton, they stll don t translate nto household s budgetary consderatons to nvest on early educaton. Identfyng how dfferent factors affect expendture n educaton decsons n households s crucal to do the proper evaluaton of the effcency and qualty of the exstng programs to promote educaton, mplementaton strateges and the role of regulatory mechansms and ncentves structures. Ths study estmates an expendture functon to dentfy the effect of some household s cultural and soco-economc characterstcs as determnants of famly s behavor when nvestng n educaton of ther chldren. The OLS estmates allow many mportant conclusons. In the frst place, the concentraton of economc actvtes, schools and tranng nsttutons n urban centers show the perennal problem n Colomba, evdence of a gap of 15% percent between urban and rural households s founded. Decentralzaton polces mght have an mportant effect n household expendtures on educaton. If qualty s mproved n rural schools, and more prepared teachers are ncentvzed to move to rural areas, parents mght also be ncentvzed to spend on the educaton of ther chldren. On the other hand, polcy actons already targets sngle-mother households, but evdence s really not robust enough to dentfy ths group as the most vulnerable populaton among poor households n Colomba. Further analyss of ntra-household nequalty and gender gap consderatons should be studed by the government to reallocate polcy programs. 22

Fndngs n ths analyss suggest that the educatonal attanment of the parents s the most statstcally sgnfcant determnant of ther expendture n the educaton of ther chldren. Polcy addressng ths factor should be prortzed through specfc programs to mprove educatonal attanment of parents and n general adult populaton. Also, from the statstcal pont of vew, the mportance of the ncdence of famly structure n many socal outcomes, further efforts should be pursued to collect nformaton about many more household characterstcs, to acheve proper statstcal analyss. For nstance, the martal status of the household head, one of the key varables for ths knd of polcy analyss, appears to have a lot of measurement concerns. Adequate surveys and methods are needed to make polcy conclusons. How can programs nowadays address sngle parents wth no relable data about ths households really strugglng wth access to basc educaton for ther chldren. Ths knd of research provdes emprcal evdence for Colomba n ths feld for the frst tme. Hopefully ths analyss would be useful for polcy makers who seek a general understandng of educatonal regulatory polcy and ts mplcatons to natonal development when promotng and structural educatonal reform. In some cases though, dentfyng determnants s not revealng all the factors behnd access, opportuntes and avalablty of resources, but nstead cultural practces and propensty of poor households to under nvest n educaton. Further studes n the feld are needed to contrbute generate a better emprcal pcture of the complextes of households restrctons and behavor when nvestng n educaton. For nstance gender gap studes regardng fertlty, educatonal attanment and labor market partcpaton. 23

8. APPENDICES 8.1 APPENDIX 1 Summary Statstcs 24

8.2 APPENDIX 2 Regressons Specfcatons The analyss ncludes the followng specfcatons; the numbers on the equatons correspond to the columns n Table 1: (1) OLS l ( exed/ chldren) = β + βsze + β urb + β fem + β marr + βed + β age + β age + + µ 0 1 2 3 4 5 6 7 2 (2) Regonal Fxed Effects l( exed / chldren) = β + βsze + β urb + β fem + β marr + β age + β age + β ed + β reg1 + β reg2 + β reg3 + β reg4 + β reg5 + β reg6 + β reg7 + β reg8 + µ 8 9 0 10 1 11 2 12 3 13 4 14 5 15 6 2 7 (3) OLS female households l( exed/ chldren ) = β + βsze + βurb + βmarr + βage + βage + βed + µ fem, 0 1 fem, 2 fem, 3 female, 4 fem, 5 2 fem, 6 fem, fem, (4) OLS male households male, male, male, male, male, 2 male, l( exed/ chldren ) = β + βnch + βurb + βhmarr + βhage + βhage + βhed + µ 0 1 2 3 4 5 6 male, male, 25

8.3 APPENDIX 3 Statstcal Lmtatons Multcolnearty: Some authors have suggested a formal detecton-tolerance or the varance nflaton factor (VIF) for multcollnearty: where s the coeffcent of determnaton of a regresson of explanator j on all the other explanators. A tolerance of less than 0.20 or 0.10 and/or a VIF of 5 or 10 and above ndcates a multcollnearty problem. 26

9. REFERENCES The Program for Internatonal Student Assessment (PISA, 2001, 2004, 2007). Unted Natons Populaton Dvson, World Populaton Prospects: The 2006 Revson, March 2007. UNESCO Insttute for Statstcs, Data Centre, January 2008. http://stats.us.unesco.org/unesco/reportfolders/reportfolders.aspx N Bose & M E Haque & D R Osborn, 2003. "Publc Expendture and Growth n Developng Countres: Educaton s the Key," Centre for Growth and Busness Cycle Research Dscusson Paper Seres 30, Economcs, The Unversty of Manchester. Hough, J (1993) Educatonal cost-beneft analyss Educaton Research Paper No 2 Overseas Development Admnstraton. www.dfd.gov.uk/pubs/fles/edcostandanedpaper02.pdf Department of Educaton and Chld Development, Fact Sheet - Educaton and Tranng Reform Act 2006. 27

Hendey, Leah and Steuerle, C. Eugene. 2010. How does early chldhood care and educaton affect cogntve development? An nternatonal revew of the effects of early nterventons for chldren from dfferent socal backgrounds. Early Chldhood Research Quarterly 25. Becker, G., and N. Tomes (1976): \Chld endowments and the quantty and qualty of chldren," The Journal of Poltcal Economy, 84(4), 143{162. Behrman, J., R. Pollak, and P. Taubman (1982): \Parental preferences and provson for progeny," The Journal of Poltcal Economy, 90(1), 52{73. Tlak, J. B. G. (1987c)." Publc fnancng of educaton n a federal framework: A case of Inda". Invted paper for the Sxth World Congress of Comparatve Educaton. Ro de Janero: Brazlan Comparatve Educaton Socety. July (mmeo). Tlak, J. B. G. (2002)." Determnants of Household Expendture on Educaton n Rural Inda Natonal Councl of Appled Economc Research, Workng Paper Seres No. 88, New Delh, Inda. Tunal, I. (2000). Determnants of school attanment of boys and grls n Turkey: ndvdual, household and communty factors Economcs of Educaton Revew 21 (2002) 455 470, Department of Economcs, Mddle East Techncal Unversty, 06531 Ankara, Turkey Receved 1 May 1999; accepted 15 February 2000. 28