Available online at Procedia Economics and Finance 2 ( 2012 )

Transcription

1 Avalable onlne at Proceda Economcs and Fnance 2 ( 2012 ) nd Annual Internatonal Conference on Qualtatve and Quanttatve Economcs Research Abstract (QQE 2012) A Survey of Tha Student Performance n Mathematcs and Englsh: Evaluatng the Effect of Supplementary Tutorng Nchamon Thongphat* Faculty of Economcs,Thammasat Unversty,2 Prachan Road, Pranakorn, Bangkok 10200, THAILAND Ths study ams to use the propensty score matchng technque to evaluate the tutorng effect n student s school performance model whch s derved from the educatonal producton functon (EPF). The study found that students who obtan tutorng has sgnfcantly dfferent grade from those who do not. Moreover, the technque usng n the study can reduce the bas estmaton problem whch s more sutable than non-matchng method The Authors. Publshed by Elsever Ltd. Selecton and/or peer-revew under responsblty of Global Scence and Selecton Technology and/or Forum peer-revew Open under access under responsblty CC BY-NC-ND of Global lcense. Scence and Technology Forum Pte Ltd Keywords: Treatment Evaluaton; School Performance Evaluaton; Prvate Supplementary Tutorng Effect; Tutorng Effect; Economc Development 1. Introducton Durng recent years, prvate supplementary tutorng schools have grown and became nterestng phenomenon for Tha educaton system. The evdence from Snlarat (2002) and Offce of Prvate Educaton Commsson (OPEC) n Thaland shows the hgher number of prvate supplementary tutorng school and tutored students, from 177 schools and 28,562 students n year of educaton 1985 to 337 schools and 229,249 students n year of educaton The prvate tutorng school busness also has the floursh beneft whch s around 300 mllon Baht for each semester n downtown of Bangkok but ths consumes large resource. Moreover, parents have to oblque hgher educatonal expendture for ther chldren and students have less * Correspondng author. Tel.: ; fax: E-mal address: nchamont@gmal.com The Authors. Publshed by Elsever Ltd. Selecton and/or peer-revew under responsblty of Global Scence and Technology Forum Open access under CC BY-NC-ND lcense. do: /s (12)

2 354 Nchamon Thongphat / Proceda Economcs and Fnance 2 ( 2012 ) tme spendng wth famly or dong other useful actvtes to create ther skll. Snce ndvdual decdes to nvest n hgher educaton based on ther return to schoolng whch wll lead them the hgher human captal accumulaton then hgher ncome (Uruyos and Teeraumpon; 2005), the competton of gettng nto the unversty becomes more severe. Besdes, the school qualty s dfference (Snlarat; 2002, and Suthus; 2010), so the demand for tutorng n Thaland s ncreasng. Moreover, the hgh demand also ensures the nequalty n Tha educaton system and poor atttude, focusng on passng the examnaton, of most students toward learnng. Despte the concernng of polcymaker, Tha students are pressured to obtan more tutorng. Thus, to understand the real nfluence of the tutorng can lead to an effectve polcy. Many educatonal polcy researches (Polydordes; 1986, Sawada and Kobayash; 1986, Kulpoo; 1998, Jukraree; 2001, and Pnyopatsorn; 2002) found the postve consequence of obtanng prvate tutorng and most students decde to obtan tutorng for academc performance goals (Bray; 1999, Bray and Kwok; 2003, and Km; 2004). To evaluate the effect of supplementary tutorng to student s school performance s a useful dea. Snce school performance s one of the mportant ndcators to measure academc achevement of students. However, the sutable method to measure s adoptng the natural experments and randomzed trals whch are less lkely to suffer from the selecton bas as Smmons and Alexander (1978) suggested n educatonal producton functon (EPF) by Glewwe and Kremer (2006). Most studes (Smmons and Alexander; 1978, Comber-Keeves and Thorndke; 1973, Carnoy and Thas; 1974, Polydordes; 1986, Fuller and Nyrongo; 1989, Rumberger and Palardy; 2005, and Glewwe and Kremer; 2006) measured student s school performance wth test or examnaton scores such as GPA, PISA, and the test score n other subjects and languages. The results from these studes could classfy determnants nto characterstcs of famly background, school nputs, peer group characterstcs, and student s ntal endowment. Hence, ths study ams to fnd the nfluence of obtanng supplementary tutorng to student s school performance whch obtaned the dea from Bray (1999) and Snlarat (2002); the survey studes wth descrptve analyss and regresson technque that found the postve relatonshp between tutorng and student s school performance. Anyway, the contrbuton of ths study s evaluatng the effect of tutorng on students academc performance by reducng selecton bas problem wth propensty score matchng (PSM) technque and adoptng the tutorng and sutable varables nto educatonal producton functon (EPF). The queston of the study s whether the score dffer between tutored and untutored students. Then, the study tres to analyze other factors for achevng n school performance by usng non-matchng method. The data n the study were obtaned by mplementng a survey technque to 590 hgh school students, both government and prvate schools, n grade 12 n Bangkok. Moreover, the study evaluates tutorng effect to student s school performance on Fundamental Mathematcs and Fundamental Englsh whch every hgh school students have to learn n manstream school. The paper s organzed as follows. In secton II, the theoretcal backgrounds behnd ths study are provded. Then, secton III states the study s methodology and framework. Fnally, all the results and conclusons are demonstrated n secton IV. 2. Theoretcal Frameworks The study mplements tutorng effect to school performance framework whch contans 2 theoretcal frameworks; school performance framework (Glewwe and Kremer; 2006) and treatment effect framework (Angrst; 2008, and Cameron and Trved; 2005). Then, the study apples the PSM technque to analyse the effect. To measure the school performance, the study has to adopt the educatonal producton functon (EPF) as mentoned by Glewwe and Kremer (2006) A a( S, QCHI,,, ) (1)

3 Nchamon Thongphat / Proceda Economcs and Fnance 2 ( 2012 ) where A s student s performance from school, S s years of schoolng, Q s a vector of school qualty, C s a vector of chld characterstcs ncludng nnate ablty, H s a vector of household characterstcs, and I s a vector of educatonal nputs under parents control (.e. chldren s daly attendance, purchases of textbooks). Assume that vector C, H and Q are exogenous. And let P s a vector of prces related to schoolng (.e. school fees, prces for school supplers purchased by parents) and P can affect educaton outcome through decsons made for the endogenous varables S and I, so t does not appear n (1). Assume that P s exogenous. Then, vectors S and I could be expressed as the general functons of the exogenous vectors S f( QCHP,,, ) (2) I g( QCHP,,, ) (3) Parents maxmze household utlty (student s performance s ncluded) wth respect to each schoolng choce by choosng S and I subject to constrants mentoned n (2) and (3). After nsertng (2) and (3) nto (1), we wll get the reduced form for A whch s a casual relatonshp, not a producton functon, because t reflects household preference and prces. A h( QCHP,,, ) (4) Snce the estmaton of (1), (2), and (4) wll cause bas n the ordnary least square (OLS) estmaton, the way out s adoptng natural experments and randomzed tral to estmate program or treatment effects. Ths treatment effect framework s contrbuted by Angrst (2008), and Cameron and Trved (2005) whch wll provde the two possble estmaton of treatment effect n order to reduce the selecton bas. Let there be a target populaton for the treatment of nterest. Whle N denotes the number of randomly selected ndvduals who are elgble for treatment, N T denotes the number of randomly selected ndvduals who are treated, then NC N N T denotes the number of nontreated ndvduals who serve as potental control group. Let y denotes the vector of observatons on the scalar-valued outcome varable, x denotes the vector of observable varable ( x ) whle 1,..., N, and D denotes a bnary ndcator of a treatment varable. Assume that anyone who s assgned treatment gets t and who s not does not get t, then y 1 denotes the outcome varable of the treated ndvdual and y 2 denotes the outcome varable of the nontreated ndvdual. Assume that no random assgnment mechansm for treatment and absence of general equlbrum n the sense of applyng to observatonal data and not evaluaton the treatment program that affects the entre sector. To measurng the effect of the treatment would be to construct a measure that compares the average outcomes of the treated and nontreated groups. However, selecton bas problem could be solved by mplementng matchng technque wth propensty score. For each ndvdual, Y Y0 D( Y1 Y 0), Y 1, and Y 0 are observed. Snce Regresson estmates of causal effects can be easly motvated by postulatng a constant-effects model as; Y Y ; s constant (5) 1 0 The condtonal ndependence assumpton s EY [ X, D] X (6) 0

4 356 Nchamon Thongphat / Proceda Economcs and Fnance 2 ( 2012 ) where Then, s a vector of regresson coeffcents E Y D E D X / E D D E D X (7) By the law of large numbers, sample regresson coeffcents estmate n ths populaton regresson coeffcent wll be consstent. For matchng technque, assume that the only source of selecton bas s the set of observed covarates, X. Then, treatment effects n matchng are constructed by matchng ndvduals wth the same covarates nstead of through a lnear model for the effect of covarates. The dentfyng assumpton s weaker as the effect of covarates on need not be lnear. Hence, the condtonal ndependence assumpton s whch can be mpled as E Yj X, D E Yj X ; for j =0,1 (8) EY Y E EY X, D 1 Y X, D 0 (8a) Hence, t can construct consstent estmators from ther sample analogs. Accordng to dffculty n computaton n matchng model, the propensty score s a useful way for matchng the condtonal probablty of treatment gven covarates. Because t bases on the fact that, f condtonng on X elmnates selecton bas, then condtonng on P D 1 X can do so. For all the theores above and the ams to verfy the consequence of obtanng supplementary tutorng, the frst hypothess of the study s the student s school performance, measured by grade, s dfferent between tutored and untutored students. For another purpose to fnd the alternatve ways of the students to acheve n school performance, the second hypothess could be stated that all control varables n the study are correlated wth school performance. 3. Research Methodology Ths study separates the methodology nto 1) data collecton and processng and 2) analyss method Data Collecton and Processng The data come from a survey study of 590 hgh school students n grade 12 n Bangkok. The number of observaton was selected, wth proportonal samplng, from each 2 types of school whch are Offce of the Prvate Educaton Commsson (OPEC) and Offce of the Hgher Educaton Commsson (OHEC) whch are classfed as prvate school type. Offce of the Basc Educaton Commsson (OBEC) and Department of Educaton Bangkok Metropoltan Admnstraton (DBMA) whch are classfed as government school type. Then 2 schools from each type of school were sampled from 2 sampled dstrcts from each 3 zones of area n Bangkok. These zones were dvded by the number of housng n each dstrct, whch are hgh, mddle, and low number of housng. Anyway, the sampled schools and dstrcts were selected by convenence samplng. The respondents had to answer the questons n three parts. Frstly, nformatonal backgrounds of students

5 Nchamon Thongphat / Proceda Economcs and Fnance 2 ( 2012 ) part, t contans demographc, characterstcs of students famles and ther school performance nformaton. Secondly, characterstcs of obtanng tutorng part whch ncludes obtanng tutorng decson nformaton. The last one s characterstcs of actvtes. They had to answer about ther chances to obtan tutorng n Fundamental Mathematcs and Fundamental Englsh wthn last 4 semesters; from the 1 st semester n grade 10 to the 2 nd semester n grade 11, ther grade of each subject wthn these perods, and ther average GPA of each subject Analyss Method To test the hypothess, the study apples the tutorng effect to school performance framework by usng PSM approach from Cameron and Trved (2005). The model represents the relatonshp between obtanng tutorng of the students, as the treatment, and the GPA of tutored subject and other varables. The propensty score was calculated by a logstc regresson, then match the most smlar score of treated ndvdual and controlled one, dependng on each PSM algorthm, to verfy whether tutorng can brng dfferent n GPA as stated n the hypothess. Moreover, the study attempts to fnd other factors for student s school performance by non-matchng technque. Therefore, the model specfcaton s YA, YNA, f ( D, X ) (9) whle Y A, denotes the average GPA of each subject of tutored students, Y NA, denotes the average GPA of each subject of untutored students. f ( D, X ) s the educatonal producton functon (EPF) whle D denotes the treatment whch s obtanng tutorng of students ( D = 1 f student obtans tutorng n each subject, and D = 0 f student does not obtan tutorng n each subject) and X denotes a vector of control varables whch s correlated wth dependent varable, and can be a factor of school performance. Snce X = (prvsch +, gender +, area +, paroc +, pared +, famncome +, dep +, gpax +, peergpax +, schclasze, jonmathpro +, jonengpro +, dstance, sklllearn +, hobby +, fambz +, housework +, parttme + ) 4. Results and Conclusons Propensty score matchng technque uses the logstc regresson to estmate the propensty score of all covarate varables and match the treated wth the control ndvdual who has the most smlar propensty score, accordng on each PSM algorthm. The more smlar score ndvdual has, the more smlar value of varable one gets and ths s the balance property of the technque whch can ease the bas estmate problem Matchng Results To verfy the bas reducton, the matchng technque should have sgnfcantly dfferent average value of score between treated and control unts n each varable (Null hypothess must be rejected) before matchng process and ndfferent of those two values (Null hypothess s not rejected) n each varable after matchng. The results n table 1 shows the best bas reducton algorthms n evaluatng Mathematcs and Englsh whch are Radus (0.05) matchng and Fve Nearest Neghbour and Radus (0.05) matchng, respectvely, comparng to Sngle Nearest Neghbour, Radus (0.01), and Mahalanobs matchng. These best matchng methods also provde less dfferent of the average value of score n each varables between the treated and the control groups than non-matchng method. Thus, matchng methods can ease the based estmate n both Mathematcs and Englsh. Table 2 also shows the results from each propensty score matchng method whch the students grade s dfferent between those who obtan tutorng and not, n both Fundamental Mathematcs and Englsh. Snce the crtcal t-value s less than t-value from the estmaton, then the null hypothess s rejected, for

6 358 Nchamon Thongphat / Proceda Economcs and Fnance 2 ( 2012 ) Mathematcs, at 99% sgnfcant level by usng Fve Nearest Neghbor and Radus (0.05) matchng; and 95% sgnfcant level by usng the Mahalanobs matchng. For Englsh effect, the null hypothess s rejected by usng the Radus (0.05) and Mahalanobs at 99% sgnfcant level and 95% sgnfcant level by usng the Fve Nearest Neghbor matchng. Therefore, Mathematcs tutorng does have sgnfcant mpact on student s school performance, or school grade of Fundamental Mathematcs. The average effect of Mathematcs tutorng of obtanng tutorng students (ATT) s , , and by usng the Radus (0.05), Fve Nearest Neghbor, and Mahalanobs matchng. Whle the average effect of Englsh tutorng of obtanng tutorng students (ATT) s , , and , by usng the Mahalanobs, Radus (0.05), and Fve Nearest Neghbor matchng, respectvely. Moreover, there s postve bas n non-matchng method n both subjects evaluaton whch can be nvestgated from later results Non Matchng Results From the non-matchng results n table 3, the female student s the only one of demographc background factors that has postve sgnfcant effect only on student s Englsh grade. Meanwhle, for famly background factors, students who have ther famly ncome hgher than 50,000 Baht/month have postve sgnfcant relatonshp wth both Mathematcs and Englsh grade, comparng to students who have less than 10,000 Baht/month of famly ncome. For the educatonal background, students from prvate school are found to have postve sgnfcant relaton wth Mathematcs and Englsh grade. By comparng to students from other majors, students from Mathematcs and Scence and Foregn Language are found to have postve effect to Englsh grade. Students who have of GPAX are found to have postve sgnfcant effect to both Mathematcs and Englsh grade, comparng to those students whose GPAX s less than Furthermore, manstream school class sze has postve sgnfcant relatonshp wth student s Mathematcs and Englsh grade. Fnally, the actvty background factors, students who had to take part n ther famly s busness are found to have postve effect to Mathematcs grade. Whle students who spent tme learnng other not-academc sklls and dong some housework are found to have negatve relaton to Mathematcs and Englsh grade, respectvely Conclusons and Polcy Implcatons From the study, most of matchng methods gve the statstcal evdence that the student s school grade s dfferent between students who obtan supplementary tutorng and who do not n both Mathematcs and Englsh. So, the prvate supplementary tutorng polcy lmtaton should be replaced by new educaton polcy whch focuses more on academc program to create student s skll and ablty on each subject. School qualty should be provded by manstream school to create the educatonal equty to the students snce famly ncome, type of school and school class sze are found to be sgnfcant factors for student s school achevement n both Mathematcs and Englsh. Fnally, students need famly support for academc and extra currculum actvtes snce students who had to work n famly busness are found to have postve sgnfcant effect on student s scores n Mathematcs. Moreover, the results n table 3 ensure postve bas n nonmatchng technque snce tutorng varables, mathtu and engtu, have hgher sgnfcant effect to students performance than PSM matchng technque. Moreover, all results explan that postve bas occurs from famly and educatonal background factors. For examples, students who have hgher level of famly ncome or/and hgher level of GPAX are able to be potental students to have hgher GPA of each tutored subject. Snce the PSM technque can analyse the dfference of students characterstcs, and then the bas s reduced.

7 Nchamon Thongphat / Proceda Economcs and Fnance 2 ( 2012 ) Table 1. Test for Based Reducton Before and After Matchng n Mathematcs and Englsh Tutorng Effect Mathematcs Effect Evaluaton Englsh Effect Evaluaton Varables Non Non Radus (0.05) 5 Nearest Neghbour Radus (0.05) Matchng Matchng mean %bas reducton mean mean %bas reducton mean %bas reducton mean prvsch % 0.09** % % 0.09** gender % % % 0.03 area % -0.11** % % -0.11** area % 0.07** % % 0.07** paroc % 0.07* % % 0.07* paroc % % % paroc % 0.06** % % 0.06** paroc % paroc % -0.14*** % % paroc % 0.04* % % -0.14*** paroc % -0.03* 0.04* -1827% % 0.04* pared % % % -0.03* pared % -0.12*** % % pared % % % -0.12*** pared % % % pared % % % 0 famncome % -0.06*** % % 0.04 famncome % -0.16*** % % -0.06*** famncome % 0.06* % % -0.16*** famncome % % % 0.06* famncome % % % dep % % % 0 dep % -0.19*** % % 0.02 dep % -0.02** % % -0.19*** gpax % -0.1*** % % -0.03** gpax % -0.06* % % -0.1*** gpax % % % -0.06* peergpax % -0.1*** % % 0.06 peergpax % 0.09** % % -0.1*** peergpax % 0.09** % % 0.09** schclasze % 2.82*** % % 0.09** jonmatpro % 0.12*** jonengpro % % 2.82*** dstance % 4.09** % % 0.12*** sklllearn % 0.06** % % 4.09** hobby % 0.09** % % 0.06** fambz % % % 0.09** housework % % % parttme % % % 0.03 ***. **. *, are sgnfcant at 1%, 5%, 10%. 15% respectvely. Acknowledgements I would lke to thank my major advsor, Dr. Supacha Srsuchart, and my co-advsor, Assstant Professor Dr. Sttsak Leelahanon, and Dr. Dlaka Lathappat for lots of teachng, and very good advces. Thanks also the school drectors, teachers, and students n hgh school level who dedcated ther tmes n my survey study.

8 360 Nchamon Thongphat / Proceda Economcs and Fnance 2 ( 2012 ) Table 2. Average Effect of Tutorng to Student s Fundamental Mathematcs and Englsh GPA Tutorng effect to student s Mathematcs grade Methods Fve Nearest Neghbour Radus (0.05) Mahalanobs Non Matchng Treated Controls ATT S.E T-stat 3.09*** 3.43*** 2.11** 9.12 % ncrease 11.10% 11.36% 8.08% Tutorng effect to student s Englsh grade Methods Fve Nearest Neghbour Radus (0.05) Mahalanobs Non Matchng Treated Controls ATT S.E T-stat 2.44** 2.71*** 2.56*** 7.56 % ncrease 7.38% 7.66% 7.92% ***. **. * are sgnfcant at 1%, 5%, 10% respectvely. Table 3. Alternatve Factors to Student s GPA n Mathematcs and Englsh Varables Demographc Background gender + area1 + area2 + area3 + Famly Background paroc1 + paroc2 + paroc3 + paroc4 + paroc5 + Descrpton Gender of student (female = 1, male = 0) Lvng n area1 (hgh number of housng) Lvng n area2 (mddle number of housng) Lvng n area3 (low number of housng) Parent occupaton s merchant Parent occupaton s farmer/fsherman Parent occupaton s publc offcal Parent occupaton s hrer Parent occupaton s Employee of publc enterprse Coeffcents (Std. Error) Maths Englsh (0.0665) (0.0686) (0.0918) (0.1101) (0.1483) (0.0993) (0.2980) ** (0.0645) (0.0662) (0.0889) (0.1066) (0.1435) (0.0961) (0.2879) Varables Descrpton paroc6 + Parent occupaton s husband/housewfe Parent occupaton s paroc7 + Employee of prvate enterprse paroc8 + Parent has other occupatons pared1 + Parent educatonal level s lower than prmary school pared2 + Parent educatonal level s prmary school degree Parent educatonal level s pared3 + hgh school degree/vocatonal certfcate Parent educatonal level s pared4 + dploma/hgh vocatonal certfcate pared5 + Parent educatonal level s bechelor degree pared6 + Parent educatonal level s master degree or hgher famncome Famly ncome s less than ,000 Baht/Month Coeffcents (Std. Error) Maths Englsh (0.0845) (0.0810) (0.1352) (0.2023) ** (0.2716) ** (0.2670) ** (0.2759) ** (0.2695) ** (0.2865) (0.1307) (0.1939) (0.2620) * (0.2575) (0.2659) (0.2599) (0.2766)

9 Nchamon Thongphat / Proceda Economcs and Fnance 2 ( 2012 ) Varables Descrpton Famly ncome s famncome ,000 20,000 Baht/Month Famly ncome s famncome ,001 30,000 Baht/Month Famly ncome s famncome ,001 40,000 Baht/Month Famly ncome s famncome ,001 50,000 Baht/Month famncome Famly ncome s hgher 6 + than 50,000 Baht/Month Educatonal Background prvsch + dep1 + dep2 + dep3 + dep4 + gpax1 + Student enrolls n school (prvate = 1, government = 0) Major study s Mathematcs and Scence Major study s Foregn Language and Scence Major study s Foregn Language Student has other majors Student s GPAX s not over 2.00 gpax2 + Student s GPAX s gpax3 + Student s GPAX s gpax4 + Student s GPAX s gpax5 + Student s GPAX s References Coeffcents (Std. Error) Maths Englsh (0.1267) (0.1422) (0.1505) (0.1635) ** (0.1441) ** (0.0837) (0.2462) (0.2585) (0.2487) * (0.2989) *** (0.2974) *** (0.2991) *** (0.3040) (0.1226) (0.1374) (0.1454) (0.1594) (0.1412) ** (0.0810) * (0.2371) (0.2493) ** (0.2402) * (0.2886) 0.840* (0.2874) 0.933** (0.2888) 1.451*** (0.2935) Varables peergpax1 + peergpax2 + peergpax3 + peergpax4 + peergpax5 + schclasze jonmatpro + jonengpro + dstance Actvty Background Descrpton Student s peer GPAX s not over 2.00 Student s peer GPAX s Student s peer GPAX s Student s peer GPAX s Student s peer GPAX s Sze of school class Attendng specal program experence n Mathematcs Attendng specal program experence n Englsh Dstance between school and housng Coeffcents (Std. Error) Maths Englsh (0.3694) (0.3585) (0.3663) (0.3553) (0.3660) (0.3544) (0.3707) (0.3603) * (0.0043) (0.0041) (0.0963) (0.0856) (0.0015) (0.0015) sklllearn + Learnng other sklls whch ** are not academc (0.0991) (0.0958) hobby + Havng any hobbes (0.0670) (0.0647) fambz + Takng part n famly s busness (0.1225) (0.1177) organze + Takng part n any ** organzaton project (0.0966) (0.0925) housework + Dong some housework * (0.0907) (0.0877) parttme + Havng any part tme jobs (0.2260) (0.2182) ***. **. *, are sgnfcant at 1%, 5%, 10%. 15% respectvely. - represents droppng varable n the model. + represents dummy varable. [1] Angrst, J. D. (2008). treatment effect. In S. N. Durlauf & L. E. Blume (Eds.), The New Palgrave Dctonary of Economcs. Basngstoke: Palgrave Macmllan. [2] Bray, M., & Plannng, U. I. I. f. E. (1999). The Shadow Educaton System: Prvate Tutorng and Its Implcatons for Planners. Pars: Unted Natons Educatonal, Scentfc and Cultural Organzaton. [3] Bray, M., & Kwok, P. (2003). Demand for Prvate Supplementary Tutorng: Conceptual Consderatons, and Soco-Economc Patterns n Hong Kong. Economcs of Educaton Revew, 22(6), [4] Bray, M. (2006). Prvate Supplementary Tutorng: Comparatve Perspectves on Patterns and Implcatons. Compare: A Journal of Comparatve and Internatonal Educaton, 36(4), [5] Cameron, A. C., & Trved, P. K. (2005). Mcroeconometrcs: Methods and Applcatons. Analyss (Vol. 100, p. 1056). Cambrdge Unversty Press. [6] Deheja, R. H., & Wahba, S. (2002). Propensty Score Matchng Methods for Non-Expermental Causal Studes. The Revew of Economcs and Statstcs, 84(1), [7] Foondun, A. R. (2002). The Issue of Prvate Tuton: An Analyss of the Practce n Maurtus and Selected South-east Asan Countres. Internatonal Revew of Educaton, 48(6), [8] Glewwe, P., & Kremer, M. (2006). Chapter 16 Schools, Teachers, and Educaton Outcomes n Developng Countres. In E. Hanushek and F. Welch (Eds.), Handbook of the Economcs of Educaton (Vol. Volume 2, pp ): Elsever. [9] Hussen, M.G. A. (1987). Prvate Tutorng: A Hdden Educatonal Problem. Educatonal Studes n Mathematcs, 18 (1),

10 362 Nchamon Thongphat / Proceda Economcs and Fnance 2 ( 2012 ) [10] Imbens, G. M. & Wooldrdge, J. M. (2008). Recent Developments n the Econometrcs of Program Evaluaton. NBER Workng Paper, No [11] Km, T. (2004). Shadow Educaton: School Qualty and Demand for Prvate Tutorng n Korea. KDI School of Pub Polcy and Management Paper No [12] Km, J.-H., & Park, D. (2010). The Determnants of Demand for Prvate Tutorng n South Korea. Asa Pacfc Educaton Revew, 11(3), [13] Khandker, S. R., Koolwal, G. B., & Samad, H. A. (2010). Handbook on Impact Evaluaton: Quanttatve Methods and Practces. The Internatonal Bank for Reconstructon and Development. [14] Lockheed, M. E., Fuller, B., & Nyrongo, R. (1989). Famly Effects on Students' Achevement n Thaland and Malaw. Socology of Educaton, 62(4), [15] Rumberger, R. W., & Palardy, G. J. (2005). Test Scores, Dropout Rates, and Transfer Rates as Alternatve Indcators of Hgh School Performance. Amercan Educatonal Research Journal, 42(1), [16] Smmons, J., & Alexander, L. (1978). The Determnants of School Achevement n Developng Countres: A Revew of the Research. Economc Development and Cultural Change, 26(2), [17] Beebout, H. S. (1972). The Producton Surface for Academc Achevement: An Economc Study of the Malaysan Secondary Educaton. Ph. D. dssertaton, Unversty of Wsconsn. Madson., Cted by Smmons, J., & Alexander, L. (1978). The Determnants of School Achevement n Developng Countres: A Revew of the Research. Economc Development and Cultural Change, 26(2), [18] Carnoy, M. (1971). Famly Background, School Inputs and Students Performance n School: The Case of Puerto Rco. Mmeographed. Palo Alto, Calf.: Stanford Unversty., Cted by Smmons, J., & Alexander, L. (1978). The Determnants of School Achevement n Developng Countres: A Revew of the Research. Economc Development and Cultural Change, 26(2), [19] Comber, L. C. & Keeves, J. P. (1973). Scence Educaton n Nneteen Countres. New York: Halsted Press., Cted by Smmons, J., & Alexander, L. (1978). The Determnants of School Achevement n Developng Countres: A Revew of the Research. Economc Development and Cultural Change, 26(2), [20] Carnoy, M. & Thas, H. H. (1974). Draft Report of Second Tunsa Educaton Research Project RP0248. Mmeographed. Washngton, D.C.: I.B.R.D., Cted by Smmons, J., & Alexander, L. (1978). The Determnants of School Achevement n Developng Countres: A Revew of the Research. Economc Development and Cultural Change, 26(2), [21] de Slva, W. A. (1994). Extra-school Tutorng In the Asan Context Wth Specal Reference To Sr Lanka. Maharagama: Department of Educatonal Research, Natonal Insttute of Educaton. [22] Kulpoo, D. (1998). The Qualty of Educaton: Some Polcy Suggestons Based on a Survey of Schools--Maurtus. SACMEQ Polcy Research Report: Unted Natons Educatonal, Scentfc, and Cultural Organzaton, Internatonal Inst. for Educatonal Plannng, 7-9 rue Eugene-Delacrox, Pars, France. [23] Polydordes, G. (1986). The Determnants of Educatonal Achevement at the End of Secondary Schoolng: The Case of Greece. [24] Schefelben, E. & Farrell, J. (1973). Factors Influencng Academc Performance among Chlean Prmary Students. Paper presented at the Annual Meetng of the Internatonal Socety of Educatonal Planners and Amercan Assocaton for the Advancement of Scence, Mexco Cty., cted by Smmons, J., & Alexander, L. (1978). The Determnants of School Achevement n Developng Countres: A Revew of the Research. Economc Development and Cultural Change, 26(2), [25] Sawada, T., & Kobayash, S. (1986). An Analyss of The Effect of Arthmetc and Mathematcs Educaton of Juku. Translated wth An Afterward by Horvath, P.; Compendum 12, Tokyo: Natonal Insttute for Educatonal Research. [26] Jukraree, C. (2001). The Effect of Coachng on the Predctve Valdty of the Entrance Examnaton of Prnce of Songkla Unversty. Unpublshed Master s Thess, Srnakharnwrot Unversty. [27] Snlarat, Patoon. (2002). Prvate Tutorng for Hgh School Students n Thaland. Offce of the Natonal Educaton Commsson. [28] Suthus, O. (2010). An Analyss of Market Structure of Tutoral Schools n Patumwan Dstrct, Bangkok. Unpublshed Master s Thess, Srnakharnwrot Unversty, School of Economcs and Publc Polcy. [29] Uruyos, Manachaya., & Teeraumpon, Swapong. (2005). Prvate Tutorng and Economc Growth. Faculty of Economcs, Chulalongkorn Unversty: The Fsrt Natonal Conference of Economst. Bangkok.