Internationnal remittances and the education of yuong generations : the case of Vietnam : Luận văn thạc sĩ

\ UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES HO CHI MINH CITY THEHUGUE VIETNAM THE NETHERLANDS VIETNAM- THE NETHERLANDS PROJECT OF M.A ON DEVELOPMENT ECONOMICS INTERNATIONAL REMITTANCES AND THE EDUCATION OF YOUNG GENERATIONS: THE CASE OF VIETNAM A thesis submitted in partial fulfillment of the requirements for the degree of MASTER OF ARTS IN DEVELOPMENT ECONOMICS BY r-·-80GIAoot~c8AorJ?.o~· · TRUONG f)~l HQC I chi2 Pseudo R2 -32.576065 141 70.40 0.0000 0.5169 Robust enroll I Odds Ratio Std Err z [95% Conf Interval] P>lzl -+ -ln re - pc 1.48276 2871809 2.03 0.042 1.014402 2.167362 ln - iner pc 2.15169 7299474 2.26 0.024 1.106662 4.183546 per sa child 9804709 -0.62 0311926 0.535 9212016 1.043554 hhage 1.008787 0235571 0.37 0.708 963657 1.056031 hhgender 4692536 3695771 -0.96 0.337 1002338 2.196854 hhschool 928508 0980361 -0.70 7549391 0.482 1.141982 hhemp 2.12801 2.362831 0.68 496 2414591 18.75442 urban 2725652 2256652 -1.57 053794 0.116 1.381041 6116435 -3.41 age 0881099 0.001 4611891 8111808 gender 262118 193607 -1.81 0.070 061628 1.114849 emp 028475 0229841 -4.41 0.000 0058533 1385239 Table 4.9 Logistic regression results of Model (restricted) (sum of wgt is 1.4939e+06) Logistic regression Log pseudolikelihood enroll Number of obs Wald chi2(7) Prob > chi2 Pseudo R2 -33.316233 Odds Ratio Robust Std Err z P>lzl 141 57.26 0.0000 0.5060 [95% Conf Interval] -+ -ln re ln_iner_pc age gender emp hhschool urban 1.522214 "2.110347 6481507 267579 0271455 9424958 2717626 2932323 7184485 0633847 1921244 0222355 1021579 2333765 2.18 2.19 -4.43 -1.84 -4.40 -0.55 -1.52 69 0.029 0.028 0.000 0.066 0.000 0.585 0.129 043529 1.082855 5350992 065506 0054508 7621089 0504916 2.220482 4.112801 7850869 1.093008 1351878 1.165579 1.462717 Table 4.10 Diagnostic test to compare Modell and Model Measures of Fit for legit of enroll (Efron's R2, Count R2, and Adj Count R2 not calculated if pweight used) Current Saved Difference Model: logit logit N: 141 141 Log-Lik Intercept Only -67.435 -67.435 0.000 Log-Lik Full Model -0.720 -33.296 -32.576 D 66.592(133) 65.152(129) 1.440(4) LR 68.277(7) 69.718(11) 1.440(4) Prob > LR 0.000 0.000 0.837 McFadden's R2 0.506 0.517 -0.011 McFadden's Adj R2 0.388 0.339 0.049 ML (Cox-Snell) R2 0.384 0.390 -0.006 Cragg-Uhler(Nagelkerke) R2 0.623 0.634 -0.010 McKelvey & Zavoina's R2 1.000 000 -0.000 Efron's R2 Variance of y* 75752.041 84394.370 -8642.329 Variance of error 3.290 3.290 0.000 Count R2 Adj Count R2 AIC 0.586 0.632 -0.047 AIC*n 82 592 89.152 -6.560 BIC -591.593 -573.238 -18.355 BIC' -33.636 -15.281 -18.355 BIC used by Stata 106.183 124.537 -18.355 AIC used by Stata 82.592 89.152 -6.560 Difference of 18.355 in BIC' provides very strong support for current model Note: p-value for difference in LR is only valid if models are nested The calculation of BIC' is based on LR chi-square The difference in the BIC' from the two models indicates which model is more likely to have generated the observed data If BIC' - BIC' < is negative, the first model is preferred In contrast, in case of obtaining positive difference of BIC', it is appropriate to choose the second model Absolute Difference Evidence 0-2 Weak 2-6 Positive 6-10 Strong >I Very Strong Source: Long & Freese (200 I, pp 82) Table 4.11 Hosmer and Lemeshow's goodness-of-fit test 70 Logistic model for enroll, goodness-of-fit test (Table collapsed on quantiles of estimated probabilities) + + I Group I Prob I Obs I Exp_1 I Obs_O I Exp_O I Total I -+ + -+ -+ -+ -+ -l I I I I I I I I I I 0.1555 0.6611 0.8484 0.9063 0.9511 I I I I I 11 12 11 I I I I I 1.2 6.8 10.8 12.4 13.0 I I I I I 14 3 I I I I I 13.8 7.2 3.2 1.6 1.0 I I I I I 15 14 14 14 14 I I I I I -+ + -+ -+ -+ -+ -l I I I I I 10 I I I I I 0.9679 0.9825 0.9909 0.9955 0.9992 I I I I I 14 14 14 14 14 I I I I I 13.4 13.7 13.8 13.9 14.0 I I I I I I I I o I I 0.6 0.3 0.2 0.1 0.0 I I I I I 14 14 14 14 14 I I I I I + + 141 10 number of observations number of groups Hosmer-Lemeshow chi2(8) Prob > chi2 5.98 0.6490 -> With a very large p-value, we can say that our model fit the data well Table 4.12 Diagnostic test for model specification error 71 (sum of wgt is 1.4939e+06) Logistic regression Log pseudolikelihood = Number of obs Wald chi2(2) Prob > chi2 Pseudo R2 -33.296117 141 56.84 0.0000 0.5062 Robust Coef Std Err z [95% Conf Interval] enroll I P>lzl -+ -1.326979 00169 1659672 6.04 0.000 6763997 _hat I 0.978 _hatsq I -.0010213 0374905 -0.03 -.0745013 0724586 -.7489865 7530341 cons I 0020238 3831756 0.01 996 0 It is necessary to check whether the suggested model in this paper has all relevant predictors and if the linear combination of them is sufficient In order to run specification error test, the author uses Pregibon' s link test Cl applied in Stata In the specification error test illustrated above, the coefficient of indicator _ hatsq has p-value of z statistic very large Therefore, we can conclude that we should not be able to find additional predictors that are statistically significant except by chance Table 4.13 Diagnostic test for Multicollinearity (2) The idea behind link test is that if the model is correctly specified, one should not able to find any additional significant predictors The link test uses the linear predicted value (_hat) and linear predicted value squared (_hatsq) as the predictors to rebuild the model The variable _hat should be a statistically significant predictor as it is the predicted value from the model In addition, if our model is properly specified, the variable _hatsq should not be a significant predictor Therefore, in this case, if _hatsq is significant, then the linktest is significant That means we face the problems of omitting relevant variables or incorrect specified model (Bruin, 2006) 72 Collinearity Diagnostics VIF SQRT VIF ln_re_pc ln _iner_pc age gender emp hhschool urban 1.06 1.16 1.24 09 1.27 1.05 1.23 03 08 1.11 04 1.13 02 1.11 Mean VIF 1.16 Variable Eigenval 5.3819 0.9532 7239 0.4681 0.3862 0.0515 0.0308 0.0045 Condition Number Tolerance 0.9435 0.8609 0.8059 0.9175 0.7849 0.9518 0.8119 RSquared 0.0565 0.1391 0.1941 0.0825 0.2151 0.0482 0.1881 Cond Index 1.0000 2.3762 7266 3.3907 3.7331 10.2260 13.2290 34.4719 34.4719 Eigenvalues & Cond Index computed from scaled raw sscp (w/ intercept) Det(correlation matrix) 0.6189 The table 4.13 above illustrates two common measures including tolerance (an indicator of how much collinearity that a regression analysis can tolerate) and VIF (Variance Inflation Factor -an indicator of how much of the inflation of the standard error could be caused by collinearity) Tolerance= 1- R2 and VIF =1 I Tolerance In the event that one variable are fully uncorrelated with others, both the tolerance and VIF indicator are equal to However, it is necessary to pay attention if the tolerance goes to 0, and the variance inflation gets very large In our case, the table shows 12 or times of the average of leverage abs(Pearson Residuals) > abs(Deviance Residuals) > 75 Table 4.14 Logistic regression results of Model3 (after excluding influential observations) (sum of wgt is 1.4778e+06) Number of obs Wald chi2(7) Prob > chi2 Pseudo R2 Logistic regression Log pseudolikelihood = -31.256704 enroll I Odds Ratio Robust Std Err z P>lzl 139 54.74 0.0000 0.5327 [95% Conf Interval] -+ -ln_re_pc ln_iner_pc age gender emp hhschool urban 1.545007 2.39067 6261224 2239274 0208594 8928885 2373395 2996108 8233401 0686412 1695276 0173668 1192891 1910207 2.24 2.53 -4.27 -1.98 -4.65 -0.85 -1.79 0.025 0.011 0.000 0.048 0.000 0.396 0.074 1.056486 1.217218 5050599 05078 0040797 687191 0490104 2.259421 4.695384 7762035 9874656 1066548 1.160158 1.14935 Table 4.14 shows that all coefficients of independent variables excluding variable HHSCHOLL are statistically significant at 1%, 5% and 10% level The author does not pick the variable of number of school years of household head out of the analyzed model because it is meaningful in reality Table 5.2 Marginal effects of explanatory variables on the probabilities of enrollment Marginal effects after logit y = Pr(enroll) (predict) 94665118 variable dy/dx Std Err z P>lzl 95% C.I X + ln_re_pc ln_iner_pc age gender* emp* hhschool urban* 0219701 0440168 -.0236459 -.0788982 -.5901581 -.0057216 -.0840042 01088 01757 00854 04552 18466 00727 06143 2.02 2.50 -2.77 -1.73 -3.20 -0.79 -1.37 0.044 0.012 0.006 0.083 0.001 0.431 0.171 000638 043302 009571 078463 -.04039 -.006902 -.168106 01031 -.952077 -.228239 -.019963 00852 -.204407 036399 (*) dy/dx is for discrete change of dummy variable from to 76 7.46277 8.81733 13.1972 514911 12641 1.81585 429499 ... overview of international remittances and education status of young generations in Vietnam 21 CHAPTER 3: OVERVIEW OF INTERNATIONAL REMITTANCES AND EDUCATIONAL ATTAINMENT OF YOUNG GENERATIONS IN VIETNAM. .. times larger than the size of the impact of other financial sources In rural areas, the effect of remittances is about 2.6 times larger than the size of the impact of other kinds of income Estimated... characteristics ofthe family Z: characteristics of the young generations This research chooses to measure the impact of international remittances on the probability of young generations to attend schools The