1. HEALTH AND GROWTH: A META-REGRESSION ANALYSIS UNIVERSITY OF IOANNINA DEPARTMENT OF ECONOMICS Nikos Benos and Georgios Giotis MAER-Net Colloquium, Prague 2015

2. -The theoretical literature on human capital and economic growth.  Human capital can be considered as those skills, abilities and knowledge embodied in individuals.  Human capital is acquired through education, health, training, migration and other investments that enhance individual productivity.  Education and health are considered to be the most significant investments in human capital. 2015 MAER-Net ColloquiumIntroduction

3. A classification of theoretical works is based on the different roles human capital in the process of economic growth. - Growth models can be divided into two categories: exogenous and endogenous growth models.  Exogenous growth models include the Solow-Swan model and its extensions (augmented neoclassical models).  Endogenous growth models consider human capital accumulation and productive knowledge as driving forces of economic growth.. 2015 MAER-Net Colloquium

4. The empirical literature on health human capital and economic growth.  The empirical results remain controversial, since they have not always been consistent with those produced by theoretical growth models.  Health human capital proxies are often not significant or enter with negative sign in growth analysis.  Empirical evidence depends on technical problems that have to do with: I.the definition of the variables, II.the methodology used, and III.the time period over which the model is estimated.. 2015 MAER-Net Colloquium

5. MRA  Given the diversity of empirical findings on the link between health and growth, we conduct Meta-Regression Analysis (MRA).  MRA is quantitative literature review of the estimates obtained from previous regression analyses.  Meta-analysis integrates the results of several studies that share a common aspect so as to be combinable in a statistical manner (Harmon et al., 2003).  MRA aims at explaining the excess study-to-study variation in empirical results and investigates the presence of publication selection bias (Stanley, 2005). 2015 MAER-Net Colloquium

6. We proceed in two steps for conducting MRA: 1.Construct meta-data set, collecting empirical studies examining the link between health and growth. 2.Define meta-regression model to distinguish between numerous factors, which influence the estimated health effect on economic growth. In all steps of our analysis we follow the guidelines and protocols expressed by the Meta-analysis of Economics Research-Network (MAER-Net). 2015 MAER-Net Colloquium

7. We have searched EconLit, Google Scholar to find English- written articles in academic journals and working papers, estimating the health-growth nexus. The keywords used in the search process were: human capital, health and economic growth. Include only macro studies in meta-sample, which estimate the coefficient of the size effect of health on growth. Only studies providing regression results where measure of growth rate is dependent variable and at least one health measure is among the explanatory variables are included in our meta-data set. 2015 MAER-Net Colloquium

8. We perform meta-regression analysis using data from 42 empirical studies. However, the coding of the studies is ongoing and the number of studies included in the meta-sample will be considerably higher. Include all reported estimates in each study, any potential dependence among estimates is captured by study identifiers. Given that most studies include many estimations, we use all of them as independent regressions, report a total of 688 observations. We calculated the partial correlations from each study. 2015 MAER-Net Colloquium

9. The meta-regression model:  β j = β 0 + Σα k Z jk + β 1 se j + u j (1), where: β j reported estimate of health coefficient of j th study, β 0 true value of health coefficient, Z jk moderator variables which explain variation in β j, α k MRA coefficients reflecting effect of particular study characteristics se j standard error of coefficient of j th study u j meta-regression disturbance term.. 2015 MAER-Net Colloquium

10. Empirical studies use varying sample sizes, econometric specifications and estimation procedures. Hence, u j are likely to be heteroscedastic. Thus, we estimate the Weighted Least Squares (WLS) version of equation (1), by dividing it by se j : t j = β 1 + Σγ i K ij + β 0 (1/se j ) + Σα k Z jk /se j + v j (2), where: t j is t-statistic which corresponds to the estimate β j. K ij are additional factors correlated with publication process itself. We follow general-to-specific modeling approach for variable selection. The estimation methods we use are: i) OLS, ii) OLS-cluster, iii) REML, iv) FE, v) Weighted-Least-Squares and vi) FE-WLS. 2015 MAER-Net Colloquium

11. Figure 1: Funnel graph (n=688) 430 estimates < 0 431 estimates > 0 5 estimates = 0

12. Table 1: Summary statistics of the studies included in meta-regression analysis Number of estimates MinimumMaximumMedian Standard deviation Mean 1 Bhargava-Jamison-Lau-Murray 2001 40.11278930.20534120.15522270.03857140.157144 2 Lorentzen-McMillan-Wacziarg 2008 470.0390137 0.5377780.29884220.0980810.2962789 3 Bloom-Canning-Sevilla 2004 20.09107180.17640160.13373670.06033730.1337367 4 Hassan-Cooray 2012 26-0.244281910.2372839260.0410489780.1065808260.027527711 5 Dauda 2011 25-0.78800760.6300946-0.19368810.3649692 -0.108819 6 Chakraborty 2004 40.4040490.63782310.47676510.10887340.4988506 7 Bloom-Malaney 1998 20.05077180.26878830.159780.1541610.15978 8 Ogunleye-Eris 2008 6-0.06776410.4447640.00793110.23866550.1354641 9 Bloom-Finlay 2008 40.13160990.17566880.1544320.02019570.1540357 10 Hamoudi-Sachs 1999 90.01396870.39600250.23113370.13541380.1832942 11 Bloom-Canning 2005 40.12934160.21416010.16537710.03485910.168564 12 Ogunleye 2011 2-0.03755910.0471830.00481190.05992170.0048119 13 Aguayo-Rico 2005 16-0.86527720.0368825-0.31622070.2976298 -0.3671903 14 Hartwig 2010 90 -0.360570.3111134-0.23786560.1429659-0.1830541 15 Aghion Howitt Murtin 2011 46-0.43494670.67651920.27058220.33385760.2495588 16 Barro 2013 20.19210570.19585180.19397870.0026489 0.1939787 17 Barro 1996 7 -0.16380540.19529040.19094040.13277090.1296216 18 Akram 2009 60.24291590.99999340.99996130.3090633 0.8737888 19 Acemoglu-Johnson 2007 8-0.09443950.52002440.03683810.23504880.1368345 20 Bloom-Canning-Fink 2013 12 -0.4894350.7236468-0.03866920.4037553 0.1249301 21 Grimm 2011 22 -0.67068180.5253376 0.17143290.33606170.019544

13. Number of estimates MinimumMaximumMedian Standard deviation Mean 22 Acemoglu-Johnson 2014 16-0.3866729-0.1136748 -0.29732360.094554 -0.2742631 23 Coorey 2013 260.00759360.52711780.10203740.1435864 0.1409324 24 Pocas-Soukiazis 2013 20-0.4323282 -0.0274014-0.26034620.1081623-0.2572113 25 Morgado 2014 40.32773240.2771657-0.01469220.2912972-0.0199878 26 Naidu-Chand 2013 2 0.01096030.13886620.0749133 0.0904431 0.0749133 27 Strittmatter- Sunde 2013 88-0.52738830.1363126-0.21499340.1563704-0.2129748 28 McCarthy-Wolf-Wu 2000 12-0.311311-0.1195245-0.17765830.0627895-0.2056611 29 Nketiah-Amponsah 2009 1 0.369881 - 30 Rivera-Currais 2004 9-0.13598570.3470660.24540380.1548530.1779704 31 Kumar-Mitra 2009 2-0.3165010350.225946776-0.045277130.383568525-0.04527713 32 Bloom-Canning-Fink-Finley 2009 30.1927532790.4668023170.3264230760.1370382070.328659557 33 Magnus-Powell-Prufer 2010 2300.0762908790.0382639370.016632810.035733569 34 Barro 1996 40.2178220790.2331886120.219155620.0072670530.222330482 35 Cervellati-Sunde 2011 18-0.0991324040.046921478-0.0326485160.038253654-0.035900685 36 Suhrske-Urban 2010 34-0.276109750.257386953-0.131022910.132438-0.094009152 37 Hansen 2014 24-0.3691669790.1667258020.0092873620.161062829-0.060852529 38 Acemoglu-Johnson 2013 26-0.2082810820.089734833-0.0675422580.078132284-0.064155108 39 Afonso-Jalles 2013 8-0.1237669260.164756211-0.0737661920.104004459-0.033609615 40 Afonso-Allegre 2007 2-0.337897177-0.243573802-0.2907354890.066696698-0.290735489 41 Miller-Russek 1997 6-0.248311345-0.010049871-0.1797270190.083391651-0.16170273 42 Devarajan-Swaroop-Zou 1996 26-0.3049971410.3537432430.0856981680.1568524920.074243354

14. Table 2. Moderators in the Multiple Meta-Regression Analysis. VariableDescription of the variable K-variablesSample size= the size of the sample Z-variablesinvse=1/standerror Health variablesLife expectancy=1 if study uses life expectancy as proxy for health adult survival rate=1 if study uses adult survival rate as proxy for health mortality=1 if study uses mortality as proxy for health healthexpend=1 if study uses health expenditure as proxy for health Additional controlspcapital= if study uses physical capital as explanatory variable political=1 if study uses political variable as explanatory variable fiscal=1 if study uses fiscal variable as explanatory variable demog=1 if study uses demographic variable as explanatory variable pcgdp=1 if study uses GDP pc as explanatory variable openness=1 if study uses openness variable as explanatory variables schooling=1 if study uses education variable as explanatory variables Effect measurement var.loghealth=1 if study uses log of health variable as explanatory variable Data variablesearlyyearFirst year of sample lastyearLast year of sample obsSample size Qualitypublic=1 if study is published in academic journal Specification variablesindependent= number of independent variables ols=1 if study uses OLS estimation IV=1 if study uses IV estimation panel=1 if study uses panel data in the sample Country moderatorsOECD=1 if study uses OECD countries in the sample

15. -Assuming that all α k and γ i are zero,we employ the Funnel Asymmetry Test or FAT for publication selection: t j =β 1 +β 0 (1/se j ) + e j (3) Table 3: FAT – PET tests 2015 MAER-Net Colloquium VariablesOLSOLS-clusterREMLFEWLSFE-WLS invse 0.9745795*** (0.0083094) 0.9745795*** (0.0483479) 0.9745817*** (0.0083095) 0.9878064*** (0.0054442) 1.024898*** (0.0019008) 0.9993504*** (0.0011459) constant -13.56022*** (0.4564856) -13.56022*** (1.815613) -13.56075*** (0.4565126) -16.92275*** (2.115023) -22.08949*** (0.6732306) -17.23793*** (2.112692) R-squared0.9525 95.24% (Adjusted) 0.99380.99760.9998 Ramsey RESET test F(3, 683) = 3151.78 Prob > F = 0.0000 F(3, 683) = 3151.78 Prob > F = 0.0000 -F(3, 642) = 689.17 Prob > F = 0.0000 F(3, 683) = 6935.32 Prob > F = 0.0000 F(3, 642) = 979.92 Prob > F = 0.0000

16. Table 4: Multiple MRA (General-to-Specific) ModeratorsOLSOLS-clusterREMLFEWLSFE-WLS invse-4.954885 (5.81282) -4.75473*** (1.131392) -38.67527*** (11.03288) -6.521372*** (2.00681) -144.245*** (26.11726) -162.2581** (69.41389) lifexpectancy se 0.5728103*** (0.025954) 0.5246986*** (0.1119178) 0.3785084*** (0.0482872) 0.5248614** (0.2292944) 0.2569807*** (0.02415730) 0.0087589** (0.0035193) asrse0.6004023*** (0.0354863) 0.6876609*** (0.1248938) 0.4857136*** (0.0818584) 0.4678013** (0.02253665) 0.2878495*** (0.1045601) mortalityse0.4786432*** (0.0378257) 0.5174235*** (0.1125756) 0.382556*** (0.0483545) 0.4387827* (0.2321043) 0.265295*** (0.0558326) 0.0047896*** (0.0012525) healthexpend se 0.4554834*** (0.0360014) 0.5185342*** (0.1134509) 0.37558*** (0.0482951) 0.4355875* (0.2311656) 0.2552161*** (0.0505899) pcapitalse0.1187521** (0.0597831) politicalse-0.0962391* (0.0580775) -0.0725786* (0.0413745) -0.1140728** (0.0553672) fiscalse0.1527059*** (0.0437664) 0.0814832* (0.0420769) demogse pcgdpse0.1109274*** (0.0300665) 0.1543276** (0.0564921) 0.0634092* (0.0351573) 0.059688** (0.024953) openessse0.0168764* (0.0090839) schoolingse-0.00661672** (0.0333537) -0.01704737*** (0.0435144) -0.0147475** (0.0720169) -0.157932*** (0.0420475) -0.207941*** (0.0581286)

17. ModeratorsOLSOLS-clusterREMLFEWLSFE-WLS loghealthse0.0853329*** (0.0288474) 0.3804074*** (0.0496019) 0.2622761** (0.1263145) 0.6536539*** (0.0522448) 0.6496928*** (0.0997399) loggdpse0.0566082** (0.0265359) 0.0840303* (0.043673) -0.0567871* (0.031742) -0.0559627* (0.031624) publicse-0.218693*** (0.053403) -0.130664** (0.0598798) -1.285459*** (0.1025633) -1.487794*** (0.3016201) earlyyearse0.0017318*** (0.0003361) 0.0023478*** (0.0005704) 0.0020655** (0.00086) 0.0034352*** (0.0009158) 0.0050321*** (0.0010224) 0.0066438*** (0.0017439) lastyearse-0.0098592** (0.0039211) obsse0.0005302*** (0.0000812) 0.0004833*** (0.0000625) 0.000717*** (0.0000836) 0.0005416*** (0.0000815) Independent se 0.0281693*** (0.0043363) 0.020515*** (0.0052838) 0.0268547* (0.0148301) 0.0145058** (0.0057576) 0.0208934* (0.0104596) olsse0.0881661*** (0.0260463) 0.1123361*** (0.0272101) 0.1251389*** (0.0242303) 0.1498104** (0.0560428) ivse0.0797895*** (0.0258407) 0.0772722*** (0.0270671) 0.0871327*** (0.0257016) 0.1161138** (0.0522497) panelse-0.0870967*** (0.0262189) -0.102632*** (0.0320081) 0.0639528* (0.0365299) 0.1726486*** (0.0354127) 0.1701726*** (0.0530974) oecdse-0.0962151* (0.0508852) obs-0.035752*** (0.0042325) -0.037476*** (0.0036561) -0.0448501*** (0.0038293) -0.035272*** (0.0070162) -0.004252*** (0.0015215) constant-2.820381*** (0.3331904) -2.711719*** (0.7327533) -2.13143*** (0.5409898) -2.419682** (1.158048) -2.12159*** (0.5020711) -1.295005 (1.138706) R-squared0.9535 95.29% adjusted0.99480.99860.9999

18. 2015 MAER-Net Colloquium

21. Thank you very much for your attention