On the within level, the only significant coefficient had a variable gross regional product, where a percent increase in the GRP, ceteris paribus, was associated with a decrease in changes in IMR by 2.15. Sectoral product shares had no significant relationships with the changes in IMR. Second regression more specifically explains which sector activities affected changes in IMR. Coefficient for the agricultural value was not significant, but the secondary and tertiary sectors had a negative correlation with the changes in IMR. A percent change in industrial and services value, ceteris paribus, is associated with a decline in changes in IMR by 2.2 and 3.4 respectfully.

On the between province level, a percent increase in the share of the Kurdish population, ceteris paribus, is associated with the increase in the negative changes of the infant mortality rates by 6.49. Additionally, a percent increase in the average total GRP value, ceteris paribus, is associated with a decrease in the negative changes of IMR by 3.09. A percent change in the share of agricultural product, ceteris paribus, correlates with a change in IMR of -10.29. That means that the increase in the share of agricultural product might've influenced the increase in IMR. These results imply that disparities in changes in infant mortality rates were related to the Kurdishness of a province, share of agriculture in total GRP, and total regional product. Higher number of Kurds in a region, and higher share of agriculture correlated with increases, or negative changes in mortality rates, while higher economic growth was correlated with positive changes, and a decrease in IMR. From the second regression on IMR, we can conclude that the increases in sectoral and industrial values imply lower infant mortality rates.

Regarding the changes in nursing staff per capita, within a province gross regional product value had a positive coefficient. A percent change in total GRP, ceteris paribus, is associated with an increase in delta number of nurses by 0.809. Additionally, a percent increase in the share of agricultural product in total GRP, ceteris paribus, is associated with the decline in the changes in nursing staff of 1.6. Among the between provinces variables, Kurdishness and total GRP had a significant coefficient, where a percent increase in the share of the Kurdish population, ceteris paribus, was associated with an increase in the changes in number of nurses by 0.211. These results imply that the differences between changes in the number of nurses were positively correlated with the higher economic activity within a province, across six years. Using the results from the second regression on nursing staff we can conclude that industry and services had the most positive effect on the increase in the number of nurses. Moreover, provinces with higher share of the Kurdish population were positively correlated with the increase in nursing staff. These results are in line with the positive changes in literacy rates, where Kurdish province most likely experienced faster improvements in both health and education indicators.

4. Discussion

Results in the previous chapter show how economic performance and ethnic characteristics correlate with the state provided goods in education and services. The first set of regressions showed that ethnic characteristics of provinces are positively correlated with illiteracy rates and infant mortality rates, and negatively with the higher education diploma holders. However, the share of Kurdish population had no significant relationship with the number of nurses and number of secondary schools per capita. Public goods provision indices which were constructed using health and education variables also showed that the latent variable measured, in this case, public goods provision, is associated with the share of the Kurdish population. This correlation was negative, meaning that the index had lower scores for provinces with higher shares of the Kurdish population. Additionally, using two indices - one for direct provision, and another for the outcomes of the government's provision of goods, findings imply that ethnic diversity does impair public goods provision. Therefore, we argue that on the provincial level, ethnic characteristics of provinces have a strong negative relationship with the state-funded public goods provision. These findings imply that historically, Kurdish provinces received less since they have lower scores in the indices, and therefore receive significantly less funds for health and education. Because of the potential bias of previous governments towards Kurdish areas, we cannot argue that the AKP-led government continued with this practice.

The economic performance variables, specifically gross regional product components, and sectoral shares in total product have a significant relationship with public goods provision, especially on the within-province level. Therefore, yearly improvements in the province, or the annual positive changes are strongly associated with the increase in the gross regional product value. Thus, we can conclude that the economic growth brought positive changes in the level of goods provided.

Across provinces, economic performance seems to vary quite a lot, depending on the proxy public provision variable used in the regression. The differences in the share of people with higher education, infant mortality rates, and the number of nurses between provinces seem to be associated with the differences in the agricultural product. Higher agricultural product leads to a lower number of medical staff per capita and lower education levels. It is also significantly correlated with the increase in infant mortality rates. That means that the higher agricultural activity in a province most likely negatively affects a provision in health. This might be the case since individuals engaged in the agricultural production tend to have less education, and therefore less knowledge when it comes to childcare. Additionally, provinces with higher agricultural share in total product are also less developed, which is in line with the theory that lower development is associated with more infant deaths. The increase in the industrial product negatively affects illiteracy rates across provinces. Lastly, a bigger service sector positively correlates with higher education and a larger number of facilities and medical staff, and negatively with infant mortality rates and the number of secondary schools.

Indices, created using the exploratory factor analysis, are positively correlated with the industry and service sectors, meaning that the public goods provision in education and health increases with the higher secondary and tertiary sector activities. On the other hand, higher share of the agricultural product negatively affects the Index scores. When we look only at the direct provision index, which is loading on the number of health and education facilities and staff employed in these two sectors, increases in all three sector values leads to higher provision within a province. However, across provinces, agricultural product negatively correlates with the direct provision of goods, while services product per capita increases direct provision. Additionally, the share of the industrial product negatively associates with Direct Provision Index scores, while the provinces with a higher share of services sector more likely have higher Direct Provision Index scores. That means that urban areas, with a high tertiary sector, most likely have more hospitals and schools per capita, as well as more staff employed in health and education. Unemployment rates between provinces are significantly associated with numbers of schools, hospitals, hospital beds and staff employed. Therefore, a province with a higher mean unemployment rate will most likely have lower Direct Provision scores. The Indirect Provision Index seems to be negatively correlated with the increase in the average agricultural product, meaning that provinces whose agricultural sector is growing, might have lower human capital scores (measured by educational attainment variables and infant mortality rates). A stronger services sector highly correlates with better human capital scores, while the industry sector does not significantly correlate with the Indirect Provision Index. However, the share of the agricultural product associates negatively with the Indirect Provision Index, meaning that provinces majorly engaged in the agriculture most likely have lower educational levels and worse health indicators.

In the period 2009 - 2014, the amount of publicly provided goods seemed to depend on the ethnic characteristics of a province. Kurdish provinces did profit less and still are. To account for the possible bias, lower economic development has been controlled for using the proxy for income level disparities. Therefore, possible heterogeneity bias is controlled for, which implies that it is not only because of the lower economic performances that Kurdish provinces have lower provision. It is because they on average have fewer facilities per capita, and due to the results of the indirect provision index, most likely have lower efficiency in the education and health sectors. All three indices have the lowest scores for Kurdish provinces, but highest scores vary between metropolitan and western provinces. For example, Ankara, Eskisehir, Istanbul, and Izmir have the best scores for the Public goods provision index and the Indirect index, but for the Direct Provision Index, provinces Isparta, Trabzon, Bolu, Tunceli, and Kirikkale. Interestingly, provinces with the highest scores for the direct provision index are homogenous Turkish provinces, except for the province Tunceli which has 29.4 percent of the population being Kurdish, and it is in the Eastern Anatolia region and thus closer to the Kurdish provinces. According to the Tunceli province official website, the population is divided into 47 tribes, with the majority being Zaza Kurds. Additionally, they have very low infant mortality rate, of 10.85, one of the lowest in Turkey (Government of Tunceli, 2017).

Therefore, results implying that the delegation of public goods was associated with the ethnic structure are most likely stemming from historically low investments and the oppression of the Kurdish population in the twentieth century. Due to the data limitations, we could not analyze the period before 2009. However, since the goal of this research is to see whether current the AKP government delegates goods based on the ethnic structure, changes in the education and health provision after AKP won the elections in 2002 are the ones of interest.

Changes in illiteracy, which is a proxy of public goods provision in education show that the AKP government delegation was positively associated with the Kurdishness of a province. That means the mechanisms of funding were designed to increase the human capital in underdeveloped Kurdish regions. Therefore, the hypothesis that Kurdish regions profited than the other regions regarding the public provision in education is rejected. Kurdish provinces profited significantly more than others. Specifically, investment incentives in the agriculture sector had a very high impact on the changes in illiteracy rates. These findings are in line with the results from the regression using illiteracy as the dependent variable, where the agricultural product had the highest coefficient out of three gross regional product components. Therefore, government incentives targeting illiteracy had a positive effect on the increase in the literate population, and we can conclude that AKP interest was pure to boost economic activity among Kurdish areas, and by that to increase human capital.

On the other hand, changes in infant mortality rates were negatively associated with the Kurdishness of a province, the share of agricultural product in a province, and gross regional product value. This means an increase in infant mortality rates was significantly correlated with higher proportions of the Kurdish population and a higher share of agricultural product in total GRP of a province. However, higher economic growth measured by total GRP value is associated with a decrease in infant mortality rates. Moreover, since most of the provinces experienced an increase in infant mortality rates since 2002, we can say that the increase in IMR was more pronounced in Kurdish areas and less pronounced if a province had higher economic growth based on service and industry sectors. Since the share of the agricultural product seems to increase IMR scores, investment incentives in the agricultural sector might've had negative effect on public goods provision in health. Additionally, infant mortality rates increased significantly in the years 2013 and 2014 in the province of Kilis, where Syrian refugee camps are located. However, the inflow of the refugees from Syria did not affect infant deaths in the Kurdish provinces that received refugees.

Contrary to findings from the IMR regression, changes in the nursing staff per capita had a positive correlation with the Kurdishness of a province. Also, increases in gross regional product value were associated with positive changes within a province. More economic activity, therefore, implies positive changes in the number of nursing staff per population. Even though R^2 was not as high as in the previous regression, results with Delta nursing staff per capita as a dependent variable confirmed the results from the illiteracy regression, where Kurdish provinces received, on average, more provision or experienced faster positive changes. Regarding the infant mortality rates results, it is important to highlight that the data on infant deaths and live births does not correspond with findings and official statistics provided by UNICEF, World Bank, and the Turkish government. Therefore, we cannot make any final conclusions based on the changes in IMR. However, changes in illiteracy rates and the size of the nursing staff in public health sector imply that the delegation of goods in health and education from the AKP-run government was positively correlated with the share of the Kurdish population. This is in line with the findings from the Jordan study (Gao, 2016), where Gao challenges the common agreement in academia that the underproduction of public goods depends not only on the different preferences, and the competition between ethnic groups, but also on the fragmentation inside the ethnic group. Additionally, in countries such as Turkey, where few political parties compete for the votes, central governments might try to attract voters by incentivizing the economic activity. Therefore, Kurdish provinces profited more than the others, which might be the result of that competition. It is important to note that Kurdish provinces are usually governed by Kurdish political parties (DTP, and HDP). Therefore, the contract-awarded firms in the Southeastern region, and provinces such as Diyarbakir, are governed to some extent by Kurdish representatives. However, AKP-awarded firms still get the highest share from the procurements, even in Kurdish provinces (Gьrakar, 2016). However, it seems that the higher share of agricultural product, and in some cases industrial product negatively impacts public goods provision, especially when it comes to the effects of that provision through infant mortality rates and illiteracy. Since incentives from the government are focused on raising the agricultural product in underdeveloped Kurdish areas, it might be of concern that because of these incentives, IMR and illiteracy do not follow the same trend as in homogenous Turkish, and industry and services based provinces.

Erdogan's goal is to increase the standard of living and, through better economy, acquire more public support. Regarding the Kurdish issue, by engaging in the peace process with PKK, and allowing Kurdish language courses in universities, he was the first Turkish leader willing to solve the ethnic conflict. In a country which stigmatizes the Kurdish population (Kuzu, 2016), and with the Kurdish population being fragmented, it is very hard to start with any radical changes related to the Kurdish question. Therefore, better economic performance accompanied by small sociopolitical changes seem to be the safest way of building the trust and cooperation between two major ethnic groups in Turkey, while both sides benefit from the economic growth. The possible obstacle, however, is the ongoing conflict in Syria, where the Syrian Kurds are being supported with the ammunition from the U.S. to fight against ISIS. Many news articles imply that Erdogan is afraid of the resolution of the Syrian conflict, mainly because Iraqi Kurds have high level of autonomy, and Syrian Kurds could be next (Totten, 2015) (BBC, 2016). If that happens, it might lead to a conflict between Turkish Kurds fighting for the autonomy, and Turks on the other side defending the territory of Turkey. Because of this, Erdogan specifically highlighted that Turkey will never allow for another country to be created on its southern border, implying to Kurds. (Totten, 2015)

Finally, it is necessary to highlight the limitations of this research. Firstly, the observed period is too short, and we could not discuss any changes between previous government practices, and the AKP. Data for the province level is poorly recorded in the official statistics. Thus, we had to use proxies for main factors. Additionally, exploratory factor analyses reduces the explained variation, and that is important to know when interpreting the index results. Most importantly, the ethnic structure is captured by the mother language variable from 2004 DHS dataset, where the actual ethnic structure might be significantly different and therefore affect beta coefficients. Further research of this topic should include a longer period, and possibly better quality and more precise data. Since public goods provision can be measured in many ways, that also means more diverse results. Using factor analysis is a common choice for clustering variables to measure the underlying factor, but it might be avoided if data could be digitalized and available for academic research.

Conclusion

In this research, I tried to answer the question of whether Kurdish provinces profited less from the provision of public goods in education and health because of the ethnic characteristics. By controlling for income levels and urban population, it was possible to account for the different development stages of provinces and focus only on the economic versus ethnic effects. The overall public goods provision is lower in the Kurdish areas, most likely because of the historical disparities and the discrimination of the Kurdish population from the central government. This means that the lower development rates of Kurdish provinces might have been an outcome of the government policies that were not designed to foster economic growth and development of Kurdish areas in the past. Conversely, using the changes in public goods provision, it was possible to estimate the trend of public goods delivery during the AKP mandate. Accounting for the changes showed that the Kurdishness of a province was positively associated with improvements in health and education in the period 2009-2014.

Therefore, Kurdish provinces historically profited less because of the ethnic characteristics. However, looking only at the mechanisms of public goods delivery during the AKP mandate they profited more, because of the investment incentives. The government initiated several projects whose aim was to increase the economic activity in Kurdish areas, which lead to the increase in public investments in health and education. Therefore, the hypothesis that Kurdish provinces received less because of the ethnic characteristics, during the AKP mandate, is rejected. Nevertheless, educational issues persist. Learning in the non-native language obviously affects lower education performances of the Kurdish population, especially in lower educational levels. In order to provide equal opportunities for all, scholars argue that Turkish government should be more actively engaged in the promotion of multiculturalism (Kuzu, 2016), together with the investments in Kurdish regions' development.

But, as this study implies, Erdogan and the AKP are working on the structural issues population is facing. When people are satisfied with their economic status, and living standards it is easier to discuss the social and political issues, because Kurdishness stopped being an ethnic issue a long time ago. Because it is a political issue, it is very sensitive and it significantly affects voting outcomes. I argue that because of a century long and exhausting ethnic conflict, and stigmatization of Kurdish population, stronger actions, such as official recognition of Kurds, are not possible simply because a political party would lose popularity. Therefore, it seems that Erdogan is strategically investing in Kurdish areas, to gain more popularity while also not overlooking the Turkish nationalists who are used to, and advocate the pro-Turkish discourse. Moreover, with the April 2017 referendum which directly increased his legitimate power, and the arrests of the Kurdish politicians based on weak links with PKK, he is sending a clear message to everyone not to intervene with his plan.

Recommendations for the further research are, firstly, to include longer period, because it would help in establishing the trend and the breaking point from the discriminatory practices in public goods provision to pro-growth promotion. Additionally, this research ignores the ethnic fragmentation and its effect on the competition for public funding. If ethnic fragmentation is properly estimated, it would be possible to track the changes in provinces where fragmentation is low, to see if the public provision was higher than it the provinces with highly fragmented population. Hence, more specific conclusions would arise, especially when it comes to the alleged Kurdish support of Erdogan, and the election results.

Bibliography

1. Acemoglu, D., & Ucer, E. (2015). The Ups and Downs of Turkish Growth, 2002-2015: Political Dynamics, the EU and the Institutional Slide. NBER Working Paper No. 21608.

2. AFAD. (2016, December 12). Current Status in AFAD Temporary Protection Centres. Retrieved May 11, 2017, from Republic of Turkey Prime Ministry: https://www.afad.gov.tr/en/2602/Current-Status-in-AFAD-Temporary-Protection-Centres

3. Agency for Healthcare Research and Quality. (2008, January ). Chapter 3: Creation of New Race-Ethnicity Codes and SES Indicators for Medicare Beneficiaries - Chapter 3. Rockville, MD. Retrieved March 18, 2017, from http://archive.ahrq.gov/research/findings/final-reports/medicareindicators

4. Alesina, A., & La Ferrara, E. (2000). Participation in heterogenous communities. Quarterly Journal of Economics, 115(3), 847-904.

5. Alesina, A., & LaFerrara, E. (2005). Ethnic Diversity and Economic Performance. Journal of Economic Literature, 43 (3), 762-800. doi:10.1257/002205105774431243

6. Alesina, A., & Zhuravskaya, E. (2009, May). Segregation and the quality of government in a cross-section of countries. Retrieved 01 11, 2017

7. Alesina, A., Baqir, R., & Easterley, W. (1999). Public goods and ethnic divisions. Quarterly Journal of Economics, 114 , 1243-1284.

8. Axiarlis, E. (2014). Political Islam and the Secular State in Turkey: Democracy, Reform and the Justice and Development Party. I.B. Tauris. Retrieved February 13, 2017

9. Baldwin, K., & Huber, J. D. (2011). Economic versus cultural difference. Forms of ethnic diversity and public goods provision. American Political Science Review, 104, 644-662.

10. Bartels, B. (2008). Beyond ``Fixed versus Random Effects'': A Framework for Improving Substantive and Statistical Analysis of Panel, Time-Series Cross-Sectional, and Multilevel Data. Political Methodology Conference.

11. BBC. (2016, August 23). Turkey v Syria's Kurds v Islamic State. Retrieved May 18, 2017, from BBC News: http://www.bbc.com/news/world-middle-east-33690060

12. Belge, C. (2011). State building and the limits of legibility: kinship networks and kurdish resistance in Turkey. Journal of Middle East Studies , 43, 95-114. Retrieved February 09, 2017, doi: 10.1017/S0020743810001212

13. Bell, A., & Jones, K. (2015, January). Explaining Fixed Effects: Random Effects Modeling of Time-Series Cross-Sectional and Panel Data. Political Science Research and Methods, 3(1), 133-153. doi:10.1017/psrm.2014.7

14. Benson, C. (2005). The Importance of Mother Tongue-based Schooling for Educational Quality. EFA Global Monitoring report. Retrieved April 23, 2017

15. Berlin, J., Kimmel, S., & Ten, T. (1999). An Empirical Comparison of Several Clustered Data Approaches under Confounding due to Cluster Effects in the Analysis of Complications of Coronary Angioplasty. Biometrics, 55(2)(2), 470-476.

16. Bierschenk, T., & de Sardan, J.-P. (1997). Local Powers and a distant state in rural Central African Republic. Journal of Modern African Studies, 35(3), 441-468.

17. Bruinessen, M. (1992). From Ottomanism to Separatism: The Religious and Ethnic Background of the Sheikh Said Rebellion. Writtings on Kurdistan ( Kurdistan Uzerine Yazular), p. 127. Retrieved 02 10, 2017

18. Bruinessen, M. (n.d.). Kurtler Arasinda Bir Siyasi Protesto Araci Oralak Naksibendi Tarikati.

19. Bulbul, M. (2016). The Transformation Of Kurdish Education in Turkey. Mardin: Mardin Artuklu University. Retrieved February 10, 2017

20. Caselli, F., & Wilbur, J. (2006, March). On the theory of ethnic conflict. NATIONAL BUREAU OF ECONOMIC RESEARCH. Retrieved from http://www.nber.org/papers/w12125.pdf

21. Constitution of the Republic of Turkey. (2002, December 27). Retrieved May 20, 2017, from The Grand National Assembly of Turkey: https://global.tbmm.gov.tr/docs/constitution_en.pdf

22. Desmet, K., Ortuсo-Ortнn, I., & Weber, S. (2009). Linguistic Diversity and Redistribution. Journal of the European Economic Association, 7(6), 1291-1318.

23. Easterly, W., & Levine, R. (1997). Africa's growth tragedy: Policies and ethnic divisions. The Quarterly Journal of Economics, 112, 1203-1250. Retrieved January 04, 2017

24. Egel, D. (2012). Tribal heterogeneity and the allocation of publicly provided goods: Evidence from Yemen. Journal of Development Economics, 101, 228-232. Retrieved February 23, 2017

25. Fabrigar, L., Wegener, D., MacCallum, R., & Strahan, E. (1999). Evaluating the use of exploratory factor analysis in psychological research. Psychological Methods, 4(3), 272-299. doi:doi:10.1037/1082-989X.4.3.272

26. Factor Analysis: A Short Introduction, Part 1 . (n.d.). Retrieved from http://www.theanalysisfactor.com/factor-analysis-1-introduction/

27. Fearon, J., & Laitin, D. (1996). Explaining interethnic cooperation. American Political Science Review, 90(4), 715-735.

28. Fernandez, R., & Levy, G. (2008). Diversity and Redistribution. Journal of Public Economics, 92, 925-943.

29. Finch, J., & West, S. (1997). The investigation of personality structure: Statistical models. Journal of Research in Personality, 31(4), 439-485. Retrieved March 25, 2017

30. Gao, E. (2016). Tribal Mobilization, Fragmented Groups, and Public Goods Provision in Jordan. Comparative Political Studies , 49(10), 1372-1403. doi:10.1177/0010414015621075

31. Gennaioli, N., & Rainer, I. (2007). The modern impact of precolonial centralization in Africa. Journal of Economic Growth, 12, 185-234.

32. Gerring, J., Thacker, S., Lu, Y., & Huang, W. (2015). Does diversity impair human development? A multi-level test of the diversity debit hypothesis. 66, 166-188.

33. Government of Tunceli. (2017, May 12). Health. Retrieved May 12, 2017,

34. Grusky, D. (2000). Social Stratification: Class, Race, and Gender in Sociological Perspective. Westview Press.

35. Эзduygu, A., Romano, D., & Sirkeci, Э. (1999). The Ethnic Question in an Environment of Insecurity: The Kurds in Turkey. Ethnic and Racial Studies, 22(6), 991-1010.

36. IFAD. (2017, May). Ardahan-Kars-Artvin Development Project. Retrieved May 13, 2017, from IFAD operations: https://operations.ifad.org/web/ifad/operations/country/project/tags/turkey/1492/project_overview

37. Kaya, N. (2015). Discrimination in Turkey's Education System. Istanbul: Tarih Vakfэ, Minority Rights Group International. Retrieved April 23, 2017, from http://minorityrights.org/wp-content/uploads/2015/10/EN-turkiye-egitim-sisteminde-ayirimcilik-24-10-2015.pdf

38. Koз, I., Hancioрlu, A., & Cavlin, A. (2008). Demographic Differentials and Demographic Integration of Turkish and Kurdish Populations in Turkey. Population Research and Policy Review, 27(4), 447-57.

39. Kohler, U., & Kreuter, F. (2009). Data Analysis Using Stata (2nd ed.). Stata Press. Retrieved March 24, 2017

40. KONDA. (2006). Social Structure Research. KONDA. Retrieved January 04, 2017, from http://www.konda.com.tr/tr/raporlar/2006_09_KONDA_Toplumsal_Yapi.pdf

41. KONDA. (2010). Kьrt Meselesini Yeniden Dьєьnmek. Istanbul: KONDA Research. Retrieved 12 15, 2016

42. KONDA. (2015). June 7 Election and Electorate Analysis. Istanbul: KONDA Research. Retrieved February 14, 2017

43. Krishnan, V. (2010). Constructing an Area-based Socioeconomic Index: A Principal Components Analysis Approach. Edmonton: University of Alberta. Retrieved March 16, 2017

Appendix