GEORG-AUGUST-UNIVERSITÄT GÖTTINGEN

Transcription

1 GEORG-AUGUST-UNIVERSITÄT GÖTTINGEN FACULTY OF ECONOMIC SCIENCES CHAIR OF MACROECONOMICS AND DEVELOPMENT Bachelor Seminar Economics of the very long run: Economics of Islam Summer semester 2017 Does Secular Education Impact Religiosity, Electoral Participation and the Propensity to Vote for Islamic Parties? Konstantin Gast Enrollment Number: Economics and Law (BA) 6 th Semester Date of submission: 29 th May 2017 Date of presentation: 16 th / 17 th June 2017

2 I. Introduction On September 25 th, 2015 the international community adopted the United Nations Sustainable Development Goals (SDGs). SDG number 4 is to ensure that by 2030 all girls and boys complete free, equitable and quality primary and secondary education. Education is one of the most important factors for the development of a society and its economy, as it impacts the productivity in and outside the labor market. People who are more educated have higher wages (Oreopoulos 2006) and are healthier (Mocan and Altindag 2013, Grossman 2008). Moreover, education influences beliefs and preferences pertaining religion. Resul Cesur and Naci Mocan contribute with their working paper (Cesur and Mocan 2013) to the discussion whether secular education impacts religiosity, electoral participation and the propensity to vote for Islamic parties. They investigate the influence of a law enacted in Turkey in 1997 which increased the compulsory years of secular education from five to eight years. The authors use a micro-census data from 2012 to show a rise in education of the cohorts of children who were affected by the school reform, and also identify a change of the affected people s religiosity and political preferences. Precisely, the amended law leads to a significant impact on education for both male and female, measured by middle school completion. This increase in secular education influenced women s religiosity and the propensity to vote for Islamic parties. Education lowers women s tendency to describe themselves as religious. It also decreases their propensity to wear head covers. Although education did not impact women s propensity to vote, it lowers the probability to cast their vote for an Islamic party in general elections. Surprisingly, in contrast to women, three additional years of secular schooling did not significantly influence neither men s religiosity, propensity to participate in general elections, nor their preference to vote for or against Islamic parties. In the following, I will, first, describe the 1997 education reform. Second, I am going to introduce the data and the variables. Third, I report the empirical methodology Cesur and 1

3 Mocan use, and I present the results. Fourth, I will describe the robustness checks and potential mechanisms through which education might impact religiosity and voting behavior. Section VI is the conclusion. In Section VII and VII, References and Appendix can be found, respectively. The 1997 Education Reform In August 1997, a new education law increased compulsory secular schooling from five to eight years in Turkey. It entered into force for the education year A person that had completed the 4 th grade or lower at the end of the school year was affected by the reform. However, the reform did not affect the start of education (the education year starts in September). In Turkey, every child must start school if they are six years in September. Furthermore, a person can go to school also prior to their 6 th birthday if they turn six the same year (October-December). Prior to the reform, every student had to attend school at least 5 years. Afterward, pupils had the possibility to, end their education, or to continue with secondary schooling at either traditional middle schools or vocational schools. The ladder includes religious schools. After the reform, primary and secondary schools were combined. Consequently, middle schools and vocational middle schools which stood alone had to close. Moreover, only after students have completed 8 years of compulsory schooling, vocational high schools were allowed to admit them. II. Data The paper used a data set provided by the KONDA Research and Consultancy which is a national and international well-respected source for date concerning both standard election poll and lifestyle questions in Turkey. The data for this empirical analysis consist of 9,590 voting-age adults who have been surveyed in The sample consists of people who were 2

4 between 18 and 32 years old in Thus, they were born between 1980 and Additionally, Cesur and Mocan excluded the 1.6 percent of the respondents that indicated to be non-muslims in order to focus on Muslims. Table 1 shows both the independent variables, which are personal attributes of individuals, and the outcomes. Respondents which have been born between 1980 and 1985 constitute the control group, as they had to complete only five years of mandatory education. On the contrary, the individuals that were born between 1987 and 1994 were obliged to attend school for at least eight years. Hence, they constitute the treatment group. It is unsure whether those born in 1986 have been treated by the education reform. For this birth cohort, three cases are possible depending on whether a person was born between January and September or between October and December A person who was six years old in September 1992 had to enter school in the education year 1992/93. Consequently, would complete 4 th grade four years later and would be affected by the education reform. A different person born in 1986 (also six years old in 1992), but in October, November, or December, might be affected by the law. It depends whether the parents decided to send their child to school, although they were not obliged, or the parents wait until next September (1993). In the latter case, the individual would be not treated by the new law. As displayed in Table 1, the proportion of both men and women being in possession of at least a middle school degree (eight years of schooling) is higher in the treatment group compared to the control group. The difference for females is with 29 percentage points almost two times bigger as the difference for males (15.3 percentage points). Measures of Religiosity Among other questions, the KONDA survey interviewed the individuals about their religious beliefs. The authors created the variable Religious which takes the value of 1 if the person chose the answer religious or devout, and the value of 0 if they indicated to be atheist or believer. The latter is categorized as non-religious because in Turkey a believer 3

5 is not regarded as Muslims. He or she only believes in the existence of some power but does not necessarily practice a religion. The additional variable Atheist takes the value of 1 if the respondent states to be a non-believer and the value of 0 if they described themselves as a believer, religious, or devout. Evident from Table 1, about 60 percent of the men and 70 percent of the women indicated to be religious if they have not been exposed to the education reform. On the contrary, the proportion of religious persons is smaller in the treatment group for both sex (51.6 percent men, 58.8 percent women). Another indicator for religiosity of women in Turkey is whether they wear a head scarf or not. More than 58 percent of the female respondents of the cohort did cover their head and only about 39 percent of the cohort. With regard to their lifestyle, the respondents were asked to indicate whether it is modern, traditional conservative, or religious conservative. The outcome of the answers generates the dummy variable Modern which takes the value of 1 if a person declared themselves as modern and a value of 0 if they chose one of the other two optional answers. As shown in Table 1, in proportion there are more individuals having a modern lifestyle than a traditional or religious conservative one. The Tendency to Cast a Vote, and the Propensity to Vote for an Islamic Party All respondents of the 2012 survey were asked: If elections were held today, which party would you vote for?. They had the possibility to choose either a political party from a comprehensive list or to vote for an independent person. Moreover, the responses not going to vote and casting a blank vote were provided, too. The variable Voter Now takes the value of 0 if the respondent indicated that they would cast a blank vote and the value of 1 otherwise. Similarly, the survey included the question which party the person voted for in the last general elections that were held in June The binary variable Voter 2011 takes the value of 1 if the individual indicated they voted for a party or for an independent candidate. The variable is 0 if the person did not cast their vote. 4

6 Based on the collected answers the dummy variable Islamic Voter Now was created. This variable is 1 if the respondent answered that they would vote for the Justice and Development Party (AKP), the Felicity Party (SP), or the People s Voice Party (HAS) if general elections took place today. Islamic Voter Now takes the value of 0 if the person stated that she would vote for any other party than the AKP, SP, or HAS. The term Islamic Party is used to describe a political party with Islamic roots according to political scientists (Fuller 2004). Although Article 2 of the Turkish constitution does not permit political parties which refer explicitly to any religion in their party statute, Islamic parties have been founded in Turkey for the last 50 years. Cesur and Mocan amplify why those three parties above-mentioned qualify as Islamic political parties. They argue that these parties are the successor or an offshoot of an Islamic party founded in 1970 by Necemettin Erbakan, the National Order Party. Supporters of Erbakan created the SP as a successor of another prohibited Islamic Party, the Virtue Party which was closed in Those leaders that did not join the SP, formed the AKP. Hence, the AKP is a political party with Islamic roots although it tries to address a wider base of voters. (Taspinar 2012, Roy 2012). The HAS, however, was formed as another offshoot of the Islamic SP in In 2012, more than 53 percent of the men and 66 percent of the women who were born between 1980 and 1985 would vote for an Islamic party as described in Table 1. The proportion decreases by 4.4 percentage points and 10.5 percentage points with regard to the men and women who have been affected by the law reform, respectively. Comparably, the variable Voted Islamic in 2011 describes whether a person voted for one of the three Islamic parties in the 2011 general elections. The difference for male and female voters fluctuates around plus or minus 1 percent point regardless of the exposure to the law. 5

7 III. Empirical Specification and the Basic Results The paper s main goal is to investigate the impact of education on religiosity, the propensity to vote, and the propensity to vote for an Islamic party. Therefore, equation (1) was set up: (1) R i = β 0 + β 1 Educ i + X i Ω + ε i In this equation, R i stands for a particular outcome, such as whether the individual has voted in the general election 2011, or whether they have a modern lifestyle. Educ indicates whether the person has at least a middle school degree and consequently at least eight years of schooling. X is a vector which represents personal characteristics of a person. It includes the religious sect, age, the location of residence, and ethnicity. The birth cohort had to attend school at least for eight years contrary to the control group ( birth cohort) who had only five years of compulsory education. Hence, Cesur and Mocan use the exposure to the law reform in 1997 as an instrument of education. As a result, the first stage regression is equation (2): (2) Educ i = γ 0 + γ 1 Law i + X i Ψ + μ i Law is a binary variable that takes the value of 1 if a person was born between 1987 and 1994 or later, and it is 0 if they were born between 1980 and Thus, this variable shows whether the person was treated by the law reform. The Impact of Law on Educational Attainment The OLS regression s results are presented in Table 2. As visible in column (1), the exposure to the education reform increases the probability to have at least a middle school diploma by 10.2 percentage points. The impact on the females education is higher than on the males one. Column (2) shows a 6 percentage point impact for men and an increase by 14 percentage points for women. Consequently, the exposure to the amended law increases the probability of having at least a middle school education for men with a sample mean of

8 by 13 percent and for women by 20 percent (sample mean of 0.67). In summary, the results affirm that the education reform increases the observed educational attainment of individuals. In addition, the reform s effect on the propensity of completing at least high school or college was examined. And Table 3 shows that the education reform indeed had spillover effects. The probability to have a higher education than a middle school diploma increased by 12 percentage points for high school digress and 7 percentage points for college degrees. As expected, the falsification test confirms the exposure to the law had no significant effect on the propensity to complete five years of primary schooling. The Impact of Education on Religiosity, Electoral Participation, and the Propensity to Vote for Islamic Parties Table 4 shows the results of the instrumental variables regression for the whole sample and having a middle school diploma is instrumented with a treatment dummy. Having at least eight years of middle school education, instead of five years of primary school, reduces the probability to be religious by over 39 percent points. This result displayed in column (1) implies that on average one extra year of schooling after primary education decreases the propensity of being religious by 13 percentage points. Column (2) presents that the probability a person declares themselves as atheist increases with additional mandatory years of education. The propensity to call one s lifestyle as modern, as opposed to traditional conservative or religious conservative, also increases by 30 percentage points. The impact of Middle School Diploma on the dependent dummy variables Voter 2011 and Voter Now is not significantly different from zero, as apparent in column (4) and (5). Hence, being exposed to the education reform does neither affect the person s participation in the last election in 2011 nor in a hypothetical election today. However, the results described in column (6), (7) and (8) indicate that having at least middle school education, as opposed to only primary school education, impacts voting preferences pertaining Islamic parties. The columns describe the results for the dependent 7

9 variables Voted Islamic in 2011, Islamic Voter Now, and Never Vote Islamic. The first two variable take the value of 1 if a person voted for an Islamic party in 2011 general elections or would cast their vote for an Islamic party if elections were held today, respectively. Both variables are 0 if the person cast or would cast their vote for a different political party. The exposure to the law decreased both propensity to vote Islamic in 2011 by 33 percentage points, as well as to vote Islamic today by about 42 percentage points. If a person has not voted for an Islamic party in 2011 will also not cast their vote for an Islamic party today the third variable Never Vote Islamic is 1, and 0 otherwise. As described in column (8), the probability to have voted and would vote against an Islamic party in 2011 and in 2012, respectively, increases by 45 percentage points if a person has at least a middle school diploma. Concluding, the probability to have a political preference against Islamic parties rises significantly by the reform generated an increase in education. In order to differentiate between the effects on men and women, the same regression as in Table 4 is run separately with both the male and the female sample. For the women the survey includes the supplementary dependent dummy variable Wears Head Cover as an additional measure of religiosity. Table 5 presents the results for the female sample s estimation. An increase in mandatory secular schooling increases modernity both by about 30 percentage points. It lowers religiosity and the propensity to wear a head cover for women by about 30 and 40 percentage points, respectively. The additional education s effect on the tendency to vote either in 2011 or today (2012) is both not significant different from zero. But education significantly impacts propensity to cast vote against an Islamic party. In contrast, Table 6 reports that in the case of men, three more years of mandatory secular schooling after primary school has no substantial influence on neither religiosity, modernity, nor the probability to vote for an Islamic party. To summarize, only in the case of women, secular education has a significant impact on religious self-perception and political sympathies which are influenced by religion. 8

10 The Impact of Economic Circumstances Islamic Voter Now is a dependent variable that indicates whether a person would vote for an Islamic party if general elections were held today. The last 2011 general elections were won by an Islamic party (AKP). This party has governed in Turkey since Consequently, Turkey was also governed by an Islamic party in 2012 when the survey was conducted. Therefore, indicating to have no intention to vote for an Islamic political party in 2012 might reflect a respondent s dissatisfaction with the past and current policies of the AKP. Unfortunately, the survey does not provide a question about government policies approval. Nonetheless, it includes information about respondents personal economic circumstances. If people answered that they were financially comfortable and could save money last month, or although not comfortably financially, at least managed to make ends meet last month, the new variable Can Make Ends Meet is 1. The dummy variable is 0 if the people stated financial difficulty, difficulty in paying the bills, or the need to borrow money to make ends meet last month. Similarly, the variable Expect Personal Economic Hardship was employed to indicate whether the respondent expect personal hardship during the next months. It takes the value of 1 if the person responded affirmatively. Cesur and Mocan examine the assumption that those people who have difficulty making ends meet or who expect personal economic hardship tend to vote against AKP and hence against an Islamic party. If the thesis proves true, the impact of education could be biased. To check the thesis the new model is created, where Islamic Voter Now is regressed on the same set of explanatory variables as before and achieving at least middle school education is instrumented with the exposure to the law. Table 7 describes the model s results. Column (1) shows those for the whole sample. Column (2) and (3) present the results of the male and female sample, respectively. The coefficient of Expect Personal Economic Hardship indicates that a person s expectation of personal economic hardship next month has a negative impact on the propensity to vote for an Islamic Party if elections were held today (2012). More 9

11 precisely, men s propensity to vote for an Islamic Party increases by 13.8 percentage points and the women s propensity by 10.2 percentage points. For the whole sample, the coefficient s value is People who expect to make ends meet are 7.7 percentage points more likely to cast their vote for either the AKP, SP, or HAS. In the female sample, the impact is higher (9 percentage points) than in the male sample (6 percentage points). These findings indicate that being concerned about ones own economic circumstances lead to a higher probability to vote against the currently governing Islamic party, AKP. Notwithstanding, controlling for the variables Expect Personal Economic Hardship and Make Ends Meet has no impact on the magnitude of the estimated coefficient of Middle School Diploma. Specifically, in Table 7 the education s coefficient is for the whole sample. In column (7) of Table 4, it was where the same specification as in Table 7, but excluding economic circumstances, were used. Comparable results can be found with regard to both the male and the female sample. Concluding, the results indicate that a person s economic well-being affects the tendency to vote for the governing political party (AKP). Additionally, education is not highly biased to influence the propensity to support Islamic parties. IV. Robustness Checks Cohort Effect As already described, the data set used in the paper is based on a survey conducted in In order to be sure that no unobserved factors could have influenced the respondents who were born between 1985 and 1994 (18-25 years old in 2012) compared to those born between 1980 and 1984 (27-35 years old in 2012) a second survey, done in 2008, was used. The KONDA provided also this 2008 data set It was a nation-wide conducted survey and included the same questions as in 2012 but is a smaller sample size. The following investigation intend to show whether the results have ben biased due to a cohort effect 10

12 Table 8 summarizes the birth cohort information related to the 2012 data, as well as the 2008 data. More precisely, columns (1) and (2) show the year of birth and the resulting exposure to the education reform. In columns (3) and (4) the respondents ages are displayed pertaining the years 2008 and 2012, respectively. For example, a person born in 1990 was exposed to the law, was 18 years old in 2008, and 22 years old in For the investigation, the authors extract from the 2008 data those individuals who were years old in 2008 and extract from the 2012 data the people who are also years old but in The so extracted same age group is connected by arrow A in Table 8. Those at the age of in 2012 are treated by the education reform, contrary to their agecounterparts from Hence, using the two data sets Cesur and Mocan create a sample of years old individuals. The dummy variable Year2012 takes the value of 1 if a person was surveyed in 2012, and it is 0 if they were surveyed in The same way, a sub-sample is created with those who are 23 or 24 in 2008 or The results for the instrumental variables regressions for both samples are displayed in Table 9 where the variables Modern and Wears Head Cover depend on the question whether a person has at least a middle school diploma. In both panels (A: ages 23-25, B: ages 23-24) eight years of secular education, opposed to five years, significantly increases the propensity for males to indicate their lifestyle as modern. Although the result for women is not significant at conventional levels with regard to the lifestyle, education reduces significantly the probability for women to wear a head cover. Besides, the results are similar in Panel A and Panel B. Concluding, the results reported in Table 9 indicate that the impact of three additional years of mandatory schooling presented in Tables 3-7 is not due to the age difference between the individuals who have been treated by the law and those who were not treated. Furthermore, the authors of the paper test whether aging of the same birth cohort might explain the variation in religiosity. Consider now only those individuals at age of in 11

13 2008 and, hence, years old in For clarity, the cohort is marked with the arrow C in column (4) in Table 8. As they are not exposed to the law, there should be no change in their propensity either to have at least eight years of education, nor to indicate having a modern lifestyle, nor to wear a head scarf. To test this the sample which consists of the year olds gets assigned a placebo treatment. The binary variable Exposure to the Placebo Law is 1 for those who are 23-25, and takes a value of 0 for the older group. Given that nobody in this analysis is actually treated by the reform, no coefficient should have a significant impact on the dependent variables Middle School Diploma, Modern, or Wears Head Cover. The results presented in Tables 10A and 10B confirm these assumptions because all decisive coefficients are small and not different from zero. Thus, being young alone is not responsible for being religious or modern. Difference-in-Difference Cesur and Mocan use the difference-in-difference method to investigate whether age effects bias the estimated impact of education. In the new model, the dichotomous variable Young is 1 if the individual is years old, and 0 otherwise. Additionally, the dummy variable Year2012 indicates whether the person was surveyed in 2012 (Year2012 = 1) or in 2008 (Year2012 = 0). For the estimation, only people at the age of or in 2012 are considered, where exclusively the first cohort (23-25 years old) were treated by the education reform. The authors estimate the model as follows. (3) R i = α + X i Φ + λ 1 Young i + λ 2 Year2012 i + λ 3 Young i *Year2012 i + τ i In equation (3) λ 1 represents the impact of being years old in 2008, as opposed to being years old, on modernity or the propensity to wear a head scarf. Coefficient λ 3 is the dif-in-dif magnitude, which is the differential effect of being young in 2012, in contrast, to being young in

14 Table 11 shows the results for the dependent variables Modern (males), Modern (females), and Wears Head Cover. The dif-in-dif estimate is not different from zero for the male coefficient. However, dif-in-dif estimates for both Modern (female) and Wears Head Cover are statistically different from zero. Accordingly, the coefficients indicate that the probability to have a modern lifestyle increases by 9.4 percentage points and the probability to wear a head cover decreases by 11.4 percentage points if a woman was years old in Only this group was exposed to the law. For clarity consider arrow C in the columns (3) and (4) in Table 8. Hence, the dif-in-dif estimates affirm the reported instrumental variables regressions results. V. Potential Mechanisms In the paper Cesur and Mocan identify the total impact of additional compulsory secular schooling on religiosity and the propensity to vote for Islamic parties in general elections in Turkey. This influence can work through many channels. Firstly, education might enhance cognitive ability and shift preferences towards scientific ways of thinking. Secondly, attending school as a teenager may lead to a higher number of friends and social interactions which could confront a person with a higher amount of different viewpoints and experiences. Those can impact lifestyle preferences. Thirdly, education might cause migration from rural to urban areas. In turn residing in a city could affect religious and political preferences. Lastly, the probability to participate in the labor force can be influenced by the number of completed years of schooling. To examine whether any of these factors impact the estimated effect of mandatory nonreligious education, the authors utilize the same benchmark model for females as used in Table 5 but add four variables. The dummy variable Migrant takes the value of 1 if the respondent has migrated out if the city they were born, and 0 otherwise. Labor Force Participation is a dichotomous variable which indicates whether a person is in the labor force. 13

15 The last two variables City and Metro Area measure the location of residence opposed to the rural area. Table 13 shows the results in Panel B. To compare the results easily, also the baseline results from Table 5 are reported in Panel A of Table 13. Women that migrated to the city where they live now are more likely to practice a modern lifestyle. Living in metro areas decreases the propensity to be religious or to wear a head cover for females and raises their probability to self-identify as modern. The same applies to women who participate in the labor force. Comparing the coefficients of Middle School Diploma between Panels A and B of Table 13 Cesur and Mocan come to the conclusion that migration, choice of residence, or labor force participation are not likely channels through which education impacts religiosity and preferences for Islamic parties (Cesur and Mocan 2013, p.41). They base their deduction on the fact that the coefficients of Middle School Diploma in Columns (7), (8) and (9), are statistically significant different from zero, and similar in Panel A and Panel B. The three columns present the impact on voting preferences pertaining Islamic parties (Voted Islamic in 2011, Islamic Voter Now, and Never Vote Islamic). Moreover, the coefficients in the same columns in Panel B are not statically significant from zero at conventional levels for Migrant, Metro Area, and Labor Force Participation. Hence, the author s conclusion is correct with regard to the preference for Islamic parties. However, the analogous reasoning pertaining religiosity cannot be approved. Consider column (1) in Panels A and B in Table 13. The Middle School Diploma s coefficient is significant from zero in Panel A, but is not in Panel B. Additionally, if a person lives in metro area, or participates in the labor force, their self-identification as religious decreases significantly (both significance at the 1 percent level or lower) by 7.4, or 11.7 percentage points, respectively. 14

16 VI. Conclusion Cesur and Mocan have shown that additional years of secular education are able to influence people s preferences in a number of different dimensions, especially women s attitude towards religion and voting preferences. The consequences of an exogenous change in education are important to analyze, because of the consequences for the political economy of development. As displayed in Table 14, countries with a predominantly Muslim population (e.g. Iraq, Iran, Morocco, Egypt, or Jordan) are characterized by low GDP per capita, low level of democracy, low number of average years of schooling, and high religiosity. If more years of compulsory education causally changes religiosity and has an effect on voter preferences, education policies can change political and economic conditions in these emerging and developing countries. As education might impact religious and political preferences through a number of channels, such as migration or labor force participation, every empirical-based contribution helps to complete the picture. However, empirical evidence should be treated with caution, because country-specific factors, such as rule of law, quality of education, or equal access for girls and women to public services, could also influence education and its impact on economic development. 15

17 VII. References Cesur and Mocan Does Secular Education Impact Religiosity, Electoral Participation and the Propensity to Vote for Islamic Parties? Evidence from an Education Reform in a Muslim Country. NBER Working Paper 19769: (last access 23/05/2017). Fuller Turkey s Strategic Model: Myths and Realities. The Washington Quarterly 27(3), pp Grossman The Relationship between Health and Schooling. Eastern Economic Journal, pp Mocan, Naci, and Altindag Education, Cognition, Health Knowledge and Health Behavior. European Journal of Health Economics 15, pp Oreopoulos Estimating Average and Local Average Treatment Effects of Education when Compulsory Schooling Laws Really Matter. American Economic Review 96(1), pp Roy The Transformation of the Arab World. Journal of Democracy 23(3), pp Taspinar Turkey: The New Model? : (last access 23/05/2017). 16

18 VIII. Appendix Table 1 Source: Cesur and Mocan 2013, p

19 Table 2 Source: Cesur and Mocan 2013, p

20 Table 3 Source: Cesur and Mocan 2013, p

21 Table 4 Source: Cesur and Mocan 2013, p

22 Table 5 Source: Cesur and Mocan 2013, p

23 Table 6 Source: Cesur and Mocan 2013, p

24 Table 7 Source: Cesur and Mocan 2013, p

25 Table 8 Source: Cesur and Mocan 2013, p

26 Table 9 Source: Cesur and Mocan 2013, p.61 25

27 Table 10 A Source: Cesur and Mocan 2013, p. 62 Table 10 B Source: Cesur and Mocan 2013, p

28 Table 11 Source: Cesur and Mocan 2013, p

29 Table 12 Source: Cesur and Mocan 2013, p

30 Table 13 Source: Cesur and Mocan 2013, p

31 Table 14 Source: Cesur and Mocan 2013, p