Advanced Master in International and Development Economics

Transcription

1 Advanced Master in International and Development Economics Jointly Organized By RELIGION AND DEVELOPMENT David Villalobos Promoter Tutor : Professor Guilhem Cassan : Ombeline De Bock Project presented as part of the requirements for the award of the Advanced Master in International and Development Economics Academic year

2 Abstract Relative to the extensive debate about the relationship between religion and development, and based on micro point of view literature reviewed, this personal project has as objective to go deep into the question about if to be religious or non-religious person has a relationship with any development attribute. Using the Generation & Gender Program Longitudinal Survey for Germany, a set of descriptive and inferential analysis -by each wave and capturing changes among different waves- is proposed to understand the relationship of religion and development based on a wide approach of religion (affiliation, attendance and intensity) and development (income, education, health, trust in others, fairness and happiness) variables proposed by the literature review and constrained by survey data availability. Results suggest that to be religious -or non-religious- people has an ambiguous association with development, depending to the sample, religion and development variables examined, confirming the relativity about the Marx and Weber Hypothesis. However, religion intensity variables seem to explain much better -than religious respondent, religious intensity or all religious variables together- the variations on development outcomes, probably because could capture the real religious or nonreligious- identity of the respondent. More research should be needed to support these results. Content 1. Introduction Literature Review Macroeconomic Level Microeconomic Level Data and Methodology Results Conclusions REFERENCES APPENDIX I NOTES

3 Tables Index Page Table 1. Summary statistics 22 Table 2. Dummy Variables 23 Table 3. Distribution of Population among waves 24 Table 4. Respondent s Religion Distribution among waves 26 Table 5. Means Analysis by wave 31 Table 6. Means Analysis among waves 32 Table 7. OLS results of wave 1 34 Table 8. OLS results of wave 2 35 Table 9. OLS results: Religious Respondents and Development Outcomes 36 Table 10. OLS results: Religious Attendance and Development Outcomes 37 Table 11. OLS results: Religious Intensity (a) and Development Outcomes 38 Table 12. OLS results: Religious Intensity (b) and Development Outcomes 39 Table 13. OLS results: Religious Intensity Index and Development Outcomes 40 Table 14. OLS results: Religiosity Index and Development Outcomes 41 Figures Index Page Figure 1. Respondents' Religion 27 Figure 2. Respondents' Religious Attendance 27 Figure 3. Respondent' Religious Intensity: Marriage Opinion 28 Figure 4. Respondent' Religious Intensity: Homosexual couples 28 Figure 5. Share of Religious and Non-Religious Respondents by age 29 3

4 1. Introduction The interest on the relationship between religion and development is widely disseminated from Social Sciences, mainly economics, sociology and politics -and recently from Medical Sciences-, that are attracted to understand the association and impacts among the both variables. Since Karl Marx and Max Weber hypothesis formulations about the negative or positive role of religion on development, there is an extensive debate about the nature of the relationship between them and a particular concentration on the diverse effects that religious aspects have in different environments. Institutional approach has retaken the interest on this connection focusing in the role of religion through colonization, property rights, and informal institutions frameworks on economic growth and development. Regarding to the religion, as Nolan (2004) mentioned, is important take into account that this concept is a very dynamic category that could present very different results depending to the time, the environment, the sample size and other conditions, and therefore the diverse results obtained by authors need to be carefully interpreted and tested continuously on the time and the space. In fact, Karl Marx and Max Weber hypothesis have been tested on similar and different contexts presenting robust results but in the contrary direction. For example, Guiso et al. (2003) tested the Marx Hypothesis about the religion as people s opium and found that religious beliefs are associated with good economic attitudes conductive to economic growth (pp. 225). Cantoni (2010) tested the Weber Hypothesis about the Protestantism effects on development on the same time when this was formulated and conclude that Protestantism has no effect when interacted with other likely determinants of economic development (pp. 1). Taking in account this context, the specific aim of this personal project is to analyze the association between religion and development. Exactly, the objective is to go deep into the question about if to be religious or non-religious person has a relationship with any development attribute. To reach this objective, first, Literature Review section summarizes the current state of the debate related of the relationship between Religion 1 and Development on macro and micro levels-, focusing in the most relevant and recent research accessible, and second, and based on the approach reviewed on literature review, a 4

5 descriptive and inferential analysis -by wave and between waves- is proposed to understand this relationship and test the associations between religious and non-religious respondents (affiliation, attendance and intensity) and different development outcomes (as income, education, health, trust in others, fairness and happiness) constrained by the data of the Generation & Gender Program Longitudinal Survey for Germany. This study is divided as follows. Section 2 presents a literature review about the relationship between religion and development with a particular focus on the difference between the macroeconomic and the microeconomic point of view, and their mechanism. Section 3 contains Data and Methodology explanations about the empirical exercise and Section 4 provides the results. Finally, Section 5 contains concluding remarks. 2. Literature Review The start points to this section arises from Iannaccone (1998) that admirably summarizes the knowledge about Religion s Economics and the most relevant literature, and from Aldashev & Platteau (2014) that systematically organized the state of literature about religion and development. A literature classification is proposed based on the analysis s scope of the studies. Moreover, the literature is also classified considering if specifies the religion s mechanism of impact on development or not Macroeconomic Level This section presents the recent literature review about the relationship between Religion and Development from a macroeconomic point of view. The majority of this kind of literature uses cross-country analysis to show the effects of religion on economic growth in both developed and developing countries. As observed by Aldashev and Platteau (2014) the most important limitation of this type of studies is that they do not consider the specific context where the connection among religion and economic behavior is conforming. Thus, the complex environment of each country about its history, geography, social, political and other relevant facts is simply overlooked and should be qualified by more in-depth withincountry studies (pp. 595). Shortly, the advantage to generalize the global effect of religion 5

6 aspects on economic performance is at the same time the major disadvantage of these studies, because they do not take into account the differences among countries contexts. Other complication with these studies is that they focus on the global effect between religion and economic outcomes but they do not precise the specific channels and mechanism whereby the religion could affect the economic performances. To overcome this problem, the macroeconomic level literature is divided in two big groups: the first one, that includes those papers that analyze the relationship between Religion and Development from a direct approach, without specify the mechanisms and the channels; and the second one, including those papers that identify and support detailed explanation about the mechanism and channels. a. Without channels One of the most relevant and cited papers that analyze the relation between Religion and Economic Growth was written by Barro and McCleary (2003). The main idea of this paper is to explain the aggregate economic outcomes through the church attendance effect on religious beliefs (especially about heaven and hell) and individual traits. While the fact that the church attendance affects religious beliefs is turned out, how the religious beliefs modify the individual traits is not clearly supported. As the authors recognize an interesting extension would examine the links between religious beliefs and individual behaviors or values (pp. 779). Concerning the methodology, the authors used a cross-country data panel, with data about church attendance, religion beliefs and economic growth for 41 countries 2, in order to estimate the effect of religion on economic growth. The data source is individual information collected in six international surveys between 1981 and About church attendance data, the authors use survey responses to generate the fractions of the population that attended church. The instrumental variables approach is used to treat the possible endogeneity problems between the analyzed variables. The main findings of the Barro and McCleary (2003) study are that, for given levels of church attendance, an increment of some religious beliefs (especially on belief in hell and in heaven) tends to increase economic growth. There are clear results showing that, for the 6

7 economic growth, the fear of hell is more important than the hope of heaven. Despite of clear mechanisms are not explained, differentiating by kind of belief could contribute to understand the role of the religion on growth: in this case, a negative belief (fear on heal) is stronger than a positive belief (hope on heaven). This result could suggest that religions that promote the fear than the hope beliefs may contribute more to economic performance. There is one important authors assumption behind all of these results: the church has a firm behavior, whose main input is the church attendance and main output is the beliefs. Thus, for given beliefs, higher church attendance (inputs) implies more available resources (production capacity) used up by the religion sector. Therefore, for given religious beliefs, an increase of the church attendance tends to reduce the economic growth. This result, at the beginning, could be surprising because seems to show a negative relationship between religion specifically religious attendance- and growth. However, the authors justify it arguing that a higher levels of church attendance depress economic growth because greater attendance signifies a larger use of resources by the religion sector, and the main output of this sector (the religious beliefs) has already been held constant (pp. 779). Therefore, these findings should not be interpreted as a negative effect of church attendance on economic growth; on the contrary, higher levels of attendance contribute to stronger beliefs, which encourage growth 3. Another relevant study is that of Sala-I-Martin, Doppelhofer, and Miller (2004). The main objective of this work is to use a set of explanatory variables explaining economic growth 4. The authors originality consists in analyzing the robustness of the variables obtained in the cross-country regressions thorough a new econometric method: the Bayesian Averaging of Classical Estimates (BACE) that put together the averaging of estimates across models (a Bayesian concept) with a classical ordinary least squares (OLS) estimation. Authors found that a good number of economic variables (around one-fifth of the 67 explanatory variables) have robust partial correlation with long-run growth, which means that these variables are highly significant to growth. The strongest evidence is found for the following explanatory variables: the initial level of income, the investment rate and various measures of education. 7

8 It is interesting to observe that three religious variables (the fractions of Confucians, Muslims and Buddhist) are included in the 20 most significant explanatory variables on growth. All religious variables have a positive relationship with growth, with exception of fraction Protestant 5, fraction Catholic and religious intensity. In the same line of research, Noland (2005) uses at the same time cross national and withincountry regressions in India, Malaysia and Ghana to analyze the behavior between religion and economic performance, and contrasts the results of other past studies. The regressions are built on the basis of a religious independent variable related to religion group s fractions (Muslim, Hindu, Buddhist, Jewish, Catholic, Orthodox, Protestant) and the dependent variable of economic performance measured by the total factor productivity (TFP) 6, and, the GDP growth. The control variables used are: initial GDP per capita, investment share, government share, openness and education level. With reference to the cross-countries regressions, the author found that the null hypothesis that religious affiliation is uncorrelated with performance can frequently 7 can be rejected (pp. 1127), confirming a global positive effect of religion on economic growth. However, this positive effect is not imputable to any particular religion. For the within-country regressions, the countries selected by the author were India, Malaysia and Ghana. The most interesting finding is related with the Muslim religion due to some authors (Bhalla, 2002 & Kuran, 2004 as cited in Noland, 2004) consider this religion and its practices as a possible obstacle the growth. However, the author found that Muslims Religion have a positive and significant effect on growth for almost all the analyzed samples, concluding that Islam does not appear to be a drag on growth (pp. 1228). This finding does not mean that other mentioned authors are wrong about Islam. For Nolan only means that Islam, as all religions, is a highly dynamic variable that change over the time and space, therefore is perfectly possible to have different results on growth analyzing the same religion. 8

9 b. With channels This section presents a selection of recognized literature on the field, based on the criteria that religion could affect development outcomes but through specific and supported channels and mechanisms that the authors explain in different level of detail. One of the most important papers in this line of literature is that of Guiso, Sapienza and Zingales (2003). The main objective of the study is to understand the effect of religious people on economic growth by explaining a chain of supported mechanism as the level of trust in others, on government and on legal system, but also analyzing how these beliefs, based on the power of trust, could improve the institutions and markets functioning and then the economic growth. The link between trust and growth is not empirically treated by the authors but is widely supported by the literature (Putman, 1993 as cited in Guiso et al., 2003). To reach this goal, the authors focused on the people attitude explained by the following variables: cooperation, women, government, legal rules, the market economy and its fairness, and thriftiness. While using cross-country regressions with an extensive data set created by the World Values Survey (WVS) 8, Guido et al. could conclude that religion has on average a positive relationship with trust variables mentioned above. Moreover, they argue that religious attendance could raise the level of trust between Christians; in particular, more in Protestants than Catholics. For other religions this effect is inexistent. Relate to the relationship between religion and attitudes conductive to growth, authors found again a higher positive association for Christians and a lower association for Muslims, in different direction that Xala-i-Martin et al. (2004) and Noland (2005). Guiso et al. (2003) also mention the relationship between religion and intolerance: religious people tend to be more intolerant and less sensitive to women s rights than no religious people 9. This association is present across all religions with the exception of the Buddhists, which are more tolerant than non-religious people. Organizing religions by higher level of intolerance towards immigrants and races we have: Hindus, Muslims, Jews, Catholics and Protestants. La Porta et al. (2007) are also interested on trust variables with a particular approach based on the institutional role of large organizations. Without enter to analyze the link between trust and development, it is amply reinforced during all the paper because trust or social 9

10 capital as a propensity of people in a society to cooperate to produce socially efficient outcomes (pp. 333). For the authors, the main mechanism to improve the performance of a large organization 10 is based on trust that promotes cooperation between the society members. Moreover, this cooperation, especially in larger organizations, could potentiate the effect on a more effective performance of the organization. However, strong heretical structures as a religion could depress this effect in the long run. To probe this mechanism, the authors tested Putman thesis: Catholic Church, imposing a hierarchical structure on the society, often in symbiosis with the state, has discouraged the formation of trust (Putman, 1993 as cited in La Porta et al., 2007). This affirmation about Catholic Church is used to other hierarchical religions and verified across different countries. Authors test this hypothesis through crosscountry regressions based on the data of the World Values Survey among 1980 and 1990 and 1000 persons of 40 countries. The main finding of the study is that there is a negative correlation between all the hierarchical religions with trust Hierarchical religion includes Catholic, Orthodox and Muslim religions. Based on this results authors conclude that there is a strong and negative relationship between hierarchical religions (large organizations) and trust. Another interesting finding, based on the same data, is that trust could contribute to cooperation, but trust is lower in countries with dominant hierarchical religions (pp. 337) due to the destruction of horizontal networks of cooperation, a key concept in the formation of trust among the countries, and in clear opposition to a vertical network or hierarchy. Therefore, authors conclude that there is a mechanism through hierarchical religions deteriorates the performance of large organizations and cause inefficient institutions on the long term, with the obvious effects on development outcomes to these countries. Based on the same approach of institution matters, but focusing more on colonization effects on current societies, the study of Grier (1997) analyzes the effect of Religion on Economic Development. Colonization is a concept much more closed than religion. This is because colonizers imposed in many places their religious systems and practices. In this 10

11 study, the authors objective is to test the positive relationship between Protestantism (measured by the growth of Protestants during ) and economic growth measures (as rates and real per capita incomes), and to explain the religion effect on economic performance comparing British and Spanish/French colonies 11. Author founds that British colonies have a significantly better performance, on average, than French and Spanish colonies, both Catholic countries 12. Moreover, that the growth rates and the level of Protestantism are positive and significantly correlated with Real GDP growth and real per capita income, respectively. These results are not surprising if they are compared with the extensive literature (Macaulay 1874, Morse 1964, Andreski 1988 and Harrison 1985, cited in Grier, 1997) 13 suggesting that Spanish colonies have a set of especial characteristics that are negative correlated with growth and development, especially due to the Catholicism -the dominant religion in the metropolis- that is focused on eternity life (over the terrestrial life) and sins punishment (over sins forgiveness). Both beliefs contribute less to promote economic performance then the Protestantism. Despite of these last results are clearly aligned with Max Weber hypothesis, Grier (1995) remarks that religion is not the sole determinant of differential development and growth (pp. 48) because controlling for Protestantism (level and growth rate) the development gap among British and French and Spanish colonies do not decrease significantly. Thus, Protestantism is not the unique factor that explains economic performance Microeconomic Level This section contains the relevant literature about the relationship between religion and development on a microeconomic point of view. In contrast with macro level studies, micro level ones take into account the specific context where the relationship between religion and development is interacting (Aldashev & Platteau, 2014). Moreover they improve the understanding of the phenomena by analyzing the relation and the impact s channels between variables. At the same time, this specificity of the micro level studies needs to be carefully treated because does not allow to make generalizations about the results to different environments and comparisons among contexts. 11

12 Another important advantage of micro level studies is that it allows to deeply analyzing the effect of religion by different development outcomes. This criterion is used to review this section prioritizing development outcomes as education, health, social insurance, life satisfaction, social attitudes and political participation. Income is not directly analyzed here. With reference to education, it is extensively know that Jewish people seems to have a higher level of income and education than other religious and not religious groups. Botticini and Eckstein (2005, 2007) have dedicated two papers to clarify and test this hypothesis. They argue that changes in religions patterns could origin changes in education patterns. For the authors, this is the case of Jewish religion, precisely from the first century of Christ Era (CE), when a religious reform was implemented and completely transformed Judaism into a religion centered around reading, studying and implementing the rules of the Torah, the Mishna, and the Talmud (pp. 12). Entirely based on historical evidence, the authors affirm that the reform compels to each Jewish man to be able to read the Torah, the Mishna and the Talmud. Thus the reform strongly contributes to increase the literacy level of the population, in comparison to non-jews people that during this period was practically illiterate. The most interesting remark of the authors is that this intensive and deeper increment on the Jewish s level of literacy could be accredited substantially to the effect of a religion s change, because there is no historical evidence about the presence of human capital investment during the analyzed period, thus education was an exogenous hedge in the formation of Jewish religion that we take as given (pp. 23). Moreover, another relevant development outcome supported by the authors is the Jews society transformation from farmers until more skilled occupations, as an effect of the religion s change. Continuing with these findings, Botticini and Eckstein (2007) now analyze the link between education levels increments and income ones, combining historical with theoretical methods 14. Authors continuing supporting evidence about the effects of religious norms changes on education and now economic individual performance. The mechanism is widely explained: changes in Judaism religious norms affect incentives on father and sons education and these micro level incentives could provoke changes on macro level economic 12

13 performance and demographic aspects. Specifically, the changes in religious norms in the first centuries of CE did incentive Jewish parents to teach Hebrew (themselves or through others) to their male children with exclusively religious objectives, but this change on education had positive repercussions on long-term economic returns (pp. 942), mainly across high skilled occupations in the different places that they were 15. This analysis are remarkable because first, explore the effect of religion on different development outcomes at the same time (education, occupancy and income level); second, explain clearly the change s mechanisms; and third, explore the affectation of micro level decisions on macro level results. Finally, authors conclude based on historical evidence- that the main events on Jewish history as: the change on urban and more skilled jobs, the decrement of the population in some periods and the diaspora (process of Jewish systematic dispersion around the world) is an effect of the changes of Jewish religion focused on literacy and education, and not due to persecutions, constraints and expulsions. Regarding to the health development outcome, Gruber and Hungerman (2008) interested on analyze the possible competition between the church and the mall since a changes on blue laws in United States (during the period ), suggested that the reduction on religious participation -due to this law change- has a significant impact on the levels of drink and drug use in the sample evaluated 16. The mechanism behind these findings consists that the new state laws flexibility about the prohibitions to work on Sunday (blue laws) discourage the religious attendance of religious people and increase their drinking and the drugs consume. Moreover, increasing their probability to commit illicit actions and contract diseases related on alcohol and drug abuse. Also, the rate distance between nonreligious and religious heavy drinkers decreases one half after the new law s applicability. As Iannaconne mentioned (Iannaconne, 1992 as cited in Gruber and Hungerman, 2008) there are strong positives externalities from religious attendance that when it decrease could maximize the individual utility but affect the social welfare, included health aspects. 13

14 Studies that analyze the direct impacts of religion on health level arrived from medical researches. Levin (1994) studied the direct impact of religion on health, regarding epidemiologic studies that remark the positive and significant impact, specifically about mortality and morbidity on people. Author tests this hypothesis and cannot reject it but clearly remarking that association not means causality, and determine that it is ambiguous. Other interesting development outcome is related to the level of social insurance 17 among religious and non-religious people. Social Insurance has a relevant role on stabilize the inter-temporal consumption of the people and protect them against shocks. Hungerman (2005) analyzing the substitution or complementarity between the Presbyterian Church and the government welfare expenditures in United States during period, found a negative and significant correlation between them. Church social activities could replace government ones through religious supporters contributions, especially to support poor people in the community. Chen (2010) analyze during the East Asian Financial Crisis ( ) if the level of religious intensity 18 of the Muslim people in Indonesia has a relationship with the level of social insurance that they received. Based on a theoretical model, author tests if religion is a risk-sharing mechanism to soften the effect on people budget as a result of external shocks like an economic crisis. Using four waves of panel data and differences and differences strategy, author concludes that the religious institutions improve the people s capacity to deal with an external shock. Moreover, religious attendance level smooth the credit constrains of the borrowers: lower religious attenders facing more obstacles to accede to credit. Regarding to the relation between religion and happiness, Clark and Lelkes (2009) examine if religious behavior of religious (Catholics and Protestants) and non-religious (Atheist) in Europe -during period- has an impact on life satisfaction, at individual and regional level. Based on regressions about the religion type, the religious attendance level and the pray personal s degree, authors found a positive and significant association between satisfied and religious people on individual and regional levels. Also, regions with a higher 14

15 share of non-religious people show -on average- lower levels of personal satisfaction for religious and non-religious people; and regions where religious are the majority, on average, present higher levels of satisfaction for both religious groups. Catholics and Protestants don t present significant differences across their life satisfaction s levels. In relation to religion and social attitudes changes, Clingingsmith, Khwaja and Kremer (2009) estimate the effect of attending to the pilgrimage to La Mecca on Pakistanis Muslims attitudes towards others. For Muslims attend to this pilgrimage is consider as one of the most important religious events in their lives. Authors built a randomized experiment based on the Saudi Arabian lottery s results that select the people that can assist or not to the pilgrimage 19 and founded on a surveys that include questions about the level of tolerance towards others, Muslims or not. Authors found that the pilgrim participation increases the attenders religious intensity after the trip but also the strength of beliefs in the equality, harmony and peace towards other religious groups and reinforce ideas in favor of women. Moreover, the religious participation in the pilgrim increases the attitudes towards the unity of Islamic religion without increase the aversion to other religious groups. The mechanism behind the pilgrim s participation changes the belief s intensity of the attenders is not explained in detail by authors. Relative to political participation and religion, Gerber et al. (2010) conclude that lower levels of church attendance are associated with diminishing levels of political participation. Based on a sample composed by Catholics and Protestants in United States during around thirty years ( ), results confirm the positive correlation between religious attendance and political participation. However, the association s significance is weak contrary to other researches mentioned by the authors (Gerber et al., 2010). This significance is stronger when authors include on the analysis the effect of blue laws 20 revoke on political participation. However, republicans and Catholics voters seem to be more affected than democrats and other religions ones. While it is possible to find a wide literature discussing the relation between religion and development, there are no papers describing the same relation with atheism. In particular, 15

16 atheism is often studied from a sociological point of view in order to understand social rules and behaviors that induce into atheism. Moreover, it is possible to find a discrete number of articles that analyses the relation between secularization and atheism. Concerning the relation between atheism and secularization, Phil Zuckerman provides some paper explaining this relation and resuming the main important characteristics of atheism. As discussed in his books and articles, the main findings concerning atheism are the following: positive relationship between atheism and the degree of individual and societal security; strong correlation between organic atheism and levels of societal health (e.g. high levels of education, low poverty and strong gender equality). Indeed, Zuckerman affirms that atheism and secularity have many positive correlates, such as higher levels of education and verbal ability, lower levels of prejudice, ethnocentrism, racism, and homophobia, greater support for women s equality, child-rearing that promotes independent thinking and an absence of corporal punishment, etc. (pp. 960). 3. Data and Methodology 3.1. Description of the survey The Generation & Gender Program (GGP) is a longitudinal survey that contains information about the same persons -with three-year intervals, or waves- among year olds in 19 European countries. This questionnaire includes questions about demographics (sex, age, marital status, etc.), self-reported characteristics (as income or education level) and answers relative to topics as fertility, partnership, the transition to adulthood, economic activity, care duties, religion and attitudes. Germany s data was selected to work due to present more precise information specifically relative to religious aspects (as religion and religious attendance) in comparison to other countries initially analyzed (as France and Belgium). This data contains two waves of information relative to the periods: 2005/2006 (wave one) and 2008/2009 (wave two). The summary statistics are presented in Table 1, with the description of our data set. In particular, per each wave, we can find the number of observations, the mean, standard deviation, minimum, and maximum values. The variables are divided into: general (that 16

17 will be used as control variables, including fixed effects), religion and development variables. The number of observations in the longitudinal study to Germany drops to observations in wave one to 3226 in wave two. Moreover, these values vary across the survey s questions in each wave because respondents could not response them. To evaluate how missing data drops on wave two and analyze the presence of a potential selection bias that could invalidate the conclusions of changes among waves results, a set of analysis were made focusing on demographic variables available on the survey. These analyses are: a. A comparison between the percentage distributions of wave one and wave two. Table 3 summarizes this descriptive analysis regarding to the data distribution in percentage for each wave by answer s groups of demographics, religious and other self-reported characteristics. All the differences between waves in all the group s characteristics are lower at ±10% and the higher variations could be justified based on the natural movements of the population over the time. For example, there are decrements in the percentage of never married people, secondary education or very good health s respondents, and there are increments in the percentage of married, retired and tertiary education respondents. b. A comparison within wave one between the respondents that was and was not surveyed on wave two. Results (available in Appendix II) again seem not be significantly different between groups. c. Analyzing the wave s means and their significance to look for differences among the distributions (available in Appendix III). Means of age, income, education and health are significantly different among waves, but could be explained as a normal demographic movements across the time in the wave two: age, education and income means increase, and health means decrease for the same respondents). Means of sex and region (both fixed characteristics of the respondents) are not significantly different among waves. These three results could suggest that missing data seems to drop not biased in wave two in comparison to wave one, regarding to the sex, age, marital status, occupation, geographical region, type of settlement, religion, and other self-reported characteristics of the 17

18 respondents. Moreover, the survey applies a weighting process on wave two with adjustments targets of wave one and original sample that could explain that the differences among waves are not significant (Generations and Gender Survey Germany, 2014). An important data s limitation is that some relevant survey s questions are only available in wave one and not in wave two. Questions with data only in wave one are religion, religious attendance, geographical region and type of settlement. When was necessary, this limitation was treated based on the assumption that respondents not change their characteristics in wave two Measures of Demographic aspects These set of variables was selected form available survey s questions about the respondents: sex, age, marital status and activity status (occupation); and about respondent s place where he/she lives: geographic region and type of settlement (population number). Appendix I contains a detailed structure about the original survey s questions and answer s options. Dummy variables were created for the following respondents characteristics: male, married, employ and urban based on sex, marital status, activity status and type of settlement questions, respectively (see Table 2) Measures of Religion Religious variables were structured based on the questions relative to the respondents about their religion (Religion of the respondent?), religious attendance (Time unit of religious service attendances of the respondent?) and religious intensity founded on Marriage/Divorce Opinion (Marriage is a lifetime relationship and should never be ended?) and Homosexual Couples Opinion (Homosexual couples should have the same rights as heterosexual couples?) respondent s opinions 21. These opinions are taken as religious intensity s proxies founded on the criteria that could reveal respondent s positions about two sensitive topics for Christians and Muslims (the majority of respondents in both waves) to specific behavior rules expressed on the Bible (for Christians) as much as on the Koran (for Muslims)

19 About religion s question is important to mention that Christian, Muslims and Other religion could include a large number of denominations with different beliefs, practices and intensity among their members. For example, Christians answer includes Catholics, Protestants, and Orthodox respondents. Moreover, Protestants also contains diverse congregations as Mormons, Jehovah Witnesses or Baptists. Regarding to None religion respondents, also can include a large number of non-religious positions as atheism, agnosticism or even God s beliefs but outside of a particular religion, among others. Dummy variables about religious respondent and religious attendance level were produced and they are explained in Table 2. Religious Respondent dummy variable is generated in based on the religion of the respondent question, differentiating between religious (when the respondent is Cristian, Muslim or be a member of other religion) and not religious respondent (none religious). This variable will allow differentiate the association between religious and not religious respondents with development outcomes levels. Religious Attendance level variable was generated in based on a specific survey s question relative to the Time unit of religious service attendances, summarizing the nine original answer s options in four simple categories (none, low, medium and high attenders). Religious attendance level (non, low, medium and high attenders) and opinion s religious intensity (strongly disagree, disagree, neither agree or disagree, agree and strongly agree) were ordered to interpret that a higher value obtained means a higher religious degree. Two indexes were created to analyze the global influence on development outcomes of religious intensity and all the religious variables analyzed: First, Religious Intensity Index simply add the results of both questions about Married/Divorce and Homosexual Couples Opinions ordered in the same sense of religious intensity degree- capturing in one indicator these point of views. For example, the most religious respondent could answer that is strong agree (value = 5) about marriage for all life ( Marriage/Divorce Opinion ) and strong disagree (value = 5) about homosexuals couples equal rights ( Homosexual Couple Opinion ), obtained a value of ten (5 + 5) as result of the Religious Intensity Index. 19

20 Second, Religiosity Index summarizes in one value per respondent his religious global behavior: religious status ( Religious Respondent ), religious attendance level ( Religious Attendance Level ) and religious intensity ( Religious Intensity Index ). For example, the most non-religious respondent could answer that he is non-religious ( Religious Respondent = 0), non-religious attender ( Religious Attendance Level =1) and has a very lower Religious Intensity value, recently explained. Each of these three variables has a weight of 1/3 in the total index, reached a maximum value of 3 (for a perfect religiosity respondent) or minimum value of 0,45 (for a perfect non-religiosity respondent). Table 4 summarizes the Religiosity Index composition is explained as follow: Religiosity Index = ( Religious Respondent * 1) + ( Religious Attendance Level / 4) + ( Religious Intensity Index / 10) Minimum Maximum Variable Weight Formula Value Value Religious Respondent 1/3 Religious Respondent * Religious Attendance Level 1/3 Religious Attendance Level / 4 0,25 1 Religious Intensity Index 1/3 Religious Intensity / 10 0,2 1 Religiosity Index 3/3 Total 0,45 3 Table 4 presents the distribution in absolute and relative terms about the religion of the respondents. In both waves, Christians have a strong presence (with around 70% of respondents) followed by None-Religious group (around 25%), Muslims (3% in wave 1 and 1% in wave 2) and other religion s respondents (around 1%). The participation in absolute and relative terms of the last two groups is insignificant and any conclusion about them needs to be especially carefully. This data is showed graphically in Figure 1. Additionally to have a first approximation to the data Figures 2, 3 and 4 present the percentage distributions per wave for the religious attendance (Figure 2) and religious intensity (Figure 3 and 4). Figure 5 presents the share of religious and non-religious respondents by age on wave one, suggesting that the religiosity of the respondents seems to be related with their age: while age group shows a higher percentage of non-religious respondents (around 30%), this relation is decreasing until the older age groups (71-75, and 81-85) with around 80% of religious respondents. 20

21 3.4. Measures of Development Development variables were chosen based on the development outcomes and beliefs analyzed during the literature review and the available data in the survey as: Income (Approximate range of monthly income of the household?), Education (Highest education level of respondent?), Health (Health general status of the respondent?), Trust (What is your opinion about thrusting people?), Fairness (What is your opinion about fairness of people?) and Happiness (Do you experience a general sense of emptiness?). Regarding to questions about the income, education and health of the respondents, there is a specific scale of eight, six and five options to answer, respectively. Appendix I summarizes these questions and answers options. About the questions about trust, fairness and happiness was generated dummy variables explained in Table 2. All development variables were ordered to be interpreted in the sense that a higher value reached means higher level of respondent s development Measures of Variation among waves To analyze the changes of a same respondent over time were created a number of variations variables based on the absolute differences between answers of wave two and wave one. Absolute variations variables are generated to all the development variables as income, education, health, trust, fairness and happiness. Absolute variations or differences determine the changes respect the respondents answers. For example one respondent can answer that has a secondary education level (option number 3 on the question) on wave 1 and that he has a tertiary education level on wave 2 (option number 4 on the question). Respondent has increased his education level in one level : wave two s value minus wave one s value (4-3 = 1). Finally absolute conversion was applied to this result to know the general variations and their degree in the same respondents over the time. These computations will be useful to measure the general impact of religion on development over the time Control Variables and fixed effects To isolate the effect of religion on development was considered as control variables the demographic variable age and the dummies male, married, employ and urban. 21

22 Moreover, the explained development variables income, health and education are taking in account as control variables when it is pertinent. Finally, to control for fixed effects by geographic region of Germany, sixteen regional dummy variables are added to the model (based on question about region of the respondent available in Appendix I). Table 1. Summary statistics Variable Wave Obs Mean Std. Dev. Min Max Demographic Variables Age ,71 16, ,14 14, Urban (dummy) ,96 0, ,95 0, Male (dummy) ,46 0, ,45 0, Married (dummy) ,58 0, ,67 0, Employ (dummy) ,52 0, ,54 0, Region ,66 3, ,71 3, Religion Variables Religion , ,66 631, Religiouos Respondant (dummy) ,74 0, ,73 0, Religious Attendance (dummy) ,95 0, ,98 0, Marriage Opinion ,10 1, ,92 1, Homosexual Coulpes Opinion ,85 1, ,56 1, Development Variables Income ,58 1, ,84 1, Education ,46 1, ,62 1, Health ,92 0, ,79 0, Trust ,35 0, ,36 0, Fairness ,62 0, ,66 0, Happiness ,82 0, ,85 0,

23 Table 2. Dummy Variables Dummy Variable Survey's Question If answer is Value Male Sex of the respondent? Male = 1 Female = 0 Married Marital status of the respondent? Married = 1 Never married, divorced, widowed = 0 Employ What is your activity status? Employed, Self-employed, Helping in a family business, Social Service = 1 Urban* Religious Respondent Religious Attendance Trust Fairness Type of Settlement where live? Unemployed, Student, Retired, Ill, Looking = 0 >= 5000 habitants = 1 < 5000 habitants = 0 Religion of the respondent? Christians, Muslims, Other Religions = 1 Time unit of religious service attendances? What is your opinion about thrusting people? What is your opinion about fairness of people? None Religion = 0 Never No Attender (=1) Rarely, Once per year, Mor than one per year Low Attender (=2) Once per month, More than once per month Normal Attend. (=3) Once per week, More than once per week, Daily High Attender (=4) Most people can be trusted = 1 Need to be very careful with people = 0 Would try to be fair = 1 People would take advantage (would not be fair) = 0 Happiness Dou you experience a general sense of emptiness? No = 1 Yes, More or less = 0 * Criteria to classify urban and non-urban populations is based on the Environmental Strategies for Cities Program of MIT. 23

24 Table 3. Distribution of Population among waves Question Answer Wave 1 Wave 2 Difference Age ,4% 0,1% -3,3% ,2% 7,9% -6,3% ,7% 13,7% -5,0% % 27% 5,4% % 20% 4,5% % 18% 2,0% % 11% 1,1% % 2% 1,6% Total 100,0% 100,0% Sex Male 46,0% 44,7% -1,4% Female 54,0% 55,3% 1,4% Total 100,0% 100,0% Marital Status Activity Status Type of Settlement (inhabit.) never married 32,1% 21,4% -10,7% married 57,6% 66,9% 9,2% divorced 6,7% 7,1% 0,4% widowed 3,6% 4,6% 1,0% Total 100,0% 100,0% employed or self-employed 50,6% 52,9% 2,2% helping family in family business 0,3% 0,3% 0,0% unemployed 7,9% 5,0% -2,9% student, in school 5,9% 2,3% -3,6% retired 23,8% 29,2% 5,4% maternity, parental, childcare leave 2,2% 1,1% -1,1% ill or disabled permanently 0,4% 0,5% 0,1% looking after the home or family 7,9% 7,8% -0,1% in military or social service 0,2% 0,1% 0,0% other 0,7% 0,7% 0,0% Total 100,0% 100,0% from 1 to 1,999 2,1% 2,7% 0,6% from 2,000 to 4,999 2,3% 2,4% 0,1% from 5,000 to 19,999 7,7% 7,7% 0,1% from 20,000 to 49,999 10,9% 12,1% 1,2% central area from 50,000 to 99,999 2,5% 2,1% -0,4% peripheral area from 50,000 to 99,999 7,8% 7,5% -0,3% central area from 100,000 to 499,999 17,5% 17,7% 0,2% peripheral area from 100,000 to 499,999 14,8% 15,9% 1,1% central area >500,000 26,5% 23,7% -2,8% peripheral area >500,000 7,9% 8,1% 0,2% Total 100,0% 100,0% 24

25 Table 3. Distribution of Population among waves (continued) Question Answer Wave 1 Wave 2 Difference Region baden-württemberg 12,1% 12,5% 0,4% bayern 15,1% 15,9% 0,8% berlin 4,9% 4,4% -0,5% brandenburg 3,4% 3,6% 0,2% bremen 0,7% 1,0% 0,2% hamburg 2,4% 2,1% -0,3% hessen 7,7% 7,3% -0,3% mecklenburg-vorpommern 2,2% 1,7% -0,5% niedersachsen 9,5% 9,5% 0,0% nordrhein-westfalen 20,5% 18,0% -2,5% rheinland-pfalz 4,9% 5,5% 0,6% saarland 1,0% 1,1% 0,1% sachsen 5,4% 5,7% 0,2% sachsen-anhalt 3,4% 4,0% 0,6% schleswig-holstein 3,6% 4,5% 0,9% thüringen 3,1% 3,1% 0,0% Total 100,0% 100,0% Income 499 Eur or less 2,9% 2,8% -0,2% 500 to 999 Eur 11,1% 9,2% -1,9% 1,000 to 1,499 Eur 18,4% 15,3% -3,1% 1,500 to 1,999 Eur 18,4% 17,8% -0,6% 2,000 to 2,499 Eur 16,6% 16,9% 0,3% 2,500 to 2,999 Eur 11,7% 14,3% 2,6% 3,000 to 4,999 Eur 15,6% 13,7% -1,9% 5,000 Eur or more 5,2% 9,9% 4,7% 100,0% 100,0% Education Primary level 1,0% 1,2% 0,2% Lower secondary level 12,6% 9,0% -3,7% Upper secondary level 54,8% 50,9% -3,9% Post-secondary non-tertiary 5,1% 7,4% 2,2% First stage of tertiary 24,2% 28,9% 4,7% Second stage of tertiary 2,2% 2,7% 0,4% Total 100,0% 100,0% Health very bad 1,1% 1,5% 0,4% bad 4,5% 4,9% 0,4% fair 20,4% 24,0% 3,6% good 49,3% 52,4% 3,1% very good 24,7% 17,2% -7,4% Total 100,0% 100,0% 25