| Original Full Text | UC BerkeleyUC Berkeley Previously Published WorksTitleHuman capital affects religious identity: Causal evidence from KenyaPermalinkhttps://escholarship.org/uc/item/02z5h7d0AuthorsAlfonsi, LiviaBauer, MichalChytilová, Julieet al.Publication Date2024-03-01DOI10.1016/j.jdeveco.2023.103215Copyright InformationThis work is made available under the terms of a Creative Commons Attribution License, available at https://creativecommons.org/licenses/by/4.0/ Peer reviewedeScholarship.org Powered by the California Digital LibraryUniversity of CaliforniaJournal of Development Economics 167 (2024) 103215Available online 27 November 20230304-3878/© 2023 Elsevier B.V. All rights reserved.Regular Article Human capital affects religious identity: Causal evidence from Kenya Livia Alfonsi a,*, Michal Bauer b, Julie Chytilová c, Edward Miguel d a Harvard University, United States b CERGE-EI (a Joint Workplace of Charles University in Prague and the Economics Institute of the Czech Academy of Sciences) and Institute of Economic Studies, Faculty of Social Sciences, Charles University, Czech Republic c Institute of Economic Studies, Faculty of Social Sciences, Charles University, And the Economics Institute of the Czech Academy of Sciences, Czech Republic d University of California, Berkeley and NBER, United States A R T I C L E I N F O JEL classification: C93 O12 Z12 A B S T R A C T We study how human capital and economic conditions causally affect the choice of religious denomination. We utilize a longitudinal dataset monitoring the religious history of more than 5000 Kenyans over twenty years, in tandem with a randomized experiment (deworming) that has exogenously boosted education and living stan-dards. The main finding is that the program reduces the likelihood of membership in a Pentecostal denomination up to 20 years later, when respondents are in their mid-thirties, while there is a comparable increase in mem-bership in traditional Christian denominations. The effect is concentrated and statistically significant among a sub-group of participants who benefited most from the program in terms of increased education and income. The effects are unlikely due to increased secularization because the program does not reduce measures of religiosity. The results help explain why the global growth of the Pentecostal movement, sometimes described as a “New Reformation”, is centered in low-income communities. 1. Introduction Religion plays a central role in many people’s social and economic lives, and this makes it crucial to understand the forces, including eco-nomic factors, that shape religiosity and the choice of religious denomination (Henrich et al., 2010; Clingingsmith et al. 2009; Bryan et al. 2021; Campante and Yanagizawa-Drott 2015; Squicciarini 2020). Motivated by the classic secularization hypothesis that dates back to the European Enlightenment, much scholarly attention has focused on exploring whether economic development, and education in particular, reduces religiosity, i.e., the extent to which people believe in supernat-ural forces and participate in religious rituals. This idea has been tested and received some support in wealthy Western countries (Brown and Taylor, 2007; Gulesci and Meyersson 2016; McCleary and Barro 2006). At the same time, the observation that religiosity remains high and stable in many parts of the world despite decades of rapid economic growth has led certain scholars to be skeptical about the broad validity of the secularization hypothesis (Berger 1999; Stark 1999; Iannaccone 1998). This reservation is particularly relevant for much of Sub-Saharan Africa and Latin America, settings with high average levels of religi-osity and thriving religious “marketplaces” in which churches may adapt to local religious preferences and people can choose between multiple denominations (Stark and Bainbridge, 1985; Iannaccone, 1992, 1998). Thus, an important open question that has (to our knowledge) so far largely escaped rigorous empirical inquiry is whether and how in such settings differences in human capital and economic living standards shape the religious landscape, by affecting individual demand for different forms of religion (i.e., religious denominations and cults) or religious identities, without necessarily affecting the level of religiosity per se. Here we empirically test the hypothesis that individual human capital causally influences the choice of religious denomination using a unique data set from an African setting, namely, Kenya. In particular, we examine the decision to convert from the traditionally locally dominant Christian churches – namely, the Catholic and Anglican churches – to recently emerging churches that are part of the “Renewal” or “Pente-costal” movement of Protestant Christianity. Pentecostal Churches are characterized by beliefs in the active and miraculous role of God and spirits in everyday life, including the power to alleviate hardship (Gif-ford 2016), and are associated with more conservative views on various social and moral issues (The Pew Forum on Religion and Public Life, 2006). The relative lack of causal evidence about how individual human capital and economic well-being shape demand for particular forms of * Corresponding author. E-mail addresses: lalfonsi@hbs.edu (L. Alfonsi), bauer@cerge-ei.cz (M. Bauer), chytilova@fsv.cuni.cz (J. Chytilová), emiguel@berkeley.edu (E. Miguel). Contents lists available at ScienceDirect Journal of Development Economics journal homepage: www.elsevier.com/locate/devec https://doi.org/10.1016/j.jdeveco.2023.103215 Received 7 July 2023; Received in revised form 14 November 2023; Accepted 15 November 2023 Journal of Development Economics 167 (2024) 1032152religion and religious identities is not necessarily surprising due to several well-known empirical challenges. The first challenge is econo-metric identification of causal impacts and the risk of reverse causality: simple comparisons of religious beliefs across people with different levels of human capital can be misleading because underlying religious beliefs may affect human capital investments and socio-economic status (Bryan et al. 2021). Further, individuals may possess unobserved per-sonal or family characteristics that affect both human capital and reli-gious choices (omitted variable bias). Similarly, certain types of churches may strategically choose to locate branches in advantaged (or disadvantaged) localities in an attempt to gain more adherents. Failure to consider these and other confounding factors could lead to spurious correlations. Addressing these concerns requires an exogenous source of variation in individual human capital and economic living standards. A key challenge is practical: most human capital investments take place during childhood and adolescence, while religious practices and identities may evolve for many decades afterwards. Therefore, estimating the causal impact of human capital investments necessitates the ability to track and survey individuals long after the original intervention. This is particu-larly challenging in low- and middle-income country (LMIC) settings without established panel data collection infrastructure. A further challenge is the limited availability of measures of stated religiosity and religious denomination or identity, which are typically not collected in administrative records (when such records even exist). This study attempts to address these core empirical challenges in Kenya. We utilize a long-term (20-year) panel dataset with detailed longitudinal information on religiosity as well as religious beliefs, practices, and identity, in tandem with a randomized experiment (school-based deworming) that exogenously boosted individual human capital (health and education) and living standards into adulthood. To our knowledge, this is the first study to estimate the causal impacts of Table 1 Church affiliation difference in views on gender roles, consumption of substances, political views, and trust in others. Variable Pentecostal (broad classification) Traditional Christian Pairwise t-test (1) (2) (1)–(2) N Mean/(SE) N Mean/(SE) N Mean difference Panel A: Views on gender roles Female can be mechanic 2119 0.907 1788 0.92 3907 − 0.013* (0.005) (0.005) Male should make family decisions 2119 0.603 1789 0.562 3908 0.041*** (0.009) (0.01) Husband should help with chores 2118 0.785 1788 0.799 3906 − 0.014 (0.007) (0.008) Female should be encouraged to join elections 2119 0.904 1789 0.922 3908 − 0.018** (0.006) (0.006) Female should be subject to traditional laws 2118 0.103 1790 0.105 3908 − 0.002 (0.006) (0.006) Husband has the right to beat wife 2119 0.113 1790 0.102 3909 0.012 (0.006) (0.007) Panel B: Consumption of tobacco and alcohol Consumed tobacco last 7 days 2067 0.036 1743 0.055 3810 − 0.019*** (0.004) (0.005) Amount of alcohol last 7 days (Trimmed Top) 1973 0.192 1634 0.611 3607 − 0.419*** (0.02) (0.04) Panel C: Political views Participated in political rallies last 12month 2119 0.086 1790 0.104 3909 − 0.019** (0.006) (0.007) Participated in demonstrations last 12month 2119 0.013 1790 0.016 3909 − 0.003 (0.002) (0.003) Discussed about politics last 12month 2119 0.235 1790 0.299 3909 − 0.064*** (0.009) (0.011) Voted in the 2017 election 2119 0.825 1790 0.846 3909 − 0.021* (0.008) (0.009) Democracy is preferable 2119 0.803 1790 0.827 3909 − 0.024** (0.007) (0.007) Politics are very important 2119 0.423 1790 0.472 3909 − 0.049*** (0.008) (0.008) Leaders should be chosen through elections 2117 0.947 1790 0.952 3907 − 0.004 (0.003) (0.004) Panel D: Trust Can trust most people 2119 0.048 1790 0.05 3909 − 0.002 (0.005) (0.005) Can trust fellow tribe members 2119 0.351 1790 0.361 3909 − 0.011 (0.01) (0.011) Can trust other tribe members 2119 0.238 1790 0.261 3909 − 0.023* (0.009) (0.01) Can trust fellow church members 2119 0.588 1790 0.54 3909 0.048*** (0.011) (0.012) Can trust other church members 2118 0.344 1790 0.321 3908 0.023 (0.01) (0.011) Notes: This table reports a series of t-tests on gender roles, consumption of substances, political views, and trust in others by church affiliation. The first column summarizes the Pentecostal (broad classification) church members’ views and activities on gender roles, consumption of substances, political views, and trust in others. The second column measures the same variables for Traditional Church members. The third column conducts t-tests on the variables and measures if members of the Pentecostal (broad classification) church and members of the Traditional Church on average differ from each other in these variables. The variables are measured in KLPS-4, and thus the table uses the KLPS-4 sample only. All variables, except alcohol consumption, are normalized into a range of 0–1 for easier comprehension. L. Alfonsi et al. Journal of Development Economics 167 (2024) 1032153human capital investments on long-run individual religious outcomes using experimental variation. As background for our setting, the Pentecostal movement empha-sizes the work of the Holy Spirit in people’s everyday lives, and promises material benefits in terms of alleviating economic hardship during one’s own lifetime through religious contributions and practice (Auriol et al., 2020; Gifford 2016). The move away from more established forms of Christianity is among the most important global religious dynamics of the last half century (The Pew Forum on Religion and Public Life, 2006), and is often described as a “New Reformation” or the “Pentecostal Reformation”. Some commentators argue that its social and political consequences could in time rival those of the 16th century Protestant Reformation (Botha 2007; Brown 2011; Jenkins 2011; Kobyliński 2017; Thelen 2017). Numbers of Pentecostal adherents have been rising rapidly: out of two billion self-identified Christians globally, around half a billion are currently members of churches that can be classified as Pentecostal or Pentecostal-like (so-called “Charismatic” churches). The movement’s growth has predominantly taken place during the last three decades, and Sub-Saharan Africa has been one of the main areas expe-riencing exploding growth (with Latin America being another on), although there are rising numbers in scores of countries. At the current rate of growth, some researchers predict there will soon be one billion followers (McClung 2006), replacing Catholicism as the world’s largest Christian denomination in terms of followers. Beyond differences in religious practices and beliefs, evidence from our sample and other data sets suggests that members of Pentecostal churches have more conservative social and moral attitudes on a range of issues, from gender roles to alcohol consumption. The Pew Forum on Religion and Public Life (2006) shows in samples from Brazil, the US, Kenya, and the Philippines that members of Pentecostal churches are more likely than members of traditional Christian churches to consider drinking alcohol, divorce, and abortion as never justified. In line with these patterns, in Table 1 we find in our sample that members of Pentecostal churches hold more conservative views on gender roles and are less likely to consume alcohol and tobacco. We also show that they are less interested in politics. Moving beyond descriptive evidence, recent work of Buccione and Mello (2021) suggests greater exposure to a Pentecostal TV channel increases fertility rates and lowers female labor force participation. Thus, the move to Pentecostal churches from other Christian faiths may have important social and economic consequences. In Kenya, we examine whether improvements in individual human capital and economic circumstances reduce or increase the appeal of converting to Pentecostal denominations and away from the Catholic and Anglican churches that have traditionally been numerically domi-nant. To do so, we utilize a unique longitudinal dataset, the Kenya Life Panel Survey (KLPS), which has tracked and regularly surveyed more than 5000 Kenyan individuals over twenty years, from early adolescence (median age 12 years at baseline) into adulthood (median age 35 years in the most recent round). This is an important period of life to focus on for the analysis, since previous work suggests that adolescence and young adulthood are formative periods featuring a large share of reli-gious conversions, as individuals search for the best “match” between their preferences and characteristics and the context in which they consume religion (Barro et al. 2010; Iannaccone 1990). The dataset contains unusually rich individual religious histories, which allows us to document the prevalence of conversions from traditional Christian churches to Pentecostal churches (and vice versa), as well as conversions to other denominations and other religious outcomes. To start, we use the 20-year KLPS dataset to generate descriptive evidence on patterns of religious conversion over time. We first show that there is a massive shift in religious affiliation during the 1998–2021 period, with approximately 30% of individuals leaving traditional churches and joining Pentecostal churches (Fig. 1). This mirrors past discussions regarding the rapid spread of Pentecostal and related beliefs and demonstrates that this is driven not just by shifts across cohorts but also by religious conversions among individuals over time. We then present a large and highly significant correlation between individual human capital and living standards – measured in terms of years of schooling, cognitive skills, and earnings – and the likelihood of these conversions, where individuals with lower levels of human capital are far more likely to switch to Pentecostal churches. At the same time, we find no systematic link between these human capital or living standards measures and levels of self-reported religiosity (e.g., as captured in stated strength of belief or by frequency of attendance at religious services). While this descriptive use of the KLPS panel data does alleviate some concerns regarding omitted variable bias, it cannot fully resolve them. Finally, and in what is (in our view) this study’s most novel contribution, we move beyond descriptive longitudinal evidence and estimate the causal role of the human capital investment that took place during participants’ youth by comparing those who were randomly assigned to receive additional years of a school-health program (a randomized school-based deworming intervention) to a control group. The Primary School Deworming Project (PSDP) has produced positive impacts on individual health, education, and economic well-being, as well as on the sector of economic employment and the likelihood of urban residence, over two decades (Baird et al., 2016; Hamory et al., 2021; Miguel and Kremer 2004a). We focus on estimating the overall effect of this human capital intervention on later individual religious denomination and religiosity, noting that any effects could operate through multiple sub-channels (e.g., health gains, income, urban residence) that we are unable to fully disentangle. The experimental estimates indicate that those individuals who exogenously received the human capital investment (deworming) are more likely to remain affiliated with traditional churches rather than switching to Pentecostal denominations, but there is no detectable change in their overall level of religiosity (mirroring the descriptive patterns noted above). These effects are concentrated and statistically significant among a subgroup of the KLPS sample – namely, those in-dividuals who were above median age at baseline – who experienced the largest deworming treatment effects in terms of education and earnings, providing further suggestive evidence that impacts are working through those channels. Taken together, the panel evidence from the longitudi-nal KLPS dataset and the experimental variation induced by the deworming treatment tell a consistent story and indicate that the persistence of relatively low levels of education and living standards could be contributing to the rapid rise of the Pentecostal movement in Kenya. As noted below, several observers of the rise of Pentecostalism in African countries have advanced related hypotheses (Martin 2002; McClung 2006; Gifford 2016). This study adds to a nascent literature utilizing randomized control trials to explore fundamental questions related to the economics of religion. Researchers have made progress in identifying the causal effect of religion on attitudes and behaviors by exploiting a lottery allocating visas for the Islamic Hajj pilgrimage (Clingingsmith et al. 2009), and by randomizing participation in a Protestant Evangelical values and the-ology education program (Bryan et al. 2021). In terms of economic determinants of religion, the randomized provision of formal insurance was shown to reduce the size of religious donations among believers of a large Pentecostal church in Ghana (Auriol et al., 2020). This latter study is particularly relevant for us, and its evidence suggests that Pentecostal adherents perceive God to be an active force in their personal economic lives and that individual church donations are seen as a way to receive divine protection against expected future negative economic shocks. Further, the investigation of what drives people and societies to be more or less religious has a rich tradition in the social sciences, but there is a relative lack of empirical studies that explore causal determinants of the demand for different forms of religion or religious identity. Existing work has made progress in analyzing individual-level data to study the relationship between education levels and religious participation, with most studies finding a negative relationship between the two (Deaton 2009; Ruiter and van Tubergen, 2009), with some exceptions (Sacerdote L. Alfonsi et al. Journal of Development Economics 167 (2024) 1032154and Glaeser 2008). More recent research has taken advantage of educational reforms that can be viewed as natural experiments (Hun-german 2014; Gulesci and Meyersson 2016), and finds support for the interpretation that the link between higher education and lower reli-gious participation is causal in Turkey and Canada. Relatedly, city-level education was found to be negatively related to church attendance in pre-WWI Germany, in an analysis that controls for city fixed effects (Becker et al. 2017). However, little well-identified evidence exists outside of rich countries (Iyer 2016). Yet, understanding how recent advances in education, health, and economic well-being affect religious demand is especially important for LMIC’s, and in Sub-Saharan Africa in particular, in which recent human capital improvements have been dramatic. Most of the limited evidence on the determinants of religious conversions is correlational and focuses on the role of country-level characteristics (Barro et al. 2010), or basic demographics such as gender and age (Smith and Lundquist Denton, 2009; Greeley 1994). Recently, non-experimental approaches have been employed to estimate the effects of economic conditions on conversions in Latin America. Costa, Marcantonio, and Rocha (2022) find that regions more exposed to negative economic shocks experience a rise in Pentecostal church membership in Brazil, which is broadly in line with our observation that economic disadvantage makes Pentecostal churches more attractive among Christians. On the other hand, in an analysis using a regression discontinuity design in the allocation of cash subsidies in Ecuador, greater income was found to be positively linked to church attendance and the likelihood of being a member of a Protestant Evangelical denomination rather than the Catholic Church (Buser 2015). Turning to non-economic determinants of Pentecostalism, Buccione and Mello (2021) exploit the expansion of church-affiliated TV channels in Brazil and document that exposure to this media fosters religiosity and con-versions to Pentecostal churches. This study provides the first field experimental evidence on the im-pacts of human capital investment on the choice of religious identity. Specifically, it aims to contribute to the existing literature by (i) focusing on data from an LMIC (Kenya); (ii) employing longitudinal data over 20 years (from the KLPS), to explore when in the life cycle the impacts of increased human capital emerge on religious choices and whether they persist; (iii) measuring individual religiosity and the choice of religious denomination over time; and finally, (iv) utilizing a randomized control trial of a child school-health intervention (deworming) to provide causal evidence on the impact of individual human capital on religious choices into adulthood. 2. Pentecostalism and religion in Kenya and beyond The set of beliefs and practices offered by Pentecostal churches1 are sometimes hypothesized as being particularly attractive for the poor and socially marginalized (Martin 2002). Although there are important dif-ferences across particular Pentecostal churches, Pentecostalism’s uni-fying and most prominent feature is that religious teachings consider the relationship with God and spiritual world as being personal and “this--worldly” (Gifford 2009, 2016; Botha 2007). Poverty, health problems, or a lack of success in business and education are attributed to individual relationships with spirits and God rather than to structural constraints. Pastors emphasize how spiritually meritorious actions, such as partici-pation in rituals and financial contributions, unleash God’s blessing, helping to address material and other problems, and to reduce the risks of negative future shocks. Pentecostal preachers across Africa describe a God who does not want His people to be poor or to suffer (Gifford 2016; Fig. 1. Development of the share of respondents affiliated with Traditional Christian, Pentecostal and Other religious denominations Notes: This figure plots the share of members of Traditional Christian Churches, Pentecostal Churches and Other religious denominations from 1998 to 2021. Using data collected in KLPS-1, KLPS-2, KLPS-3 and KLPS-4 on current religious denominations as well as the history of religious conversions in the five years preceding the survey, we construct a yearly measure of the share of respondents belonging to each denomination. “Traditional Christian Churches” includes the Catholic and Protestant Anglican denominations. “Pentecostal Churches” includes churches that belong to one of the historically large Pentecostal denominations as well as newer and typically smaller churches closely resembling Pentecostalism. “Other religious denominations” include other Christian churches that are neither traditionally dominant in Kenya nor part of the Pentecostal movement (e.g., Baptists, Methodists, Jehovah’s Witnesses, Seventh Day Adventists, etc.), as well as followers of Islam and traditional local religions. For a more granular list of religious denominations and more details on classification, see Table A4. 1 This diverse branch of Christianity is difficult to label with a single term. For simplicity, we will use the term “Pentecostal churches” broadly, when referring to churches which belong to a historical Pentecostal denomination (such as the Assemblies of God and the Church of God in Christ) as well as to newer, largely independent and often smaller churches with similar beliefs and practices to the more established Pentecostal churches. L. Alfonsi et al. Journal of Development Economics 167 (2024) 1032155Auriol et al., 2020).2 The time frame of expected benefits is short, in contrast to Catholic, Anglican and other Christian denominations, where religious behavior is generally thought to influence mainly “afterlife consumption” (Azzi and Ehrenberg 1975). In addition to these “material” benefits of having access to an interventionist God, Pentecostal worship is generally less formal and more emotionally expressive than that of other Christian traditions. Preaching relies more on stories and less on textual analysis, and Pentecostal pastors are often untrained lay people who use simpler language than Catholic clergy (Botha 2007). Much Pentecostal worship is designed to bring about an experience of God’s presence and services often incorporate experiences such as divine healings, speaking in tongues and other miraculous signs of the Holy Spirit. Finally, Pente-costal churches do not have a formal or fixed theological orthodoxy, allowing for more flexibility to relate worship patterns, rituals and practices to traditional local belief systems and thus for more syncretic approaches (Botha 2007; Gifford 2016); see Online Appendix A for more background on Pentecostal churches. Kenya is a particularly relevant setting for studying the individual decision to change religious denomination and its determinants in relation to the rise of Pentecostalism. Kenya is currently in the process of a dramatic shift away from traditional Christian churches (Catholic and Anglican) towards Pentecostal churches, as we document below. At the same time, Kenya is a highly religious country, in which religion plays a crucial role in people’s everyday lives. In our sample, 95% of people report that religion is very important in their lives (as compared to, for example, 53% in the US (Pew Research Center 2018), 75% report attending church regularly, and respondents spend a non-negligible share of their household budgets on church donations. The religious landscape is also characterized by high levels of local diversity and pluralism: Pentecostal churches are present throughout the country, together with nearby Catholic and Anglican churches, and thus most Kenyans can, in principle, relatively easily switch between denominations. 3. Experimental design and data We use data from the longitudinal Kenya Life Panel Survey (KLPS), which tracks individuals from childhood into adulthood. Sample in-dividuals attended the 75 rural primary schools in Busia district (in western Kenya) that participated in the Primary School Deworming Project (PSDP) starting in 1998. This project was an early experimental study in development economics, and used a list randomization to assign the order in which these 75 schools received deworming drugs, where the phase-in of treatment was necessitated by the financial and logistical constraints of the implementing NGO. Pupils in the 50 randomly selected Group 1 and Group 2 schools (which we refer to as the treat-ment group here) were assigned to receive deworming starting in 1998 or 1999, and thus received on average 2.4 more years of deworming than pupils from the 25 Group 3 (or control) schools, which started receiving deworming in 2001.3 Table A1 documents that the treatment and control groups were well-balanced along a range of baseline char-acteristics; see (Miguel and Kremer 2004b; Baird et al., 2016; Hamory et al., 2021) for more details. KLPS was designed to follow and survey a representative sample of approximately 7,500 participants of the deworming program. Survey tracking rates have remained high across rounds, with effective survey rates of approximately 85% in all four rounds, balanced across the treatment and control groups, and nearly 90% of individuals being surveyed at least once across the four survey rounds collected to date, namely, KLPS-1 (2003–2005), KLPS-2 (2007–2009), KLPS-3 (2011–2014), and the 20-year follow-up round KLPS-4 (2017–2021). Data collection thus covers a period from childhood (PSDP baseline median age 12 years, 10–90 age range 9–16 years old) into adulthood (median age 35 years, 10–90 age range 32–39 years old in KLPS-4). High tracking rates are, in part, due to the decision to track migrants beyond the original study region, to other parts of Kenya, East Africa, and beyond, including a large share who have moved to urban areas, including the large cities of Nairobi and Mombasa, over time (see Fig. A5).4 While the KLPS sample is not nationally-representative, Busia district is close to the Kenyan national median along several leading socio- economic measures (Kenya National Bureau of Statistics 2010), and the PSDP’s school-based sample captures the vast majority of local children since 96% of Busia children 6–17 years old had “ever attended” school at baseline. The percentage of the Busia population with sec-ondary education in 2009 (10%) was comparable with the national median (11%), in 2005 75% of Busia adults were literate compared to 80% nationally, and 62% of Busia households fell below the poverty line compared to 41% nationally. The fact that Busia was somewhat poorer than average for Kenya arguably makes the KLPS population more representative of Sub-Saharan Africa as a whole, because Kenyan in-come levels are somewhat higher than the Sub-Saharan African average. In terms of the religious landscape at the outset of our study, the 1999 Census shows that the high level of religious plurality is not specific to Busia but characterizes Kenya as a whole, although there were some differences in the relative size of different religious groups between Busia and Kenya as a whole. In Busia county there was a higher fraction of Christians (94% vs. 84%), especially Catholics (49% vs. 28%), and a lower fraction of Muslims (2% vs. 8%). Online Appendix B contains more details about the PSDP experimental design and the KLPS sample. Several previous studies have estimated the short- to long-run im-pacts of the PSDP on individual human capital and economic life out-comes. Over the first two years of the deworming program, there were gains in self-reported health and pupil absenteeism in treatment schools fell by one quarter (Miguel and Kremer 2004b). Ten years after the intervention, individuals in the treatment group (as compared to the control group) had better self-reported health, higher educational attainment (by 0.3 years on average), test scores and secondary schooling attainment, as well as higher income among wage earners (20% gains) (Baird et al., 2016). Up to 20 years later, individuals in the treatment group experienced higher household consumption expendi-tures (by 14%), hourly earnings (by 13%), non-agricultural work hours (by 9%) and were more likely to live in urban areas (by 9%). The pro-gram was shown to have a high social internal rate of return given the very low cost of purchasing and delivering deworming drugs (Hamory et al., 2021). Importantly, the observed deworming effects are driven by a sub- sample of respondents who were above the median age (12 years old) 2 For example, Olukoya, the pastor of the Mountain of Fire and Miracle Ministries in Nigeria, writes “Every born again Christian is destined for an all- round success. Success is your kingdom right. It is your covenant and redemptive right” (Olukoya, 2002, p. 15) … but we are all prey to spiritual forces deter-mined “to pollute, trap and destroy people’s destiny” (Olukoya, 2005, p.24). 3 All schools were public secular schools and none of them were private. In some cases, there is a traditional affiliation with the local church but in practice they are all teaching the same government curriculum. Thus, it is unlikely that religious teaching favoring either Catholic or Anglican churches would drive our results. 4 The tracking of respondents involves two stages. In the second stage when the pace of locating respondents slows down, a representative random sub- sample containing approximately one quarter of still-unfound respondents is drawn and those are tracked very intensively. We re-weight those chosen for the intensive sample to maintain representativeness of the sample. L. Alfonsi et al. Journal of Development Economics 167 (2024) 1032156at baseline,5 experiencing greater gains in schooling levels, consumption and earnings (see Columns 1 and 2, Table 3). For instance, the deworming treatment effect on schooling attainment is 0.45 additional years of schooling for the older group (significant at 99% confidence), while for younger individuals it is close to zero and not significant. Hamory et al. (2021) contains a detailed discussion of why this pattern could prevail. To briefly recap here, this pattern is consistent with the marginal benefit of deworming declining with each additional year of treatment received. Among the younger sub-sample, pupils in both the treatment and the control groups received some deworming, with those in the treatment group receiving on average 2.4 years more. While this gap between treatment and control is the same among the older sub-sample, many older pupils in the control group did not receive any deworming at all because they graduated or left their school (as a drop-out) before 2001, when control schools were phased in. Thus, among the older subgroup, the treatment versus control effect is driven by those who received some deworming versus those (in the control group) who received close to no deworming on average. In the analysis below, we take advantage of these earlier findings and, in addition to estimating the average effects, test whether program effects on religious choices are concentrated among the older subgroup, which experiences larger gains in human capital and living standards. The main outcomes of interest in the analysis are (i) choice of reli-gious denomination and (ii) measures of religiosity. In each KLPS round, the participants reported their religious denomination at the time of the interview by choosing from an extensive list of possible churches and religions (although only a small share of respondents, at 6.3%, report non-Christian affiliations, mainly Islam or no affiliation). They also re-ported changes in religious denomination and the year and month in which they occurred. We construct two main variables of interest regarding religious affiliation and identity, namely, first, a variable indicating the religious denomination at the time of the interview (e.g., a Pentecostal or a traditional Church affiliation), and second, a variable indicating the share of time a respondent has belonged to a Pentecostal denomination since the last interview. While there might be somewhat more “noise” in the second variable due to imperfect recall about the specific timing of any religious switches, this measure does conceptually provide a more comprehensive picture of individuals’ religious denomination history, as compared to the snapshot measure of religious denomination elicited at the time of each interview round. In any case, as discussed below, patterns are qualitatively the same using both measures. To assess the level of religiosity, respondents answered a set of seven questions on the importance of religion (importance of religion in life, importance of religious identity, changes in religiosity in the past year) and on religious behavior (regular church attendance, recent church attendance, monetary and labor donations to church). We construct a mean effects index of “Religiosity” by taking the sum of the standardized values of each of these components (and then re-standardizing it so that it has mean zero and standard deviation one). The first set of questions on the importance of religion can also be seen as measures of intrinsic religiosity, which refers to a private relationship with God, while the second set of questions on religious behavior can be seen as measures of extrinsic religiosity, capturing the social dimension of being religious (Bentzen 2019). In some of the analysis, we construct two sub-indices in order to analyze the effects on intrinsic and extrinsic religiosity sepa-rately. Online Appendix C provides the survey wording of all the rele-vant questions and more detailed information on how the index is constructed. Since we aim to shed light on the transition from traditionally dominant Christian churches to Pentecostal ones, we classify churches Table 2 Human capital, the choice of religious denomination and religiosity: Descriptive evidence. Traditional Christian Pentecostal (broad classification) Pentecostal (conservative classification) Religiosity (index) (1) (2) (3) (4) Panel A: Sample - all participants Education (years) .177*** − .165*** − .125*** .073*** Raven’s test score .089*** − .075*** − .053*** .040*** Father’s education (years) .124*** − .104*** − .070*** .010 Mother’s education (years) .116*** − .101*** − .055*** .004 Total earnings (USD) .088*** − .090*** − .044*** − .015 Panel B: Sample - older participants Education (years) .163*** − .150*** − .116*** .078*** Raven’s test score .068*** − .050*** − .037** .027 Father’s education (years) .108*** − .079*** − .066*** .018 Mother’s education (years) .121*** − .088*** − .041** .015 Total earnings (USD) .116*** − .115*** − .080*** − .017 Panel C: Sample - younger participants Education (years) .185*** − .176*** − .129*** .065*** Raven’s test score .099*** − .090*** − .060*** .045*** Father’s education (years) .132*** − .121*** − .069*** − .002 Mother’s education (years) .107*** − .108*** − .060*** − .011 Total earnings (USD) .069*** − .072*** − .018 − .014 Notes: This table reports pairwise correlation coefficients between relevant measures of human capital and living standards and the main outcomes of interest using data pooled across KLPS-3 and KLPS-4. Panel A reports the correlations for the full sample. Panel B reports estimates for the sub-sample of respondents who were older than 12 years at baseline. Panel C reports estimates for the sub-sample of respondents who were 12 years old or younger at baseline. Education is calculated as the respondent’s years of schooling measured in KLPS-3. The Raven’s test score is the result of a standard Raven’s matrix test the respondents took in KLPS-3. Father’s and mother’s education measure the highest years of schooling attained by each parent. Total earnings are annual amounts calculated as the sum of wage employment across all jobs, non-agricultural self-employment profit across all businesses, and individual farming profit. Pentecostal Churches (both broad and conservative classifications) and Traditional Christian Churches are survey-round varying indicator variables equal to one if the last reported religious denomination fell into the corresponding category. Online Appendix C provides additional details on the construction of variables. * denotes significance at 10%, ** denotes significance at 5%, and *** denotes significance at 1%. 5 Note that for some respondents there are inconsistencies in the reported year of birth across different rounds of data collection. We construct the vari-able “year of birth” by using the most recent response available. An alternative way would be to use the earliest responses available. We believe using later responses is more reliable, since adults are more likely to know and report their year of birth precisely. The uncertainty about year of birth among respondents in the initial waves is also indicated by a relatively large fraction of missing responses in the baseline survey. For completeness, we report the main esti-mates when using the alternative approach in Table A.18 and the results are qualitatively similar. L. Alfonsi et al. JournalofDevelopmentEconomics167(2024)1032157Table 3 The effect of the deworming treatment on the choice of religious denomination and religiosity. Human Capital Religious Identity Religiosity Index Educational Attainment Annual individual earnings Traditional Christian Pentecostal (broad classification) Pentecostal (conservative classification) Share of time in Pentecostal (broad classification) Share of time in Pentecostal (conservative classification) Religiosity (index) (1) (2) (3) (4) (5) (6) (7) (8) Panel A: Sample - all participants Deworming treatment .258 124.114** .034 − .025 − .020 − .028 − .022* − .011 (.173) (57.570) (.020) (.017) (.012) (.017) (.012) (.025) Control Mean 9.30 1026.66 .53 .40 .25 .38 .25 − .00 Control SD 2.90 2158.93 .50 .49 .44 .46 .4 .97 Treatment Effect (%) 2.77 12.09 6.39 − 6.12 − 7.71 − 7.24 − 8.85 – R-squared .09 .17 .06 .05 .06 .06 .06 .03 Number Individuals 4588 6253 6740 6740 6740 6740 6740 5823 Number Observations 4588 12,929 18,958 18,958 18,958 18,958 18,958 10,713 Panel B: Sample - older participants Deworming treatment .449*** 250.085*** .073*** − .065*** − .064*** − .077*** − .072*** − .051 (.157) (83.433) (.027) (.024) (.018) (.026) (.017) (.042) Control Mean 8.34 1060.28 .48 .45 .29 .43 .29 − .00 Control SD 2.60 1973.75 .50 .50 .45 .47 .42 1.00 Treatment Effect (%) 5.38 23.59 15.27 − 14.43 − 22.18 − 17.90 − 25.23 – R-squared .25 .16 .06 .05 .06 .06 .07 .05 Number Individuals 1938 2663 2917 2917 2917 2917 2917 2496 Number Observations 1938 5468 7926 7926 7926 7926 7926 4588 Panel C: Sample - younger participants Deworming treatment .044 39.326 .008 .001 .008 .004 .009 .011 (.208) (64.032) (.022) (.017) (.013) (.017) (.012) (.032) Control Mean 9.98 1004.29 .56 .37 .23 .35 .22 .00 Control SD 2.92 2273.85 .50 .48 .42 .45 .38 .95 Treatment Effect (%) .44 3.92 1.36 .39 3.51 1.09 4.11 – R-squared .17 .20 .06 .06 .07 .07 .07 .03 Number Individuals 2650 3590 3823 3823 3823 3823 3823 3327 Number Observations 2650 7461 11,032 11,032 11,032 11,032 11,032 6125 Notes: This table reports treatment effects for numerous outcomes, using data pooled across KLPS-1, KLPS-2, KLPS-3, and KLPS-4 unless otherwise indicated. In Column 1, the dependent variable is educational attainment, measured in years of schooling at KLPS-3. In Column 2, the dependent variable is annual individual earnings (measured at KLPS-2, KLPS-3 and KLPS-4), calculated as the sum of wage employment across all jobs, nonagricultural self-employment profit across all business, and individual farming profit. Columns 3–7 report treatment effects on religious identity. Pentecostal (broad and conservative) and Traditional Christian are survey-round varying indicator variables constructed using data pooled across KLPS-1, KLPS-2, KLPS-3 and KLPS-4. Each indicator variable is equal to one if the last reported religious denomination falls into the corresponding categorization. For a more granular list of religious denominations reported by respondents and more details on classification, see Table A4. The share of time in Pentecostal (broad and conservative classification) is computed as the share of the five years preceding each survey date in which the respondent belonged to a Pentecostal denomination. Both variables are built from information collected in KLPS-1, KLPS-2, KLPS-3 and KLPS-4. In Column 8, the dependent variable is a religiosity index constructed as the sum of the standardized values of answers to a set of seven questions (importance of religion in life, importance of religious identity, changes in religiosity in the past year, regular church attendance, recent church attendance, monetary and labor donations to church). Since four questions from this set were added to the survey in the middle of KLPS-2, Column 8 uses data from KLPS-2 (wave 2), KLPS-3 and KLPS-4. Deworming treatment is an indicator variable equal to one for PSDP worm groups 1 and 2, which received an additional 2.4 years of deworming, on average, compared to group 3. Online Appendix C provides additional details on variable construction and question wording. Panel A reports the overall treatment effects for the full sample. Panel B reports estimates for the sub-sample of respondents who were older than 12 years at baseline. Panel C reports estimates for the sub-sample of respondents who were 12 years old or younger at baseline. Covariates follow (Hamory et al., 2021) and include controls for baseline 1998 primary school population, geographic zone of the school, survey wave, month and year of interview, a female indicator variable, baseline 1998 school grade fixed effects, the average school test score on the 1996 Busia District mock examinations, total number of primary school pupils within 6 km, and a cost-sharing school indicator. Those treated in a separate vocational training intervention (Technical and Vocational Vouchers Program, VocEd) which occurred prior to KLPS-3 are dropped from the KLPS-3 and KLPS-4 samples. Those treated in a separate small grant intervention (Startup Capital for Youth, SCY) that occurred after KLPS-3 are dropped from the KLPS-4 sample. Observations are weighted to be representative of the original PSDP population, and include KLPS population weights, SCY and VocEd control group weights, and KLPS intensive tracking weights. Standard errors are clustered at the 1998 school level. * denotes significance at 10%, ** denotes significance at 5%, and *** denotes significance at 1%. L. Alfonsi et al. Journal of Development Economics 167 (2024) 1032158into three broad and analytically relevant categories. The first category, labeled “Traditional Christian”, includes the Catholic and Protestant Anglican denominations. To define the second category of churches, labeled “Pentecostal”, we use two approaches. “Pentecostal (broad classification)” includes churches that belong to one of the historically large Pentecostal denominations (e.g., Church of God, Assembly of God Church, Pentecostal Church) and newer indigenous churches closely resembling Pentecostalism (e.g., Roho Church, Legio Maria Church). We also include many smaller churches in this category, since in most cases their name and results of internet searches suggest that they likely belong to the Pentecostal movement (e.g., Jesus Praise Center, Jesus Restoration Gospel, Miracle Church, among many others). We alterna-tively also employ a more conservative approach in constructing the “Pentecostal (conservative classification)”, in which we exclude this latter group of smaller and usually independent churches, which are more difficult to unambiguously classify. We again show that the main results are similar when using the more conservative classification. Given our interest in the distinction between traditional versus Pentecostal churches in Kenya, in the most recent round of data collection (KLPS-4) we added a survey module with a more detailed set of questions on individual religious beliefs and practices (see Online Appendix C). As expected, individuals who self-identify as members of churches that we classify as Pentecostal are far more likely to report that their church employs characteristic Pentecostal practices during ser-vices, to report personally experiencing miraculous signs of the Holy Spirit (e.g., prophecy, exorcism, speaking in tongues), and to believe in salvation (Table A2). These beliefs and practices are similar for members of larger Pentecostal churches, as well as of smaller churches in the Pentecostal category (Table A3). Interestingly, a non-trivial share of members of Traditional Christian churches also report related beliefs and practices, suggesting that these practices have to some extent been adopted by Traditional churches as well, but rates are substantially lower than for members of Pentecostal churches. Finally, a third analytical category are “Other” denominations, including other Christian churches that are neither traditionally dominant in Kenya nor part of the Pentecostal movement (e.g., Baptists, Methodists, Jehovah’s Witnesses, Seventh Day Adventists, etc.), as well as followers of Islam and the very small number of those who state they follow traditional local religions or do not state a religious affiliation (see Table A4). This category represents just 6% of the sample and the proportion is relatively stable during the study period, in contrast to the other two categories, as we describe below. 4. Results: descriptive evidence We find clear evidence that religious denomination is not fixed in the KLPS sample but evolves dynamically during adolescence and early adulthood: 38% of respondents report having changed their religious denomination from Traditional Christian to Pentecostal churches, or from Pentecostal to Traditional Christian churches, while 39% report always belonging to a Traditional Christian church and 15% to a Pentecostal denomination. Note that these numbers provide a somewhat conservative perspective on the prevalence of all switches among reli-gious denominations because they do not include changes within the group of Traditional Christian denominations (i.e., from Catholic to Anglican or vice versa, at 7% of the sample) or among Pentecostal churches (a sizeable 25%). We find a strong shift from Traditional Christian churches to Pentecostal churches over time, as shown graphically in Fig. 1: the share of respondents reporting they belong to a Traditional Christian church dropped from 72% in the first follow-up round collected during 2003–5 (KLPS-1) down to 42% in KLPS-4, while the share of those reporting belonging to a Pentecostal church increased sharply by an almost identical amount, from 22% to 50%. The drop is observed for both those initially affiliated with the Catholic and Anglican denominations. The increased membership in Pentecostal Churches is mainly driven by in-dividuals joining the larger and more established Pentecostal churches (see Fig. A1). As noted above, the share of respondents reporting they belong to the residual “Other” category is relatively constant, between 5% and 8% across the four KLPS rounds. The real action in terms of Fig. 2. The effects of the deworming treatment on the likelihood of being a member of Pentecostal churches, across the four survey rounds Notes: This figure plots estimated deworming treatment effects by survey round on the choice of religious denomination. The dependent variable is being a member of Pentecostal churches (conservative classification), which is an indicator variable equal to one if the last reported religious denomination fell into this category. The whiskers denote 95% confidence intervals. The estimation strategy and the set of control variables are the same as in Table 3. For each wave, we report estimates for (i) all observations (red diamond), (ii) the sub-sample of respondents who were older than 12 years at baseline (blue square), and (iii) the sub-sample of respondents who were 12 years old or younger at baseline (green triangle). Tables A9-11 present the regression results in detail. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) L. Alfonsi et al. Journal of Development Economics 167 (2024) 1032159religious identity in this sample is between the traditionally dominant churches and upstart Pentecostal denominations. In terms of proxies for religiosity, we find that the self-reported importance of religion is high and stable over time, while church attendance decreases slightly with age (Fig. A2). We document robust statistical associations between individual human capital and earnings measures and religious denomination: re-spondents with less human capital (e.g., years of schooling) are signif-icantly more likely to belong to a Pentecostal church and they are less likely to belong to Traditional Christian churches. Table 2 presents the pairwise correlations between respondents’ religious denomination and years of schooling measured in KLPS-3, i.e., at a point in time when most respondents are in their mid-20’s and the vast majority have completed their education. The correlations are sizable and highly statistically significant. In a linear regression framework, each additional year of completed schooling is associated with a reduced likelihood of belonging to a Pentecostal church of 3.1 percentage points (and an increased likelihood of belonging to a Traditional Christian church of 3.3 points) in KLPS-3. This link holds for other human capital proxies, namely respondent’s cognitive skills as measured by a Raven’s Matrices test and the education of the respondent’s parents (both mother and father), which is informative about the socioeconomic status of their childhood household. In line with these patterns, higher income is also negatively associated with being a member of a Pentecostal church and positively associated with being a member of a Traditional Christian church. When exploring a more detailed classification of churches (Table A5), we find similar patterns for both types of Traditional Christian churches (Catholic and Anglican), and also for both types of Pentecostal churches (larger versus smaller ones). Yet, in contrast to the choice of religious denomination, we do not detect a systematic link between the various measures of human capital, economic well-being and the stated level of religiosity: the overall index of religiosity is positively correlated with education, but not correlated with income (Column 4 of Table 2), and both sets of correlations tend to be far smaller in magnitude than those documented for religious denomination. Similarly, when examining specific components of the index (Table A6), we find that most are positively correlated with years of schooling but negatively correlated with total earnings, but most re-lationships are not statistically significant. Taken together, there is not strong evidence that higher levels of individual human capital and living standards translate into reduced religiosity in this setting. A next natural question is whether the observed statistical correla-tion between individual human capital and religious identity is causal. To shed light on this, we exploit the experimental deworming treatment program. 5. Results: causal evidence from a school health program The estimation strategy builds on (Baird et al., 2016; Hamory et al., 2021).6 We focus on intention-to-treat (ITT) effects of the deworming program, i.e., the difference in outcomes for individuals from treatment schools (who received approximately 2.4 additional years of deworming treatment in childhood) versus from control schools, and we begin with pooled regressions which utilize data from all four KLPS rounds to improve the precision of the estimates. The main coefficient of interest is the coefficient on the indicator variable for assignment to a treatment school, and we further control for covariates used in the stratification for the randomization and in KLPS sampling, including baseline school characteristics (average test score, population, number of students within 6 km, and administrative zone indicators), baseline individual characteristics (gender and grade), indicators for the KLPS survey cal-endar month, calendar year, wave and round, and an indicator for in-clusion in the control group of two other interventions not examined here (namely, vocational training and cash grants); details are provided in Online Appendix D. Following (Baird et al., 2016; Hamory et al., 2021) we present the results for the entire sample and then separately for the older (baseline age greater than 12) and younger respondents. To start, and mirroring the observational analysis above, there is no evidence of systematic impacts of the deworming human capital in-vestments on the overall index of self-reported religiosity: the coefficient estimate on deworming treatment is small in magnitude and not sig-nificant, for the full sample as well as for the sub-samples of older and younger respondents (Column 8 of Table 3). Further, there are no meaningful effects on the indices measuring intrinsic and extrinsic religiosity, as well as on separate components of the religiosity index (Table A7). Once again, in parallel to the correlations in Table 2 above, the data do suggest that the exogenous deworming human capital treatment af-fects the choice of religious denomination (Columns 3–7 of Table 3, Panel A): treatment group respondents are somewhat more likely to report to belong to a Traditional Christian denomination (by 3.4 per-centage points, Column 3) and less likely to belong to a Pentecostal church (by 2.5 percentage points, Column 4), although these full sample effects are not significant at traditional confidence levels. We find meaningful heterogeneity across older and younger subgroups and a pattern that lines up closely with these subgroups’ experienced educa-tion and living standards gains. The effects of deworming on affiliation with a Traditional Christian church are far larger in magnitude and statistically significant among the older respondents (Panel B) – the same subgroup that showed the largest gains in education and earnings – while they are close to zero for younger respondents (Panel C). Specif-ically, for the older respondents, being allocated to the treatment group reduces the probability of belonging to a Pentecostal church by 6.5 percentage points (p-value = .024), while the probability of belonging to a Traditional Christian denomination increases by 7.3 percentage points (p-value = .027).7 The deworming treatment also leads to a reduction in the share of time respondents report belonging to a Pentecostal church, based on the monthly history of their religious denomination con-structed from the answers to retrospective survey questions (Columns 6–7, p-value <.01). Effects are, if anything, somewhat stronger statistically when the more conservative classification of the Pentecostal churches is used: among the older participants, the program reduces the likelihood of being a member of a large Pentecostal church by 6.4 percentage points (Column 5 of Table 3, p-value <.01), while the effects are positive but smaller and not statistically significant when considering membership in small Pentecostal churches alone (Table A8). Thus, the dynamics of the conversions over time as well as the mitigating effect of the human capital intervention are mainly driven by reducing conversions to the larger and more established Pentecostal churches. 6 Since the deworming intervention and collection of most of the data we use (namely, rounds KLPS-1, KLPS-2 and KLPS-3) took place before pre-analysis plans became common in the social sciences, we did not pre-specify the following econometric analysis in a pre-analysis plan, similarly to (Baird et al., 2016). Thus, the readers should treat the analysis as follows as more explor-atory compared to studies that utilize a pre-analysis plan. 7 Estimated effect magnitudes are large. For a speculative calculation, recall that the deworming human capital intervention leads to an increase in educa-tional attainment of 0.449 years on average for the older respondents. If we were to impose the strong (and untestable) assumption that the effect of deworming on the choice of religious denomination operates solely via increased years of education (and that this effect is linear), one additional year of education among the older subgroup would lead to a reduction in affiliation with a Pentecostal church by 14.5 percentage points on a base rate (over the entire study period) of 45%, in other words a sizeable reduction of 32% percent. This calculation is for illustrative purposes only since it is plausible that mul-tiple channels are operative, e.g., higher earnings, urban residence, and possibly even changes in health status, beyond the effects of schooling alone. L. Alfonsi et al. Journal of Development Economics 167 (2024) 10321510While the previous analyses pooled data from all four KLPS rounds, we next take advantage of longitudinal structure of KLPS to gauge the evolution of these effects over time and their persistence by separately estimating the effects for each round (Fig. 2, Tables A9-11). Among the older respondents, the effects emerge during adolescence and persis-tently increase in size over the next fifteen years that we follow re-spondents, i.e., into their mid-thirties on average. Specifically, the deworming treatment reduces the probability of belonging to a Pente-costal church (broad classification) among older respondents by 6 per-centage points in KLPS-1 (p-value = .044), by 4.1 p. p. in KLPS-2 (p- value = .219), by 9.2 p. p. in KLPS-3 (p-value = .016) and by 8.5 p. p. in KLPS-4 (p-value = .004). We observe very similar patterns when focusing on the conservative classification of Pentecostal churches (Fig. A3): deworming treatment reduces the likelihood of belonging to one of these churches by 4.1 percentage points in KLPS-1 (p-value =.145), by 4.2 p. p. In KLPS-2 (p-value = .151), by 8.8 p. p. In KLPS-3 (p- value = .000) and by 9.5 p. p. In KLPS-4 (p-value = .000). We conclude that program effects on the choice of religious denomination are long- lasting, spanning the adolescent and early adult periods of the life cycle, during which most people make a host of crucial economic, family and social decisions, including regarding their religious identity, as we find. As expected, for the sub-sample of younger participants, we do not detect any meaningful treatment effects across the four rounds. More-over, there is no meaningful heterogeneity by respondent gender (Table A12). In order to gauge whether the treatment may operate via indirect effects through a peer group that is exposed to the same treatment, we tested whether the religious conversions are clustered within school cohorts. We estimated an intra-cluster correlation that measures the proportion of the overall variance that is explained by within-group variance for the following variables: (i) conversions to Pentecostal churches in general; (ii) conversions to Pentecostal churches specifically from Traditional churches and (iii) being a current member of a Pentecostal church. For all these variables, we find that the intra-cluster correlation is very low (below 0.005) (Table A.16). Further, we find that the number of schoolmates who later became pastors, and thus were potentially influential peers in the religious domain, is not positively related to the likelihood of converting (Table A.17). These patterns suggest that peer influences are unlikely to play an important role in explaining the treatment effects. Expert prior beliefs. Are these results surprising? We measure prior beliefs elicited from expert predictions (related to the approach in Del-laVigna and Pope (2018). Before sharing any empirical results with the scientific community, we elicited predictions in 2019 from a large number of academic experts on religion across various disciplines (N =149), mostly sociologists (37%), economists (32%), political scientists (13%) and psychologists (13%), as well as other research fields (5%). We explained the deworming intervention and its 10- and 15-year impacts on human capital and economic outcomes, and provided summary sta-tistics on the average reported importance of religion, church atten-dance, church membership and the frequency of conversions among respondents. The experts were asked to predict the effects on conver-sions and on two components of the religiosity index (church attendance and the stated importance of religion) fifteen years after the intervention (in KLPS-3). In general, the expert survey does not indicate a clear collective prior belief that the deworming intervention would reduce conversions from Traditional to Pentecostal churches. The experts’ predictions vary sub-stantially: approximately one third expected no impact at all, one third expected positive effects and another third negative effects. The ex-pected effects (both positive and negative) were often relatively small in magnitude, and they are broadly similar across scholars from different disciplines. In our view, this ex ante lack of consensus among experts suggests that the impacts of improved human capital and living stan-dards on choices of religious denomination in contexts like ours are currently not well understood, and that our evidence provides a novel input to the scholarly debate. Online Appendix E provides further details about the design and results of the expert survey.8 6. Discussion Despite extensive academic and popular debates over many decades, there remains limited econometrically well-identified, experimental evidence on how improvements in individual human capital and eco-nomic well-being affect people’s religious lives in low- and middle- income countries. We provide causal evidence indicating that improved individual human capital does not lead to a reduction in the stated importance of religion in people’s lives, using a rich set of proxies of religiosity. The findings align with the observation that most Kenyans in our sample remain very religious throughout the entire study period: the fraction of respondents who report not adhering to any religion re-mains in the range of 1%. The results are consistent with the view that belief in supernatural forces and participation in religious rituals are deeply rooted in human societies and will not automatically fade away with economic development, in line with aggregate trends showing that religiosity has not diminished in many settings outside of Western Europe despite the massive global increase in income and wealth in recent decades. At the same time, we provide several new pieces of evidence indi-cating that individual human capital can shape the demand for partic-ular forms of religion and religious identities. The unusual KLPS panel data on individual religious histories reveals that switching between different Christian denominations is extremely prevalent in Kenya, and that during the 20-year period we study, the proportion of the sample affiliated with Pentecostal churches has increased by more than 20 percentage points, while the share in Catholic and Anglican churches has declined nearly one-for-one. The exogenous variation in health, education and living standard generated by the Primary School Deworming Project indicates that human capital plays a causal role in this religious transformation, echoing the strong observational correla-tions documented in the KLPS data: in both cases, there is a strong negative association between the level of human capital and the likeli-hood of being a member of a Pentecostal church. The deworming im-pacts are driven by the sub-group of older sample individuals who benefited the most from the program in terms of increased education and adult income, suggesting that these channels are key. The process of economic development is characterized by advances along various di-mensions, especially in education, health, income levels, and ultimately in greater migration to cities. It is noteworthy that the deworming intervention causes changes in all of these dimensions, and thus the detailed micro-level evidence that we provide here from KLPS may also be relevant for thinking about how economic development more broadly could affect the decision to join Pentecostal churches. At the same time, the experimental design here cannot credibly separate which particular channel – e.g., better health, higher education, income, or urban resi-dence – or which combination of channels, is the key driver of the choice of religious identity9 It is also possible that the effects we observe are partly due to a supply-side response: Pentecostal Churches may specif-ically target relatively disadvantaged households, perhaps because of 8 Further, approximately half of the experts surveyed expected negative (mostly small) effects of the treatment on both measures of religiosity, one quarter predicted zero effects, and a quarter positive impacts. This pattern suggests that, on average, the experts believe the treatment might slightly reduce stated religiosity, broadly in line with the classic secularization hypothesis. 9 The effects are unlikely to be due to a direct effect of the deworming treatment on generally higher confidence in modern science and medicine, rather than miracles. In Table A15 we show that the treatment has virtually no influence on the likelihood the respondents ensure their own children receive vaccinations and use bed nets. L. Alfonsi et al. Journal of Development Economics 167 (2024) 10321511awareness of a greater receptiveness of this segment of the population to Pentecostal belief sets. These are interesting areas for future research to explore.10 The results are consistent with the view that religious identity is a dynamic, adaptive cultural construct, which may evolve differently across societies. The findings suggest that people who are relatively disadvantaged in terms of human capital and economic well-being are more receptive to forms of religion, like Pentecostalism, that feature an interventionist God who rewards spiritually meritorious people over immediate timeframes and more conservative values. Thus, the findings may help to explain arguably the most seismic shift in global religion during the last half century, namely, the rapid spread of Pentecostalism in the LMIC’s of Sub-Saharan Africa and Latin America, and among disadvantaged communities within more affluent societies. It is noteworthy that, at the aggregate level, improvements in human capital and economic conditions coincide with the spread of Pentecos-talism in most LMIC’s. Our results suggest this aggregate relationship might be misleading if interpreted causally and that other factors are driving the growth of Pentecostalism. In fact, our results lend some support to a speculation that the progress in terms of human capital and economic well-being may have slowed down the growth of this new variant of Christianity. It is also possible that one’s relative standing in society, in terms of human capital and economic vulnerability, rather than absolute levels, may affect the appeal of Pentecostal denominations. We end with some speculation about how awareness of the shifts in religious preferences caused by human capital could affect the strategic behavior of the leadership of various denominations. If church leaders were aware that low levels of education and living standards make in-dividuals more likely to move from Traditional Christian churches to Pentecostal churches, they may face different incentives to invest in clinics, schools and other components of the social safety net that pro-mote living standards. In particular, the findings may help to explain why Catholic and Anglican churches are seen as taking the lead in supporting education and health programs in Africa, including building schools, clinics and organizing child sponsorship programs, while Pentecostal churches are much less active in organized economic development programs (Gifford 2016). The relative lack of concern among Pentecostal leaders about structural barriers to economic growth (Gifford 2016) may in part originate in fears that rising education and living standards will stymie their church’s own growth: recall that (Auriol et al., 2020) found that providing greater access to insurance products among Pentecostal members in Ghana led to a reduction in their church contributions. While the micro-empirical approach we take in this study has some advantages, most importantly in allowing us to exploit the randomized deworming experiment, it does present several important limitations. First and foremost, changes in religious practice occur not just at the individual level but also in terms of the broader norms and expectations of society as a whole. We are unable to examine how human capital changes at the aggregate level – say, for Kenya as a whole – are affecting the evolution of religious practice, identity and religiosity. Other studies that examine city-wide or regional variation in education policies may be more promising approaches to gauge these changes (Gulesci and Meyersson 2016; Hungerman 2014; Becker et al. 2017). Finally, while the 20-year longitudinal dataset we exploit is unusually rich for an in-dividual panel survey, two decades is a short timeframe in the evolution of religious and cultural norms, and as such longer-term data will be needed to continue to assess these changes in Kenyan and other soci-eties. That said, the rapid changes we document since 1998 in Kenya are noteworthy and suggest that a massive realignment of religious identity is already well underway. Data availability The de-identified data will be shared. Acknowledgements We thank Michelle Layvant, Eric Ochieng, Hao Wang and Matthew Krupoff among others, for providing excellent research assistance on the KLPS project. We gratefully acknowledge our collaborators (Innovations for Poverty Action), and funding from the Dioraphte Foundation, Givewell, US NIH Awards (R01-TW05612, R01-HD044475, R01- HD090118, and R03-HD064888), US NSF Awards (SES-0418110 and SES-0962614), the Berkeley Population Center, Economic Research on Identity, Norms, and Narratives (ERINN), the Czech Science Foundation (20-11091S), and the Czech Ministry of Education, Youth, and Sports (ERC CZ program, LL2303). The data collection was approved by UC Berkeley Committee for the Protection of Human Subjects. The content is solely the responsibility of the authors and does not necessarily reflect the views of any of our funders. Corresponding author: lalfonsi@hbs. edu. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi. org/10.1016/j.jdeveco.2023.103215. References Auriol, Emmanuelle, Lass, Julie, Panin, Amma, Raiber, Eva, Paul, Seabright, 2020. God insures those who pay ? Formal insurance and religious offerings in Ghana. Q. J. Econ. 135 (4), 1799–1848. Azzi, Corry, Ehrenberg, Ronald, 1975. Household allocation of time and church attendance source. J. Polit. Econ. 83 (1), 27–56. Baird, Sarah, Hamory Hicks, Joan, Kremer, Michael, Miguel, Edward, 2016. Worms at work: long-run impacts of a child health investment. Q. J. Econ. 131 (4), 1637–1680. Barro, Robert J., Hwang, Jason, McCleary, Rachel M., 2010. Religious conversion in 40 countries. J. Sci. Stud. Relig. 49 (1), 15–36. Becker, Sascha O., Nagler, Markus, Woessmann, Ludger, 2017. Education and religious participation: city-level evidence from Germany’s secularization period 1890-1930. J. Econ. Growth 22 (3), 1–39. Bentzen, Jeanet Sinding, 2019. Acts of God? Religiosity and natural disasters across subnational world districts. Econ. J. 129 (622), 2295–2321. Berger, Peter L., 1999. The Desecularization of the World: Resurgent Religion and World Politics. Eerdmans Publishing Co, Michigan. Botha, Eugene, 2007. The new reformation: the amazing rise of the pentecostal- charismatic movement in the 20th century. Stud. Hist. Eccles. 33 (1), 295–325. Brown, Candy Gunther, 2011. Global Pentecostal and Charismatic Healing. Oxford University Press, New York. Brown, Sarah, Taylor, Karl, 2007. Religion and education: evidence from the national child development study. J. Econ. Behav. Organ. 63 (3), 439–460. Bryan, Gharad, Choi, James J., Dean, Karlan, 2021. Randomizing religion: the impact of protestant evangelism on economic outcomes. Q. J. Econ. 136 (1), 293–380. Buccione, Giulia, Mello, Marcela, 2021. The Effect of Media on Religion: Evidence from the Rise of Pentecostals in Brazil. Buser, Thomas, 2015. The effect of income on religiousness. Am. Econ. J. Appl. Econ. 7 (3), 178–195. Campante, Filipe, Yanagizawa-Drott, David, 2015. Does religion affect economic growth and happiness? Evidence from ramadan. Q. J. Econ. 130 (2), 615–658. 10 Empirically separating supply and demand channels is notoriously difficult. In this context, a couple of observations are noteworthy. First, Pentecostal churches are not concentrated only in poorer, rural areas. We find that the proportion of respondents who reported being members of Pentecostal churches was 43% among those living in rural areas at the time of interview and 39% among those living in urban areas, suggesting that Pentecostal churches are available across localities. Second, in Table A16 we show that intra-cluster correlation in conversions to Pentecostal churches is very low, indicating that the treatment effects are not primarily driven by factors common for class-mates, for example, by Pentecostal churches specifically targeting and opening new churches in localities where respondents who received less treatment lived. Nevertheless, we acknowledge this evidence is only indirect and to the best of our knowledge there is no database of churches, their location and date of establishment, which would allow us to test more directly whether new Pentecostal churches are targeting disadvantaged localities. L. Alfonsi et al. Journal of Development Economics 167 (2024) 10321512Clingingsmith, David, Ijaz Khwaja, Asim, Kremer, Michael, 2009. Estimating the impact of the Hajj: religion and tolerance in islam’s global gathering. Q. J. Econ. 124 (August), 1133–1170. Costa, Francisco, Angelo, Marcantonio, Rocha, Rudi, 2022. Stop Suffering! Economic Downturns and Pentecostal Upsurge. Journal of the European Economic Association (no. forth). Deaton, Angus S., 2009. Aging, Religion and Health. NBER, 15271. DellaVigna, Stefano, Pope, Devin, 2018. Predicting experimental results: who knows what? J. Polit. Econ. 126 (6), 2410–2456. Gifford, Paul, 2009. Christianity, Politics and Public Life in Kenya. Hurst, London. Gifford, Paul, 2016. Christianity, Development and Modernity in Africa. Oxford University Press, New York. Greeley, Andrew, 1994. A religious revival in Russia? J. Sci. Stud. Relig. 33 (3), 253–272. Gulesci, Selim, Meyersson, Erik, 2016. For the Love of the Republic: Education, Secularism and Empowerment (Working Paper). Hamory, Joan, Miguel, Edward, Walker, Michael, Kremer, Michael, Baird, Sarah, 2021. Twenty-year economic impacts of deworming. Proc. Natl. Acad. Sci. Mimeo 118 (14), e2023185118. Henrich, Joseph, Ensminger, Jean, Mcelreath, Richard, Barr, Abigail, Barrett, Clark, Alexander, Bolyanatz, Juan Camilo Cardenas, Gurven, Michael, Gwako, Edwins, 2010. Markets, religion, community size, and the evolution of fairness and punishment. Science 1480 (March), 1480–1484. Hungerman, Daniel M., 2014. The effect of education on religion : evidence from compulsory schooling laws. J. Econ. Behav. Organ. 104, 52–63. Iannaccone, Laurence, R., 1992. Sacrifice and stigma: reducing free-riding in cults , communes , and other collectives. J. Polit. Econ. 100 (2), 271–291. Iannaccone, Laurence, 1990. Religious practice: a human capital approach. J. Sci. Stud. Relig. 29 (3), 297–314. Iannaccone, Laurence, 1998. Introduction to the economics of religion. J. Econ. Lit. 36 (3), 1465–1495. Iyer, Sriya, 2016. The new economics of religion. J. Econ. Lit. 54 (2), 395–441. Jenkins, Philip, 2011. The Next Christendom: the Coming of Global Christianity. Oxford University Press, New York. Kenya National Bureau of Statistics, 2010. The 2009 Kenya Population and Housing Census. Kobyliński, Andrzej, 2017. Is pentecostalization the new reformation? The causes and consequences of the contemporary pentecostalization of christianity. Relig. Stud. Rev. 4 (266), 105–116. Martin, David, 2002. Pentecostalism: the World Their Parish. Blackwell Publishers Ltd, Oxford, UK. McCleary, Rachel M., Barro, Robert J., 2006. Religion and economy. J. Econ. Perspect. 20 (2), 49–72. McClung, Grant, 2006. Pentecostals: the Sequel. Christianity Today, 2006. Miguel, Edward, Kremer, Michael, 2004a. Worms : identifying impacts on education and health in the presence of treatment externalities. Econometrica 72 (1), 159–217. Miguel, Edward, Kremer, Michael, 2004b. Worms: identifying impacts on education and health in the presence of treatment externalities. Econometrica 72 (1), 159–217. Olukoya, D.K., 2002. The Star Hunters. Lagos: Mountain of Fire and Miracles Ministries. Olukoya, D.K., 2005. Praying against the Spirit of the Valley. Lagos: the Battle Cry Christian Ministries. Pew Research Center, 2018. The Age Gap in Religion Around the World. Ruiter, S., van Tubergen, F., 2009. Religious attendance in cross-national perspective: a multilevel analysis of 60 countries. Am. J. Sociol. 115 (3), 863–895. Sacerdote, Bruce I., Glaeser, Edward L., 2008. “Education and religion.” Journal of human capital. NBER 2, 188–215. Smith, Christian, Lundquist Denton, Melinda, 2009. Soul Searching: the Religious and Spiritual Lives of American Teenagers. Oxford University Press. Squicciarini, Mara, 2020. Devotion and development: religiosity, education, and economic progress in 19th-century France. Am. Econ. Rev. 110 (11), 3454–3491. Stark, Rodney, 1999. Secularization, R.I.P. Sociol. Relig. 60 (3), 249–273. Stark, R., Bainbridge, W.S., 1985. The Future of Religion: Secularization, Revival and Cult Formation. University of California Press, Berkeley and Los Angeles. The Pew Forum on Religion and Public Life, 2006. Spirit and Power: A 10-Country Survey of Pentecostals. Thelen, Mathias D., 2017. The explosive growth of pentecostal-charismatic christianity in the global south, and its implications for catholic evangelization. Homiletic & Pastoral Rev. 28. L. Alfonsi et al. |
|---|