Gospel or curse: the impact of religious beliefs on energy poverty in rural China

The study investigates whether and how religious beliefs affect rural energy poverty in China, defined along two dimensions: accessibility to clean energy and affordability. While the energy ladder hypothesis posits that rising incomes drive transitions to cleaner energy, evidence is mixed, suggesting socio-cultural factors also matter. Religion is widespread in rural China and could shape household energy choices and constraints. The research question is whether religious beliefs exacerbate (curse) or alleviate (gospel) rural energy poverty, and through which mechanisms. The authors propose three hypotheses: H1: Religious beliefs exacerbate rural energy poverty; H2: Low income mediates the relationship between religious beliefs and energy poverty; H3: Education mediates the relationship between religious beliefs and energy poverty.

Three strands are reviewed. (1) Energy poverty in China: Despite near-universal electrification, rural households often continue to rely on solid fuels, facing both access and affordability challenges. Measures include thresholds for electricity use and energy expenditure shares. Recent work highlights persistent rural energy poverty. (2) Non-economic determinants: Beyond income, demographics (education, gender), ethnicity, social capital, and culture affect energy choices and energy poverty. Education generally reduces energy poverty; gender and ethnic disparities persist. (3) Religion and environment/energy: Evidence on religion’s environmental effects is mixed (pessimism, optimism, indifference). Two prior studies on religion and energy poverty (not rural-focused) report opposing effects. Given China’s distinct rural religious landscape (Buddhism, Protestantism, Islam prevalent) and severe rural energy poverty, the net effect in rural China remains unclear, motivating this study.

Data: CLDS 2012, 2014, 2016 mixed cross-sections; 29 mainland provinces (excluding Tibet, Hainan); rural adult sample N=13,773 after cleaning. Variables: Dependent variables are energy poverty measures: (a) Accessibility: binary indicator of primary reliance on solid fuels (firewood/coal) for cooking; 1 indicates accessibility energy poverty. (b) Affordability: continuous ratio of household energy expenditure to total income (following Danziger et al. 2019); affordability poverty also referenced at the 10% threshold. Key independent variable: Religion (1 if respondent reports any religious belief; 0 otherwise), with information on denomination (Buddhism, Protestantism, Islam, etc.). Mediators: Low income (below official poverty line) and education (years/level). Controls: demographics (sex, age and age^2, marital status), household characteristics (family size), CPC membership, agricultural experience, income (log), infrastructure (e.g., concrete pavement share), plus year and regional fixed effects (Eastern/Central/Western). Baseline models: For accessibility, Probit and IV-Probit: Accessibility_i = F(Religion_i) + X_i β + ε_i. For affordability, Tobit and IV-Tobit: Affordability_i = F(Religion_i) + X_i β + ε_i. Endogeneity strategy: Primary instrument: village/community-level density (count per thousand residents) of places of worship (churches, temples, mosques). Relevance: proximity encourages belief/activity through local religious milieu and peer effects. Exclusion: established prior to survey and not determined by a single household; unlikely to directly affect household energy use except via belief. Additional historical IV: number of temples at the provincial level in 1820 (captures intergenerational transmission of religiosity; plausibly exogenous to current household energy choices). First-stage F-statistics reported confirm instrument strength. Mediation analysis: Baron and Kenny (1986) framework with Sobel tests to assess indirect effects through low income and education on both accessibility and affordability energy poverty. Robustness: Selection-on-observables diagnostics using Altonji–Elder–Taber (2005)/Oster (2019) to gauge the role of unobservables (compute ratios of restricted vs. full-control estimates). Additional heteroskedasticity-based IV (Lewbel-type) identification; tests for heteroskedasticity (Breusch–Pagan, White). Heterogeneity: interactions by sex, ethnicity, and income groups. Intensity: frequency of religious activities as proxy for religiosity intensity (effect on energy poverty). Denominational analysis restricting to believers to compare Buddhism, Protestantism, Islam on accessibility.

- Baseline effects: Religion increases energy poverty. - Accessibility: IV-Probit marginal effect for Religion ≈ 0.564 (p < 0.01); Probit ≈ 0.232 (p < 0.01). - Affordability: IV-Tobit marginal effect for Religion ≈ 0.070 (p < 0.05); Tobit ≈ 0.212 (p < 0.05). - Instrument strength and validity: - Village-level worship density strongly predicts religiosity (first-stage coefficient ≈ 0.016, p < 0.01; F ≈ 109.69). - Historical IV (temples in 1820) also relevant (first-stage F ≈ 11.6; exogeneity tests significant in specifications reported). - Robustness to unobservables (selection-on-observables ratios): - Accessibility ratios up to ~29.742 (mean ~18.212); affordability ~3.110–3.643 (median ~3.428). Unobservables would need to be multiple times stronger than observables to nullify effects. - Heteroskedasticity-based IV indicates positive, significant religion effects; heteroskedasticity tests strongly significant. - Heterogeneity (interactions): Effects are stronger among males, ethnic minorities, and low-income groups (significant positive interaction terms). - Mechanisms (mediation via low income and education): - Income channel: Religion → Low income: 0.048***; Low income → Accessibility: 0.114***; Low income → Affordability: 0.071***. Direct Religion effects on energy poverty remain positive. Sobel tests significant, indicating mediation. - Education channel: Religion → Education: −0.240***; Education → Accessibility: −0.020**; Education → Affordability: −0.007***. Sobel tests significant, confirming education mediates the effect. - Religiosity intensity: Each additional religious activity per year is associated with ~0.1 percentage point higher probability of energy poverty (Probit marginal effects ~0.001–0.002; significant). - Denominational differences (accessibility among believers): Buddhism associated with lower accessibility energy poverty (−0.079***), Islam associated with higher (0.364***), Protestantism not statistically significant. - Controls and patterns: U-shaped relation of age with energy poverty; marriage and larger family size associated with higher risk; greater infrastructure (concrete pavement) and plains location reduce risk; agricultural experience associated with higher accessibility energy poverty; lower income strongly associated with higher energy poverty.

The findings confirm that religiosity is positively associated with both accessibility- and affordability-based energy poverty among rural Chinese households, addressing the study’s central question and supporting H1. The positive effect persists across multiple identification strategies (IV using local worship density and historical temples), robustness checks, and heteroskedasticity-based identification, suggesting a causal interpretation. The results indicate that religions shape household behavior and constraints beyond pure economic determinants, aligning with literature that socio-cultural factors affect energy choices. Mediation analysis supports H2 and H3: religiosity increases the likelihood of being low-income and lowers educational attainment; both channels raise the probability and intensity of energy poverty. Education not only facilitates adoption and use of modern energy technologies (reducing accessibility poverty) but also enhances income-generating capacity (reducing affordability poverty). The stronger effects among males, ethnic minorities, and low-income groups highlight socio-demographic and cultural intersections that exacerbate energy vulnerability. Denominational heterogeneity suggests religion-specific norms and community structures differentially influence energy choices (e.g., Buddhism’s negative association vs. Islam’s positive association with accessibility energy poverty among believers). The positive association between religious activity intensity and energy poverty adds credibility to the mechanism that deeper religiosity further limits economic capacity and adoption of clean energy. Collectively, the results emphasize that policies targeting rural energy poverty in China should account for religious and cultural contexts, complementing economic interventions with education, income support, and engagement with religious communities to promote clean energy adoption.

Using nationally representative CLDS data (2012–2016) and multiple econometric strategies, the paper shows that religious beliefs increase the likelihood of rural energy poverty in China across both accessibility and affordability dimensions. The effects are stronger among males, ethnic minorities, and low-income households. Mediation analyses identify low income and education as key pathways through which religiosity affects energy poverty. Religiosity intensity further elevates risk, and denominational analysis reveals heterogeneous impacts (Buddhism mitigates, Islam exacerbates accessibility energy poverty among believers; Protestantism shows no significant effect). Policy implications include: (1) promoting prudent religious consumption to reduce financial burdens that contribute to affordability poverty; (2) integrating religious communities with modern technology and education to enhance skills, employability, and acceptance of clean energy; (3) leveraging religious institutions’ social influence to advocate for clean energy and environmentally responsible practices; and (4) prioritizing investments in clean energy infrastructure and targeted support in minority and vulnerable rural areas. Future work could deepen understanding of religion-specific mechanisms, refine measures of religiosity and energy poverty, and explore longitudinal dynamics as rural energy transitions evolve.