Target trial emulation  ·  NZAVS 2018–2023  ·  Lab report

Religious Attendance and Selective Gains in Well-being

Target trial emulation shows that supported causal effects of religious attendance on well-being are selective

Joseph A. Bulbulia · Don E. Davis · Crystal Park · Kenneth G. Rice · Geoffrey Troughton · Daryl R. Van Tongeren · Chris G. SibleyEvolutionary Human Sciences · 2026

Religious service attendance is associated with better well-being. We ask how well-being would change if monthly religious service attendance became more probable among people who were non-attenders at the start of the study. Across six annual waves of the New Zealand Attitudes and Values Study (NZAVS), we estimate effects on 24 well-being indicators. The supported effects concentrate in meaning, forgiveness, and sexual satisfaction. Estimates for most physical-health outcomes, psychological distress, and social support are near zero.

46,377participants at baseline, NZAVS

6annual waves, 2018–2023

24well-being indicators

2–3%of non-attenders initiate attendance per year

The emulated trial

A target trial emulation organises an observational study around the question and design of a hypothetical randomised trial.

What is our question?

Among baseline non-attenders, how would making monthly religious service attendance more probable affect 24 well-being indicators? We compare the probability intervention defined below with the natural course, which allows attendance to change as it ordinarily would. The intervention operates in 2019 and 2022, with outcomes measured in 2023.

What is our population?

The attendance study follows 46,377 NZAVS participants across six annual waves, 2018–2023. The primary contrast concerns 38,477 baseline non-attenders. Censoring adjustment accounts for people leaving the study, aiming to estimate outcomes for this baseline group under complete follow-up.

What interventions do we compare?

Initial question · universal attendance

Among baseline non-attenders, compare monthly attendance for everyone with zero attendance for everyone.

Adopted · δ = 5

In 2019 and 2022, divide each baseline non-attender's conditional probability of non-attendance by five. Here, conditional means that the probability depends on the person's measured characteristics and earlier survey responses. The parameter δ specifies this divisor. Compare that intervention with the natural course, which allows attendance to change as it ordinarily would. The contrast targets outcomes under complete follow-up.

Can the data support the comparison?

Many participants, few new attenders

To estimate what would happen if everyone began attending, we need evidence about people who make that change. Positivity requires that each attendance pattern assigned by an intervention has a positive probability in the observed population among people with the relevant measured history. In practice, we also need enough comparable examples in the study to reliably estimate the effects.

Only 2–3% of baseline non-attenders begin monthly attendance in a given year. In this study, the universal-attendance comparison would require substantial extrapolation: using models to predict outcomes for attendance histories with very few comparable observations. This is the practical positivity failure. A large sample can still contain very few examples of the change we want to study.

We therefore ask what would happen if attendance became more probable. The adopted intervention changes attendance probabilities, allowing attendance to vary across people. The study provides better empirical support for this comparison. It answers a different causal question: the effect of making attendance more likely among baseline non-attenders.

The central result

Small, supported gains appear in meaning and purpose, forgiveness, and sexual satisfaction; estimates for distress and most physical-health outcomes are near zero.

When we compare people at the same survey wave, attenders generally report better well-being across most outcomes. These are cross-sectional associations. By contrast, estimated gains under the probability intervention among baseline non-attenders concentrate in fewer outcomes. This difference reflects both confounding adjustment and the change in causal question. Confounding occurs when common causes of attendance and well-being distort their apparent relationship. We describe gains as supported when they meet the study's sensitivity criterion: the E-value for the confidence limit closest to zero exceeds 1.10. Larger E-values mean that stronger unmeasured confounding would be needed to move that limit to zero.

Supported gains

Meaning, forgiveness, and sexual satisfaction

As noted above, the attendance probability intervention yields small estimated gains in meaning and purpose, sexual satisfaction, and forgiveness. Standard deviations express changes relative to the variation in an outcome across people. Estimated increases are 0.085 standard deviations for meaning and purpose (E-value bound 1.25), 0.086 for sexual satisfaction (1.25), and 0.075 for forgiveness (1.21). Smaller supported gains appear for meaning and sense (0.071), short-form health (0.061), and body satisfaction (0.058).

Estimates near zero or sensitive to confounding

Small estimates for distress and most physical-health outcomes

The attendance probability intervention yields a small estimated reduction in body mass index. Its confidence interval excludes zero, while its E-value bound falls below the study's threshold. Estimates for sleep, alcohol use, anxiety, depression, fatigue, and rumination are near zero, with intervals spanning zero. The E-value bounds for social belonging and perceived social support also fall below the study's threshold of 1.10.

The estimates

Estimated effects of dividing non-attendance probability by five (δ = 5), compared with the natural course among baseline non-attenders. Effects are in standard-deviation units, which express each change relative to the variation in that outcome across people. Bonferroni-adjusted intervals control family-wise error at 5% across 24 outcomes (approximately 99.79% individual intervals). This adjustment accounts for testing several outcomes together. Source: Supplementary Table 25. Green marks effects whose E-value bounds exceed the study's threshold of 1.10. Gold marks gratitude and body mass index, whose intervals exclude zero but whose E-value bounds fall below that threshold. E denotes the E-value for the confidence limit closest to zero. Estimates for the remaining outcomes are near zero.

Inference & robustness

01

Estimation

We estimate the attendance intervention effects with a sequentially doubly robust estimator and cross-validated machine learning. We adjust for 62 baseline covariates, characteristics measured at the start of the study, and 32 time-varying confounders, common causes of attendance and well-being measured over time. Inverse-probability-of-censoring weights account for dropout. Participants who remain in the study receive weights based on their estimated probability of remaining, given their measured histories. Causal interpretation depends on the study's identification assumptions.

02

Sensitivity

The six attendance effects marked green have E-values from 1.16 to 1.25 for the confidence limits closest to zero. These E-values summarise sensitivity to unmeasured confounding beyond the measured covariates, the participant characteristics included in the analysis. Larger E-values mean that stronger unmeasured confounding would be needed to move the confidence limit to zero.

03

Measurement

Religious service attendance is self-reported. Reporting errors may bias the estimated intervention effects. The direction of bias depends on how those errors relate to attendance, outcomes, and confounders.

04

Scope

We estimate population-average total effects of the attendance probability intervention among baseline non-attenders. Those averages may conceal variation by tradition, belief strength, upbringing, and relationship status. Moreover, the study concerns adult attendance; childhood religious socialisation requires a different causal question.

What this means

Making religious attendance more likely may bring small gains in meaning, forgiveness, and sexual satisfaction among adults who begin the study as non-attenders. Most estimated changes in physical health, distress, and social support are close to zero. These are population averages under a specified attendance intervention, and their causal interpretation depends on the study's assumptions.

Citation

Bulbulia, J. A., Davis, D. E., Park, C., Rice, K. G., Troughton, G., Van Tongeren, D. R., & Sibley, C. G. (2026). Target trial emulation shows that supported causal effects of religious attendance on well-being are selective. Evolutionary Human Sciences, 8, e16. doi:10.1017/ehs.2026.10043

Data: the New Zealand Attitudes and Values Study. Estimation: supported stochastic interventions with a sequentially doubly robust estimator.