Beyond Averages: Discovering Who Benefits from Religion?

Day 2: Advanced Methods in Causal Inference

Author

Affiliation

Joseph Bulbulia

Victoria University of Wellington, New Zealand

Venue

Date: Tuesday, 12 August 2025, 09:00 – 17:00
Location: Seventh College, 15th Floor Meeting Rooms
Building: Tower West, Building 1
Located on the northwest end of campus.

Page Contents

On this page you’ll find:

Technical Setup - R, RStudio, and package installation
Workshop Agenda - Detailed schedule and learning goals
Workshop Objectives - What you’ll learn and prerequisites
Workshop Presentations - Case study PDFs (ATE and CATE)
Reference Materials - Links to glossary and DAGs

Quick links: Setup | Agenda | Presentations

Install R and R Studio

Visit the r archive network (cran) at https://cran.r-project.org/ Select the version of r suitable for your operating system (windows, mac, or linux) Download and install it.

Then install RStudio: https://www.rstudio.com/products/rstudio/download/ Choose the free version of rstudio desktop,

Then install the workshop package. The margot package is optional.

# Needed for installation
if (!requireNamespace("devtools", quietly = TRUE)) {
  install.packages("devtools")
}

# Simulations and analysis
devtools::install_github("go-bayes/causalworkshop")

# Install margot for advanced workflows
# warning, this package is being refactored: for demonstration purposes only.
devtools::install_github("go-bayes/margot")
# explanations
devtools::install_github("go-bayes/boilerplate")

This will install causalworkshop along with all required dependencies.

After installation, verify everything works:

# load the package
library(causalworkshop)

# check prerequisites
check_workshop_prerequisites()

Next run `get_workshop_scripts()` to get the scripts. Run them in order.

library(causalworkshop)

# copy workshop scripts to a working directory
# default will create a directory called `workshop-scripts`
get_workshop_scripts()

# you can set a different path, e.g.
# get_workshop_scripts("different-directory-name")


# see what scripts are available
list_workshop_scripts()

# or if you placed them in a directory other thna the default `workshop-scripts` directory
list_workshop_scripts("different-directory-name")


# Work through the scripts in order
# 01-baseline-adjustment.R    - Generate data
# 02-causal-forest-analysis.R - Simple causal forest
# 03-rate-qini-curves.R       -  Evaluate heterogeneity using RATE metrics
# 04-policy-trees.R           - Decision rules
# 05-margot-workflow.R        - Better Data Viz

:::

Agenda

Table 1: Workshop Schedule

Time	Session	Learning Goals & Activities
09:00–09:15	Recap of Day 1	Potential Outcomes and Average Treatment Effect: Review of (Day 1)
09:15–09:45	How to Ask a Causal Question	Goal: Understand why causal questions in the scientific study of religion require a structured approach: Stating your causal question precisely (and what can go wrong if you don't). Example: Does religious service attendance increase charitable giving?
09:45–10:00	How to Answer a Causal Question	Goal: Overview of the workflow steps: state question, state target population, check assumptions (with experts), data, estimation, sensitivity analysis, communicate to audience Illustration: Which causal effect of religious service?
10:00–10:30	Activity & Workflow Mapping	Activity: Formulate a precise causal question. Following this, select one question and map it through the key workflow steps, identifying potential challenges at each stage.
10:30–10:45	Morning Tea
10:45–11:30	Beyond Averages: Conditional Average Treatment Effects (CATEs)	Goal: Introduction to effect heterogeneity. Activity: Identify potential effect modifiers of religion → Donations, Volunteering
11:30–12:30	Why Time-Series Data Are Tricky	Goal: Describe why sequential exposures/time-varying exposures (such as causal mediation) is hard . Activity: Imagine cases where the confounder of a mediator might also be affected by the exposure.
12:30–13:30	Lunch
13:30–14:15	Bias in Data: A Structural Approach to Measurement Error	Goal: How to use causal diagrams to investigate structural sources of bias Activity: Whiteboard session using graphs to illustrate measurement error (and bias correction).
14:15–15:00	Worked Example: Who does Religious Attendance Affect?	Goal: Hands-on demonstration using causal forests to discover subgroups in the religious attendance-prosociality relationship. Activity: Live walkthrough using NZAVS analysis
15:00–15:15	Afternoon Tea
15:15–16:00	Group Exercise: Interpreting Heterogeneous Treatment Effects	Goal: Develop skills in translating causal forest results into actionable insights for stakeholders (academic, government, religious groups, etc.). Activity: Teams analyse policy graphs showing which groups show strongest religion-charity, discuss implications.
16:00–16:45	Workflow Synthesis: Your Research Questions	Goal: Clarify complete workflows for participants' own research questions about the social consequences of religion. Activity: Participants map their questions through each workflow stage.
16:45–17:00	Wrap-up & Resources	Goal: Review: What have we learnt? Activity: Q&A, What Next? Brainstorm next steps: how might we work together?

Workshop Objectives

What You’ll Learn

By the end of this workshop, you will be able to:

Formulate clear causal questions about the social consequences of religion
Apply the causal inference workflow from question formulation through to communication of results.
Understand Effect Modification and distinguish it from Moderation
Identify and address common pitfalls in causal inference with longitudinal data, including time-varying confounding and measurement error.
Understand modern machine learning methods (such as causal forests) to discover both average and heterogenious causal effects in religion $\to$ prosociality relationships
Understand how to translate causal findings into meaningful insights for diverse stakeholders (government, institutional leaders, religious communities)
Apply insights to your flagship SPARCC Proposal.

Prerequisites

Basic familiarity with R and regression analysis (someone on your team)
Expertise in the scientific study of religion
No prior knowledge of causal inference required beyond SPARCC Day 1

What to Bring

Laptop.
Download R, Rstudio, and the workshop package (as per the above instructions.)

Reference Materials

Glossary & DAGs Reference Page.

Note

PDF Compatibility Note: These presentations work best in Safari. Chrome users may experience issues with the PDF preview - if so, please use the “Open in New Tab” links for the best viewing experience.

Case Study 1: ATE

Open in New Tab

Case Study 2: CATE

Open in New Tab

--- title: "Beyond Averages: Discovering Who Benefits from Religion?" subtitle: "Day 2: Advanced Methods in Causal Inference" author: - name: "Joseph Bulbulia" id: "JB" orcid: "000 0-0002-5861-2056" email: "joseph.bulbulia@vuw.ac.nz" corresponding: true affiliations: - name: "Victoria University of Wellington, New Zealand" id: "VUW" format: html: highlight-style: kate code-fold: true code-overflow: scroll code-line-numbers: true execute: echo: false warning: false error: false editor_options: chunk_output_type: console --- ::: callout-important ### Venue **Date:** Tuesday, 12 August 2025, 09:00 – 17:00\ **Location:** Seventh College, 15th Floor Meeting Rooms\ **Building:** Tower West, Building 1\ Located on the northwest end of campus. ::: ::: welcome-banner SPARCC Day 2 Beyond Averages: Heterogeneous Treatment Effects Tuesday, 12 August 2025 \| 09:00 – 17:00 ::: ## Page Contents **On this page you'll find:** - **Technical Setup** - R, RStudio, and package installation - **Workshop Agenda** - Detailed schedule and learning goals - **Workshop Objectives** - What you'll learn and prerequisites - **Workshop Presentations** - Case study PDFs (ATE and CATE) - **Reference Materials** - Links to glossary and DAGs Quick links: [Setup](#install-r-and-r-studio) | [Agenda](#agenda) | [Presentations](#case-study-1-ate) ### Install R and R Studio Visit the r archive network (cran) at https://cran.r-project.org/ Select the version of r suitable for your operating system (windows, mac, or linux) Download and install it. Then install RStudio: https://www.rstudio.com/products/rstudio/download/ Choose the free version of rstudio desktop, Then install the workshop package. The `margot` package is optional. ```r # Needed for installation if (!requireNamespace("devtools", quietly = TRUE)) { install.packages("devtools") } # Simulations and analysis devtools::install_github("go-bayes/causalworkshop") # Install margot for advanced workflows # warning, this package is being refactored: for demonstration purposes only. devtools::install_github("go-bayes/margot") # explanations devtools::install_github("go-bayes/boilerplate") ``` This will install `causalworkshop` along with all required dependencies. After installation, verify everything works: ```r # load the package library(causalworkshop) # check prerequisites check_workshop_prerequisites() ``` ### Next run `get_workshop_scripts()` to get the scripts. Run them in order. ```r library(causalworkshop) # copy workshop scripts to a working directory # default will create a directory called `workshop-scripts` get_workshop_scripts() # you can set a different path, e.g. # get_workshop_scripts("different-directory-name") # see what scripts are available list_workshop_scripts() # or if you placed them in a directory other thna the default `workshop-scripts` directory list_workshop_scripts("different-directory-name") # Work through the scripts in order # 01-baseline-adjustment.R - Generate data # 02-causal-forest-analysis.R - Simple causal forest # 03-rate-qini-curves.R - Evaluate heterogeneity using RATE metrics # 04-policy-trees.R - Decision rules # 05-margot-workflow.R - Better Data Viz ``` ::: ## Agenda ```{r} #| label: tbl-schedule-flex #| tbl-cap: "Workshop Schedule" #| eval: true #| include: true # load required libraries library(tidyverse) library(flextable) library(kableExtra) # create the schedule data schedule <- tibble( Time = c( "09:00–09:15", "09:15–09:45", "09:45–10:00", "10:00–10:30", "10:30–10:45", "10:45–11:30", "11:30–12:30", "12:30–13:30", "13:30–14:15", "14:15–15:00", "15:00–15:15", "15:15–16:00", "16:00–16:45", "16:45–17:00" ), Session = c( "Recap of Day 1", "How to Ask a Causal Question", "How to Answer a Causal Question", "Activity & Workflow Mapping", "Morning Tea", "Beyond Averages: Conditional Average Treatment Effects (CATEs)", "Why Time-Series Data Are Tricky", "Lunch", "Bias in Data: A Structural Approach to Measurement Error", "Worked Example: Who does Religious Attendance Affect?", "Afternoon Tea", "Group Exercise: Interpreting Heterogeneous Treatment Effects", "Workflow Synthesis: Your Research Questions", "Wrap-up & Resources" ), `Learning Goals & Activities` = c( "Potential Outcomes and Average Treatment Effect: Review of (Day 1)", "Goal: Understand why causal questions in the scientific study of religion require a structured approach: Stating your causal question precisely (and what can go wrong if you don't).\nExample: Does religious service attendance increase charitable giving?", "Goal: Overview of the workflow steps: state question, state target population, check assumptions (with experts), data, estimation, sensitivity analysis, communicate to audience\nIllustration: *Which* causal effect of religious service?", "Activity: Formulate a precise causal question. Following this, select one question and map it through the key workflow steps, identifying potential challenges at each stage.", "", "Goal: Introduction to effect heterogeneity.\nActivity: Identify potential effect modifiers of religion → Donations, Volunteering", "Goal: Describe why sequential exposures/time-varying exposures (such as causal mediation) is hard .\nActivity: Imagine cases where the confounder of a mediator might also be affected by the exposure.", "", "Goal: How to use causal diagrams to investigate structural sources of bias\nActivity: Whiteboard session using graphs to illustrate measurement error (and bias correction).", "Goal: Hands-on demonstration using causal forests to discover subgroups in the religious attendance-prosociality relationship.\nActivity: Live walkthrough using NZAVS analysis", "", "Goal: Develop skills in translating causal forest results into actionable insights for stakeholders (academic, government, religious groups, etc.).\nActivity: Teams analyse policy graphs showing which groups show strongest religion-charity, discuss implications.", "Goal: Clarify complete workflows for participants' own research questions about the social consequences of religion.\nActivity: Participants map their questions through each workflow stage.", "Goal: Review: What have we learnt?\nActivity: Q&A, What Next? Brainstorm next steps: how might we work together?" ) ) # create the table with rainbow-themed styling schedule %>% kbl(escape = FALSE) %>% kable_styling( bootstrap_options = c("striped", "hover"), full_width = TRUE, font_size = 14 ) %>% column_spec(1, width = "15%", bold = TRUE) %>% column_spec(2, width = "25%", bold = TRUE) %>% column_spec(3, width = "60%") %>% # row_spec(c(3, 5, 7), background = "gold") %>% # breaks in pink row_spec(0, background = "#1e3a8a", color = "white", bold = TRUE) # header in navy # create the flextable with simpler formatting # ft <- schedule %>% # flextable() %>% # # set column widths # width(j = 1, width = .5) %>% # width(j = 2, width = 1.5) %>% # width(j = 3, width = 4.5) %>% # # format headers # bold(part = "header") %>% # # bold session names # bold(j = 2, part = "body") %>% # # set font size # fontsize(size = 10, part = "all") %>% # # remove all borders first # border_remove() %>% # # add only top border above header # hline_top(part = "header", border = officer::fp_border(color = "black", width = 2)) %>% # # add border below header # hline_bottom(part = "header", border = officer::fp_border(color = "black", width = 1)) %>% # # add bottom border at the end of table # hline_bottom(part = "body", border = officer::fp_border(color = "black", width = 2)) %>% # # set padding # padding(padding = 4, part = "all") %>% # # align text # align(align = "left", part = "all") %>% # valign(valign = "top", part = "all") # # # display the table # ft ``` ## Workshop Objectives ::: card ### What You'll Learn By the end of this workshop, you will be able to: 1. **Formulate clear causal questions** about the social consequences of religion 2. **Apply the causal inference workflow** from question formulation through to communication of results. 3. **Understand Effect Modification** and distinguish it from **Moderation** 4. **Identify and address common pitfalls** in causal inference with longitudinal data, including time-varying confounding and measurement error. 5. **Understand modern machine learning methods** (such as causal forests) to discover both average and heterogenious causal effects in religion $\to$ prosociality relationships 6. **Understand how to translate causal findings** into meaningful insights for diverse stakeholders (government, institutional leaders, religious communities) 7. **Apply insights** to your flagship SPARCC Proposal. ::: ::: callout-tip ### Prerequisites - Basic familiarity with R and regression analysis (someone on your team) - Expertise in the scientific study of religion - No prior knowledge of causal inference required beyond SPARCC Day 1 ::: ::: callout-note ### What to Bring - Laptop. - Download R, Rstudio, and the workshop package (as per the above instructions.) ::: ::: {.callout-tip} ### Reference Materials [**Glossary & DAGs Reference Page**](glossary-and-dags.qmd). ::: ::: callout-note **PDF Compatibility Note:** These presentations work best in Safari. Chrome users may experience issues with the PDF preview - if so, please use the "Open in New Tab" links for the best viewing experience. ::: # Case Study 1: ATE ::: {.presentation-embed} ```{=html} <iframe src="https://www.dropbox.com/scl/fi/op60chx25eypm36u30zo8/presentation.pdf?rlkey=ya244bljbj7y953p37sejgq0w&dl=0&raw=1" width="100%" height="500px" style="border: 1px solid #ddd; border-radius: 4px;"> </iframe> <p style="text-align: center; margin-top: 1rem;"> <a href="https://www.dropbox.com/scl/fi/op60chx25eypm36u30zo8/presentation.pdf?rlkey=ya244bljbj7y953p37sejgq0w&dl=0&raw=1" target="_blank" style="display: inline-block; padding: 0.75rem 1.5rem; background: #0066cc; color: white; text-decoration: none; border-radius: 4px;"> Open in New Tab </a> </p> ``` ::: ## Case Study 2: CATE ::: {.presentation-embed} ```{=html} <iframe src="https://www.dropbox.com/scl/fi/bbbs02vb0y09icxdhn3v6/presentation.pdf?rlkey=e51s9c4661io5xvzle9awnjsg&dl=0&raw=1" width="100%" height="500px" style="border: 1px solid #ddd; border-radius: 4px;"> </iframe> <p style="text-align: center; margin-top: 1rem;"> <a href="https://www.dropbox.com/scl/fi/bbbs02vb0y09icxdhn3v6/presentation.pdf?rlkey=e51s9c4661io5xvzle9awnjsg&dl=0&raw=1" target="_blank" style="display: inline-block; padding: 0.75rem 1.5rem; background: #0066cc; color: white; text-decoration: none; border-radius: 4px;"> Open in New Tab </a> </p> ``` ::: ------------------------------------------------------------------------ ::: {.callout-note appearance="minimal"} © 2025 Joseph Bulbulia. This work is licensed under a [Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License](https://creativecommons.org/licenses/by-nc-sa/4.0/). :::

Page Contents

Install R and R Studio

Next run get_workshop_scripts() to get the scripts. Run them in order.

Agenda

Workshop Objectives

What You’ll Learn

Case Study 1: ATE

Case Study 2: CATE

Next run `get_workshop_scripts()` to get the scripts. Run them in order.