battenr

# Title: Posterior Predictive Checks for Bayesian & Frequentist Models

# Description: When fitting a model, we need to check assumptions. In a Bayesian framework, 
# one excellent tool is the posterior predictive check. A similar thing can be done with 
# frequentist models by simulating data with the model then comparing to the observed data 

# This code demonstrates that and how it's useful

# Note: This code was created with the help of Claude Sonnet 4.5 and reviewed by me (R Batten)

battenr

# Title: Average Treatment Effect in the Overlap 

# Description: The average treatment effect in the overlap (ATO) can be a tremendously helpful 
# causal estimand. However, it can be tricky to define because it's hard to identify
# before an analysis. 

# This code demonstrates what the target population is for the ATO using code 

# Note: Claude was used to help with parts of this code, in particular the formatting of the graphic.  

battenr

# Title: Bias-corrected and Accelerated Bootstrap Interval for Causal Inference 

# Description: Sometimes the confidence interval can be tricky to estimate for 
# causal inference. In particular, there may be situations where we can't simply estimate
# it with a formula. 

# In these cases, bootstrapping can help. However, when the sampling distribution 
# is skewed or biased this can cause problems. The BCa interval can help by accounting 
# for this bias and skewness. This code demonstrates how the BCa bootstrap 
# interval works with an example. 

battenr

# Title: Why Dichotomizing a Continuous Confounder is a Bad Idea 

# Description: Dichotomizing a continuous confounder is a bad idea. 
# This code demonstrates why it's a bad idea, and how it can lead to 
# bias. 

# Setup ----

#... Packages ----

battenr

# Title: Plotting Causal Estimands using Potential Outcomes

# Description: Understanding different causal estimands is a key part of causal inference. 
# Using the potential outcomes framework can help with this! 

# This code simulates data under the following conditions: 

# Research Question: Does coffee cause happiness?

# Three variables: 

battenr

# Title: Marginal Structural Models for Time-Varying Confounding

# Description: Time-varying confounding can be tricky. If it's not properly accounted 
# for the results can be misleading. 

# This code demonstrates how to fit a marginal structural model, using 
# simulated data and the WeightIt package. It also demonstrates how 
# we can get an estimate at each time point but also an overall 
# marginal effect. 

battenr

# Title: Entropy Balancing

# Description: Demonstrating how entropy balancing can be useful for causal inference. 
# In particular, it can be useful when there is some overlap between the groups. 

# For further reading, I highligh recommend the paper by Jans Hainmuller 
# "Entropy Balancing for Causal Effects"

# Setup ----

battenr

# Title: Linear Mixed Effects Models for Causal Inference 

# Description: Mixed effects models can be useful for causal inference, in particular
# when there are clusters of groups that we want to account for. If we do not account
# for these clusters, we could get an incorrect result 
# (specifically we need to account for the differences within and between clusters)

# Note: This code was created with the help of Gemini

# Setup ----

battenr

# Title: Clone-Censor-Weight Method 

# Description: Immortal time bias can be a problem for causal inference. 
# The typical approach is to deal with this through study design. However, sometimes
# that's not possible. In particular when using secondary data. 

# One method that can be useful is the clone-censor-weight method. The process of doing that is 
# demonstrated below. 

# Notes: 

battenr

# Title: Coverage Probability 

# Description: Demonstrating how coverage probability is useful for causal inference. 
# This code uses the example of calculating variance based on three methods: 
# a) sandwich estimator, b) M-estimation, c) bootstrapping. 
# The coverage probability is calculated for each method then compared. 

# Notes: 
# 1. The bootstrap step may take a while to run
# 2. This code was generated with the help of Gemini