A/B Testing Notebooks

This repository contains two notebook-based analyses:

• AB_Testing_Website_Conversion_Simulation.ipynb: an A/B test of website background color and conversion.
• Medicaid_DinD_testing.ipynb: a Difference-in-Differences analysis of Medicaid expansion and insurance coverage.

Both notebooks use simulated or synthetic data and are intended for statistical analysis practice, not production decision-making.

Repository Contents

File	Description
AB_Testing_Website_Conversion_Simulation.ipynb	A/B testing workflow for website conversion analysis
ab_testing.csv	User-level A/B testing dataset
NJ-PA_DinD_testing.ipynb	Difference-in-Differences workflow for Medicaid expansion analysis
synthetic_medicaid_did_micro_v1.csv	Individual-level synthetic Medicaid expansion dataset used by the DiD notebook
synthetic_medicaid_did_dataset_v2.csv	Additional synthetic Medicaid expansion dataset included in the repo

Notebook 1: Website A/B Testing

Objective

Evaluate whether website background color affects user conversion rates between a control variant and a treatment variant.

Dataset

The notebook uses ab_testing.csv, with user-level experiment data:

• User ID: unique user identifier
• Group: control or treatment assignment
• Page Views: pages viewed during the session
• Time Spent: session duration in seconds
• Conversion: whether the user converted
• Device: access device type
• Location: user location

Methods

• Data loading and basic cleaning
• Null checks and unique user assignment checks
• Sample Ratio Mismatch check using chi-square goodness-of-fit
• Device and location balance checks
• Two-proportion z-test for conversion rate difference
• Absolute lift, relative lift, and confidence interval calculation
• Guardrail metric review for time spent and page views

Main Takeaway

The notebook finds a statistically significant conversion difference between variants, with the treatment group outperforming the control group while guardrail engagement metrics remain stable.

Notebook 2: Medicaid Expansion DiD Analysis

Objective

Estimate whether Medicaid expansion increased insurance coverage in expansion states relative to non-expansion states.

Dataset

The notebook uses synthetic_medicaid_did_micro_v1.csv, with individual-level state-year observations:

• state: U.S. state identifier
• year: observation year
• expanded_state: indicator for states that eventually expanded Medicaid
• expansion_year: expansion year, or Never for non-expansion states
• post: indicator for observations at or after expansion
• age: individual age
• female: gender indicator
• low_income: low-income indicator
• insured: insurance coverage indicator

Methods

• Data quality checks for expansion-year assignment and treatment coding
• Exploratory summaries of adoption timing and coverage rates
• Difference-in-Differences interaction term creation
• Baseline DiD regression
• State and year fixed effects regression
• Adjusted fixed effects regression with individual covariates
• Coverage trend visualization
• Low-income subgroup analysis

Main Takeaway

The fixed-effects DiD estimate is positive, suggesting Medicaid expansion is associated with higher insurance coverage in this synthetic dataset. The low-income subgroup estimate is larger but less precisely estimated.

Requirements

The notebooks use Python 3 and the following packages:

• pandas
• numpy
• matplotlib
• seaborn
• scipy
• statsmodels

Install dependencies with:

pip install pandas numpy matplotlib seaborn scipy statsmodels

If you want to run the notebooks interactively, also install Jupyter:

pip install notebook ipykernel

How to Run

1. Open the repository folder.
2. Confirm the CSV files are in the same directory as the notebooks.
3. Start Jupyter:

jupyter notebook

4. Open either notebook and run cells from top to bottom.

Notes

• The datasets appear to be simulated or synthetic, so results should be interpreted as examples of analytical workflows.
• The A/B testing notebook focuses on experiment validation and conversion lift.
• The Medicaid notebook focuses on causal inference using a Difference-in-Differences design.
• Outputs may differ if the notebooks are rerun after package upgrades or local display settings change.