Causal Inference Dashboard

Interactive Dash app for comparing a suite of causal measurement approaches using a publicly available randomised marketing experiment as an example (Hillstrom, 2008).

Live demo: Hugging Face Space

This project provides a dashboard to:

• estimate average treatment effects (ATE) with uncertainty
• inspect heterogeneity and targeting value
• show where each method agrees and disagrees, and what assumptions drive the result

What this dashboard shows

Teams might ask two different questions:

• "Did the campaign work on average?" (causal effect / ATE)
• "Who should we target next?" (HTE / uplift policy)

This dashboard puts both views side-by-side so the methodological choices and any business implications are easily comparable.

Methods Covered

Tab	Method	Role in this project
1	Overview	Dataset summary, arm balance, headline effects
2	Bayesian A/B (PyMC hurdle model)	Probabilistic effect estimation with posterior uncertainty
3	Multi-Arm OLS with interactions	Precision-adjusted average effects and subgroup patterns
4	Uplift / HTE (S- and X-Learner)	Ranking customers by estimated incremental value, X-learner is default
5	PSM (propensity matching + caliper)	Pedagogical: the workflow you'd use on observational data, run on this RCT for comparison
6	Method Comparison	Side-by-side estimate reconciliation and takeaway

Dataset

Source: MineThatData Email Analytics (Hillstrom)

Randomised experiment across ~64,000 customers:

• Men's, Women's and Control (split roughly equal-sized three ways)
• Primary outcome: 2-week post-campaign spend (USD)
• Key covariates: recency, history, mens/womens indicators, zip code, newbie, channel

Quick Start

Dependencies are managed with uv.

Requirements

• uv installed
• Python 3.13+

Install

From the project root:

uv sync

This creates .venv (if needed) and installs the locked dependency set.

Run

uv run app.py

Open http://localhost:8050.

First-run behavior

• First run precomputes models and caches results in .cache/results.pkl.
• Subsequent runs load from cache and start quickly.
• Depending on machine speed, initial build can take several minutes.

Force recompute

• Delete .cache/results.pkl, or set USE_CACHE = False in causal_utils.py.
• Restart the app once to rebuild the cache.
• Set USE_CACHE back to True after a deliberate rebuild (optional, deleting the pickle has the same effect if USE_CACHE stays True).

Hugging Face Spaces (Docker)

Live Space: huggingface.co/spaces/jordancheney89/causality

This repo includes a Dockerfile configured for the Docker Spaces SDK

Methodology Notes and Caveats

• The underlying dataset is randomized, so causal identification of average effects comes from random assignment.
• Covariate-adjusted analyses are included for precision and interpretability, propensity matching is included as a pedagogical workflow.
• Average CATE is reported with a bootstrap interval, though that interval is estimation-conditional and reads as a lower bound on the true uncertainty for high-stakes targeting.
• Propensity matching (PSM) is included as a demo workflow for non-randomised data, and isn't used as a headline estimate for this dashboard.

Results Snapshot

• Men's email lifts two-week spend by about $0.77 per recipient (95% CI $0.50 to $1.05), Women's by about $0.42 ($0.17 to $0.67).
• The methods agree: Bayesian, OLS and both uplift learners all land between $0.74 and $0.79 (Men's) and $0.42 and $0.45 (Women's).
• The lift is a conversion effect. The emails roughly double the share of customers who buy (1.25% / 0.88% vs 0.57% in control) while spend per buyer stays around $114.
• Both uplift rankings beat random targeting (permutation p < 0.002), and the policy view turns the ranking plus a send cost and margin into an optimal mailing share.

Project Structure

.
├── Dockerfile             # Hugging Face Spaces (Docker SDK), gunicorn on port 7860
├── .dockerignore          # Smaller build context (excludes .venv, caches of dev tools)
├── app.py                 # Thin entrypoint: Dash app, theme registration, layout, callback wiring
├── causal_utils.py        # Data prep, caching, and all causal estimation logic
├── dashboard/
│   ├── theme.py           # Design tokens, Plotly template, shared style dicts
│   └── data.py            # Loads cache → exposes RESULTS, DF, PSM, BAYESIAN, UPLIFT, OLS
├── pages/                 # Dash Pages route registration
│   ├── overview.py        # /
│   ├── bayesian.py        # /bayesian
│   ├── ols.py             # /ols
│   ├── uplift.py          # /uplift
│   ├── psm.py             # /psm
│   └── comparison.py      # /comparison
├── layouts/
│   ├── shell.py           # Masthead + section nav + Dash Pages container
│   ├── components.py      # Reusable UI helpers
│   ├── overview.py        # Overview layout
│   ├── psm.py             # PSM layout
│   ├── bayesian.py        # Bayesian A/B layout
│   ├── uplift.py          # Uplift / HTE layout
│   ├── ols.py             # Multi-Arm OLS layout
│   └── comparison.py      # Method Comparison layout
├── callbacks/
│   ├── __init__.py        # register_callbacks(app)
│   ├── psm.py             # PSM callbacks
│   ├── bayesian.py        # Bayesian A/B callbacks
│   ├── uplift.py          # Uplift / HTE callbacks
│   ├── ols.py             # Static OLS figure builder
│   └── comparison.py      # Method Comparison callbacks
├── figures/
│   └── overview.py        # Static Plotly helpers for Overview tab
├── content/
│   └── methodology.py     # Long-form copy separated from layout code
├── assets/
│   └── style.css          # Global styles (Dash serves /assets automatically)
├── .cache/                # Precomputed outputs (e.g. results.pkl)
├── pyproject.toml
├── uv.lock
├── .python-version
└── README.md

Roadmap

• Add data ingestion wizard

License

MIT. See LICENSE.