Advanced Data Analytics for Causal Inference and Experiment Design Training Course

Course overview

Business teams routinely make high-stakes decisions from observational data: changing prices, targeting customers, redesigning workflows, introducing product features, or reallocating marketing spend. Standard dashboards and predictive models can show associations, but they cannot reliably answer whether an intervention caused an outcome. This course equips experienced analysts to distinguish correlation from causation, challenge weak claims, and design experiments that produce evidence decision-makers can defend.

Across five days, participants build an end-to-end causal analysis workflow. They formulate causal questions, construct directed acyclic graphs (DAGs), identify confounders and invalid controls, define estimands, and select appropriate experimental or quasi-experimental designs. The programme covers randomised controlled trials, power and sample-size calculations, stratification, cluster and sequential experiments, difference-in-differences, regression discontinuity, instrumental variables, propensity-score methods, and sensitivity analysis. Participants use Python, R, SQL and DAGitty to prepare data, analyse treatment effects and communicate uncertainty.

Teaching combines instructor-led method demonstrations with applied labs based on realistic product, marketing, operations and policy scenarios. Participants critique flawed analyses, diagnose common sources of bias, design an experiment, and complete a capstone causal-inference case. They leave with a documented causal analysis and experiment-design pack containing a DAG, estimand statement, data requirements, design choice, power calculation, analysis plan, and stakeholder-ready recommendation.

The course is designed for analysts and data professionals who already work with business data and need to move from descriptive or predictive reporting to credible intervention measurement. It is also valuable for analytics leaders responsible for approving test plans, assessing vendor claims, and setting standards for evidence-based decisions.

Course objectives

By the end of this course, participants will be able to:

• Construct directed acyclic graphs to represent causal assumptions, confounders, mediators and colliders
• Define treatment effects and estimands, including average treatment effects, intention-to-treat and treatment-on-the-treated
• Design randomised controlled trials using randomisation, blocking, stratification and cluster assignment
• Calculate statistical power, minimum detectable effect and sample size for A/B and multivariate experiments
• Analyse experimental results with regression adjustment, confidence intervals, multiple-testing controls and heterogeneous treatment effects
• Apply difference-in-differences, regression discontinuity and instrumental-variable methods to observational business data
• Evaluate propensity-score matching and weighting models using covariate-balance diagnostics and sensitivity analysis
• Produce a defensible causal analysis and experiment-design pack for stakeholder review

Benefits of attending

Target competencies

Who should attend

• Senior Data Analysts — who need to substantiate whether business interventions changed outcomes
• Data Scientists — who must complement predictive models with causal estimates for product and commercial decisions
• Product Analysts — who design feature experiments and interpret A/B test results for product roadmaps
• Marketing Analytics Managers — who need credible incrementality evidence for campaign, channel and offer investment
• Business Intelligence Leads — who set measurement standards and challenge correlation-based performance claims
• Operations and Strategy Analysts — who evaluate policy, process and service changes when randomisation is constrained

Requirements and prerequisites

Participants should be comfortable working with tabular data and writing or reviewing basic SQL queries. They need practical familiarity with either Python or R, including data frames, visualisation and fitting a basic regression model. The course assumes understanding of descriptive statistics, probability, confidence intervals and hypothesis testing, plus experience interpreting business metrics. Prior exposure to A/B testing is useful but not essential. Participants do not need prior knowledge of causal inference, DAGs, econometrics, advanced machine learning or Bayesian methods. No specialist mathematics beyond applied algebra and statistical reasoning is required.

Training methodology

The instructor uses short technical briefings to introduce each method, then guides participants through notebook-based analysis in Python or R and SQL data preparation tasks. Case studies include feature roll-outs, campaign targeting and operational policy changes, allowing participants to compare randomised and observational approaches. Small groups review causal diagrams, identify invalid controls and defend design decisions to a mock steering group. Each day closes with a practical output, culminating in an individual application plan and a documented causal analysis and experiment-design pack.

Course outline

Day 1: Causal Questions, Assumptions and Data Structure

• Correlation, prediction and causal-effect questions
• Potential outcomes and counterfactual reasoning
• Treatment, outcome, unit and estimand definitions
• Directed acyclic graph construction in DAGitty
• Confounders, mediators, colliders and selection bias
• Backdoor criterion and valid adjustment sets
• SQL extraction patterns for treatment and outcome datasets

Workshop: Participants map a business intervention with a DAG and produce an estimand statement, adjustment-set rationale and initial data specification.

Day 2: Randomised Experiments and Test Planning

• Randomised controlled trial architecture
• Unit of randomisation and interference risks
• Simple randomisation, blocking and stratified assignment
• Cluster randomisation and intracluster correlation
• Primary metrics, guardrails and success criteria
• Power, minimum detectable effect and sample-size calculation
• Pre-registration and statistical analysis plans

Workshop: Participants create a powered A/B test plan for a product or campaign decision, including assignment logic, metrics, sample target and stopping rules.

Day 3: Experimental Analysis and Decision Rules

• Intention-to-treat and treatment-on-the-treated estimation
• Difference in means and regression-adjusted treatment effects
• Confidence intervals, p-values and practical significance
• Covariate adjustment and precision improvement
• Multiple comparisons and false discovery control
• Sequential testing and peeking bias
• Heterogeneous treatment effects and subgroup analysis

Workshop: Participants analyse an A/B test in Python or R and produce a decision memo that reports effect size, uncertainty, guardrail results and limitations.

Day 4: Quasi-Experimental Methods for Observational Data

• Identification strategy selection for non-randomised interventions
• Propensity-score matching and inverse-probability weighting
• Covariate-balance diagnostics and common-support checks
• Difference-in-differences and parallel-trends assessment
• Regression discontinuity design and bandwidth choice
• Instrumental variables and exclusion-restriction assumptions
• Sensitivity analysis for unobserved confounding

Workshop: Participants evaluate a non-randomised programme using two candidate quasi-experimental approaches and recommend the more credible identification strategy.

Day 5: Causal Evidence Governance and Applied Capstone

• Data-quality checks for causal measurement
• Missing data, attrition and non-compliance handling
• Spillovers, novelty effects and external-validity threats
• Causal model diagnostics and assumption documentation
• Visual communication of treatment effects and uncertainty
• Stakeholder challenge sessions and decision framing
• Causal analysis and experiment-design pack structure

Workshop: Participants complete and present a capstone causal analysis and experiment-design pack containing a DAG, estimand, method choice, diagnostic evidence, findings and implementation recommendation.

Tools & standards covered

Python, R, SQL, DAGitty

A typical training day

Live online deliveries follow the same structure in the East Africa Time zone, with shorter screen blocks and longer breaks.

What the fee includes

How you can take this course

Certification

Participants who complete the full five days receive the Skillset Development Certificate of Completion, stating the course title, course code, dates and delivery format — suitable for professional-development records and employer reimbursement.

Frequently asked questions