Course Introduction
The Propensity Scores, Matching and Treatment Effect Estimation Training Course provides an advanced framework for estimating causal effects from observational data when randomized treatment assignment is unavailable, impractical, or ethically inappropriate. Participants will develop a strong understanding of potential outcomes, counterfactual reasoning, confounding, selection bias, treatment assignment mechanisms, and the assumptions required to construct credible comparisons between treated and untreated populations.
The course provides comprehensive coverage of propensity-score methods, including propensity-score matching, subclassification, inverse probability weighting, covariate adjustment, entropy balancing, overlap weighting, and doubly robust approaches. Participants will learn how each method attempts to balance observed characteristics across treatment groups, how treatment probabilities are estimated, and how analytical choices influence the resulting treatment-effect estimates. Particular emphasis is placed on understanding when these methods strengthen causal inference and when their assumptions remain vulnerable to unmeasured confounding.
Advanced matching techniques are explored in detail, including nearest-neighbor matching, caliper matching, exact and coarsened exact matching, optimal matching, Mahalanobis distance matching, full matching, and replacement strategies. Participants will examine matching specifications, common support, overlap, balance diagnostics, treatment-to-control ratios, matching quality, and the trade-offs between bias and precision. Practical exercises will demonstrate how poorly specified matching procedures can create misleading estimates even when the resulting datasets appear statistically balanced.
The course develops advanced treatment-effect estimation skills covering average treatment effects, average treatment effects on the treated, average treatment effects on the untreated, conditional treatment effects, and treatment-effect heterogeneity. Participants will learn how to distinguish the estimand from the estimator, select appropriate methods for different research questions, and interpret estimates in the context of observational-study assumptions. Attention is also given to standard errors, clustered data, weighting, uncertainty, finite samples, and sensitivity to analytical specifications.
Participants will also examine important threats to observational causal inference, including hidden confounding, positivity violations, poor overlap, treatment-selection bias, measurement error, missing data, treatment contamination, spillovers, and model misspecification. Robustness and sensitivity analysis will be integrated throughout the course, including covariate-balance assessment, placebo analyses, Rosenbaum-style sensitivity concepts, alternative propensity models, trimming, weighting diagnostics, negative controls, and comparisons across multiple estimation strategies.
By the end of the course, participants will be able to design, implement, validate, and communicate rigorous observational treatment-effect studies using propensity scores, matching, weighting, and modern causal inference techniques. Emerging methods such as doubly robust estimation, targeted learning, causal machine learning, generalized propensity scores, overlap weighting, high-dimensional confounding adjustment, and machine-learning-assisted nuisance estimation will prepare participants to work with complex administrative, survey, healthcare, social-policy, development, and programme-evaluation datasets.
Duration
10 days
Who Should Attend
-
Impact evaluation specialists conducting causal analysis using observational programme and policy data.
-
Economists and econometricians working with treatment-effect estimation and non-randomized studies.
-
Monitoring and evaluation professionals seeking advanced methods for estimating programme impacts.
-
Government analysts evaluating public interventions using administrative and survey datasets.
-
Policy researchers assessing treatment effects where randomized experiments are unavailable.
-
Development practitioners conducting observational impact evaluations across programmes and populations.
-
Statisticians and data scientists interested in propensity scores, matching, weighting, and causal inference.
-
Academic researchers conducting applied research using observational or longitudinal datasets.
-
Health and social research professionals evaluating interventions, services, exposures, and treatment programmes.
-
Consultants providing programme evaluation, impact assessment, economic analysis, and evidence-generation services.
-
Programme managers responsible for understanding and interpreting evidence on intervention effectiveness.
-
Professionals preparing empirical research papers, evaluation reports, policy briefs, and analytical studies.
Course Objectives
-
Develop advanced knowledge of propensity scores, matching methods, weighting approaches, and treatment-effect estimation for observational studies.
-
Understand potential outcomes, counterfactuals, causal estimands, confounding, selection bias, and the assumptions required for credible treatment-effect estimation.
-
Estimate propensity scores using appropriate statistical and machine-learning models while assessing specification quality and predictive performance.
-
Apply multiple matching techniques and determine how matching specifications influence balance, sample composition, bias, and statistical precision.
-
Diagnose common-support and overlap conditions and identify observations whose treatment probabilities make reliable causal comparison difficult.
-
Evaluate covariate balance before and after matching or weighting using appropriate graphical, statistical, and substantive diagnostics.
-
Estimate average and conditional treatment effects using matching, weighting, regression adjustment, and doubly robust approaches.
-
Apply inverse probability weighting, stabilized weights, overlap weights, entropy balancing, and related methods for constructing comparable analytical populations.
-
Identify and address threats caused by hidden confounding, positivity violations, treatment misclassification, missing data, and model misspecification.
-
Conduct sensitivity and robustness analyses to assess how treatment-effect conclusions change under alternative assumptions and estimation strategies.
-
Integrate modern causal machine-learning methods into treatment-effect analysis while maintaining transparent identification strategies and appropriate validation.
-
Design complete observational impact studies that connect research questions, estimands, treatment models, analytical methods, diagnostics, uncertainty, and causal interpretation.
Comprehensive Course Outline
Module 1: Foundations of Observational Treatment Effect Estimation
-
Potential outcomes, counterfactual reasoning, treatment effects, causal estimands, and the fundamental challenge of estimating causal effects from observational data.
-
Differences between randomized experiments, observational studies, quasi-experimental designs, and descriptive analyses in terms of identification and interpretation.
-
Confounding, selection bias, treatment assignment mechanisms, and the role of measured covariates in constructing credible counterfactual comparisons.
-
Defining treatment, control, target population, outcomes, intervention timing, estimands, and analytical populations before selecting an estimation method.
Module 2: Propensity Scores and Causal Identification
-
Definition and interpretation of the propensity score as the conditional probability of treatment given observed pre-treatment characteristics.
-
Conditional exchangeability, positivity, consistency, and related assumptions required for propensity-score-based causal inference.
-
Propensity-score balancing properties and the relationship between treatment assignment, observed covariates, and comparable analytical populations.
-
Assessing whether the available covariates adequately capture important treatment-selection mechanisms and potential sources of observed confounding.
Module 3: Propensity Score Estimation and Model Specification
-
Logistic regression and alternative statistical approaches for estimating treatment probabilities from observed baseline characteristics.
-
Selecting clinically, substantively, and theoretically relevant covariates rather than relying exclusively on automated statistical significance procedures.
-
Nonlinearities, interactions, splines, transformations, rare treatment exposure, high-dimensional covariates, and treatment-model specification challenges.
-
Evaluating propensity-score distributions, model diagnostics, predicted probabilities, extreme scores, and implications for subsequent matching or weighting.
Module 4: Propensity Score Matching Methods
-
Nearest-neighbor matching, matching with and without replacement, treatment-to-control ratios, and selection of appropriate distance measures.
-
Caliper matching, propensity-score distance restrictions, matching quality, discarded observations, and the trade-off between bias reduction and sample retention.
-
Mahalanobis distance, exact matching, coarsened exact matching, optimal matching, full matching, and alternative matching algorithms.
-
Evaluating how matching choices affect covariate balance, effective sample size, precision, generalizability, and interpretation of treatment effects.
Module 5: Common Support, Overlap, and Covariate Balance
-
Understanding common support and positivity as essential conditions for credible comparison between treated and untreated observations.
-
Graphical and statistical assessment of propensity-score overlap before and after matching or weighting.
-
Standardized mean differences, variance ratios, distributional diagnostics, balance plots, and other methods for assessing covariate comparability.
-
Trimming, restricting analytical populations, handling extreme propensity scores, and documenting the implications of overlap decisions.
Module 6: Matching Diagnostics and Quality Assessment
-
Evaluating whether matching successfully reduces systematic differences in observed baseline characteristics between treatment groups.
-
Identifying residual imbalance, influential observations, poor matches, unmatched units, and problematic regions of the propensity-score distribution.
-
Sensitivity of matching results to calipers, matching ratios, replacement, distance metrics, covariate sets, and matching algorithms.
-
Developing transparent matching protocols that separate design-stage decisions from outcome-model analysis and reduce opportunities for specification-driven bias.
Module 7: Treatment Effect Estimands and Estimation
-
Average treatment effect, average treatment effect on the treated, average treatment effect on the untreated, and conditional treatment-effect concepts.
-
Selecting the appropriate estimand according to the research question, intervention population, policy objective, and available comparison group.
-
Estimating treatment effects after matching using appropriate outcome models, weighting structures, standard errors, and finite-sample considerations.
-
Interpreting treatment-effect estimates, confidence intervals, uncertainty, effect sizes, and practical significance without extending conclusions beyond the identified population.
Module 8: Inverse Probability Weighting and Advanced Weighting
-
Inverse probability of treatment weighting and the construction of pseudo-populations with improved comparability between treatment groups.
-
Stabilized weights, truncation, trimming, extreme weights, effective sample size, and the practical consequences of poor positivity.
-
Overlap weighting, entropy balancing, covariate balancing propensity scores, and alternative balancing approaches for observational treatment-effect estimation.
-
Comparing weighting approaches according to balance, efficiency, target estimand, overlap characteristics, treatment prevalence, and substantive evaluation objectives.
Module 9: Doubly Robust and Semiparametric Estimation
-
Combining treatment models and outcome models to create estimators with robustness properties under specific model-misspecification conditions.
-
Augmented inverse probability weighting, doubly robust estimators, targeted estimation concepts, and the relationship between nuisance models and treatment-effect estimation.
-
Regression adjustment combined with propensity scores, matching, or weighting and the conditions under which combined approaches can improve estimation.
-
Cross-fitting, sample splitting, nuisance-function estimation, and modern approaches for reducing overfitting in flexible causal analyses.
Module 10: Hidden Confounding and Sensitivity Analysis
-
Understanding unmeasured confounding and why balance on observed variables does not establish balance on unobserved determinants of treatment and outcomes.
-
Sensitivity-analysis frameworks for evaluating how strong an unmeasured confounder would need to be to materially change treatment-effect conclusions.
-
Rosenbaum-style sensitivity concepts, omitted-variable sensitivity, negative controls, alternative covariate sets, and benchmarking hidden bias.
-
Reporting sensitivity findings transparently and distinguishing robust empirical evidence from conclusions that depend strongly on untestable assumptions.
Module 11: Missing Data, Measurement Error, and Treatment Misclassification
-
Missing-data mechanisms, incomplete covariate information, treatment records, outcome observations, and implications for propensity-score estimation.
-
Multiple imputation, weighting, complete-case analysis, missingness indicators, and sensitivity approaches for incomplete observational datasets.
-
Measurement error in treatment exposure, covariates, outcomes, eligibility, and intervention timing and its potential impact on causal estimates.
-
Treatment contamination, switching, partial exposure, adherence, and misclassification and their implications for treatment-effect interpretation.
Module 12: Longitudinal and Time-Varying Treatment Analysis
-
Propensity-score methods for longitudinal observational studies where treatment, covariates, and outcomes change over time.
-
Time-varying confounding, treatment-confounder feedback, marginal structural models, and inverse probability weighting for repeated treatment decisions.
-
Dynamic treatment regimes, treatment histories, switching, censoring, and the construction of longitudinal causal estimands.
-
Practical challenges involving repeated observations, administrative panels, changing treatment eligibility, and time-dependent programme participation.
Module 13: Heterogeneous Treatment Effects and Causal Machine Learning
-
Estimating treatment-effect heterogeneity across demographic groups, baseline risk, geography, institutional context, treatment intensity, and other characteristics.
-
Causal forests, generalized random forests, meta-learners, and machine-learning approaches for exploring conditional treatment effects.
-
Double machine learning, orthogonalization, cross-fitting, high-dimensional confounding adjustment, and flexible nuisance-function estimation.
-
Avoiding overfitting, data leakage, selective subgroup analysis, multiple testing, and unsupported individualized treatment-effect conclusions.
Module 14: Advanced Observational Study Design and Diagnostics
-
Integrating matching and propensity-score methods with regression, panel data, difference-in-differences, instrumental variables, and other causal designs.
-
Developing covariate-selection strategies based on causal diagrams, subject-matter knowledge, institutional processes, and treatment-assignment mechanisms.
-
Conducting placebo analyses, negative-control tests, falsification exercises, alternative treatment definitions, and alternative analytical populations.
-
Comparing estimates across multiple causal methods while distinguishing differences caused by estimands, assumptions, target populations, and model specifications.
Module 15: Applications in Programme, Policy, Health, and Development Evaluation
-
Applying propensity-score and matching methods to education, health, social protection, labor-market, development, and public-sector intervention evaluations.
-
Evaluating programme participation, service utilization, treatment exposure, policy eligibility, and intervention intensity using administrative and survey data.
-
Integrating observational treatment-effect evidence with implementation information, programme theory, outcome measurement, and cost-effectiveness considerations.
-
Communicating findings to technical and non-technical stakeholders while clearly documenting assumptions, uncertainty, limitations, and identified populations.
Module 16: Integrated Treatment Effect Estimation Projects and Emerging Methods
-
Designing complete observational treatment-effect studies from research question, estimand, causal framework, treatment model, matching strategy, and outcome analysis through reporting.
-
Developing reproducible workflows for propensity-score estimation, matching, weighting, balance diagnostics, sensitivity analysis, treatment-effect estimation, and visualization.
-
Emerging approaches including generalized propensity scores, continuous treatments, causal representation learning, targeted learning, automated balancing, and high-dimensional causal inference.
-
Preparing transparent technical reports that document analytical decisions, balance evidence, identification assumptions, robustness checks, limitations, and implications for programme evaluation.
Training Approach
This course will be delivered by our skilled trainers who have vast knowledge and experience as expert professionals in the fields. The course is taught in English and through a mix of theory, practical activities, group discussion and case studies. Course manuals and additional training materials will be provided to the participants upon completion of the training.
Tailor-Made Course
This course can also be tailor-made to meet organization requirement. For further inquiries, please contact us on: Email: training@upskilldevelopment.com Tel: +254 721 331 808
Training Venue
The training will be held at our Upskill Training Centre. We also offer training for a group (at a discount of 10% to 50%) at requested location all over the world. The Onsite course fee covers the course tuition, training materials, two break refreshments, buffet lunch, airport transfers, Upskill gift package, and guided tour.
Visa application, travel expenses, dinners, accommodation, insurance, and other personal expenses are catered by the participant
Certification
Participants will be issued with Upskill certificate upon completion of this course.
Airport Pickup and Accommodation
Airport pickup and accommodation is arranged upon request. For booking contact our Training Coordinator through Email: training@upskilldevelopment.com, +254 721 331 808
Terms of Payment:
Unless otherwise agreed between the two parties’ payment of the course fee should be done 3 working days before commencement of the training so as to enable us to prepare better.
