Statistics for Policy Makers Training Course

Participants can assess whether data arrangements meet policy, quality and confidentiality requirements.

• Data inventory mapping administrative, survey, transactional and geospatial sources
• GSBPM stages used to assign production controls and accountable owners
• SDMX concepts for classifying indicators, codelists, metadata and releases
• Data protection impact assessments covering purpose limitation, retention and access
• Quality dimensions assessed through relevance, accuracy, timeliness, coherence and accessibility
• Disclosure control choices for small cells, suppression and statistical disclosure risk

Participants can identify measurement weaknesses and specify transparent data-preparation rules.

• Measurement scales determine valid summaries, charts and model specifications
• Construct validity assessed through definitions, proxies, coding rules and field procedures
• Missing-data mechanisms classified as MCAR, MAR and MNAR
• Outliers investigated using domain rules, distributions, influence and recording checks
• Reproducible data cleaning logs preserve source files, transformations and versions
• Data dictionary records variables, units, permissible values and derivation rules

Participants can select and interpret descriptive statistics for policy monitoring and comparison.

• Frequency tables distinguish counts, proportions, rates, ratios and percentages
• Mean, median and mode selected for distribution shape and decision use
• Variance, standard deviation and interquartile range quantify dispersion appropriately
• Rate standardisation separates population composition from underlying outcome differences
• Index numbers compare price, service and welfare changes over time
• Cross-tabulations expose subgroup patterns while avoiding denominator and base-rate errors

Participants can judge whether survey estimates represent the intended population.

• Sampling frames evaluated for coverage, duplication, eligibility and update quality
• Simple random, systematic, stratified and cluster sampling designs compared
• Sample-size calculations use confidence level, precision, prevalence and design effect
• Probability weights correct unequal selection probabilities and sampled population structure
• Nonresponse bias diagnosed through response rates, paradata and follow-up patterns
• Margin of error explained without treating estimates as exact counts

Participants can assess data-collection methods and administrative records for policy use.

• Question wording avoids double-barrelled, leading, loaded and ambiguous items
• Cognitive testing identifies respondent interpretation, recall and response-category problems
• Questionnaire flow uses screening, routing, recall periods and sensitive-item protocols
• Administrative data fitness assessed against policy definitions, coverage and timeliness
• Record linkage methods match entities using deterministic and probabilistic rules
• Fieldwork monitoring tracks completion, refusal, break-off and interviewer effects

Participants can create clear visual evidence that supports accurate policy interpretation.

• Chart selection matches comparison, composition, distribution, relationship and change questions
• Axis scales, labels and annotations prevent misleading visual interpretations
• Maps use appropriate rates, classification schemes and geographic denominators
• Accessible graphics apply colour contrast, direct labels and alternative text
• Dashboard indicators distinguish operational monitoring from strategic outcome measurement
• Policy charts communicate uncertainty using intervals, bands and scenario ranges

Participants can explain statistical uncertainty and distinguish it from policy significance.

• Sampling distributions explain why repeated samples produce different estimates
• Confidence intervals express plausible population values under stated sampling assumptions
• Standard errors reflect design, weighting, clustering and estimation method
• p-values interpreted as evidence against specified null hypotheses, not policy importance
• Statistical significance separated from practical significance and decision thresholds
• Multiple testing adjustments control false discoveries across indicator portfolios

Participants can select, interpret and report common statistical tests and effect measures.

• Null and alternative hypotheses specified before inspecting outcome data
• t-tests compare group means with assumptions checked and reported
• Chi-square tests assess association in categorical policy outcome tables
• Nonparametric tests address ordinal outcomes and severely skewed distributions
• Effect sizes report magnitude using mean differences, risks and odds
• Power analysis links detectable effects, sample size and decision risk

Participants can interpret linear regression models and assess their suitability for policy questions.

• Linear regression estimates adjusted associations between continuous outcomes and predictors
• Model specifications distinguish confounders, mediators, colliders and post-treatment variables
• Regression coefficients interpreted in original units with uncertainty intervals
• Residual diagnostics check linearity, heteroskedasticity, influence and functional form
• Interaction terms test whether policy associations differ across population groups
• Model tables report estimates, sample definitions, controls and limitations

Participants can interpret model outputs for common policy and service-delivery outcomes.

• Logistic regression estimates probabilities for binary programme and service outcomes
• Odds ratios translated cautiously into predicted probabilities and risk differences
• Poisson and negative binomial models handle counts and exposure periods
• Ordinal models analyse ranked satisfaction, severity and service-quality responses
• Classification performance assessed with sensitivity, specificity, calibration and ROC curves
• Predicted values communicate model implications for defined policy profiles

Participants can distinguish causal claims from association and identify core validity threats.

• Causal questions distinguish association, prediction and attributable policy effects
• Directed acyclic graphs identify confounding paths and adjustment sets
• Potential outcomes frame treatment effects, counterfactuals and selection bias
• Threats to internal validity include selection, history, maturation and contamination
• Target trial protocol specifies eligibility, treatment, follow-up and estimand
• Evidence claims calibrated to design strength, assumptions and uncertainty

Participants can assess the design, analysis and reporting of randomised policy evaluations.

• Randomised controlled trial designs allocate interventions using transparent randomisation procedures
• Cluster randomisation addresses contamination within schools, clinics and municipalities
• Intention-to-treat estimates preserve randomised group comparisons after noncompliance
• Attrition analysis examines differential loss and consequences for estimates
• Pre-analysis plans reduce specification searching and selective outcome reporting
• CONSORT reporting items document trial flow, outcomes, harms and deviations

Participants can select and challenge quasi-experimental designs for non-randomised interventions.

• Difference-in-differences compares policy changes against credible comparison trends
• Parallel-trends diagnostics use pre-period plots, event studies and placebo tests
• Regression discontinuity exploits eligibility thresholds with bandwidth and manipulation checks
• Instrumental variables address endogenous participation using relevance and exclusion assumptions
• Propensity-score methods balance observed covariates without resolving hidden confounding
• Synthetic control constructs weighted comparators for single-unit policy interventions

Participants can develop and assess forecasts for planning, demand management and resource allocation.

• Time-series plots separate trend, seasonality, cycles, shocks and data breaks
• Seasonal adjustment supports fair monthly and quarterly policy comparisons
• Moving averages and exponential smoothing generate short-term operational forecasts
• ARIMA models specify autoregressive, differencing and moving-average components
• Forecast accuracy assessed using MAE, RMSE, bias and prediction intervals
• Scenario forecasts state assumptions, trigger points and contingency responses

Participants can compare policy options using value, affordability and equity evidence.

• Cost-benefit analysis identifies costs, benefits, discount rates and time horizons
• Cost-effectiveness ratios compare interventions using defined outcome units
• Budget impact analysis estimates affordability under uptake and price scenarios
• Microsimulation models project eligibility, caseloads and fiscal effects
• Distributional analysis disaggregates gains, losses and service access by subgroup
• Equity weights and ethical criteria make trade-offs explicit

Participants can use performance statistics to identify variation and direct management action.

• Indicator dashboards connect strategic objectives, measures, targets, tolerances and actions
• Control charts distinguish common-cause variation from special-cause performance signals
• Data review routines assign escalation, investigation and corrective-action responsibilities
• Outcome indicators paired with process, output and equity measures
• Performance targets adjusted for baseline conditions, capacity and measurement error
• Metadata notes prevent false comparisons after definition, boundary or system changes

Participants can present statistical evidence in a form suited to senior policy decisions.

• Evidence briefs state the decision, findings, uncertainty, options and recommendation
• Executive summaries lead with implications rather than methods or caveats
• Plain-language statistics retain denominators, time periods, comparators and uncertainty
• Risk communication explains rare events, absolute risks and expected ranges
• Technical annexes document sources, specifications, assumptions and reproducibility details
• Stakeholder challenge sessions test claims, framing and foreseeable misinterpretations

Participants can establish safeguards for ethical, reproducible and publishable policy analysis.

• Statistical disclosure review protects confidentiality before public indicator release
• Algorithmic bias audits test differential errors across protected and policy-relevant groups
• Data ethics considers consent, purpose limitation, proportionality and public benefit
• Independent replication checks code, data lineage, specifications and reported results
• Peer review distinguishes analytical error, interpretation risk and policy feasibility
• Publication protocols separate preliminary estimates from revised official statistics

Participants can build, peer-review and present a complete policy evidence brief with a defensible recommendation.

• Policy evidence brief integrates question, data, methods, findings and recommendation
• Analysis plan documents estimand, variables, comparison group and assumptions
• Descriptive tables and charts establish baseline patterns and subgroup differences
• Appropriate model or evaluation design supports the stated policy claim
• Uncertainty, limitations and equity implications qualify proposed policy action
• Peer review tests reproducibility, decision relevance and communication quality
• Oral presentation defends findings against technical and senior-management challenge