Stata Econometrics and Research Training Course

Participants can import common file formats and assess data structure before analysis.

• Import Excel, CSV and delimited files using import and insheet commands
• Codebook and describe commands used to inspect variable properties
• Data types, storage formats and display formats distinguished before transformation
• Variable labels and value labels created for interpretable datasets
• Tabulations identify coding anomalies, out-of-range values and sparse categories
• Data documentation records source, extraction date, units and known limitations

Participants can clean, recode and transform raw data into analysis-ready variables.

• Missing-value codes identified, recoded and distinguished from valid zero values
• Duplicate records detected using isid, duplicates report and merge diagnostics
• String variables standardised with trim, lower, substr and regular expressions
• Generate and replace commands create transparent derived analytical variables
• Encode, destring and recode convert categories into usable numeric formats
• Winsorisation, trimming and outlier flags documented without obscuring source values

Participants can combine related files and reshape records for cross-sectional and longitudinal analysis.

• Merge operations join master and using files through validated identifiers
• Append combines structurally aligned datasets while preserving source indicators
• Collapse produces grouped statistics for reporting and aggregate-level modelling
• Reshape wide and reshape long reorganise repeated measures correctly
• Contract and expand commands manage frequency data and replicated records
• Merge result codes reviewed before retaining matched analytical observations

Participants can summarise distributions and create graphics that reveal analytical patterns.

• Summarize, detail and tabstat report distributional properties and subgroup comparisons
• Cross-tabulations use row percentages, column percentages and chi-square statistics
• Histograms, density plots and box plots assess distribution shape
• Scatterplots with fitted lines inspect bivariate relationships and influential observations
• Twoway graphics combine confidence intervals, markers and reference lines
• Graph schemes, titles and notes produce report-ready visual outputs

Participants can apply sampling-based inference and explain uncertainty in estimates.

• Sampling distributions connect sample estimates with population inference
• Standard errors, confidence intervals and p-values interpreted together
• One-sample and two-sample tests compare means, proportions and variances
• Paired tests assess within-unit changes before and after interventions
• Nonparametric tests address ordinal outcomes and non-normal distributions
• Statistical significance distinguished from practical and economic significance

Participants can estimate and interpret ordinary least squares models for continuous outcomes.

• Linear regression specifies conditional mean relationships between outcomes and predictors
• Regress command estimates coefficients, standard errors and model fit
• Coefficient interpretation distinguishes unit changes, percentage changes and predicted differences
• Adjusted R-squared and root mean squared error assess model fit
• Control variables selected using theory, causal structure and measurement quality
• Regression tables report estimates, uncertainty, sample size and model specifications

Participants can specify nonlinear relationships and interpret conditional effects from regression models.

• Logarithmic transformations model elasticities, growth rates and skewed monetary variables
• Quadratic terms capture curved relationships and calculate turning points
• Factor-variable notation specifies categorical predictors and interaction effects correctly
• Marginal effects translate interactions into conditional substantive interpretations
• Margins and marginsplot commands display adjusted predictions with confidence intervals
• Functional-form choices assessed through theory, residual patterns and interpretability

Participants can diagnose common ordinary least squares problems and select appropriate remedies.

• Residual-versus-fitted plots identify nonlinearity, heteroskedasticity and unusual observations
• Variance inflation factors assess multicollinearity among explanatory variables
• Breusch-Pagan and White tests examine heteroskedastic error variance
• Robust standard errors address heteroskedasticity without changing coefficient estimates
• Influence statistics identify high-leverage observations and influential cases
• Ramsey RESET tests indicate possible omitted nonlinear terms or misspecification

Participants can model binary and ordered outcomes using appropriate nonlinear estimators.

• Logit and probit models estimate binary outcome probabilities
• Odds ratios distinguished from probability changes and marginal effects
• Margins calculates average marginal effects after nonlinear estimation
• Classification tables, ROC curves and AUC assess predictive discrimination
• Ordered logit models analyse ranked satisfaction, risk or credit categories
• Linear probability models compared critically with logit and probit alternatives

Participants can model count outcomes and choose estimators for dispersion patterns.

• Poisson regression models counts of events occurring within defined exposure periods
• Exposure and offset terms convert counts into comparable event rates
• Incidence-rate ratios communicate proportional changes in expected counts
• Overdispersion assessed using residual patterns and negative binomial comparisons
• Zero-inflated models considered when structural zeros have substantive justification
• Predicted counts and marginal effects reported for meaningful covariate values

Participants can analyse survey data while accounting for weights, strata and clustered sampling.

• Survey weights correct unequal selection probabilities and nonresponse adjustments
• Sv yset declares primary sampling units, strata and finite population corrections
• Design-based standard errors reflect clustering and stratification in inference
• Weighted means, proportions and regression estimates use survey prefixes
• Subpopulation analysis uses subpop rather than dropping observations prematurely
• Survey documentation records target population, fieldwork design and weighting method

Participants can assess missingness mechanisms and apply defensible missing-data treatments.

• Missingness patterns examined with misstable summarize and graphical diagnostics
• MCAR, MAR and MNAR assumptions distinguished for analytical decisions
• Complete-case analysis evaluated for bias, precision loss and transparency
• Multiple imputation by chained equations generates plausible missing values
• Imputation models include outcomes, predictors and auxiliary variables appropriately
• Mi estimate pools coefficients and standard errors across imputed datasets

Participants can structure longitudinal data and estimate basic panel-data models.

• Xtset declares panel identifiers and time variables for longitudinal estimation
• Balanced and unbalanced panels distinguished through observation-pattern diagnostics
• Within-unit and between-unit variation interpreted before model selection
• Pooled OLS compared with fixed-effects and random-effects estimators
• Fixed effects control for time-invariant unobserved unit characteristics
• Clustered standard errors account for within-panel error correlation

Participants can select panel estimators and test key assumptions in longitudinal studies.

• Hausman tests compare fixed-effects and random-effects consistency assumptions
• Time fixed effects control common shocks and period-specific conditions
• First-difference models estimate change relationships across adjacent periods
• Dynamic panels introduce lagged outcomes and serial-correlation concerns
• Serial correlation tests assess error dependence within panel units
• Panel results reported with entity definitions, periods and clustering choices

Participants can identify endogeneity risks and estimate instrumental-variable models when justified.

• Causal diagrams clarify confounding, mediators, colliders and treatment assignment
• Endogeneity arises through omitted variables, simultaneity and measurement error
• Instrument relevance and exclusion restrictions assessed using substantive evidence
• Two-stage least squares estimates causal parameters with valid instruments
• First-stage F-statistics assess weak-instrument risks before interpretation
• Overidentification tests and sensitivity discussion qualify instrumental-variable conclusions

Participants can estimate and communicate difference-in-differences designs for programme or policy evaluation.

• Difference-in-differences compares changes between treated and comparison groups
• Parallel-trends assumptions assessed using institutional evidence and pre-treatment outcomes
• Treatment-by-post interactions estimate average programme or policy effects
• Event-study specifications display dynamic effects before and after treatment
• Clustered inference matches the level of treatment assignment
• Staggered adoption designs require careful estimator and comparison-group selection

Participants can prepare time-series data, test stationarity and develop basic forecasting models.

• Tsset declares time indexes and identifies gaps in observations
• Line plots, lags and seasonal patterns inform initial series diagnosis
• Augmented Dickey-Fuller tests assess unit roots and stationarity
• Autocorrelation and partial autocorrelation guide ARIMA model identification
• ARIMA models estimate autoregressive, moving-average and differencing components
• Out-of-sample forecasts evaluated using forecast errors and benchmark models

Participants can produce clear econometric outputs and document methodological choices for review.

• Estimates store preserves model results for structured comparison tables
• Esttab and putdocx create reproducible tables and Word outputs
• Model narratives state estimands, samples, assumptions and uncertainty clearly
• Coefficient plots compare specifications without obscuring confidence intervals
• Research limitations distinguish data constraints from identification limitations
• Replication packages include code, data provenance and output documentation

Participants can build, review and present a complete reproducible econometric analysis addressing a defined research question.

• Research question translated into outcomes, predictors, estimand and decision context
• Analysis dataset cleaned, labelled and documented using reproducible Stata code
• Descriptive evidence and graphics establish sample characteristics and key patterns
• Econometric model selected with stated assumptions and identification rationale
• Diagnostic tests, sensitivity checks and limitations recorded in review notes
• Findings presented through tables, graphics and decision-focused research conclusions