Complex Survey Data Analysis with Stata Training Course

Duration 10 days 16 modules · 65 contact hours
Level Intermediate Certificate on completion
Who attends Monitoring and Evaluation Specialists, Quantitative Researchers and others
Fee from USD 2,200 KSh 190,000 for Kenyan organisations
Where 52 cities + online See dates by city
Next start 16 Nov 2026 Course code SD-ME-112

Before you join: Participants should be comfortable working with rectangular datasets and conducting basic descriptive analysis in Stata or another statistical package.

Dates & prices

Two start dates a month in every city, and live online on weekdays or weekends. Fees are per participant; classroom fees include tuition, materials, lunch and refreshments, and the certificate.

Suggest a date or city

Showing the next six months. Later dates, other cities and private deliveries for your team are available on request.

Course overview

Household, beneficiary, health, education and livelihoods surveys commonly use stratification, clustering, unequal selection probabilities and multi-stage sampling. Treating these data as simple random samples can produce biased population estimates, incorrect standard errors and misleading significance tests. For monitoring and evaluation teams, this can affect beneficiary coverage figures, poverty estimates, programme indicators, equity analyses and decisions about where resources should be directed.

This course teaches participants to analyse complex sample survey data correctly in Stata. Participants prepare survey datasets, inspect sample designs, calculate and evaluate weights, declare designs with svyset, estimate weighted descriptive statistics and produce valid confidence intervals. They use Stata's survey estimation commands for cross-tabulations, means, proportions, ratios, linear regression, logistic regression and population subgroups. The course also addresses design effects, finite population corrections, post-stratification, non-response adjustment, replicate weights, missing data and clear reporting of weighted results.

Teaching combines focused technical briefings, guided Stata demonstrations, individual coding work, group interpretation tasks and realistic cases based on development and humanitarian surveys. Participants work with survey microdata from the first week and progressively build an analysis file, survey design declaration, indicator tables and regression outputs. They leave with a reusable Stata analysis workflow, annotated do-file structure, quality-control checklist and a completed survey analysis report suitable for adaptation to their own monitoring and evaluation work.

The course is designed for M&E specialists, researchers, statisticians, data analysts, programme staff and evaluation consultants who already work with quantitative survey data. It is particularly relevant for teams using DHS-style, LSMS-style, MICS-style, KAP, post-distribution monitoring, needs assessment, beneficiary feedback or outcome-monitoring surveys.

What you will be able to do

  • Declare complex survey designs in Stata using svyset, weights, strata and primary sampling units
  • Calculate and assess design weights, non-response adjustments and calibrated survey weights
  • Estimate weighted means, proportions, totals, ratios and confidence intervals using svy commands
  • Interpret design effects, standard errors, confidence intervals and survey-adjusted hypothesis tests
  • Produce valid domain and subgroup estimates without incorrectly subsetting survey data
  • Fit and interpret survey-weighted linear and logistic regression models in Stata
  • Document a reproducible survey analysis workflow using annotated Stata do-files and output tables
  • Present a complete complex-survey analysis report with defensible methods and findings

Benefits of attending

For you

  • Produce survey estimates that account correctly for weights, clustering and stratification
  • Identify when simple random-sample assumptions invalidate reported standard errors
  • Build reusable Stata do-files for beneficiary, household and assessment surveys
  • Interpret survey-weighted regression findings for programme and policy decisions
  • Review external survey analysis outputs with greater technical confidence

For your organisation

  • Improve the statistical validity of reported programme indicators and survey findings
  • Reduce the risk of incorrect conclusions from clustered or weighted datasets
  • Standardise survey analysis procedures across M&E, research and assessment teams
  • Strengthen quality assurance for donor reports, evaluations and learning products
  • Create reproducible analysis records that support audit, review and handover

Course modules

16 modules over 10 days

Participants can distinguish complex probability samples from simple random samples and identify their analytical implications.

  • Probability sampling concepts: target populations, frames, elements and sampling units
  • Simple random sampling assumptions compared with clustered survey samples
  • Stratification, clustering and unequal selection probabilities in field surveys
  • Multi-stage household sampling from enumeration areas to respondents
  • Design-based inference versus model-based inference for survey estimates
  • Common M&E survey designs: baseline, endline, KAP and post-distribution monitoring

Participants can organise survey microdata and establish a reproducible Stata analysis workspace.

  • Stata datasets, variable types, value labels and missing-value codes
  • Project folder structure for raw data, scripts, outputs and documentation
  • Annotated do-files for reproducible survey data preparation and analysis
  • Importing Excel, CSV and SPSS survey files into Stata
  • Using codebook, describe and tabulate to inspect incoming datasets
  • Creating analysis variables from questionnaires, skip patterns and response codes

Participants can reconstruct selection stages and identify the information required for valid weighting.

  • Sampling stages recorded through strata, clusters, households and respondents
  • Selection probabilities calculated at each stage of multi-stage designs
  • Primary sampling units and secondary sampling units distinguished clearly
  • Systematic sampling intervals and implicit stratification in survey designs
  • Oversampling vulnerable groups to improve subgroup estimate precision
  • Sampling documentation reviewed for probabilities, replacements and frame changes

Participants can calculate and evaluate base weights and non-response-adjusted survey weights.

  • Base weights calculated as inverse probabilities of selection
  • Household and individual weights distinguished within household survey files
  • Non-response adjustment cells defined using response propensity characteristics
  • Weight trimming applied to reduce extreme-weight influence on estimates
  • Distributional diagnostics using summarize, histogram and percentile comparisons
  • Weighted totals checked against population counts and sample documentation

Participants can explain and apply calibration concepts when external population controls are available.

  • Post-stratification aligns weighted sample totals with known population margins
  • Raking adjusts weights across age, sex and geographic control totals
  • Calibration variables selected from reliable census or registry sources
  • Weight adjustment factors documented for transparent analytical reporting
  • Comparing unweighted, base-weighted and calibrated indicator estimates
  • Assessing precision trade-offs introduced by calibration and weight variation

Participants can declare single-stage and multi-stage survey designs correctly using Stata's svyset command.

  • svyset syntax specifies primary sampling units, strata and probability weights
  • Multi-stage designs declared with nested sampling-unit identifiers
  • Finite population corrections entered for sampled fractions where available
  • Single-unit strata options selected for certainty or collapsed strata
  • Survey declarations verified using svyset output and design documentation
  • Common svyset errors diagnosed from identifier and weight inconsistencies

Participants can estimate and interpret weighted population totals, means and proportions in Stata.

  • svy: mean estimates weighted averages with design-adjusted standard errors
  • svy: proportion calculates population proportions and confidence intervals
  • svy: total estimates beneficiary counts and population-level service totals
  • Ratio estimators constructed for coverage, dependency and service-use indicators
  • Confidence intervals interpreted alongside point estimates and sample sizes
  • Unweighted counts reported with weighted estimates for transparent interpretation

Participants can produce survey-adjusted cross-tabulations and test associations between categorical variables.

  • svy: tabulate produces weighted two-way tables and design-adjusted tests
  • Row percentages, column percentages and population totals selected appropriately
  • Rao-Scott adjusted tests replace ordinary Pearson chi-square assumptions
  • Binary outcome comparisons estimated across sex, location and disability groups
  • Confidence intervals compared rather than relying solely on p-values
  • Table layouts prepared for donor reports and evaluation findings sections

Participants can explain how complex designs affect variance estimation and assess estimate precision.

  • Taylor series linearization estimates variances for complex survey statistics
  • Design effects compare actual variance with simple random-sample variance
  • Intraclass correlation explains precision loss from clustered sampling
  • Stratification can improve precision when strata are internally homogeneous
  • Relative standard errors flag unstable estimates for reporting review
  • Effective sample size calculated from design effects and nominal samples

Participants can produce valid estimates for population domains without compromising variance estimation.

  • Domain estimation retains full sample structure during subgroup analysis
  • subpop() specifies analysis populations without dropping non-domain observations
  • Incorrect dataset subsetting understates variance for domain estimates
  • Disaggregation by sex, age, wealth and displacement status
  • Small-domain estimates assessed using sample size and relative standard errors
  • Equity indicator tables distinguish statistical uncertainty from substantive differences

Participants can fit, interpret and report survey-weighted linear regression models for continuous outcomes.

  • svy: regress fits linear models with design-adjusted standard errors
  • Continuous outcomes modelled for expenditure, dietary score and travel time
  • Categorical predictors entered using Stata factor-variable notation
  • Adjusted mean differences interpreted using coefficients and confidence intervals
  • Interaction terms test whether programme associations differ between groups
  • Regression diagnostics reviewed alongside survey design and measurement assumptions

Participants can estimate and communicate survey-weighted logistic regression results for binary outcomes.

  • svy: logistic estimates adjusted odds ratios for binary outcomes
  • Binary indicators modelled for service access, adoption and food insecurity
  • Odds ratios converted into predicted probabilities for decision audiences
  • margins estimates adjusted probabilities after survey-weighted regression models
  • marginsplot displays subgroup contrasts with confidence interval bands
  • Model findings reported with reference groups, uncertainty and limitations

Participants can identify when replicate-weight methods are required and configure available survey designs.

  • Replicate weights support variance estimation when design variables are restricted
  • Jackknife replication removes sampled units across successive replicate estimates
  • Balanced repeated replication uses paired half-samples and Fay adjustments
  • Bootstrap replicate weights resample clusters under specified survey designs
  • Stata svyset settings configured for jackknife, BRR and bootstrap methods
  • Published survey documentation checked before selecting variance estimation methods

Participants can assess missingness and apply defensible treatment decisions in survey analysis.

  • Item non-response distinguished from unit non-response and structural skips
  • Missingness patterns examined by strata, clusters and respondent characteristics
  • Complete-case analysis assessed against missing completely at random assumptions
  • Multiple imputation concepts considered for covariates and outcome variables
  • Survey weights and imputation decisions documented in analysis protocols
  • Data-quality checks identify duplicates, impossible values and denominator errors

Participants can communicate weighted survey findings with methods statements, tables and interpretable graphics.

  • Methods statements report sampling design, weights and variance estimation
  • Indicator tables include weighted estimates, confidence intervals and unweighted denominators
  • Bar charts display subgroup proportions with confidence interval whiskers
  • Survey findings distinguish descriptive patterns from causal programme effects
  • Technical annexes document svyset declarations, exclusions and analytic decisions
  • Peer review checklists test consistency between tables, text and conclusions

Participants can build, review and present a complete Stata analysis of a complex sample survey.

  • Survey documentation translated into a defensible Stata svyset declaration
  • Analysis dataset prepared with documented recodes, exclusions and derived indicators
  • Weighted descriptive tables produced for priority monitoring and evaluation indicators
  • Domain estimates generated for selected equity and vulnerability groups
  • Survey-weighted regression model developed to examine a programme outcome
  • Quality review checks weights, variance methods, outputs and interpretation
  • Completed findings presented with methods, limitations and operational recommendations

Exercises and examples are adapted to each cohort's sectors and roles.

What you take back to work

  1. Complex survey analysis do-file template
  2. Survey design documentation checklist
  3. Weight construction and review worksheet
  4. svyset declaration reference guide
  5. Weighted indicator table template
  6. Complex survey analysis report

Who should attend

  • Monitoring and Evaluation Specialists — need defensible survey indicators for programme reporting
  • Quantitative Researchers — require valid inference from clustered and stratified survey samples
  • Data Analysts — need to apply Stata survey commands to operational datasets
  • Evaluation Managers — need to assess the credibility of survey-based findings
  • Humanitarian Assessment Officers — analyse needs, coverage and post-distribution monitoring surveys
  • National Statistics and Ministry Staff — produce weighted estimates for official reporting

Skills and tools covered

Survey design declarationSampling weight constructionDesign-based inferenceSurvey regression modellingDomain estimationStata workflow documentation

Tools and standards

Stata svyset, Taylor series linearization, Rao-Scott adjusted test, Multiple imputation

How the course runs

Training methodology

The course uses short technical briefings followed by instructor-led Stata demonstrations and structured analysis work. Participants inspect sample designs, prepare weighted datasets, write commands, interpret output and compare correct and incorrect analytical approaches. Cases reflect household, beneficiary, health, education and humanitarian assessment surveys. Group discussions focus on methodological choices, reporting implications and common review findings. Participants progressively assemble an analysis workflow during the course, then apply it to a complete case in the final capstone.

Requirements and prerequisites

Participants should be comfortable working with rectangular datasets and conducting basic descriptive analysis in Stata or another statistical package. They should understand variables, observations, percentages, means, cross-tabulations and basic regression concepts. Prior experience with household surveys, monitoring data or evaluation datasets is helpful. Participants should be able to read Stata output and write or adapt simple commands. Advanced statistical theory, sampling-frame construction, programming expertise and prior experience with replicate-weight methods are not required. The course introduces the survey-sampling concepts needed for analysis.

A typical training day

08:30 – 10:30First session
10:30 – 10:45Refreshment break
10:45 – 12:30Second session
12:30 – 13:30Lunch and networking
13:30 – 15:00Third session
15:00 – 15:15Refreshment break
15:15 – 16:30Workshop and daily review

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

What the fee includes

  • Instruction by a practitioner facilitator
  • Full course workbook and materials
  • Exercise files, templates and case studies
  • Refreshments and lunch on classroom deliveries
  • Certificate of completion
  • Post-course application plan

Frequently asked questions

Participants should be able to open datasets, inspect variables and run basic Stata commands. The course teaches the survey-specific commands, weighting methods and variance concepts required for complex sample analysis.

Participants should bring a laptop with Stata installed if attending classroom delivery, subject to local course arrangements. Bringing a de-identified survey dataset, questionnaire and sampling documentation is useful for application planning.

It suits M&E professionals, quantitative researchers, data analysts, evaluators and assessment teams analysing probability-sample surveys. It is most useful for participants working with clustered, stratified or weighted survey data.

General Stata courses often focus on standard statistical procedures that assume simple random samples. This course focuses on survey design declarations, weights, design-based variance estimation and survey-adjusted modelling.

Participants can apply the workflow to baseline, endline, household, beneficiary, KAP, needs assessment and post-distribution monitoring datasets. They will be able to check whether existing reports use correct weights and standard errors.

The first week establishes survey sampling concepts, data preparation, weighting, svyset and descriptive estimation. The second week covers regression, replicate methods, missing data, reporting and a final integrated analysis capstone.

Questions about this course?

Ask about dates, fees, group bookings or running it for your team. Our training team replies by email.

Ask on WhatsApp info@skillsetdevelopment.com

More courses in Monitoring & Evaluation

Monitoring & Evaluation Case Study Design and Analysis Training Course 5 days · from USD 1,100 · next 16 Nov Monitoring & Evaluation Change Management Monitoring and Evaluation Training Course 10 days · from USD 2,200 · next 16 Nov Monitoring & Evaluation CommCare for Development Programmes Training Course 10 days · from USD 2,200 · next 16 Nov Monitoring & Evaluation CommCare for Field Data Management Training Course 10 days · from USD 2,200 · next 16 Nov Monitoring & Evaluation Data Collection Methods Training Course 10 days · from USD 2,200 · next 16 Nov Monitoring & Evaluation Digital Technologies for Monitoring and Evaluation Training Course 5 days · from USD 1,100 · next 16 Nov Monitoring & Evaluation Educational Programme Impact Evaluation Training Course 10 days · from USD 2,200 · next 16 Nov Monitoring & Evaluation Evaluation Management Training Course 10 days · from USD 2,200 · next 16 Nov