Survey Analysis Studio (complex samples)

The Survey Analysis Studio analyzes complex-sample survey microdata the statistically correct way — applying sampling weights and computing design-adjusted standard errors that account for stratification and clustering. It is the right tool for national and large-scale surveys (DHS, PISA, labour-force, household surveys), where ignoring the design under- or over-states precision.

Define your survey design

1. 1
   Load data

   Choose a project + dataset, or click "Try a sample dataset" for a ready-made survey (national, clustered health, or stratified education).
2. 2
   Set the design

   The sampling weight is required. Add strata and a PSU/cluster column for correct design-based SEs; optionally add replicate weights (BRR, Fay, Jackknife, Bootstrap) or a finite-population-correction (FPC) column. Sample datasets auto-detect these from column names.
3. 3
   Create the design

   A design summary reports observations, strata, PSUs, sum of weights, effective n, weight CV, mean DEFF, and degrees of freedom.
4. 4
   Analyze, interpret, export

   Run any method, generate a privacy-safe AI interpretation, and export tables to CSV or charts to PNG.

Methods

• Weighted descriptives — means, totals, proportions, and quantiles with design-based SEs, DEFF, and effective n (with optional by-group breakdowns)
• Crosstab — Rao-Scott design-adjusted chi-square test of association
• Survey t-test — weighted mean vs a value, or between two groups
• Survey ANOVA — design-adjusted comparison of a mean across 3+ groups (Wald F)
• Domain analysis — correct subpopulation estimates (the full design drives the SEs)
• Survey regression — linear / logistic / Poisson with weighted coefficients and design-robust SEs
• Weighted correlation — survey-weighted Pearson r with a Fisher-z CI on the effective n
• Ratio estimation — R = ΣWy/ΣWx with a Taylor-linearized SE
• Weight calibration — raking (post-stratification) to known population totals; saves a calibrated-weight column

Built-in sample datasets

• National Survey (3,000) — stratified + clustered with 8 jackknife replicate weights
• Clustered Health Survey (~1,500) — DHS-style: region strata, cluster PSUs, household weight
• Stratified Education Survey (~1,300) — PISA-style: school strata, school PSUs, student weight

AI interpretation & publication-ready output

Every results tab can generate a plain-language AI interpretation. It is leak-safe: only the rounded, aggregated results are sent to the model — never raw respondent rows. Tables export to CSV and charts (95% CI error bars) export to high-resolution PNG for direct use in papers and reports.