Methodology

Conjoint Survey is built around paired-profile choice tasks. Researchers define attributes, levels, optional impossible-profile restrictions, an optional neither choice, and a respondent-facing prompt.

The app randomizes profiles at the respondent level and records the profiles shown, the choices made, start and completion timestamps, URL parameters, duplicate flags, and optional attention/comprehension-check outcomes.

In-app ATE results report differences in observed choice probability for non-reference levels relative to the researcher-designated reference level. These summaries are intended for inspection, diagnostics, and early fielding checks.

For publication work, researchers should use the full-model AMCE/LPM table and export the profile-level data with the generated R or Stata replication script. The scripts include all non-reference randomized attribute levels jointly and use respondent-clustered standard errors.

If a design uses impossible-profile restrictions, estimates should be interpreted as conditional on the allowed profile space. The app flags exposure imbalance and attribute association when restrictions make a raw difference-in-means summary less appropriate as a primary estimate.

For open-ended survey outcomes, the app can export word counts and VADER compound sentiment scores as respondent-level text variables. In randomized survey experiments, those text variables can be compared by treatment arm as experimental quantitative text analysis. Read the text-analysis method page.

The app also records randomization seed metadata and exports preregistration text, an IRB description template, a codebook, design diagnostics, randomization audit tables, methods text, and reviewer appendix material.