Research Guide

Survey Data Analysis and Reproducible Export

Analysis-ready data starts while the questionnaire is being authored. Stable variable names, explicit recodes, recorded randomization, documented exclusions, and a versioned codebook prevent the cleanup file from becoming the study's undocumented source of truth.

A practical workflow

1
Name variables before launch
Use stable, meaningful names and explicit recodes that can survive wording changes and translations.
2
Record design metadata
Export treatment assignment, display order, conjoint profiles, endings, timing, and panel identifiers needed to reproduce the sample flow.
3
Review quality without overwriting
Keep raw responses intact and record exclusions, reasons, and reviewer decisions in a cleaning log.
4
Generate code and documentation
Export a codebook and starter scripts alongside native or tabular data rather than reverse-engineering labels later.
5
Archive one replication package
Bundle the instrument, flow diagram, data, codebook, analysis code, and provenance information under the same version.

Key design decisions

Decision	Guidance
Labels or coded values?	Use coded values for analysis stability and keep labels in the codebook; never rely on column position to preserve meaning.
Wide or long data?	Wide files are convenient for one-row-per-respondent models; repeated tasks often need a long table keyed by respondent and task.
Delete or exclude?	Prefer reversible exclusions with recorded reasons. Delete only when retention is not lawful or the record should not exist.