Analytics & Statistics

Every submitted response is streamed into an analytics pipeline that recomputes per-question statistics in near real time. Numeric and Likert questions surface mean, median, standard deviation, and a 95% confidence interval; NPS questions render the classic split into promoters (9-10), passives (7-8), and detractors (0-6) with the score and trend over time. Categorical relationships are explored with Chi-square tests and Cramer's V to quantify association strength, while numeric pairs use Pearson and Spearman correlations rendered as an interactive matrix with significance shading.

Segment breakdowns let analysts pivot any question by age band, gender, region, license type, or discipline, and the dashboard plots side-by-side distributions with a small-cell suppression rule (k-anonymity, n<5) that automatically masks segments too small to publish without re-identification risk. Open-ended text answers feed a text-analysis module that produces TF-IDF term scores, k-means clusters of similar verbatims, and word-frequency clouds so qualitative themes surface alongside quantitative results. Ranking questions are aggregated with the Borda count to produce a robust collective preference order independent of vote splitting.

For ad-hoc work a query builder exposes filters, breakdowns, and crosstab axes in a no-code form; queries can be saved, shared, or pinned to the dashboard. Exports are available as CSV, XLSX, and JSON, and they re-apply the survey's privacy mode at export time — anonymous datasets strip user_id columns, pseudonymous exports replace it with the hash, and attributed exports require explicit role permission and are logged for audit. All charts include sample-size badges and confidence-interval whiskers so consumers can judge reliability at a glance.