cohort studio
Cohort analysis

Loading batch…

Download sample CSV Raw data (CSV)
Ingest report — —

Analysis parameters

— ↓ Excel workbook ↓ PowerPoint deck

Off — the Greenwald (2003) default.

Trials slower than this are dropped. Trial-level only.

Participant-level: excludes the whole respondent.

Participant-level, on the combined blocks.

Statistical method

Distribution-free tests on D.

Sample funnel

—

Each respondent is counted at the first rule they fail. Participant-level screens run on the untrimmed trials, so moving the trim slider cannot hide a fast responder.

Executive takeaways

permutation

    Written by rules from the computed statistics — no model in the loop for client deliverables. The deck and workbook carry these exact sentences.

    Association strength by attribute and segment

    —

    Cell = mean D across respondents in that group. ● significant at q < .05 (Benjamini–Hochberg, selected method) · ○ not significant · dashed = low base (n < 30), indicative only. Hover a cell for the interval and both methods' p- and q-values.

    Effect sizes with 95% bootstrap intervals

    —

    Mean D per group with a percentile-bootstrap interval over participants (2,000 resamples). Filled point = significant under the selected method; hollow = not. An interval crossing zero is not a finding, whatever its point estimate.

    Reaction-time distributions

    Respondent-level spread of D

    One D per respondent. A mean can hide a split audience; the shape of these does not. Blue bins lean to the first brand, orange to the second.

    Segment comparisons

    —

    Audit — participant-level screening open to load

    Method