Learn from finished studies

Discourse frames — how the world press writes about AI

Derived from world_press data · 12 press excerpts on AI

The same technology, framed differently by region — Entman’s four functions coded, then crossed.

Phenomenology — what it is to live recovery

Synthetic interviews with 5 people · 35 utterances about recovery

From the bracketing memo to the description of essential structure — a phenomenological study carried through.

Narrative — a life story of recovery

One synthetic life-history interview · 40 transcribed turns (about 90 minutes)

A timeline rebuilt and two turning points — where one person’s story bends and turns back.

Grounded theory — the conditions of civic participation

Synthetic interviews with 8 people · 48 utterances

From data to theory — open coding through axial and selective coding, ending at a saturation check.

CQR — thesis stress among graduate students

Synthetic interviews, 6 cases · double-coded by two coders · auditor check included

How two coders working independently pass through kappa, a consensus record and an auditor check to reach a general/typical/variant frequency table.

Sociometry — a team retrospective, from text to a map of relations

Synthetic team retrospective interviews · 7 members · 16 confirmed relation tuples

The bridge from sentences in a retrospective to relation tuples, a 7×7 matrix, and four structures: star, bridge, clique and isolate.

Mixed methods — civic participation, from grounded theory to a survey

Exploratory sequential design · 240 synthetic respondents (categories → items → α → EFA → regression)

Qualitative categories become survey items, and 240 responses carry through alpha, EFA and regression.

A small survey, start to finish: from frequencies to regression

60 synthetic respondents · six steps — frequencies → descriptives → t-test → one-way ANOVA → correlation → regression

Watch one quantitative analysis run in order, and read the tables and APA sentences the engines produce. Data and results are in English.

One resident survey, end to end: questionnaire to report

Synthetic data, 1,000 respondents, 32 columns, six steps — intake, preprocessing, reliability, factor analysis, group comparison, regression

A file shaped like what a research firm receives, followed all the way through the engines into tables and sentences.

Twelve months of service requests: daily, weekly and monthly ask different questions

Synthetic data, 17,305 records, 14 columns, five steps — intake structure, daily, weekly, monthly, explanation

A file shaped like what a local government already collects, followed through the questions each reporting cycle can and cannot answer.

Synthetic demonstration data — not a real local government

Did it change before and after — one pre/post set

Synthetic data, 120 respondents, 6 columns, six steps — who answered, the shape of the score, a paired t-test, the nonparametric alternative, a group comparison that controls for the pre-test, and a reshape to long for time × group

See what measuring the same people twice lets you do, and how the question “did the participants gain more?” is actually settled.

Synthetic practice data — not a real survey

Why people leave — burnout between workload and turnover intent

Synthetic data, 200 respondents, 6 columns, four steps — the shape of the variables, correlations, a mediation analysis, and a hierarchical regression (controls, cause, mediator)

Go past “these two move together” to “what does it travel through?”

Synthetic practice data — not a real survey