Project 45 / Qualitative Research

Research Overview

Transparent qualitative coding, intercoder agreement, and traceable thematic analysis.

● LOCAL-FIRST

Demo dataset

Theme distribution

Coder A labels, among the coded excerpts

Coder agreement

Agreement and chance correction

Disagreement spotlight

Excerpts requiring researcher review

Source traceability

Segments classified by document/source type

Interpretation note: Cohen's κ evaluates coder agreement for exclusive categorical codes. It does not establish validity, objectivity, or the quality of the codebook. The example studies are entirely fictional.

Coding Workbench

Edit both coders' labels independently. Results update immediately.

Disagreements only

ID / source	Research excerpt	Coder A	Coder B	Consensus

Codebook

Codes apply to the full dataset. New codes can be used immediately.

Coder confusion matrix

Rows = Coder A, columns = Coder B. Only paired labels.

Reliability diagnostics

Nominal Cohen's kappa, complete cases

Per-code agreement

Recall of Coder A's labels in Coder B, not a measure of truth

Language & source analysis

Coding proportions and coder agreement across subsets

Break down by

Small subsets and kappa near its limits can be unstable. These comparisons are descriptive, not statistical tests of differences.

Theme patterns by group

Interpretive limitations

Language effects

Differences in coding frequencies may reflect the dataset design, the coding manual, or translation and contextual choices.

Comparability

Do not equate a higher κ with better research across groups of different category distributions.

Adjudication queue

Resolve disagreement through explicit researcher consensus. The original two labels remain unchanged.

Import research excerpts

CSV with columns: id,source,language,excerpt,coder_a,coder_b. Optional: consensus,notes. Import replaces the current in-browser session after validation.

Client-side only

Your research text is parsed locally and is not uploaded by this application. Use a downloaded copy to retain edits.

↑ Choose CSV

Maximum 1 MB / 1,000 rows / 2,000 characters per excerpt. UTF-8. Quoted multiline fields supported.

Reproducible exports

Export your complete coding decisions and agreement metrics, with methodological caveats.

Methods & limitations

Explicit assumptions and scope

Unit of analysis

Each row represents one excerpt with one exclusive thematic code assigned independently by each coder. Multiple themes in the same excerpt are not supported in this version.

Agreement

Percent agreement and unweighted nominal Cohen's κ are calculated only on rows with both coder labels. Expected agreement is the sum of marginal probability products. When expected agreement is 100%, κ is undefined.

Bootstrap uncertainty

A seeded, paired bootstrap provides a descriptive 95% percentile interval when at least 8 paired excerpts produce adequate finite resamples. This does not correct selection biases or validate the categories.

Missing and adjudicated labels

Blank labels are excluded from κ. Consensus decisions are reported separately and never used to inflate agreement between the original coders.

Research ethics

All built-in text is fictional. Remove personally identifying information before importing real interviews or publicising data. The MIT license covers the software, not third-party interview transcripts.