# Released dataset — theory-to-measurement audit in natural-hazards adaptation research

This release contains the **sampling frame and structural metadata** for every study in the
review, together with the **coding template** needed to reproduce the analysis. The measurement
assessments themselves (instrument-detail, psychometric-quality, replicability, and documentation
ratings, and the coder-written descriptions) are **not** distributed here. A researcher wishing to
reproduce or extend the analysis re-codes the empty fields using the accompanying coding scheme and
then runs the analysis script.

## Files

| File | Unit | Rows |
| --- | --- | --- |
| `released_dataset_paper_level.csv` | one row per study | 222 |
| `released_dataset_theory_application_level.csv` | one row per study × theory application | 285 |

The two files share `file_id`. The paper-level file is the list of included studies with
bibliographic metadata; the theory-application file expands each study into its individual theory
applications (the unit at which measurement is coded). Of the 285 theory applications, 224 are
applications of one of the ten core theories (`is_core` = TRUE).

## Columns

**Provided (metadata — filled):** identifiers and bibliographic fields (`file_id`, `title`,
`author`, `year`, `journal`, `doi`), hazard classification (`hazard_corrected`,
`hazard_harmonised`), and theory
identification/structure (`n_theories`, `n_core_theories`, `multi_theory`, `combination_type`,
`theories_abbrev`, `theories_full`, `core_theories_abbrev`, `raw_theories`; and, in the
theory-application file, `theory_abbrev`, `theory_full`, `is_core`).

Two hazard columns are supplied. `hazard_corrected` holds the coder-recorded label after basic
correction; `hazard_harmonised` holds the harmonised label set used in the reported analyses
(label variants, family grouping, and a small number of paper-specific corrections), and is the
column the analysis script uses. The harmonisation procedure is described in Supplementary Methods
S2. Both are retained so the harmonisation is auditable. Multi-hazard studies carry several labels
separated by semicolons.

**To be coded (empty — headers retained):**

- `survey_instrument_level`
- `theory_constructs_measured`
- `psychometric_quality`
- `measurement_replicable`
- `measurement_documentation_level`

(The paper-level file additionally retains empty descriptive/administrative headers —
`instrument_description`, `items_shown`, `items_location`, `response_scale_provided`,
`qualitative_validation`, `coding_notes`, `coder` — so the input schema is unchanged.)

## Reproduction workflow

1. For each row, obtain the study PDF and code the four assessment fields
   (`survey_instrument_level`, `theory_constructs_measured`, `psychometric_quality`,
   `measurement_replicable`) following the coding scheme.
2. Derive `measurement_documentation_level` from the three component fields
   (`survey_instrument_level`, `psychometric_quality`, `measurement_replicable`) using the fixed
   decision table in the coding scheme — this field is **computed**, not judged independently.
3. Run the analysis script on the completed files to regenerate the figures and statistics.

## Note

Empty assessment fields are intentional. Because independent coders will apply the rubric with some
variation, re-coded values are expected to approximate rather than exactly reproduce the reported
figures.