# WP21 — Extraction audit (generated)

Studies extracted: 297; eligible: 242; ineligible at full text: 55.
Coefficient rows: 18831; usable in the pooled synthesis: 7693 from 149 studies.

## Confidence of extracted rows

| confidence | rows |
|---|---|
| high | 14776 |
| medium | 3941 |
| low | 114 |

## Source of the standard error

| source | rows |
|---|---|
| reported | 10227 |
| none | 3901 |
| |b/t| | 3794 |
| |b/z(p)| (two-sided normal) | 710 |
| (ci_hi-ci_lo)/(2*1.96) | 199 |

## Harmonisation flags (rows excluded from pooling)

| flag | rows |
|---|---|
| none | 13535 |
| nonstandard_x_transform | 2555 |
| unit_not_harmonisable | 2351 |
| year_built_not_age | 226 |
| implausible_magnitude | 110 |
| year_built_not_age;nonstandard_x_transform | 54 |

## Full-text exclusion reasons (first 80 characters)

| reason | studies |
|---|---|
| Aggregate unit of analysis: the regression (Tables 4-5, p. 9) is a monthly time- | 1 |
| Aggregate unit of analysis: dependent variable is the (log) mean price per squar | 1 |
| No hedonic coefficient table. The paper estimates a generalized propensity score | 1 |
| Aggregate unit of analysis: cross-section of 985 Italian railway station areas;  | 1 |
| Aggregate unit of analysis: regressions use Seoul's 25 autonomous districts (Gu) | 1 |
| Not a hedonic regression study: the paper proposes a theoretical 'dynamic impact | 1 |
| Hedonic model (Table 5, p.11) reports only the appraisal-anchoring variables (ln | 1 |
| Conference essay reviewing anchoring effects; the only estimates (Tables 1-3, p. | 1 |
| Aggregate unit of analysis: panel of 30 Chinese cities x 10 years (2006-2015) re | 1 |
| Regression tables (Tables 4 and 6, p. 14) regress price per m2 on three expert-w | 1 |
| Aggregate unit of analysis: 1498 residential price samples averaged into 653 sam | 1 |
| Aggregate unit of analysis: the regression explains the MEDIAN house value of ce | 1 |
| Not an empirical housing-market study: the author states 'the dataset used was s | 1 |
| No regression coefficients are printed: the paper compares predictive performanc | 1 |
| No hedonic regression coefficients on property characteristics are printed. The  | 1 |

## Independent double-extraction validation

```json
{
  "papers_double_extracted": 20,
  "eligibility_agreement": 0.95,
  "coef_recall": 0.788720246856144,
  "coef_precision": 0.7731166520957543,
  "beta_agreement_1pct": 1.0,
  "se_agreement_1pct": 1.0,
  "sign_agreement": 1.0,
  "taxonomy_agreement": 1.0,
  "coefficients_compared": 647
}
```

Interpretation: numerical agreement on matched coefficients is the relevant accuracy measure; recall/precision below one reflect
differences in which auxiliary variables (composite indices, planning dummies, local-regression summaries) each extractor recorded.
Eligibility disagreements are listed in data/validated/validation_results.csv and were resolved by the protocol rule (duplicate-of-prior-publication → excluded).