construct.safety_climate predicts construct.accidents
normal · informative · 2 studies · N = 19,485
Distribution
Intervals
- Confidence interval (95%) — uncertainty about the mean ρ
- [-0.44, -0.15]
- Credibility interval (95%) — distribution of the true effect across settings (the Bayesian prior)
- [-0.48, -0.10]
- I² (heterogeneity) — share of total variance from between-study differences
- 81%
The true effect varies across settings — moderators likely matter.
SD_ρ>0 — true effect varies across settings; likely moderated (observed-score scale until artifact correction, PRN-058)
Evidence provenance
published ρ=-0.30 (k=2, high heterogeneity (I²=0.81)); no primary-deployment evidence yet
| Source | ρ (r) | Scope |
|---|---|---|
| Published literature | -0.30 | k = 2 · N = 19,485 |
- high heterogeneity (I²=0.81)
- replication: meta-analytic
No primary-deployment evidence yet — this prior rests on published literature alone. As anonymized, aggregated effect sizes from real deployments are contributed, they appear here as a distinct, publication-bias-free source, fused with the literature into a posterior estimate.
Code
Drop this prior straight into your model. Snippets generated from the synthesized distribution + parameters.
target += normal_lpdf(beta | -0.312434, 0.0819945);beta = pm.Normal("beta", mu=-0.312434, sigma=0.0819945)brms::prior(normal(-0.312434, 0.0819945), class = "b")# base R sample
rnorm(N, mean = -0.312434, sd = 0.0819945)np.random.normal(loc=-0.312434, scale=0.0819945, size=N)Parameters
| Family | Parameters |
|---|---|
| normal | I2 = 0.8065, mu = -0.3124, sigma = 0.08199, r_mean = -0.3026, k_studies = 2.000, tau_squared = 0.01125 |
Synthesis
- Method
- random_effects_meta
- Informativeness
- informative
- Replication status
- meta-analytic
- k_studies
- 2
- n_total
- 19,485
- Last updated
- 2026-05-30T23:52:05.066Z
Quality distribution
| Grade | Count |
|---|---|
| A | 2 |
| B | 0 |
| C | 0 |
| D | 0 |
Contributing effect sizes
Effect-size detail pages land with a later sub-ticket; for now, ids link to the filtered list. Browse all rows via /registry/effects.