Raw records
463,152
404,508 modeling rows after documented exclusion
Logistic-regression analysis
Adjusted fatal-crash associations in 404,508 Brazilian records.
A multivariable logistic regression on Brazilian crash records compared fatal-crash odds across reported causes while controlling for time of day, road delineation, and weather. The published artifacts expose the candidate models, selected specification, coefficient result, input checksum, and runtime environment.
Adjusted odds ratio
11.52×
95% Wald CI 10.23–12.98
Change after adjustment
−4.9%
from 12.12× unadjusted to 11.52× adjusted
Model decision
| Specification | Controls added to crash cause | AIC | Decision |
|---|---|---|---|
| Model 5 | Time of day, road delineation, weather | 177,765.3 | Selected: minimum AIC |
| Model 4 | Time of day, road delineation | 177,917.6 | Not selected |
| Model 3 | Time of day, weather | 178,988.3 | Not selected |
Interpretation
Adjusted association, not a causal effect.
The reported 11.52× compares crashes attributed to pedestrians walking in the road with alcohol-related crashes, conditional on the selected covariates. It does not mean pedestrian behavior causes fatality odds to increase by that amount. The interval is model-based and assumes independent records and a correctly specified logistic model.
Methods and artifacts
Exact result row Unadjusted and adjusted estimates, interval, p-value, sample counts, and input MD5.Candidate model table Every compared formula, degrees of freedom, and AIC.Analysis script Data cleaning, model fitting, AIC selection, and artifact generation.Runtime environment R version, platform, locale, and attached packages.Evidence memo Counts, comparison, supported wording, boundary, and reproduction command.Original report The full academic analysis as a PDF.
Data provenance
The source file is not republished. The reproduction script records its MD5 checksum (8b2672d2451e806cd7a71231ecad4cd7) so the exact input can be checked. The model excludes 58,644 rows whose road delineation was recorded as “Not Reported”; that exclusion and the resulting 404,508-row cohort are visible in the result artifact.