Which of the following strategies can be used to improve upon a basic before / after analysis for safety effectiveness evaluation?

• A. Simple before-after
• B. Empirical Bayes method
• C. Regression discontinuity
• D. Randomized Controlled Trial

Answer: (B)

When evaluating safety effectiveness, the main challenge with a basic before/after comparison is that the observed changes can be driven by random variation and regression to the mean, especially across multiple sites or treatments. The empirical Bayes method tackles this by borrowing strength from the overall experience across sites and shrinking each site's observed change toward the overall mean. This creates more stable, reliable estimates of true safety effects, particularly for sites with small amounts of data, because extreme results are tempered by the broader data pattern.

Conceptually, empirical Bayes treats the true effect sizes across sites as arising from a common distribution. It uses the data to estimate that distribution and then combines the site-specific observed changes with the prior information, producing posterior estimates. The result is an improvement over a simple before/after by reducing the impact of random fluctuations and regression to the mean, giving a clearer signal about whether the intervention actually improved safety.

Other options aren’t as well suited here. A simple before/after analysis lacks adjustment for random variation and regression to the mean, so its results can be misleading. Regression discontinuity requires a well-defined cutoff determining who receives the intervention, which isn’t always available in road safety programs. A randomized controlled trial, while ideal in theory, is often not feasible for real-world traffic safety interventions due to ethical, logistical, and practical constraints.

So, using the empirical Bayes method provides a more robust, reliable way to evaluate safety effectiveness across multiple sites than a basic before/after approach.