|
| Strategies | Challenges | Proposed By |
|
| Before-after Method |
|
| Naïve before-after | (i) Simplest method, but fails to consider regression toward mean effect
(ii) Overestimates results of countermeasure | Abdel-Aty et al., 2014 [16] |
|
| Before-after with comparison group | (i) Similar as Naïve before-after, it compares with untreated sites to reduce effects of external casual factors
(ii) Fails to account naturally expected reduction in crashes, i.e. regression toward the mean effect | Park & Abdel-Aty, 2016 [17] |
|
| Empirical Bayes | (i) Captures a true effect of countermeasure by considering the regression to mean
(ii) Lacks in considering uncertainty in data and model parameters
(iii) Not flexible due to reliance on assumption of negative binomial distribution
(iv) Affected significantly by site selection bias | Frank et al., 2010 [18], Park et al., 2016 [19], Zou et al., 2018 [20] |
|
| Full Bayes | (i) Application complexity and requires a high level of statistical training
(ii) Fails to identify confounding factors and their impacts | El-Basyouny and Sayed, 2011 [21], Lord & Kuo, 2012 [41] |
|
| Cross-sectional Method | (i) Overestimates effect of countermeasure due to presence of confounding variables
(ii) Fails to omit unobserved heterogeneity (i.e. variable bias) | Lord & Mannering, 2010 [22] |
|
| Case-control Method | (i) Control impacts of confounding variables
(ii) Data collection and sample selection are complex and time consuming | Fitzpatrick et al., 2008 [24], Gross & Donnell, 2011 [23] |
|
| Cohort Method | (i) Requires a large distribution of samples data
(ii) Changes in site parameters during study period impacts results of countermeasures | Cummings et al., 2003 [34], Zhu et al., 2007 [42] |
|
| Meta-analysis Method | (i) Recommended to perform sensitivity analysis to validate assumptions | Phillips et al., 2011 [35] |
|
| Traffic Simulation | (i) Built-in evasive algorithms in simulation platforms prevent modeling crash scenarios | Sacchi et al., 2013 [39], Shahdah et al., 2014 [40], Giuffrè et al., 2018 [43] |
|
