1. The Empirical Bayes Method for Safety Estimation Doug Harwood MRIGlobal Kansas City, MO

2. Key Reference Hauer, E., D.W. Harwood, F.M. Council, M.S. Griffith, “The Empirical Bayes method for estimating safety: A tutorial.” Transportation Research Record 1784, pp. 126-131. National Academies Press, Washington, D.C.. 2002 http://www.ctre.iastate.edu/educweb/CE55 2/docs/Bayes_tutor_hauer.pdf

3. The Problem  You are a safety engineer for a highway agency. The agency plans next year to implement a countermeasure that will reduce crashes by 35% over the next three years. To estimate the benefits of this countermeasure, what safety measure will you multiply by 0.35?

4. What Do We Need To Know?  You need to know – or, rather, estimate – what would be expected to happen in the future if no action is taken  Then, you can apply crash modification factors (CMFs) for the known effects of planned actions to estimate their effects quantitatively

5. Common Approach: Use Last 3 Years of Crash Data 200820092010 Observed Crashes 301921

6. More Data Gives a Different Result 2001200220032004200520062007200820092010 Observed Crashes 222316 91417301921

7. RTM Example with Average Observed Crashes

8. “True Safety Impact of a Measure” Long-term average (m) 3-year average ‘before’ (Xa) True safety effect Observed safety effect 3-year average ‘after’

9. Regression to the mean problem …  High crash locations are chosen for one reason (high number of crashes!) – might be truly high or might be just random variation  Even with no treatment, we would expect, on average, for this high crash frequency to decrease  This needs to be accounted for, but is often not, e.g., reporting crash reductions after treatment by comparing before and after frequencies over short periods

10. The “imprecision” problem … Assume 100 crashes per year, and 3 years of data, we can reliably estimate the number of crashes per year with (Poisson) standard deviation of about… However, if there are relatively few crashes per time period (say, 1 crash per 10 years) the estimate varies greatly … or 5.7% of the mean or 180% of the mean!

11. Things change…  BEWARE about assuming that everything will remain the same ….  Future conditions will not be identical to past conditions  Most especially, traffic volumes will likely change  Past trends can help forecast future volume changes

12. Focus on Crash Frequency vs. AADT Relationships: Use of Crash Rates May Be Misleading

13. The Empirical Bayes Approach  Empirical Bayes: an approach to estimating what will crashes will occur in the future if no countermeasure is implemented (or what would have happened if no countermeasure had been implemented)  Simply assuming that what occurred in a recent short-term “before period” will happen again in the future is naïve and potentially very inaccurate  Yet, this assumption has been the norm for many years

14. The Empirical Bayes Approach  The observed crash history for the site being analyzed is one useful and important piece of information  What other information do we have available?

15. The Empirical Bayes Approach  We know the short-term crash history for the site  The long-term average crash history for that site would be even better, BUT…  Long-term crash records may not available  If the average crash frequency is low, even the long- term average crash frequency may be imprecise  Geometrics, traffic control, lane use, and other site conditions change over time  We can get the crash history for other similar sites, referred to as a REFERENCE GROUP

16. Empirical Bayes  Increases precision  Reduced RTM bias  Uses information from the site, plus …  Information from other, similar sites

17. 3-17 Safety Performance Functions SPF = Mathematical relationship between crash frequency per unit of time (and road length) and traffic volumes (AADT)

18. 3-18 How Are SPFs Derived?  SPFs are developed using negative binomial regression analysis  SPFs are based on several years of crash data  SPFs are specific to a given reference group of sites and severity level  Different road types = different SPFs  Different severity levels = different SPFs

19. The overdispersion parameter  The negative binomial is a generalized Poisson where the variance is larger than the mean (overdispersed)  The “standard deviation-type” parameter of the negative binomial is the overdispersion parameter φ  variance = η[1+η/(φL)]  Where …  μ=average crashes/km-yr (or /yr for intersections)  η=μYL (or μY for intersections) = number of crashes/time  φ=estimated by the regression (units must be complementary with L, for intersections, L is taken as one)

20. 3-20 Regression model for total crashes at rural 4-leg intersections with minor-road STOP control SPF Example where: N p = Predicted number of intersection-related crashes per year within 250 ft of intersection ADT 1 = Major-road traffic flow (veh/day) ADT 2 = Minor-road traffic flow (veh/day) N p = exp(-8.69 + 0.65 lnADT 1 + 0.47 lnADT 2 )

21. 3-21 Calculating the Long-Term Average Expected Crash Frequency The estimate of expected crash frequency: N e = w (N p ) + (1 – w) (N o ) Weight (w; 0<w<1) is calculated from the overdispersion parameter Expected Accident Frequency Predicted Accident Frequency Observed Accident Frequency

22. Weight (w) Used in EB Computations w = 1 / ( 1 + k N p ) w = weight k = overdispersion parameter for the SPF N p = predicted accident frequency for site 3-22

23. 3-23 Graphical Representation of the EB Method

24. Predicting Future Safety Levels from Past Safety Performance N e(future) = N e(past) x (N p(future) / N p(past) ) N e = expected accident frequency N p = predicted accident frequency 3-24

25. Predicting Future Safety Levels from Past Safety Performance 3-25  The N p(future) /N p(past) ratio can reflect changes in:  Traffic volume  Countermeasures (based on CMFs)

26. CMFs—How to Use Them  CMFs are expressed as a decimal factor:  CMF of 0.80 indicates a 20% crash reduction  CMF of 1.20 indicates a 20% crash increase

27. CMFs—How to Use Them  Expected crash frequencies and CMFs can be multiplied together: N e(with) = N e(without) CMF Crashes Reduced = N e(without) - N e(with)

28. 3-28 CMFs—Single Factor  CMF for shoulder rumble strips  Rural freeways (CMF TOT = 0.79) N e(with) = N e(without) x 0.79

29. 3-29 CMF Functions CMFs for Lane Width (two-lane rural roads) (Harwood et al., 2000)

30. CMFs for Combined Countermeasures  CMFs can be multiplied together if their effects are independent: N e(with) = N e(without) CMF 1 CMF 2 Are countermeasure effects independent?

31. EB applications  HSM  IHSDM  Safety Analyst

32. EB applications HSM Part C  Estimate long-term expected crash frequency for a location under current conditions  Estimate long-term expected crash frequency for a location under future conditions  Estimate long-term expected crash frequency for a location under future conditions with one or more countermeasures in place HSM Part B  Evaluate countermeasure effectiveness using before and after data

33. EB applications Site-Specific EB Method  Based on equations in this presentation Project-Level EB Method  If project is made up of components with different SPFs, then there is no single value of k, the overdispersion parameter EB Before-After Effectiveness Evaluation  See Chapter 9 in HSM Part B

34. Questions?