1. Fall 2015

2.  Crashes are “independent” and “random” events (probabilistic events)  Estimate a relationship between crashes and covariates (or explanatory variables)  Determine the long-term average of crash occurrences for transportation facilities  Have a wide variation of applications in safety analyses: ◦ Prediction ◦ Variable screening ◦ Risk factors ◦ Before-after study

3.  Understanding the System: The first application consists of developing models with the objective of learning something about the system from which the data are taken. Examining the sign of a coefficient is an example of such application.  Screening Variables: The second application consists of developing models for screening purposes, where the objective is to determine which covariates have specific or significant effects on the risk of collisions. For this application, most of the modeling effort is devoted to the analysis of the covariates of the statistical models.  Predictive Tool: The third application aims at developing models for prediction purposes. In this application, the goal is to develop models with the best predictive capabilities. These models are usually estimated using one dataset, but are applied or evaluated using a completely new dataset.

4. Definition Definition: A model is an abstraction of reality in that it provides an approximation of some relatively complex phenomenon. A model can be deterministic or probabilistic. Most common types of probabilistic models: ● Linear Models (oka Multivariate Linear Models) Y = b+aX ● Non-Linear Models Y = aX bX ● Generalized Linear Models (GLM) (generalization of linear models) (appropriate for crash data)

5. Functional Form (most common) The probabilistic linear models have the following form: where, y = outcome or response variable x 1, x 2, …, x k = covariates or explanatory variables β 1, β 2, …, β k = unknown coefficients or regressor variables ε = random error term If ε is assumed to be normally and independently distributed with constant variance, then statistical tests on the model parameters, confidence intervals on coefficients, variables and predictions can be easily obtained.

6. Types of Model  Linear regression models ◦ Least squared/maximum likelihood ◦ Normal errors (or non-normal errors)  Count data Models ◦ Discrete non-negative events ◦ Over-dispersion ◦ Poisson, NB, COM-Poisson, etc. (many more) ◦ Very frequently used in safety (see Lord and Mannering, 2010)  Discrete choice models ◦ Provide probabilities (between 0 and 1) ◦ Logistic (logit), multinomial, ordered probit, etc. (see Savolainen et al., 2011) ◦ Used for modeling crash severity

7.  Models used as part of predicting to estimate the expected average crash frequency (by total crashes, crash severity, or collision type) of a site, facility or roadway network for a given time period, geometric design and traffic control features, and traffic volumes (AADT).  The basic elements of the predictive method are: ◦ Predictive model estimate of the average crash frequency for a specific site type. This is done using a statistical model developed from data for a number of similar sites. The model is adjusted to account for specific site conditions and local conditions; ◦ The use of the empirical Bayes (EB) Method to combine the estimation from the statistical model with observed crash frequency at the specific site. A weighting factor is applied to the two estimates to reflect the model’s statistical reliability. When observed crash data is not available or applicable, the EB Method does not apply.

8.  The predictive models in Part C vary by facility and site type, but all have the same basic elements: ◦ Safety Performance Functions (SPFs)—Statistical “base” models are used to estimate the average crash frequency for a facility type with specified base conditions. ◦ Crash Modification Factors (CMFs)—CMFs are the ratio of the effectiveness of one condition in comparison to another condition. CMFs are multiplied with the crash frequency predicted by the SPF to account for the difference between site conditions and specified base conditions; ◦ Calibration Factor (C)—multiplied with the crash frequency predicted by the SPF to account for differences between the jurisdiction and time period for which the predictive models were developed and the jurisdiction and time period to which they are applied by HSM users.

9. N predicted = predicted model estimate for site x (crashes/year) N SPFs = predicted model estimate for base conditions for site x (crashes/year) CMF yx = Crash Modification Factors for site x (type y) C x = Calibration Factor for local conditions for site x

10.  Advantages of the predictive method are that: ◦ Regression-to-the-mean bias is addressed as the method concentrates on long-term expected average crash frequency rather than short-term observed crash frequency. ◦ Reliance on availability of limited crash data for any one site is reduced by incorporating predictive relationships based on data from many similar sites. ◦ The method accounts for the fundamentally nonlinear relationship between crash frequency and traffic volume. ◦ The SPFs in the HSM are based on the negative binomial distribution, which are better suited to modeling the high natural variability of crash data than traditional modeling techniques based on the normal distribution.

11.  Crash Modification Factors (CMFs) represent the relative change in crash frequency due to a change in one specific condition (when all other conditions and site characteristics remain constant). CMFs are the ratio of the crash frequency of a site under two different conditions. Therefore, a CMF may serve as an estimate of the effect of a particular geometric design or traffic control feature or the effectiveness of a particular treatment or condition.

12. The values of CMFs in the HSM are determined for a specified set of base conditions. These base conditions serve the role of site condition ‘a’. No change in expected crash frequency Reduction in expected crash frequency Increase in expected crash frequency

13.  Safety Performance Functions (SPFs) are regression equations that estimate the average crash frequency for a specific site type (with specified base conditions) as a function of annual average daily traffic (AADT) and, in thecase of roadway segments, the segment length (L). Base conditions are specified for each SPF and may include conditions such as lane width, presence or absence of lighting, presence of turn lanes, etc.

14. N SPFs = predicted model estimate for base conditions for rural two-lane roadway segments (crashes/year) AADT = average annual daily traffic (vehicles/day) L = length of roadway segment (miles)

19.  In order to apply an SPF, the following information about the site under consideration is necessary: ◦ Basic geometric and geographic information of the site to determine the facility type and to determine whether a SPF is available for that facility and site type. ◦ Detailed geometric design and traffic control features conditions of the site to determine whether and how the site conditions vary from the SPF baseline conditions (the specific information required for each SPF is included in Part C. ◦ AADT information for estimation of past periods or forecast estimates of AADT for estimation of future periods.

20.  SPFs are developed through statistical multiple regression techniques: ◦ using observed crash data collected over a number of years at sites with similar characteristics and covering a wide range of AADTs. ◦ The regression parameters of the SPFs are determined by assuming that crash frequencies follow a negative binomial distribution. ◦ The degree of overdispersion in a negative binomial model is represented by a statistical parameter, known as the overdispersion parameter that is estimated along with the coefficients of the regression equation. ◦ The overdispersion parameter is used to determine the value of a weight factor for use in the EB Method.

22. Evaluate and compare expected crash frequency:  Existing facilities  Alternative design modifications  New facility designs  Estimated countermeasure effectiveness 4-22

23. System Planning Preliminary Design Final Design Construction Preliminary Design Final Design Construction Operations Maintenance Operations Maintenance Project Planning Predict Future Performance - Existing Facility With Vol. 1 (Part B): Diagnose crash frequency and severity Select countermeasures Conduct economic evaluation Predict Future Performance - Existing Facility With Vol. 1 (Part B): Diagnose crash frequency and severity Select countermeasures Conduct economic evaluation Predict Future Performance of Alternative Actions With Vol. 1 (Part B): Countermeasure selection Countermeasure economic evaluation Predict Future Performance of Alternative Actions With Vol. 1 (Part B): Countermeasure selection Countermeasure economic evaluation 4-23

24.  Formats ◦ Intersections ◦ Roadway segments  Applications ◦ Rural two-lane two-way roads (Chapter 10) ◦ Rural multilane highways (Chapter 11) ◦ Urban and suburban arterials (Chapter 12) 4-24

25. 4-25

26. 4-26

27.  Expected crash frequency  Sites  Facility  Network

28. Predicted Crash Frequency: N predicted = SPF x (CMF 1 x CMF 2 x ….) x C where: SPF = Safety Performance Function CMF = Crash Modification Factor C = Local Calibration Factor

29. Road segment SPF: N (Rural 2-lane) = AADT x L x (365 x 10 -6 ) x e -0.312 Intersection SPF: N (Rural multi-lane 4-way stop) = e [a + b ln(AADTmaj) + c ln (AADTmin)] Where a, b and c vary for intersection type and severity

30. AADT of Minor Approach AADT of Major Approach

31. Intersections  MV collisions  SV crashes  Vehicle-pedestrian crashes  Vehicle-bicycle crashes Roadway Segments  MV non-driveway collisions  MV driveway related collisions  SV crashes  Vehicle-pedestrian crashes  Vehicle-bicycle crashes

32. 1. Determine data needs 2. Divide locations into homogeneous segments or intersections 3. Identify and apply the appropriate SPF 4. Apply CMFs to calculated SPF values 5. Apply local calibration factor

33.  Study limits  Facility type  Study period  Site conditions (geometry, traffic control, etc.)  Traffic volume (vehicles/day)

34.  Type of intersections  Number of lanes  Cross section dimensions (LW, SW)  Alignment change (Horiz, Vertical)  Change in roadside conditions  Change in traffic volume Step 2. Divide Locations Homogeneous Segments or Intersections

35. Roadway Segment and Intersection Segment Length (Center of Intersection to Center of Intersection) A All crashes in this region are intersection crashes B Crashes in region may be segment or intersection related

36. Chapter 10 2-Lane Rural Highway SPFs Chapter 12 Urban Arterial SPFs Chapter 11 Multilane Rural Highway SPFs

37.  Review applicable SPF “base case” or typical features  Determine how study site differs from “base case”  Select CMFs for road type and atypical features from Volume 2 (Part C)  Multiply SPF value by applicable CMFs

38. Desired Level of Confidence Confidence Interval Multiple of Standard Error (MSE) Low65% - 70%1 Medium95%2 High99.9%3

39. Intersections  90 o angle (0 0 skew)  No left turn lanes  No right turn lanes  No lighting Road segments  12-ft lane widths  6-ft shoulder widths  Roadside Hazard Rating -- 3  5 driveways per mile  Tangent, flat alignment (No vertical grade)  No centerline rumble strips  No passing lanes  No two-way left turn lanes  No lighting  No automated speed enforcement

40. Intersections  90 o angle (0 0 skew)  No left turn lanes  No right turn lanes  No lighting Road segments  12-ft lane widths  8-ft shoulder widths  30-ft median (4D)  No lighting  No automated speed enforcement

41. Intersections  No left-turn lanes  Permissive left-turn signal phasing  No right-turn lanes  Right-turn on red permitted  No lighting  No automated enforcement  No bus stops, schools or alcohol sales establishments near intersection Road segments  No on-street parking  No roadside fixed objects  15-ft median (4D)  No lighting  No automated speed enforcement

42.  “C” adjusts HSM SPF-derived crash estimates to reflect local conditions ◦ Reporting levels ◦ Weather and other similar factors  Each SPF requires its unique “C”  See HSM Vol. 2 (Part C) Appendix

43. i. Repeat basic steps (time period) ii. Apply site-specific EB method iii. Repeat basic steps (next study site) iv. Apply project-Level EB method v. Calculate total expected crashes vi. Evaluate alternate design vii. Evaluate and compare results