The results described in this dissertation represent a culmination of the efforts in the study of safety performance at rural intersections with atypical design characteristics. The specificity of this topic does not limit its relevance to intersection safety and the science of relating infrastructure traits to failure risks. Statistical modelling is used to predict average crash frequency. The model specification uses attributes such as the average annual daily traffic (AADT) on the major and minor road approaches, among other characteristics, to account for the exposure to crashes of various defined types and severities. The atypical intersection geometries studied herein consist of offset-T, curved corner, highly skewed, and multi-leg with five or more intersection legs. All of the intersections studied are minor road stop-controlled, along two-lane two-way rural highways. An effort is made to summarize the outcomes in practically applicable terminology such as Crash Modification Factors (CMFs) so that the conclusions of this study can lend toward a safer transportation future.The investigation into offset-T intersections utilizes a series of random intercept negative binomial models for crash occurrence that was generated based on 10 years of crash data from a sample of 299 offset-T intersections and 301 four-leg intersections with minor stop-control along rural two-lane highways in Michigan. The search for candidate intersections was exhaustive, considering that both major and minor road AADT was desired for each site. The modeling technique used a random effect for each site (location). Models were developed for total (non-animal) intersection crashes, as well as for single motor vehicle, angle, and rear-end crash types. The effects of offset distance and direction were analyzed and incorporated into the models. Compared to conventional four-leg intersections, offset-T intersections exhibited 35 percent more crashes regardless of the offset distance or direction. Considering crash types, single motor vehicle crashes occurred more frequently at offset-T intersections, and increased as the offset distance increased. Rear end crashes also occurred more frequency at offset-T intersections, with left offsets having more crash occurrence than right offsets. However, angle crashes are 40 to 69 percent lower at offset-T intersections due to the elimination of the direct crossing maneuver. The CMF for converting an existing offset-T into a conventional four-leg intersection is 0.74 within the studied ranges of offset direction and distance. This equivalates to a 26 percent reduction in total (non-animal) crash frequency At curved corner intersections, a random intercept negative binomial models for crash occurrence used geographic region as a random effect. A total intersection crash (non-animal) model was generated based on 10 years of crash data from a sample of 227 three-leg sites and 65 four-leg sites among curved corner intersection geometry only. Reliance on the availability of minor road AADT so severely limited the sample population that an estimation of minor road AADT was made in order to proceed with modeling. A minor road AADT estimation model uses national functional classification as well as surface type, population density, and major road traffic volume to estimate the minor road traffic. A table of CMFs is presented for potentially converting an existing configuration into a combined/merged approach that is consistent with a traditional countermeasure from MDOT Geometric Design Guidelines. At curved corner intersections, installing a combined/merged intersection approach near the midpoint of the curve is a potential countermeasure that can be expected to reduce the average intersection crash frequency by 25 percent for three-leg configurations. A larger radius of curvature along the curved segment at these types of intersections is also very favorable for safety performance. Each 100-foot increase in the radius of a three-leg or four-leg curved corner intersection is estimated to reduce crash occurrence by 5 percent and 8 percent respectively. The safety influence of intersection skew angle on rural two-lane two-way facilities was evaluated by calibrating crash modification factors. Ten years of crash history among federal aid and non-federal aid highways was used to develop crash modification functions at three-leg and four-leg stop-controlled intersections. Skew angle was investigated as a parameter in the SPF models both as a continuous variable, with observed values ranging from 0 to 80 degrees, and categorized into ranges. A few transformations of the skew parameter were considered such as the flexible form model having skew interaction with AADT (annual average daily traffic), and a Hoerl curve. Both three-leg and four-leg intersections exhibited an initially increasing trend of crash rates followed by a decreasing trend as skew angle increased. A categorical model was found to best describe the skew relationship using discrete skew angle ranges. Among three-leg intersections, a skew angle between 17 to 27 degrees experienced 22 percent more crashes than perpendicular intersections. However, more highly skewed three-leg intersections exhibited a decreasing relationship to increasing skew angle. Among four-leg intersections, a skew angle between 17 to 27 degrees experienced 40 percent more crashes, while intersections with a skew angle greater than 45 degrees did not have significantly different crash occurrence than perpendicular intersections. The implications of assuming a monotonic increasing relationship to skew angle are challenged as a result of this study. Multi-leg intersections with more than four approach legs constitute a rare circumstance, yet these sites experience higher than average crash frequencies compared to conventional four-leg intersections. Single-vehicle as well as angle and rear-end crash types are most likely to occur at multi-leg intersections based on the history of crashes observed. The procedures described in this study are consistent with the Highway Safety Manual (HSM) and subsequent state of the art research for the procurement of safety performance models for any variety of circumstances. By utilizing an expansive data set and geospatial mapping techniques, the analysis is extended to previously unexplored site types. The extension of predictive safety analysis to atypical intersection types with unique geometric characteristics helps to fill a gap in the current field of practice with the hope of achieving an ultimate goal toward zero deaths in motor vehicle transportation.
Read
