Design and control elements 
Related literature 
Major findings 
Study type 
Data used 
Type of site 
Intersection Geometric Design Features

Leftturn lanes 
Harwood et al. [2000]^{(25)} 
• Based upon the judgement of an expert panel, installation of a leftturn lane along one major approach reduces intersectionrelated accidents by 18 to 24%, depending upon the type of traffic control and number of legs at the intersection.
• Based upon the judgement of an expert panel, installation of leftturn lanes along both major approaches to a fourleg intersection reduces intersectionrelated accidents by 33 to 42 percent, depending upon the type of traffic control.

Accident prediction algorithm using negative binomial distribution and accident modification factors developed by expert panel. 
Prediction algorithm combines elements of historical accident data, predictions from statistical models, results of beforeafter studies, and expert judgments made by experienced engineers. 
Rural intersections along twolane highways. 
Vogt [1999]^{(24)} 
For a fourlane by twolane STOPcontrolled rural intersection, the predictive model indicates installation of leftturn lanes along the major approach reduced total accidents by 38 percent. 
Accident prediction model using negative binomial distribution. 
72 fourleg intersections in California and Michigan. 
Fourlane by twolane STOPcontrolled rural intersections. 
Gluck et al. [1999]^{(6)} 
Installation of leftturn lanes reduced the accident rates per million entering vehicles at unsignalized intersections by 50 percent and at signalized intersections by 18 percent. Combined the presence of leftturn lanes reduced the accident rate by 35 percent. 
Synthesis of previous research conducted by Tamburri and Hammer^{(13)} and Wilson et al.^{(14)} 
53 intersections in California. 

Gluck et al. [1999]^{(6)} 
Intersections without turn lanes had an accident rate of 1.65 accidents per million entering vehicles, while intersections with leftturn lanes had an accident rate of 0.59. 
Synthesis of previous research conducted by Shaw and Michael.^{(65)} 
Eight intersections without lanes; three with leftturn lanes. 

Gluck et al. [1999]^{(6)} 
Installation of leftturn lanes reduced the accident rate per million entering vehicles by 38 percent. 
Synthesis of previous research conducted by BenYakov and Craus^{(11)} and Craus and Mahalel.^{(12)} 
25 intersections. 

Leftturn lanes (continued) 
Gluck et al. [1999]^{(6)} 
Installation of leftturn lanes reduced the accident rates per million leftturning vehicles at unsignalized intersections by 77 percent and at signalized intersections by 54 percent. 
Synthesis of previous research conducted by Agent.^{(10)} 


Gluck et al. [1999]^{(6)} 
Restriping the lane assignments to provide leftturn lanes reduced the number of accidents at eight intersection locations. The leftturn lanes reduced leftturn accidents by 62 percent and all accidents by 58 percent. 
Synthesis of previous research conducted by Greiwe.^{(9)} 
Eight intersections in Indiana. 

Gluck et al. [1999]^{(6)} 
1.8mile section of fourlane roadway was converted to threelane crosssection. Total number of accidents before conversion was 109, and 67 accidents occurred during after period. 
Synthesis of previous research conducted by New Jersey Department of Transportation.^{(7)} 
1.8 miles of Route 47 in New Jersey converted from fourlane road to threelane road. 

Gluck et al. [1999]^{(6)} 
• Installation of leftturn lanes along eightmile southern section of Route 130 reduced the accident rate per million entering vehicles by 35 percent.
• Installation of leftturn lanes along 28mile northern section of Route 130 reduced the accident rate per million entering vehicles by 51 percent.

Synthesis of previous research conducted by New Jersey Department of Transportation.^{(8)} 
Eight mile southern section of Route 130 in New Jersey and 28mile northern section. 

Bauer and Harwood [1996]^{(20)} 
Leftturn channelization resulted in an increase in total multiplevehicle accidents and fatal injury accidents. 
Statistical modeling with negative binomial regression. 
14,432 rural intersections in California. 
Rural and urban signalized and unsignalized intersections. 
Poch and Mannering [1995]^{(22)} 
Total accident frequencies were found to be higher on intersection approaches with a shared throughleft lane and two or more total lanes than on approaches with other conditions. Approaches with a leftturn lane, a through lane, and a shared throughright lane had more rearend accidents than those with other conditions. 
Accident prediction model using negative binomial distribution. 
63 intersections in Bellevue, Washington. 
Urban areas. A large number of intersections were in residential areas which are characterized by low traffic volume. All intersections had some sort of operational improvement during 198892. 
Leftturn lanes (continued) 
Maze et al. [1994]^{(23)} 
Predictive models indicate that a leftturn lane with permitted phasing at a signalized intersection has a positive effect on safety. A typical example developed by the authors indicates an anticipated reduction in leftturn accident rate of approximately 5.5 percent from installation of a leftturn lane with permitted phasing. 
Statistical modeling based on multiple regression. 
63 signalized intersections, including 248 intersection approaches. Five years of accident data were considered for each intersection. 
Atgrade signalized intersections in Iowa. 
Maze et al. [1994]^{(23)} 
Predictive models indicate that a leftturn lane with protected/permitted phasing at a signalized intersection has a positive effect on safety. A typical example developed by the authors indicates an anticipated reduction in leftturn accident rate of approximately 35 percent from installation of a leftturn lane with protected/permitted phasing. 
Statistical modeling based on multiple regression. 
63 signalized intersections, including 248 intersection approaches. Five years of accident data were considered for each intersection. 
Atgrade signalized intersections in Iowa. 
McCoy and Malone [1989]^{(4)} 
On urban fourlane roadways, leftturn lanes at signalized and unsignalized intersections significantly reduced rearend, sideswipe, and leftturn accidents. At unsignalized intersections with leftturn lanes, there was also a significant increase in rightangle accidents. 
Comparative. 
63 intersections on urban, fourlane roadways. 
Urban signalized and unsignalized intersections in Nebraska. 
Lau and May [1988]^{(40)} 
Leftturn channelization on the crossroad was found to be a significant factor in predicting injury accidents at unsignalized intersections. 
CART analysis of residuals from base model. 
17,000 unsignalized intersections. Seven years of injury accident data for each intersection. 
Unsignalized intersections on California state highways. 
Leftturn lanes (continued) 
Hauer [1988]^{(15)} 
• Provision of leftturn channelization at unsignalized intersections reduced accidents by 70 percent in urban areas when combined with curbs or raised bars. Likewise, accidents were reduced by 65 and 60 percent, respectively, in suburban and rural areas.
• When channelization was painted at unsignalized intersections, accidents decreased by 15, 30, and 50 percent in urban, suburban, and rural areas, respectively.

Synthesis of previous research conducted by McFarland et al. [1979].^{(16)} 
Not available. 
Not available. 
Hauer [1988]^{(15)} 
At signalized intersections, leftturn channelization with a leftturn phase reduced accidents by 36 percent and without the leftturn phase by 15 percent. 
Synthesis of previous research conducted by McFarland et al. [1979].^{(16)} 
Not available. 
Not available. 
Hauer [1988]^{(15)} 
Adding leftturn lanes reduced accidents by varying amounts depending on the type of intersection, whether it was signalized or unsignalized, and whether the intersection was rural or urban. 
Synthesis of previous research conducted by R. Jorgensen and Associates, Inc. [1978].^{(66)} 
Not available. 
Not available. 
McCoy et al. [1985]^{(21)} 
At unsignalized intersections on rural twolane highways, there was no significant difference in rearend and leftturn accident rates between intersections with leftturn lanes and those without leftturn lanes. 
Comparative. 
Intersections on rural twolane highways in Nebraska. 
Unsignalized intersections. 
Parker et al. [1983]^{(3)} 
Passingrelated accidents at rural intersections along twolane highways do not represent a major safety problem, but when a leftturn lane is provided at new or reconstructed intersections, potential for passingrelated accidents is greatly reduced. 
Benefit/cost analysis not reviewed directly. Overview based on synthesis of previous research by Kuciemba and Cirillo [1992].^{(49)} 
Not available. 
Not available. 
Leftturn lanes (continued) 
David and Norman [1976]^{(26)} 
Signalized intersections with opposing leftturn lanes were found to have significantly more accidents than intersections without opposing leftturn lanes. Provision of opposing leftturn lanes at fourleg signalized intersections was found to increase accident frequencies by 2.4 to 6.1 accidents per year. 
Comparative. 
22 fourleg intersections with opposing leftturn lanes; it is not clear how many fourleg intersections without leftturn lanes were available. Three years of accident data were obtained for each intersection. 
Urban intersections in the San Francisco Bay Area of California. 
Dale [1973]^{(19)} 
At intersections along rural twolane highways, installation of a traffic signal and leftturn lane reduced the total number of accidents by 19.7 percent, while the installation of a traffic signal without leftturn channelization reduced the total number of accidents by 6 percent. 



Foody and Richardson [1973]^{(5)} 
• For signalized intersections, the accident rate was reduced 38 percent with the addition of leftturn lanes.
• For unsignalized intersections, the accident rate was reduced 76 percent with the addition of leftturn lanes.

Comparative. 
Not available. 
Not available. 
Lacy [1972]^{(18)} 
Several improvements to the intersection, which included extending and rearranging the channelization and adding separate leftturn lanes, reduced the accident frequency by 35 percent and the accident severity by 80 percent. Additional improvements to the intersection included widening the approaches and modifying the traffic signals. 
Before/After. 
One urban intersection in Peoria, Illinois. 
Urban intersection. 
Caltrans [1967]^{(17)} 
Reduction in accident rates at unsignalized intersections was much higher with use of raised barrier leftturn lanes than with painted leftturn lanes. 
Before/After. 
53 safety improvement projects in California. 
Urban and rural areas, including signalized and unsignalized intersections. 
Offset leftturn lanes 
Harwood et al. [1995]^{(27)} 
A field review of traffic operations and an office review of three years of accident data were conducted for two signalized intersections with tapered offset leftturn lanes and one signalized intersection with parallel offset leftturn lanes. This review found no operational or accident problems at the intersections related to the offset leftturn lanes. However, no measures of effectiveness comparing offset and conventional leftturn lanes were developed. 
Operational and safety review of intersection performance. 
Field observation of undesirable driver behavior and three years of accident data for three intersections. 
Signalized intersections on divided highways with tapered and parallel offset leftturn lanes. 
McCoy et al. [1992]^{(28)} 
Developed guidelines concerning the amount of offset between opposing leftturn lanes to provide adequate sight distance, but performed no accident studies. 
Engineering analysis. 
Typical intersection geometrics. 
Intersections with opposing leftturn vehicles. 
Joshua and Saka [1992]^{(29)} 
Developed guidelines concerning the amount of offset between opposing leftturn lanes to provide adequate sight distance, but performed no accident studies. 
Engineering analysis. 
Typical intersection geometrics. 
Intersections with opposing leftturn vehicles. 
Rightturn lanes 
Harwood et al. [2000]^{(25)} 
• Based on the judgement of an expert panel, presence of a rightturn lane along one major approach to a rural STOPcontrolled intersection reduces intersectionrelated accidents by 5 percent.
• Based on the judgement of an expert panel, presence of rightturn lanes along both major approaches to a rural STOPcontrolled intersection reduces intersectionrelated accidents by 10 percent.

Accident prediction algorithm using negative binomial distribution and accident modification factors developed by expert panel. 
Prediction algorithm combines elements of historical accident data, predictions from statistical models, results of beforeafter studies, and expert judgments made by experienced engineers. 
Rural intersections along twolane highways. 
Rightturn lanes (continued) 
Harwood et al. [2000]^{(25)} 
• Based on the judgement of an expert panel, presence of a rightturn lane along one major approach to a rural signalized intersection reduces intersectionrelated accidents by 2.5 percent.
• Based on the judgment of an expert panel, presence of rightturn lanes along both major approaches to a rural signalized intersection reduces intersectionrelated accidents by 5 percent.

Accident prediction algorithm using negative binomial distribution and accident modification factors developed by expert panel. 
Prediction algorithm combines elements of historical accident data, predictions from statistical models, results of beforeafter studies, and expert judgments made by experienced engineers. 
Rural intersections along twolane highways. 
Vogt and Bared [1998]^{(30)} 
Presence of rightturn lanes at threeleg rural unsignalized intersections increases the total number of intersectionrelated accidents by 27 percent. 
Poisson and negative binomial modeling. 
389 rural threeleg intersections in Minnesota. 
Unsignalized intersections on rural twolane highways. 
Bauer and Harwood [1996]^{(20)} 
Rightturn channelization resulted in an increase in total multiplevehicle accidents and fatal injury accidents. 
Statistical modeling with negative binomial regression. 
14,432 rural intersections in California. 
Rural and urban signalized and unsignalized intersections. 
Channelization 
Hauer [1988]^{(15)} 
Channelization was found to reduce accidents by 32 percent and injury accidents by 50 percent. The average benefitcost ratio of channelization was 2.3. 
Synthesis of previous research conducted by Hagenaur et al. [1982]^{(67)} 
Not available. 
Not available. 
David and Norman [1976]^{(26)} 
Raised pavement markings tended to decrease accidents, especially at cross intersections. 
Comparative. 
558 intersections with 4,372 accidents in three years in the San Francisco Bay Area of California. 
Urban areas. 82 percent of intersections had some form of delineation. 
Channelization (continued) 
Exnicios [1967]^{(33)} 
Several safety measures, including rechannelization, were implemented at three intersections. The improvements resulted in a 31 percent reduction in total accidents (over two years), a 58 percent reduction in total accidents (over one year), and a 100 percent reduction in total accidents (over 26 months) at the respective intersections. 
Before/After. 
Three intersections. 
The intersections were located in a residential suburb in Chicago, in metropolitan New Orleans, and in Shreveport, Louisiana. 
Rowan and Williams [1966]^{(34)} 
Accident rates, personal injuries, and rearend type accidents were reduced due to the introduction of channelization. 
Before/After. 
US Route 290 in northwest Houston. 
Arterial, fourlane, at signalized intersections. 
Channelization
• Island design

Forrestel [1994]^{(37)} 
The pedestrian accident rate at an unsignalized intersection on a fourlane arterial was reduced by 11.5 percent when raised median islands were installed. 
Synthesis of previous research efforts. Related article entitled "A Comparison of the Pedestrian Safety of Median Islands and Marked Crossings." [1978]^{(49)} 
One intersection in Western Australia. 
Fourlane, unsignalized intersection. 
Washington et al. [1990]^{(36)} 
Intersection approaches with raised medians have a 40 percent lower accident rate than those with flush medians. 
Comparative. 
40 intersections in California. 
Not available. 
Templer [1980]^{(32)} 
Raised medians reduced the number of conflicts between pedestrians and vehicles. However, the difference was not statistically significant. 
Before/After. 
Two intersections in Clearwater, Florida. 
Signalized Tintersections, located in business and recreational areas. 
Number of intersection legs (e.g., three, four, five) 
Bauer and Harwood [1996]^{(20)} 
Rural fourleg STOPcontrolled intersections have about twice as many accidents as rural threeleg STOPcontrolled intersections (1.1 vs. 0.6 accidents per intersection per year). A similar pattern was found for urban STOPcontrolled intersections (2.2 accidents per intersection per year for fourleg intersections vs. 1.3 for threeleg intersections). 
Comparative. 
8,525 atgrade intersections in California. 
Rural fourleg, rural threeleg, urban fourleg, and urban threeleg intersections, all with STOP control. 
Number of intersection legs (e.g., three, four, five) (continued) 
Harwood et al. [1995]^{(27)} 
Predictive relationships were developed for number of multiplevehicle intersection accidents per year as a function of majorroad ADT, crossroad ADT, majorroad median width, majorroad lane and shoulder widths, majorroad design speed, presence of leftturn lanes, and terrain. Results show that typical divided highway intersections with four legs have about twice as many accidents as threeleg intersections for narrow medians and more than five times as many accidents as threeleg intersections for wide medians. 
Statistical modeling with Poisson regression. 
1,200 intersections on California state highways. 
Urban/suburban unsignalized intersections on divided highways in California. 
Hanna et al. [1976]^{(38)} 
In rural areas, fourleg intersections have higher accident rates than T intersections (69 percent increase). 
Comparative. 
232 intersections in rural municipalities in Virginia. 
Includes both STOPcontrolled and signalized intersections. 
David and Norman [1976]^{(26)} 
In urban areas at STOPcontrolled intersections, accident frequencies were very similar for fourleg intersections and T/Ytype intersections with ADT under 20,000 veh/day. Once above 20,000 veh/day, the accidents doubled for fourleg intersections. 
Comparative. 
558 intersections with 4,372 accidents in three years in the San Francisco Bay Area of California. 
Of 558 intersections, 269 were threeleg intersections, and 289 were fourleg intersections. 
Intersection type (e.g., cross, T, Y, offset) 
Lau and May [1988]^{(39)} 
Intersection type was found to be a significant factor in predicting injury accidents at signalized intersections. 
CART analysis of residuals from base model. 
2,488 signalized intersections. seven years of injury accident data for each intersection. 
Signalized intersections on California state highways. 
Lau and May [1988]^{(40)} 
Intersection type was found to be a significant factor in predicting injury accidents at unsignalized intersections. 
CART analysis of residuals from base model. 
17,000 unsignalized intersections. Seven years of injury accident data for each intersection. 
Unsignalized intersections on California state highways. 
Intersection type (e.g., cross, T, Y, offset) (continued) 
Hanna et al. [1976]^{(38)} 
For threeleg intersections, Y intersections were found to have accident rates approximately 50 percent higher than T intersections. 
Comparative. 
232 intersections in rural municipalities in Virginia. 
Includes both STOPcontrolled and signalized intersections. 
Hanna et al. [1976]^{(38)} 
For fourleg intersections, offset intersections had accident rates that were approximately 43 percent of the accident rate of conventional fourleg intersections. 
Comparative. 
232 intersections in rural municipalities in Virginia. 
Includes both STOPcontrolled and signalized intersections. 
Angle of intersection (e.g., skew) 
Harwood et al. [2000]^{(25)} 
AMFs for intersection skew angle were derived from statistical modeling and apply to total intersectionrelated accidents. For a threeleg STOPcontrolled intersection, the AMF was calculated as:
AMF = exp (0.0040 SKEW)
For a fourleg STOPcontrolled intersection, the AMF was calculated as:
AMF = exp (0.0054 SKEW)
where:
SKEW = intersection skew angle (degrees), expressed as the absolute value of the difference between 90 degrees and the actual intersection angle.

Accident prediction algorithm using negative binomial distribution and accident modification factors developed by expert panel. 
Prediction algorithm combines elements of historical accident data, predictions from statistical models, results of beforeafter studies, and expert judgments made by experienced engineers. 
Rural intersections along twolane highways. 
Bauer and Harwood [1996]^{(20)} 
Angle of intersection was found to have a statistically significant relationship to multiplevehicle accident frequency at urban, fourleg, signalized intersections, but the direction of the effect was opposite to that expected. Skewed intersections were found to have accident frequencies approximately 20 percent less than 90 intersections. This finding may represent a surrogate effect of some uncontrolled variable. 
Statistical modeling with negative binomial regression. 
198 intersections on California state highways. 
Urban fourleg signalized intersections in California. 
Angle of intersection (e.g., skew) (continued) 
McCoy et al. [1994]^{(42)} 
At twoway STOPcontrolled intersections on rural twolane highways, the number of accidents per year increases with traffic volume and skew angle. Thus, more accidents will occur with higher volumes and/or greater skew angles. Threeleg intersections have fewer accidents than fourleg intersections with equivalent traffic conditions and skew angles. 
Comparative. 
29 skewed and 39 nonskewed rural intersections in Nebraska. 
Twoway STOPcontrolled intersections on rural twolane highways. Included threeleg and fourleg intersections. Volumes on the major and minor roadways ranged from 400 to 5,200 veh/day and from 150 to 1,500 veh/day, respectively. 
Hauer [1988]^{(15)} 
Stated as an important safety factor. No safety studies conducted. 



Roundabouts 
See section of this table on type of traffic control. 




Curb return radius 
Hauer [1988]^{(15)} 
Stated as an important safety factor. No safety studies conducted. 



Sight distance
• Intersection sight distance (clear sight triangles in intersection quadrants)

Harwood et al. [2000]^{(25)} 
Based upon the judgement of an expert panel, the AMFs are as follows for intersection sight distance at intersections with STOP control on the minor leg(s):
• 1.05 if sight distance is limited in one quadrant of the intersection.
• 1.10 if sight distance is limited in two quadrants of the intersection.
• 1.15 if sight distance is limited in three quadrants of the intersection.
• 1.20 if sight distance is limited in four quadrants of the intersection.
Sight distance in a quadrant is considered limited if the available sight distance is less than the sight distance specified by AASHTO policy for a design speed of 20 km/h less than the major roaddesign speed and the sight distance restrictions are due to roadway alignment and/or terrain.

Accident prediction algorithm using negative binomial distribution and accident modification factors developed by expert panel. 
Prediction algorithm combines elements of historical accident data, predictions from statistical models, results of beforeafter studies, and expert judgments made by experienced engineers. 
Rural intersections along twolane highways. 
Sight distance
• Intersection sight distance (clear sight triangles in intersection quadrants) (continued)

David and Norman [1976]^{(26)} 
Developed estimates of the reduction in annual accident frequency from improving sight distance as a function of the initial sight distance (termed "sight radius" in the study) from the minorroad approach and total entering ADT. The results indicated that, in most cases, the worse the initial sight distance, the greater the accident reduction obtained from a sight distance improvement. Magnitudes of the sight distance improvements were not specified. 
Comparative. 
558 intersections with 4,372 accidents in three years in the San Francisco Bay Area of California. 
Urban areas where foliage and buildings obstructed the view of intersections. 
Hanna et al. [1976]^{(38)} 
In this study, the average accident rate for all intersections was 1.13, while the average accident rate for intersections with "poor sight distance" is 1.33 accidents per million entering vehicles. 
Comparative. 
Examined 41 intersections in rural area of Virginia with total of 366 accidents. 
Rural municipalities, including both STOPcontrolled and signalized intersections. 
Mitchell [1972]^{(43)} 
• Total accidents at intersections dropped 67 percent when intersection sight obstructions were removed.
• The greatest percentage of reduction in accidents was experienced at the intersections where the sight distance was improved.

Before/After. 
Five intersections in Concord, California. 
Sight distance at five intersections that had been improved. 
Sight distance
• Stopping sight distance

Fambro et al. [1989]^{(44)} 
Accident rates were high for intersections located on crest vertical curves with limited sight distance. Similar results were obtained in NCHRP Project 342 [1996].^{(28)} 
Comparative. 

Rural twolane roadways. 
Sight distance
• Sight distance to traffic control device (e.g., STOP sign, signal)

None found 




Approach width 
Bauer and Harwood [1996]^{(20)} 
It was found that as lane width decreases, the total number of multiplevehicle accidents and fatal injury accidents increases. 
Statistical modeling with negative binomial regression. 
2,999 intersections in California. 
Urban fourleg signalized and unsignalized intersections. 
David and Norman [1976]^{(26)} 
Higher accident occurrence for narrow streets was not evident. 
Comparative. 
558 intersections with 4,372 accidents in three years in the San Francisco Bay Area of California. 
269 Tintersections, 289 fourleg intersections. 298 of the intersections were STOPcontrolled. 
Lacy [1972]^{(18)} 
Several improvements to the intersection, which included widening the approaches, reduced the accident frequency by 35 percent and the accident severity by 80 percent. Other improvements to the intersection included: extending and rearranging the channelization, adding separate leftturn lanes, and modifying the traffic signals. 
Before/After. 
One urban intersection in Peoria, Illinois. 
Urban intersection. 
Number of approach lanes 
Bauer and Harwood [1996]^{(20)} 
In the models for total multiplevehicle accidents and fatal injury accidents at rural and urban unsignalized intersections, approaches with one lane were associated with higher accident frequencies, while approaches with two or more lanes were associated with lower total accident frequencies. The opposite appears to be the case for urban, fourleg, signalized intersections. 
Statistical modeling with negative binomial regression. 
2,262 rural intersections in California. 
Rural fourleg STOPcontrolled intersections. 
Lau and May [1988]^{(39)} 
The number of lanes on the major roadway and the crossroad were found to be a significant factor in predicting injury accidents at signalized intersections. 
CART analysis of residuals from base model. 
2,488 signalized intersections. Seven years of injury accident data for each intersection. 
Signalized intersections on California state highways. 
Number of approach lanes (continued) 
Lau and May [1988]^{(40)} 
The number of lanes on the major roadway was found to be a significant factor in predicting injury accidents at unsignalized intersections. 
CART analysis of residuals from base model. 
17,000 unsignalized intersections. Seven years of injury accident data for each intersection. 
Unsignalized intersections on California state highways. 
David and Norman [1976]^{(26)} 
For roadways with ADT under 10,000 veh/day, accident frequencies can be reduced by providing through lanes. 
Comparative. 
558 intersections with 4,372 accidents in three years in the San Francisco Bay Area of California. 
Of 558 intersections, 71 percent had 2x2 through lanes, 26 percent had 2x4, and 3 percent had 4x4 on the crossroad and major road, respectively. 
Median width and type 
Harwood et al. [1995]^{(27)} 
• At rural fourleg unsignalized intersections, accident frequency decreases as median width increases.
• At urban/suburban intersections (unsignalized and signalized), accident frequency increases with increasing median width.

Comparative study and statistical modeling with Poisson regression. 
2,140 divided highway intersections in urban and rural areas of California. 
Intersections include rural and urban/suburban unsignalized intersections (fourleg and threeleg), as well as urban/suburban fourleg signalized intersections. 
Van Maren [1977]^{(47)} 
Found no statistically significant relationship between median width and intersection accident rate. 



Priest [1964]^{(46)} 
Except at very low volume levels, intersection accident frequencies decrease as the median width increases. The difference in intersection accident rate between medians less than 20 ft wide and medians 20 to 30 ft wide is greater than the difference in intersection accident rate between medians with widths of 20 to 39 ft and those of 40 ft or more. 
Statistical modeling with regression analysis. 
316 intersections in Ohio. Three years of accident data were available for each intersection. 
Intersections on divided highways with partial or no access control in Ohio. 
Vertical alignment on intersection approaches 
Hanna et al. [1976]^{(38)} 
Rural intersections with steep grades (greater than 5 percent) "generally operate safely." These intersections had an accident rate of 0.97 accidents per million entering vehicles, compared to an overall accident rate of 1.13. 
Comparative. 
232 intersections in Virginia. 
Rural areas. 
Horizontal alignment on intersection approaches 
None found 




Design speed 
Bauer and Harwood [1996]^{(20)} 
As design speed decreases, there is an increase in total multiplevehicle accidents and fatal injury accidents. 
Statistical modeling with negative binomial regression. 
1,434 rural intersections in California. 
Rural and urban intersections. 
Traffic Control and Operational Features

Type of traffic control
• Uncontrolled
• YIELD control

Poch and Mannering [1995]^{(22)} 
With no control on an intersection approach, total and angle accidents decrease. 
Statistical modeling with negative binomial regression. 
63 intersections in the city of Bellevue, Washington. 
Urban areas. A large number of intersections are in residential areas which are characterized by low traffic volume. All intersections had some sort of operational improvement during 198892. 
Hauer [1988]^{(15)} 
• 44 percent and 52 percent fewer accidents after conversion to YIELDcontrol.
• Another study gives accident reduction of 23 and 63 percent after conversion of uncontrolled intersections to YIELDcontrol.

Synthesis of previous studies. 
Not available. 
Not available. 
Lau and May [1988]^{(39)} 
Traffic control type was found to be a significant factor in predicting injury accidents at unsignalized intersections. However, the traffic control type variable was formulated in such a way that it is not easy to distinguish among the traffic control types used in this table. 
CART analysis of residuals from base model. 
17,000 unsignalized intersections. Seven years of injury accident data for each intersection. 
Unsignalized intersections on California state highways. 
Hall et al. [1978]^{(50)} 
Accidents can be reduced by 20 to 60 percent through proper use of YIELD signs at lowvolume intersections. Little additional reduction is obtained if YIELD signs are replaced by STOP signs. 
Not available. 
Not available. 
Not available. 
Agent and Deen [1975]^{(51)} 
At YIELD signs, over half of the accidents were rearend collisions, while angle collisions made up over half the accidents at STOP signs. 
Comparative. 
Data for intersections in Kentucky. Three years of accident data were available. 
Intersections in rural areas. 
Type of traffic control
• STOP control

Hanna et al. [1976]^{(38)} 
Accident rates at STOPcontrolled intersections were lower at those intersections having high traffic flow. 
Comparative. 
232 intersections in Virginia. 
Rural areas. 
Type of traffic control
• STOP control with flashing beacons

None found 




Type of traffic control
• Signal control (phasing, timing, and operation)

Poch and Mannering [1995]^{(22)} 
With signal control intersections, the total and angle accidents decrease. 
Statistical modeling with negative binomial distribution. 
63 intersections in city of Bellevue, Washington. 
Urban areas. All intersections had some sort of operational improvement during 198892. 
Maze et al. [1994]^{(23)} 
Predictive models indicate that a protected leftturn signal phase without a leftturn lane has a positive effect on safety. A typical example developed by the authors indicates an anticipated reduction in leftturn accident rate of approximately 50 percent from installation of a protected leftturn signal phase. 
Statistical modeling based on multiple regression. 
63 signalized intersections, including 248 intersection approaches. Five years of accident data were considered for each intersection. 
Atgrade signalized intersections in Iowa. 
Lau and May [1988]^{(39)} 
A controltype variable based on signal phasing and actuation was found to be a significant factor in predicting injury accidents at signalized intersections. However, the control type variable was defined in such a way that explicit effects of phasing and actuation cannot be determined. 
CART analysis of residuals from base model. 
17,000 unsignalized intersections. Seven years of injury accident data for each intersection. 
Unsignalized intersections on California state highways. 
Hanna et al. [1976]^{(38)} 
Installation of traffic signal controls could result in slight increase in accident rates, significant increase in rearend accidents, and comparable decreases in angle collisions. 
Comparative. 
232 intersections in Virginia. 
Rural municipality area. 
Type of traffic control
• Signal control (phasing, timing, and operation) (continued)

David and Norman [1976]^{(26)} 
In urban areas, multiphase traffic signals appear to have lower percentages of fatal and injury accidents than twophase signals. 
Comparative. 
558 intersections with 4,372 accidents in 3 years in the San Francisco Bay Area of California. 
Of 558 intersections, 269 were T intersections and 289 were fourleg intersections. 298 of the intersections were STOPcontrolled. 
King and Goldblatt [1975]^{(52)} 
• Signalization leads to a reduction in rightangle accidents and an increase in rearend accidents.
• Signalized intersections have higher accident rates, but this is usually offset by less severity per accident.

Comparative analyses and review of related research. 
Used a large nationwide accident data base. 
Not available. 
Type of traffic control
• Roundabouts

Persaud et al. [2001]^{(53)} 
Converting intersections with conventional traffic control (i.e.,STOP, signal) to roundabouts reduces all accidents by 40 percent, injury accidents by 80 percent, and fatal and incapacitating injury accidents by 90 percent. 
Before/After. 
23 intersections located in 7 states. 
Mix of urban, suburban, and rural environments. 

Robinson et al. [2000]^{(41)} 
Converting intersections with conventional traffic control (i.e.,STOP, signal) to roundabouts reduces all accidents by 37 percent and injury accidents by 51 percent. 
Synthesis of research conducted in the U.S. and internationally. 
Not available. 
Not available. 
Turn prohibitions 
Lau and May [1988]^{(39)} 
Leftturn prohibitions were found to be a significant factor in predicting injury accidents at signalized intersections. 
CART analysis of residuals from base model. 
2,488 signalized intersections. Seven years of injury accident data for each intersection. 
Signalized intersections on California state highways. 

Lau and May [1988]^{(40)} 
Leftturn prohibitions were found to be a significant factor in predicting injury accidents at unsignalized intersections. 
CART analysis of residuals from base model. 
17,000 unsignalized intersections. Seven years of injury accident data for each intersection. 
Unsignalized intersections on California state highways. 
Presence and type of crosswalks 
Hauer [1988]^{(15)} 
Marked crosswalks had more exposure, but fewer accidents, than unmarked crosswalks. 
Synthesis of previous research conducted by Knoblauch et al. [1984]^{(68)} 
Not available. 
Not available. 
Hauer [1988]^{(15)} 
Pedestrian accidents increased 86 percent after crosswalks were marked, and rearend collisions increased 32 percent. 
Synthesis of previous research entitled "What Not To Expect from Crosswalk Signals." [1976]^{(69)} 
Not available. 
Not available. 
Hauer [1988]^{(15)} 
Pedestrian accidents may be reduced by approximately 50 percent by marking crosswalks. 
Synthesis of previous research conducted by Untermann. [1984]^{(70)} 
Not available. 
Not available. 
Hauer [1988]^{(15)} 
Painted crosswalks reduced violation of the pedestrian's right of way. 
Synthesis of previous research cited in Traffic Engineering Handbook. [1965]^{(71)} 
Not available. 
Not available. 

Smith and Knoblauch [1987]^{(53)} 
Accident analyses suggest use of crosswalks at all signalized intersections. 
Comparative. 
Not available. 
Intersections with marked and unmarked crosswalks. 

Herms [1970]^{(55)} 
More pedestrian accidents occurred in marked crosswalks than in unmarked crosswalks by a ratio of about 6 to 1. A crosswalk usage count showed the crosswalk use ratio was approximately 3 to 1, marked vs. unmarked. In terms of usage, approximately twice as many pedestrian accidents occurred in marked crosswalks as compared to unmarked crosswalks. 
Comparative. 
400 unsignalized intersections in San Diego, California. 
Each intersection had one marked and one unmarked crosswalk crossing the major flow of traffic. 
Posted speed limits on approaches 
None found. 




Advance warning signs 
Gattis and Iqbal [1994]^{(56)} 
The "Do Not Block the Intersection" sign was found to be ineffective in preventing drivers from blocking intersections. 
Before/After. 
Four intersections. 
Two street intersections and two commercial driveway intersections. 
Klugman et al. [1992]^{(58)} 
Accident rates for intersections equipped with advance warning signs with flashers (AWFs) decreased from 1.22 to 1.09 accidents per million entering vehicles for all accident types and decreased from 0.68 to 0.63 for rightangle and rearend accidents. 
Comparative and Before/After Study. 
14 intersections. 
Signalized intersections. 
Pant and Huang [1992]^{(57)} 
• On tangent approaches to intersections, the Prepare to Stop When Flashing (PTSWF) sign showed no significant increase or decrease in conflict rate.
• On curved approaches to intersections, the PTSWF sign increased the conflict rate by at least 15 percent.
• The Flashing Symbolic Signal Ahead (FSSA) sign showed no effect on vehicle conflict rates.

Before/After. 
Not available. 
Highspeed signalized intersections on rural or suburban highways in Ohio. 
Washington et al. [1991]^{(36)} 
Implementation of AWFs can reduce approach accident rates at highspeed isolated signalized intersections by as much as 50 percent. 
Comparative. 
40 signalized intersections in California. 
Highspeed isolated signalized intersections. 

Washington et al. [1991]^{(36)} 
• Rightangle accidents were significantly reduced with the presence of route markers and/or advance warning signs.
• Accident rates increased on horizontal approaches with a skew that contained advance warning signs.

Synthesis of prior research conducted by Van Maren et al. [1980]^{(47)} 
Not available. 
Not available. 

Styles et al. [1982]^{(59)} 
Red Signal Ahead signs reduced:
• Rightangle accident rates by 42 percent on approaches with sharp vertical crests.
• Total accident rates on horizontal curve approaches by 14 percent.
• Total accident rates on flat approaches by 41 percent.

Before/After. 
20 intersections. 
Highspeed signalized intersection approaches. 
Advance warning signs (continued) 
Styles et al. [1982]^{(60)} 
All intersections showed a reduction in rightangle accidents with the implementation of a flashing red strobe light. One of the intersections showed reductions in rightangle, rearend, and total accidents of 83 percent, 60 percent, and 61 percent, respectively. 
Before/After. 
Four intersections in Maryland. 
Highspeed signalized intersection approaches. 
Lighting 
Bauer and Harwood [1996]^{(20)} 
At rural, fourleg, STOPcontrolled intersections, lighted intersections had 21 percent fewer total and injury accidents than unlighted intersections. However, no similar effect was observed for total intersection accidents, and an effect in the opposite direction, indicating that lighted intersections had more accidents than unlighted intersections, was observed for urban fourleg STOPcontrolled intersections. These results were based on accidents for all times of day (daytime plus nighttime). 
Statistical modeling with negative binomial regression. 
2,262 rural fourleg STOPcontrolled intersections and 1,551 urban fourleg STOPcontrolled intersections. 
Atgrade intersections on California state highways. 
Box [1989]^{(61)} 
Lighting improvements along a suburban arterial street were found to reduce the percentage of total intersection accidents that occur at night as follows:
Accident Percent reduction
• Pedestrian/bicycle 50 to 33
• Fixed object 57 to 25
• Sideswipe 32 to 11
• Other 33 to 25
Only nighttime headon accidents increased as a proportion of total headon accidents from 33 percent to 43 percent. There is no indication whether any of the observed changes in nighttime accident proportions are statistically significant.

Before/after comparison of the proportion of nighttime accidents. 
Two years of accident data before and Two years after the improvement. 
One 2.8mi suburban arterial highway in Naperville (suburban Chicago, Illinois). 
Lighting (continued) 
City of Los Angeles [1980]^{(62)} 
Found no statistically significant reduction in nighttime accidents due to lighting improvements at intersections. Significant reductions in nighttime accidents were found for a few intersections. The assessment of this study is based on review by Keck [1990].^{(70)} 
Before/after evaluation based on regression analysis. 
528 urban intersections; 2 years of data for nighttime accidents and persons injured at the study intersections were used. 
Urban intersections on city streets in Los Angeles, California. 
Traffic Characteristics

Average daily traffic
• Total entering ADT, all approaches

Many studies used total entering ADT as traffic exposure in their studies (e.g., to calculate intersection accident rates). 





Hauer [1988]^{(15)} 
Explicitly examined safety effects of signalized intersections on traffic flow impacts. He concluded that logically sound models require that the accidents be related to the traffic flows to which the colliding vehicles belong and not the sum of the entering volumes. 
Comparative and review of past studies. 
Sample data from past studies and simulated data. 
Mostly urban settings. 

Lau and May [1988]^{(39)} 
The total entering ADT on all approaches combined was found to have a statistically significant relationship to injury accidents at signalized intersections. 
Linear regression model. 
2,488 signalized intersections. Seven years of injury accident data for each intersection. 
Signalized intersections on California state highways. 

Lau and May [1988]^{(40)} 
The total entering ADT on all approaches combined was found to have a statistically significant relationship to injury accidents at unsignalized intersections. 
Linear regression model. 
17,000 unsignalized intersections. Seven years of injury accident data for each intersection. 
Unsignalized intersections on California state highways. 
Average daily traffic
• Entering ADTs for major and minor approaches

Bauer and Harwood [1996]^{(20)} 
Majorroad and crossroad ADT variables were present in all models as significant predictors of accident frequency. The relative effects for majorroad ADT ranged from 1.77 to 2.68, depending on intersection settings (rural/urban, fourleg/threeleg, signalized/STOPcontrolled). The relative effects of minorroad ADT ranged from 1.24 to 1.80. 
Statistical modeling with negative binomial regression. 
14,432 intersections in California. 
Rural and urban, fourleg and threeleg intersections, STOPcontrolled and signalized. Intersection majorroad ADT above 400 veh/day, minorroad ADT above 100 veh/day. 

Lau and May [1988]^{(39)} 
Percentage of total entering traffic on crossroad was found to be a significant factor in predicting injury accidents at signalized intersections. 
CART analysis of residuals from base model. 
2,400 signalized intersections. Seven years of injury accident data for each intersection. 
Signalized intersections on California state highways. 

Lau and May [1988]^{(40)} 
Percentage of total entering traffic on crossroad was found to be a significant factor in predicting injury accidents at unsignalized intersections. 
CART analysis of residuals from base model. 
17,000 unsignalized intersections. Seven years of injury accident data for each intersection. 
Unsignalized intersections on California state highways. 
Turning movements 
Hauer et al. [1988]^{(63)} 
Developed relationships between accident frequency for specific accident types (e.g., left turn) and associated turning volumes. 
Predictive modeling with negative binomial regression. 
Three years of accident data for 145 intersections in Toronto, Ontario, Canada; turning movement data by approach and time of day. 
Urban, fourleg, fixedtime, signalized intersections with twoway traffic on all approaches and no turn restrictions. 
Peak hour approach volumes 
None found 




Vehicle mix / percent trucks 
None found 




Distribution of total entering volume by hour of the day 
None found 




Distribution of approach volume by hour of the day 
None found 




Average approach speed 
None found 




Volume of bicycle traffic 
None found 




Volume of pedestrian traffic 
None found 



