Development of Road Diet Segment and Intersection Crash Modification Factors

Abstract

Road diets can offer potential safety improvements for both pedestrians and vehicles. The additional space provided by reducing the number of vehicular through lanes can be reallocated to other uses such as bicycle lanes, parking, sidewalks, transit use, turn lanes, curb extensions, parklets, or pedestrian refuge islands. This study evaluated the safety effectiveness of road diets in Virginia using the empirical Bayes method, focusing on the common road diet conversion from a four-lane roadway to a three-lane roadway with added bike lanes. A total of 36 segment sites and 39 intersections were identified in Virginia where road diet installations were implemented between the years 2009 to 2018. The analysis showed segment crash modification factors (CMFs) of 0.62 for total crashes and 0.36 for fatal and injury (FI) crashes. Across all intersection types, the CMFs were 0.65 for total crashes and 0.54 for FI crashes. All CMFs were found to be statistically significant at a 95% confidence level. When intersections were separated into signalized and unsignalized intersections, no significant safety benefit was found for unsignalized intersections. Based on the results, it was concluded that road diets can potentially reduce crashes, and public agencies should consider the safety benefits of road diets when justifying roadway improvements.

Keywords

safety safety performance and analysis before-and-after safety studies crash analysis crash data crash modification factors (CMFs)general transportation safety management systems countermeasure selection network screening safe systems safety planning

A road diet is a technique that reconfigures the roadway by reducing the number of vehicular lanes in hopes of achieving traffic calming and improving safety. Common road diets involve changing a roadway from a four-lane undivided roadway to a three-lane roadway with a center two-way turn lane ( 1 ). On a typical four-lane roadway, drivers change lanes to pass slower moving vehicles (vehicles wanting to make a left turn, etc.). With a road diet, drivers’ speed on two-lane roadways is limited by the vehicle in front of them. This reduction in speed and vehicle interactions during lane changes could potentially reduce the number and severity of vehicle crashes ( 2 ). The flow of traffic becomes more uniform since vehicles wanting to make a left turn can transition to the center lane without disrupting the flow of traffic. The Federal Highway Administration (FHWA) reported that pedestrians might also benefit from road diets because vehicles are likely to be moving more slowly and because there are fewer lanes of traffic to cross ( 3 ). More specifically, they found that pedestrian crash risk was reduced when pedestrians crossed two- and three-lane roads compared with roads with four or more lanes. Likewise, bicycle safety could be improved if the roadway reconfiguration allows for marked bicycle lanes to be installed.

Literature Review

Several studies have evaluated the safety effects of road diets using various methods such as naive before-and-after studies, before–after studies with comparison sites, the empirical Bayes (EB) method, and fully Bayesian (FB) methods. This section reports on previous research that has studied road diets using these various methods and will discuss the research gap that this study sought to address.

Several studies used a simple before-and-after method to analyze the safety effects of a road diet. Knapp and Giese relied on analyzing the before–after conditions at existing road diet projects and documented several benefits of road diet installations ( 4 ). They found a crash modification factor (CMF) of 0.71 for road diet conversions. The results showed that the 85th-percentile speeds were generally reduced by less than 5 mph, and there was a significant decrease in speeds above the posted speed limit. The authors also noted a 17% to 62% reduction in total crashes. It was recommended that a road diet conversion could be considered feasible for roads with average daily traffic between 15,000 and 17,500 vehicles per day (vpd). However, the study only evaluated side street volumes at 40% of mainline volumes using simulation scenarios. This limited variation in volume distribution does not provide adequate consideration of potential sites where road diets may be beneficial or even feasible. The use of a simple before-and-after study is not statistically robust, however, as it assumes that all crash changes are attributable to the treatment being examined and does not correct for other external factors or regression to the mean bias.

Other studies analyzing road diets have used comparison sites by taking the before-and-after period of the treatment site and comparing it with the before-and-after period of the untreated comparison site. An example of this would be the study conducted by FHWA in which road diets were evaluated in Bellevue, WA and various urban cities throughout California ( 5 ). They found that crash frequencies at road diets were 6% lower than the corresponding comparison sites. They also found that crash rates did not change significantly from the before-period to the after-period. The treatment sites did not perform better or worse relative to the comparison sites, despite it having lower crash rates. While comparison sites offer a more promising analysis than a simple before-and-after study, it has the disadvantage of assuming that the untreated sites are perfect indicators of the impact of all other externalities on the treatment site.

Studies that use the EB method have also found some promising safety results from road diets. A study in Minnesota conducted EB analysis for seven sites where road diets were implemented ( 6 ). They reported crash reductions between 37.3% and 54.3% with an overall total crash reduction of 44.3%. This study also looked at collision types and injury severity using a grouped comparison procedure, which showed a net reduction of 45.7% for noninjury crashes and a net reduction of 37% for right-angle crashes. Another study that used the EB method was based in Louisiana, where they modeled the effects of road diet conversions to three- and five-lane highways ( 7 ). Their results produced segment CMFs of 0.61 and 0.70 for the three- and five-lane highways, respectively. They also conducted an analysis for both segment and intersection crashes combined, which yielded CMF values of 0.69 and 0.76 for the three- and five-lane highways, respectively. Although this research yielded an individual segment CMF, it did not provide a separate intersection CMF, which may be useful when evaluating future road diet safety effects at intersections.

Other studies have used the FB approach when conducting road diet studies. A case study in Iowa evaluated 30 sites (15 treatment and 15 comparison sites) over a 23-year period ( 8 ). Their research objective was to assess whether road diets resulted in crash reductions on Iowa roads. The sites were located in smaller urbanized areas where the volumes of vehicles ranged from 2,030 to 15,350 vpd. The study concluded that there was a 25.2% reduction in crash frequency per mile and an 18.8% reduction in crash rate after the road diet conversion was made. They reported that their results produced a much higher reduction in crash frequency per mile than a previously publicized study that reported only a 6% reduction in crash frequency per mile. Persaud et al. conducted a study comparing both the EB and FB methods and found that the estimated safety effects from the two approaches were comparable ( 9 ). The EB method found the CMF result to be 0.53 whereas the FB method found the CMF to be 0.55.

Based on the literature review of the various studies that have used different methods, the CMF values for road diets range from 0.51 to 0.94. Although several studies have examined the safety effectiveness of road diets, several uncertainties remain. First, there is a great deal of variation in the documented safety effects. This can be attributed to the range of road diet conditions evaluated and the varying robustness of the safety evaluation methods used. Secondly, several of these studies have only generated overall CMFs and have not created separate CMFs for segments and intersections. It was anticipated that intersection CMFs would be critical when evaluating future road diet safety effects, given the complex interactions and conflicts present at these locations. This research addresses this gap in the literature by providing separate CMFs for segments and intersections.

Objectives and Scope

The objectives of this project were to develop segment and intersection CMFs for road diets using data from 15 road diet conversion projects in Virginia. CMFs were developed for signalized and unsignalized intersections separately to determine whether there were differential safety effects. CMFs were also developed for total crashes and fatal and injury (FI) crashes for each facility type. This was undertaken to provide the required CMFs to assess future road diet conversion projects using the Highway Safety Manual (HSM) predictive method ( 10 ). Pedestrian and bicycle crashes were not examined separately owing to the low volume of pedestrian and bicycle crash data available.

Methodology

To address the above research gaps, this study applied the EB method recommended by the HSM ( 10 ). The EB method is an observational before–after study that examines the safety impact of installed countermeasures. It accounts for regression to the mean bias as well as changes in traffic characteristics by combining safety performance function (SPF) estimates with the observed crashes ( 11 ). The sites evaluated in this study are all classified as urban and suburban roadways.

Data Collection and Site Characteristics

The Virginia Department of Transportation (Virginia DOT) provided a list of road diet conversions that have been implemented in Virginia. This study focused on the most common design of road diet in Virginia, which is when roadways are converted from a four- to a three-lane roadway with a two-way left-turn lane and added bike lanes or buffered bike lanes. Virginia DOT identified 15 road diet conversions in the state, which had installation dates ranging from 2009 to 2018. Crash, geometric, and annual average daily traffic (AADT) data were collected from Virginia DOT databases for the study sections.

Table 1 summarizes the characteristics of the 15 road diet installations in Virginia that were examined. Mainline traffic volumes before road diet installation ranged from about 2,600 to 17,000 vpd. Minor roads for the intersections along the road diet installations had an AADT that varied from about 325 to 13,000 vpd. Table 1 also denotes the lane configuration that was present before and after installation at the sites, along with the number and type of intersections.

Table 1.

Virginia Road Diet Installations

Locality	Roadway	Length (mi)	Cross section before*	Cross section after*	Install year	Avg. before AADT	Avg. after AADT	Number of signalized intersections	Number of unsignalized intersections	Avg. crashes/year before	Avg. crashes/year after
Alexandria	King St.	1.1	AAAA	BuATABu	2016	13,811	12,388	1	1	2	2.33
Danville	W. Main/Main St.	0.4	AAAA	BATAB, BAABP	2018	8,162	7,962	3	0	2	6
Norfolk	E. Ocean View Ave.	1.6	AAAA	BuATABu	2018	16,988	17,897	1	0	2	0
Williamsburg	Second St.	0.4	AAAA	BATAB	2018	12,942	12,948	1	0	7	3
Fairfax	Armistead Rd.	0.2	AAAA	BuATABu	2018	8,394	8,371	2	1	1.33	3
Fairfax	Bluemont Way	0.3	AAAA	BuATABu	2018	9,054	7,777	1	0	0	0
Fairfax.	Colts Neck Rd.	0.8	AAAA	BATAB	2017	8,786	8,733	3	0	6.33	4
Fairfax	Commerce St.	0.5	AATAA	BuATABu	2017	10,819	12,597	2	1	5.33	2
Fairfax	Fountain Dr.	0.4	AAAA	BuATABu	2018	6,526	6,526	2	1	3	4
Fairfax	Greensboro Dr.	0.5	AAAA	BATAB	2015	14,736	16,433	2	1	5.66	3
Fairfax	Kingstowne Village Pkwy.	1.3	AAAA	BATAB	2015	4,928	4,942	2	1	2.33	4
Fairfax	Lawyers Rd.	2.0	AAAA	BATAB	2009	9,947	8,767	2	4	13.67	7
Fairfax	Oak St.	0.1	AAAA	BATAB	2013	3,109	3,401	1	1	0.33	0.33
Fairfax	Ridge Top Rd.	0.5	AAAA, AATAAP	BuATABu, BuATABuP	2018	4,844	3,897	1	2	1.67	4
Fairfax	Westbranch Dr.	0.3	AATAA	BuATABu	2015	2,654	2,735	0	2	3	4

Note: *A = automobile/general purpose travel lane; B = standard bicycle lane; Bu = buffered or separated bicycle lane; P = parking lane; T = one-way or two-way left-turn lane; Avg. = average; AADT = annual average daily traffic.

Since Virginia DOT has regionally developed SPFs, each of the sites in Virginia were matched to one of three regions based on the applicable SPFs. These regions were termed Region 1, Region 10, and Region 100. Region 1 is classified as the northern region of Virginia. Region 10 is the western region of Virginia. Region 100 covers the eastern region of Virginia. The following is a summary of the regions:

Region 1: Sites in Alexandria and Fairfax County,

Region 10: Sites in Danville, and

Region 100: Sites in Norfolk and Williamsburg.

Next, the roadways within the defined project limits were divided into individual homogenous segments. Chapter 12 of the HSM defines homogeneous segments for urban and suburban arterials as segments in which all of the following characteristics are uniform: AADT, number of through lanes, presence of medians, presence of on-street parking, roadside fixed-object density, presence of lighting, and speed category ( 10 ). Thus, each site was divided into a series of homogeneous segments based on these factors. Crashes that were within 250 ft of an intersection were considered intersection crashes. All other crashes that were outside the 250 ft range of an intersection were considered segment crashes in this analysis. In some cases, minor roads at intersections lacked traffic volume data. When this occurred, those sites were removed from the analysis. Following this process, a total of 36 segment sites and 39 intersections were identified from the 15 road diet conversions in Virginia as having complete data for CMF development.

Safety Performance Functions and CMFs for Urban and Suburban Arterials

The EB method utilizes the observed before-and-after crashes and applies SPFs to determine the expected number of crashes for the after period ( 12 ). All the sites in this study were classified as urban or suburban arterials. All sites were originally a four-lane street that was converted into a three-lane roadway with a center two-way left-turn lane. Virginia DOT has previously adopted Virginia-specific SPFs for segments and intersections on urban and suburban roadways, and these were applied in this study along with the relevant CMFs for segments and intersections ( 12 , 13 ). The EB method allows for such crash-related attributes to be controlled through the use of these relevant CMFs to obtain an adjusted SPF value. This adjusted SPF is also known as the adjusted predicted crash frequency. Equation 1 shows the formula that was used to obtain the adjusted predicted crash frequency after applying the Virginia-specific SPFs and CMFs.

N_{predicted} = Y x N_{spf} x (CM F_{1 x} x CM F_{2 x} x \dots . . x CM F_{yx})

(1)

where

Y = Virginia-specific annual calibration factors developed by Virginia DOT for each region,

$N_{predicted}$ = adjusted predicted crash frequency at the site,

$N_{spf}$ = safety performance functions, and

$CM F_{yx}$ = crash modification factors.

Segment SPFs and CMFs

All the segment sites assessed were classified as either urban multilane undivided or multilane divided arterial segments. The two sites that had multilane divided arterial segments were in Kingstowne Village Parkway and Lawyers Road in Fairfax County. A summary of the Virginia SPFs for urban and suburban segments are shown using Equation 2. Regional models were not developed for FI crashes owing to limited sample sizes and an acceptable fit of the regional models, therefore, Virginia statewide models were used for the FI portion of the segment crash analysis.

SPF = e^{α} {* AADT}_{Two Directions}^{β} * Segment Lengt h_{Centerline}

(2)

Total crashes

1. Urban multilane undivided arterial segments:

Region 1: α = −6.89, overdispersion factor (k) = 5.23

Region 10: α = −10.97, $β = 1.27$ , overdispersion factor (k) = 5.23

Region 100: α = −6.89, $β = . 84$ , overdispersion factor (k) = 5.23

2. Urban multilane divided arterial segments:

Region 1: α = −7.28, $β = . 89$ , overdispersion factor (k) = 3.54

Region 10: α = −10.70, $β = 1.22$ , overdispersion factor (k) = 3.54

Region 100: α = −5.97, $β = . 70$ , overdispersion factor (k) = 3.54

FI crashes

1. Virgina statewide urban multilane undivided arterial segments:

α = −10.36, $β = 1.09$ , overdispersion factor (k) = 4.25

2. Virginia statewide urban multilane divided arterial segments:

α = −10.19, $β = 1.06$ , overdispersion factor (k) = 3.40

Relevant CMFs applied for the segment analysis were on-street parking, roadside fixed objects, median width, and lighting ( 10 ). The automated speed enforcement CMF was not applied in this study since there was no automated speed enforcement present at any of the sites. The following is a summary of the segment CMFs and their base conditions:

1. On-street parking = $1 + p_{pk} x (f_{pk} - 1.0)$

Base condition: Absence of on-street parking.

2. Roadside fixed objects = $f_{offset} x D_{fo} x p_{fo} + (1.0 - p_{fo})$

Base condition: Absence of fixed objects.

3. Median width = Table 12-22 of HSM

Base condition: Absence of medians or median width = 15 ft

4. Lighting = $1.0 - (p_{nr} x (1.0 - . 72 x p_{inr} - . 83 x p_{pnr})$

Base condition: Absence of roadway lighting

where

p_pk = proportion of curb length with on-street parking,

f_pk = factor from Table 12–19 of the HSM,

f_offset = fixed-object offset factor from Table 12–20 of the HSM,

D_fo = fixed-object density (fixed objects/mi),

p_fo = fixed-object collisions as a proportion of total crashes from Table 12–21 of the HSM,

p_inr = portion of total nighttime crashes for unlighted roadway segments that involve a fatality or injury,

p_pnr = proportion of total nighttime crashes for unlighted roadway segments that involve property damage only, and

p_nr = proportion of total crashes for unlighted roadway segments that occur at night.

Summary statistics for the segments are presented in Table 2. These data reflect the number of segment sites analyzed for which data were available.

Table 2.

Segment Summary Statistics

Locality	Roadway	Number of segments	Avg. segment length (mi)	Number of segments with on-street parking	Avg. fixed-object density, D_fo**	Number of segments with medians/Avg. median width***	Number of segments with roadway lighting
Alexandria	King St.	6	0 .072	0	42.24	0	0
Danville	W. Main/Main St.	3	0.037	0	28.16	0	0
Norfolk	E. Ocean View Ave.	3	0.088	2	84.48	0	0
Williamsburg	Second St.	2	0.134	0	84.48	0	0
Fairfax	Armistead Rd.	1	0.02	0	42.24	0	0
Fairfax	Bluemont Way	1	0.003*	0	22	0	0
Fairfax	Colts Neck Rd.	2	0.331	0	0	0	1
Fairfax	Commerce St.	1	0.367	0	84.48	0	0
Fairfax	Fountain Dr.	2	0.115	0	0	0	0
Fairfax	Greensboro Dr.	1	0.199	0	42.24	0	0
Fairfax	Kingstowne Village Pkwy.	5	0.222	0	17.6	3, 10 ft–15 ft	0
Fairfax	Lawyers Rd.	5	0.266	0	84.48	0	5
Fairfax	Oak St.	1	0.04	0	84.48	0	0
Fairfax	Ridge Top Rd.	2	0.1603	0	84.48	0	0
Fairfax	Westbranch Dr.	1	0.158	0	22	1, 10 ft	0
Total		36		2		4, 10 ft–15 ft	6

Note: Avg. = Average.

Small value owing to intersections being near to the 250 ft intersection crash radius.

Values for D_fo (fixed objects/mi) are based on distance between utility poles for both sides of the road combined.

***

First value represents number of segments with medians and second value represents the range of the median widths.

Intersection SPFs and CMFs

Virginia-specific SPFs were also used for the intersection analysis ( 13 ). The format of the SPF is shown in Equation 3,

N_{predicted} = Y x e^{α} x Major AAD T^{β 1} x Minor AAD T^{β 2}

(3)

where

$N_{predicted}$ = predicted crash frequency per intersection per year,

Y = Virginia-specific annual calibration factors developed by Virginia DOT for each region,

Major AADT = AADT on the major road (vpd),

Minor AADT = AADT on the minor road (vpd),

α = intercept,

β1 = coefficient of major AADT, and

β2 = coefficient of minor AADT.

Table 3 presents the values for the intercept and coefficients for the three regions in Virginia that were classified when collecting the data ( 13 ). It also lists the corresponding intersection type, crash type, and overdispersion factor.

Table 3.

Intersection Safety Performance Function Parameters

Intersection type	Crash type	Intercept	Major β	Minor β	Overdispersion parameter
Region 1
4SG	Total	−8.307	0.752	0.328	0.212
4ST	Total	−6.059	0.505	0.277	0.511
3SG	Total	−4.999	0.556	0.155	0.393
3ST	Total	−5.361	0.496	0.168	0.497
4SG	FI	−9.655	0.760	0.360	0.206
4ST	FI	−8.144	0.581	0.319	0.735
3SG	FI	−7.398	0.650	0.209	0.415
3ST	FI	−7.688	0.620	0.181	0.523
Region 10
4SG	Total	−12.391	1.063	0.457	0.162
4ST	Total	−6.703	0.399	0.500	0.193
3SG	Total	−9.614	0.868	0.330	0.372
3ST	Total	−5.773	0.488	0.231	0.335
4SG	FI	−11.428	0.866	0.441	0.149
4ST	FI	−8.033	0.345	0.622	0.038
3SG	FI	−11.010	0.908	0.323	0.504
3ST	FI	−6.946	0.510	0.221	0.351
Region 100
4SG	Total	−8.855	0.7825	0.371	0.135
4ST	Total	−6.538	0.469	0.43	0.285
3SG	Total	−6.752	0.616	0.297	0.334
3ST	Total	−5.164	0.434	0.231	0.413
4SG	FI	−9.958	0.748	0.402	0.127
4ST	FI	−8.317	0.525	0.470	0.337
3SG	FI	−7.266	0.5508	0.311	0.337
3ST	FI	−6.988	0.517	0.231	0.446

Note: 4SG = four-way signalized intersection; 4ST = four-way stop-controlled intersection; 3SG = three-way signalized intersection; 3ST = three-way stop-controlled intersection; FI = fatal and injury.

For intersection CMFs, the following geometric design and traffic control features were considered: left-turn lanes, left-turn signal phasing, right-turn lanes, right-turn-on-red, and lighting ( 10 ). The CMFs and their base conditions are summarized as follows:

1. intersection left-turn lanes = Table 12-24 of HSM

Base condition: Absence left-turn lanes.

2. Intersection left-turn signal phasing = Table 12-25 of HSM

Base condition: Permissive left-turn signal phasing.

3. Intersection Right-Turn Lanes = Table 12-26 of HSM

Base condition: Absence of right-turn lanes

4. Right-turn-on-red = $0 . 98^{(n_{prohib .})}$

Base condition: Permitting right-turn-on-red at all approaches.

5. Lighting = $1 - 0.38 x p_{pi}$

Base condition: Absence of intersection lighting.

where $n_{prohib .}$ = number of signalized intersection approaches for which right-turn-on-red is prohibited, and $p_{i}$ = proportion of total crashes for unlighted intersections that occur at night. Values for $p_{i}$ were taken from Table 12–27 of HSM based on the intersection type.

Summary statistics for intersections are given in Table 4. These data reflect the number of intersection sites analyzed for which data were available.

Table 4.

Intersection Summary Statistics*

Locality	Roadway	Number of signalized intersections	Number of unsignalized intersections	Minor road AADT	Number of left-turn lanes	Number of protected left-turn approaches	Number of permissive/protected approaches	Number of right-turn lanes	Number of lighted intersections
Alexandria	King St.	1	1	1500-7300	2	0	1	1	1
Danville	W. Main/Main St.	3	0	990-4300	7	4	4	4	3
Norfolk	E. Ocean View Ave.	1	0	3300-4200	2	0	1	2	1
Williamsburg	Second St.	1	0	11745-13091	2	2	0	2	1
Fairfax	Armistead Rd.	2	1	60-8482	7	4	4	4	3
Fairfax	Bluemont Way	1	0	7777-9110	4	1	3	2	1
Fairfax.	Colts Neck Rd.	3	0	4916-8869	10	0	6	5	3
Fairfax	Commerce St.	2	1	1200-12597	6	3	3	1	3
Fairfax	Fountain Dr.	2	1	3600-6526	7	0	6	5	3
Fairfax	Greensboro Dr.	2	1	3100-16456	7	2	6	2	3
Fairfax	Kingstowne Village Pkwy.	2	1	2200-7546	8	2	6	6	3
Fairfax	Lawyers Rd.	2	4	330-4546	7	0	0	2	4
Fairfax	Oak St.	1	1	240-5855	3	2	2	1	2
Fairfax	Ridge Top Rd.	1	2	2283-5633	7	4	0	3	3
Fairfax	Westbranch Dr.	0	2	2654-2777	2	0	0	0	2
Total		24	15		81	24	42	40	36

Note: AADT = annual average daily traffic.

Sites in Danville on Main St. and in Fairfax County on Armistead Rd. were the only sites that had a right-turn on red signal approach for which a 0.98 CMF was applied.

Empirical Bayes Analysis

The EB method recommended by the HSM was used to develop CMFs for road diet installation ( 10 ). Chapter 12 of the HSM provides a systematic procedure to implement the EB method, which is briefly described here. After obtaining the adjusted predicted crash frequency (N_predicted) using SPFs and applicable CMFs, the EB correction was applied, using the observed crash data to calculate the expected crash frequency for the before period. This was done first by calculating the weight factor as shown in Equation 4 and applying this in Equation 5 to obtain the expected number of crashes.

w = \frac{1}{1 + k x (\sum_{all study years} N_{predicted})}

(4)

N_{expected} = w x N_{predicted} + (1 - w) x N_{observed}

(5)

where

w = weight factor,

k = overdispersion factor for applicable SPF,

$N_{expected}$ = expected crash frequency at the site, and

$N_{observed}$ = observed crash frequency at the site.

Next, the expected crash frequency in the after period without the treatment was computed by taking the values of the expected before period and multiplying these by an adjustment factor. The adjustment factor, r_i, is the ratio of the summation of expected values in the after period divided by the summation of the expected values from the before period. The adjustment factor, r_i and the expected crash frequency, N_(expected,A), of the after period without the treatment are shown in Equations 6 and 7,

r_{i} = \frac{\sum_{predicted, A}}{\sum_{predicted, B}}

(6)

N_{expected, A} = w x N_{expected, B} x r_{i}

(7)

where

r_i = adjustment factor,

$Σ_{predicted, A}$ = summation of the predicted crashes in the after period,

$Σ_{predicted, B}$ = summation of the predicted crashes in the before period,

N_(expected,A) = expected crash frequency in the after period, and

N_(expected,B) = expected crash frequency in the before period.

The overall road diet treatment effectiveness for both total- and FI crashes was determined by calculating a CMF value for each. To obtain the CMF, an odds ratio was computed using Equation 8, which takes the observed crashes of the after data and divides them by the expected crash frequency of the after data. Once the odds ratio value was obtained for the overall sites, the CMF, (also known as the unbiased odds ratio) was computed using Equation 9,

OR' = \frac{\sum_{All Sites} N_{observed, A}}{\sum_{All Sites} N_{expected, A}}

(8)

OR = \frac{OR'}{1 + \frac{\sum_{All Sites} [{(ri)}^{2} x N_{expected, B} x (1 - w_{i})]}{{(\sum_{All Sites} N_{expected, A})}^{2}}}

(9)

To calculate the statistical significance of the results, the standard error of the odds ratio was computed using Equation 10 and the HSM guidelines on significance.

SE (OR) = \sqrt{\frac{O {R^{'}}^{2} [(\frac{1}{(N_{observed, A}}) + \frac{\sum_{All Sites} [{(r_{i})}^{2} x N_{expected, B} x (1 - w_{i})]}{{(\sum_{All Sites} N_{expected, A})}^{2}}]}{1 + \frac{\sum_{All Sites} [{(r_{i})}^{2} x N_{expected, B} x (1 - w_{i})]}{{(\sum_{All Sites} N_{expected, A})}^{2}}}}

(10)

Results and Discussion

Segments

Segment Crash Types

A crash type analysis was conducted to get a better understanding of the crash data and the effects of road diet installation. About 41.4% of the total crashes in the before period were FI crashes, which declined to 28.2% after road diet installations. Figures 1 and 2 show the proportions of each crash type on road segments for the periods before and after road diet installation, respectively. The most frequent crash type in the before period were angle crashes at 35%, followed by rear-end crashes at 27%. The most frequent crash type in the after period were rear-end crashes at 41%, followed by angle crashes at 30%. Rear-end collisions saw a 14-percentage point increase in the proportion of total crashes for the after period, however, the overall frequency of rear-end crashes decreased from the before period. The site on Greensboro Dr. in Fairfax County had the single most angle collisions in the before period (10 crashes) followed by the site on Commerce St. (6 crashes) in Fairfax County. Based on the crash descriptions, these crashes were largely a result of drivers not properly yielding when making a left turn from driveways, or they occurred near the intersections.

Figure 1.

Segment crash types for the before period.

Figure 2.

Segment crash types for the after period.

Segment CMFs

Using the steps described above for the EB analysis, the predicted and expected values were calculated. Figure 3 shows the results of the predicted and expected total and FI crashes for before and after road diet implementation. The expected before crashes for total and FI saw a 50% median quartile of 0.878 and 0.361 crashes, respectively. The expected after crashes for total and FI without treatment both saw a 50% median quartile of 0.25.

Figure 3.

Predicted and expected crashes for segments: (a) Boxplot of segment predicted and expected crashes, and (b) Data of segment predicted and expected crashes.

The values of the expected after crashes were then compared with the observed after period crashes to obtain individual CMF values. The majority of the sites saw a reduction in crashes after the road diets were installed. These results account for the common cross sections for both the bike lanes and the buffered bike lanes. Based on the EB analysis, the recommended segment CMF was 0.62 for total crashes (standard error = 0.13) and 0.36 for FI crashes (standard error = 0.16). The results were significant at a 95% confidence level. The results from this analysis were comparable with the results conducted in Louisiana for which the segment CMF was reported to be 0.61 ( 7 ). Both these results yielded positive CMF values that imply that implementing road diets can lead to an approximate 38% reduction in total crashes and a 64% reduction in FI crashes.

Intersections

Intersection Crash Types

Some 39.5% of the total crashes were FI crashes in the before period, and 45.3% of the total crashes were FI crashes in the after period. Although the proportion of the FI crashes went up by 5.8%, the overall crash frequency decreased in the after period.

Figures 4 and 5 show the proportions of each crash type for the before and after periods, respectively. The most frequent crash type in the before period were angle crashes at 48%, followed by rear-end crashes at 30%. The most frequent crash type in the after period were angle crashes at 59%, followed by rear-end crashes at 23%. Angle collisions saw an 11-percentage point increase in the proportion of total crashes for the after period. Three counties in Fairfax saw the largest numbers of collisions in the before period (each at 11 collisions): they were the road diet installments on Greensboro Dr., Armistead, and Colts Neck Rd. The site on Colts Neck Rd. had the single most angle collisions in the before period (9 collisions) followed by the road diet on West Branch Dr. (8 collisions).

Figure 4.

Intersection crash types for before period.

Figure 5.

Intersection crash types for after period.

Intersection CMFs

The predicted and expected values were also calculated for intersections. Figure 6 shows the results of the predicted and expected total and FI crashes for before and after road diet implementation. The expected before total and FI crashes saw a 50% median quartile of 2.80 and 1.94 crashes, respectively. The expected after total and FI crashes without treatment saw a 50% median quartile of 1.53 and 0.82, respectively.

Figure 6.

Predicted and expected crashes for intersections. (a) Boxplot of intersection predicted and expected crashes. (b) Data of intersection predicted and expected crashes

CMFs were calculated for all intersections combined, as well as distinct CMFs for both signalized and unsignalized intersections. The overall CMF for intersections showed a positive result, with CMFs for total crashes being 0.65 and FI crashes being 0.54. Both of the CMFs were significant at a 95% confidence level. What was more interesting, there appeared to be significantly different safety effects of the road diet at signalized and unsignalized intersections. For signalized intersections, the total crashes CMF was 0.53 whereas the CMF for FI crashes was 0.41. Both of these were significant at a 95% confidence level. Unsignalized intersection CMFs were not statistically significant. The unsignalized intersections had relatively low crash counts owing to their lower volumes, which contributed to large standard errors in the calculation of those CMFs. Given the sample size constraints in this study, it was difficult to say definitively whether there was a large difference in safety effect between the intersection types, but this is a difference that has not been addressed in the literature and merits future study.

The overall intersection CMF in this study was 0.65, which is comparable to the results from the study done in Louisiana for which the reported CMF for both segments and intersections combined was 0.69 ( 7 ). What distinguishes the results of this study from previous research is that the segment and intersection analyses were separated to obtain CMF values that are more precise. Although the reported CMFs from previous literature may provide promising results, knowing the CMF that is specifically tailored for intersections could be critical when evaluating future road diet safety effects. This study also evaluated signalized intersections separately from unsignalized intersections to better assess the safety effectiveness of road diets on intersection approaches. Signalized intersections saw a much more positive CMF result whereas the unsignalized intersection CMF yielded results closer to 1, which was not statistically significant. This could be because of the nature of a stop-controlled intersection, where vehicles are already forced to come to a stop at every approach, thus not seeing much of a difference in safety improvements when compared with the improvements of a signalized intersection, or it may simply be an artifact of the smaller sample size for unsignalized intersections. CMFs were also examined for intersections with marked and unmarked crosswalks. Unfortunately, the sample was insufficient to determine any statistically significant CMFs for these two cases.

Conclusions

This paper quantitatively evaluated the safety effectiveness of implementing road diets on Virginian roads for the common roadway conversion of a four- to a three-lane roadway with two-way left-turn lanes and added bike lanes and buffered bike lanes. A total of 36 segment sites and 39 intersections were examined using the EB method. Based on model results, the segment CMF for total crashes was 0.62 and for FI crashes was 0.36. For intersection CMFs, the total crashes CMF was 0.65 and the FI crashes CMF was 0.54. A significant finding of this study was that safety effects differed between signalized and unsignalized intersections. Signalized intersections experienced large, statistically significant reductions in crashes, whereas unsignalized intersections did not experience a statistically significant change. This finding has not been previously reported in the literature and needs additional study.

A limitation to this study was having to omit some of the intersections within the road diet corridors owing to the lack of minor road AADT data that are necessary to compute intersection SPFs. These sites were simply not included in this study; only the sites that had the necessary AADT for both major and minor approaches were analyzed. This is likely to have disproportionately excluded extremely low volume roads, since those of all routes were most likely to have missing volume data. Another potential limitation was that many of the road diet sites evaluated were relatively short. It is possible that safety results could differ when road diets are deployed on longer corridors or as part of a larger regional effort.

This study has generated several recommendations for future research. First, this project focused on the most common form of road diet—converting a four-lane road to a two-lane road with bike lanes and a two-way left-turn lane. It would be desirable to increase the sample size by including additional cross sections such as roadways that include medians, parking, and widened sidewalks to broaden the scope of road diet safety research. Second, because a good portion of the sites had been implemented recently, it will be important to collect more crash data for the after period to better quantify the road diet safety effects and their relationship with the geometric design and AADT. This requirement was particularly evident in the results obtained for unsignalized intersections. Future research should focus on assessing the safety effects of road diets at unsignalized intersections to determine whether this study’s findings remain stable with the addition of more data and sites.

Footnotes

Acknowledgements

The authors thank the Virginia Department of Transportation for providing the safety data, traffic data, and geometric data used in this study.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: L. Lim, M. Fontaine; data collection: L. Lim; analysis and interpretation of results: L. Lim, M. Fontaine; draft manuscript preparation: L. Lim, M. Fontaine. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Linda Lim

Michael D. Fontaine

References

Huang

H. F.

Stewart

J. R.

Zegeer

C. V.

Evaluation of Lane Reduction “Road Diet” Measures on Crashes and Injuries. Transportation Research Record: Journal of the Transportation Research Board, 2002. 1784: 80–90.

Lyles

R. W.

Siddiqui

M. A.

Taylor

W. C.

Malik

B. Z.

Siviy

Haan

Safety and Operational Analysis of 4-Lane to 3-Lane Conversions (Road Diets) in Michigan. Michigan Department of Transportation, Lansing, 2012.

Neuner

Chandler

Atkinson

Donoughe

Knapp

Welch

Retting

Meekins

Widstrand

Crowe

Road Diet Case Studies. FHWA, Washington, D.C., 2015.

Knapp

Giese

Guidelines for the Conversion of Urban Four-Lane Undivided Roadways to Three-Lane Two-Way Left-Turn Lane Facilities Final Report. Center for Transportation Research and Education, Iowa State University, Ames, 2001.

Federal Highway Administration. Evaluation of Lane Reduction Road Diet Measures on Crashes. FHWA-HRT-10-053. FHWA, Washington, D.C., 2010.

Gates

T. J.

Noyce

D. A.

Talada

Hill

The Safety and Operational Effects of “Road Diet” Conversion in Minnesota. Transportation Research Board 86th Annual Meeting: Compendium of Papers CD-ROM, Vol. TRB#07-1918. Washington, D.C., 2007.

Sun

Rahman

Investigating Safety Impact of Center Line Rumble Strips, Lane Conversion, Roundabout and J-Turn Features on Louisiana Highways. Report No: FHWA/LA.18/597. Louisiana Department of Transportation and Development, Baton Rouge, 2019.

Pawlovich

Carriquiry

Welch

Iowa’s Experience With “Road Diet” Measures: Impacts on Crash Frequencies and Crash Rates Assessed Following a Bayesian Approach. Transportation Research Record: Journal of the Transportation Research Board, 2006. 1953: 163–171.

Persaud

Lana

Lyon

Bhim

Comparison of Empirical Bayes and Full Bayes Approaches for Before-After Road Safety Evaluations. Accident Analysis & Prevention, Vol. 42, No. 1, 2010, pp. 38–43.

10.

AASHTO. Highway Safety Manual. American Association of State Highway and Transportation Officials, Washington, D.C., 2010.

11.

Hauer

Observational Before—After Studies in Road Safety: Estimating the Effect of Highway and Traffic Engineering Measures on Road Safety. Emerald Group Publishing Limited, Bingley, 2008.

12.

Kweon

Y.-J.

Lim

I.-K.

Development of Safety Performance Functions for Multilane Highway and Freeway Segments Maintained by the Virginia Department of Transportation. Virginia Center for Transportation Innovation and Research, Charlottesville, 2014.

13.

Garber

N. J.

Rivera

Safety Performance Functions for Intersections on Highways Maintains by the Virginia Department of Transportation. VTRC 11-CR1. Virginia Transportation Research Council, Charlottesville, 2010.