Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 19
19
before period (for each treatment site). The EB procedure not considered as reliable as CMFs derived from well-designed
also produces an estimate of the variance of . before-after studies, unless they can be corroborated with
7. The estimate of is then summed over all sites in a treat- results from rigorous before-after studies.
ment group of interest and compared with the count of Further discussion of these methods is provided in the
crashes during the after period in that group. The variance appendices that discuss the results of each evaluation in detail
of is also summed over all sites in the strategy group. which can be found online at http://apps.trb.org/TRBNet
8. These parameters (the summation of and its variance) are ProjectDisplay.asp?ProjectID=461. Another resource is a re-
then used, along with the summation of crash counts after cent publication from the Federal Highway Administration
treatment, to estimate an effect of the treatment (CMF). entitled A Guide to Developing Quality Crash Modification
The standard deviation of the CMF is also estimated, which Factors (Gross, Persaud, and Lyon, 2010) that includes more
makes it possible to determine if the CMF is statistically information concerning different methods for developing
different from 1.0 for a specific level of significance. CMFs. The rest of this chapter provides a summary of the re-
sults obtained from each evaluation.
Before-After Analysis Using
the Comparison Group Method Evaluation Summaries
This method does not account for regression to the mean but Installation of Dynamic Signal
can be effective in accounting for other non-treatment effects Warning Flashers
such as those due to trends in crash reporting and changes in
Description of Treatment and Crash Types of Interest
traffic volume. This method can make use of an untreated com-
parison group of sites that are similar to the treatment sites used This analysis examines the safety impacts of installing
to estimate an SPF to account for changes in traffic volume and dynamic signal warning flashers (DSWF) in advance of sig-
temporal trends in crash occurrence. Steps 1 through 3 that nalized intersections. DSWF provides drivers with advance
were discussed for the EB method could potentially be the notice of the phase change. Specifically, the DSWF is linked
same for the comparison group method as well. The departure to the signal, and flashers are actuated when the signal is
from the EB method is that, instead of using steps 4 and 5 to about to change from green to yellow. The flashers are located
estimate the expected number of crashes in the before period, in advance of the intersection and are actuated at a time when
the observed crashes in the before period is used for this esti- the driver would not be able to clear the intersection before
mate. This estimate could be biased if crashes are selected for the onset of the red phase.
treatment because of a randomly high observed crash count. The basic objective was to estimate the change in crashes.
Target crash types considered included the following:
Cross-Sectional Regression Models
· All crash types (all severities);
Cross-sectional studies derive CMFs by comparing the · Rear-end crashes (all severities);
crash statistics from sites with and without the treatment. · Angle crashes (all severities);
If it is possible to find sites that are similar to each other · Fatal and injury crashes (all crash types); and
(apart from having or not having the treatment), then the · Truck-related crashes (all severities).
CMF could be defined as the ratio of the average number of
crashes in sites with the treatment to the average number of The change in crash frequency was analyzed by employing
crashes in the sites without the treatment. In practice, it is multiple methods using data gathered from three states.
very difficult to find sites that are similar to each other, and Appendix A provides the details associated with this evalua-
so regression models are typically used. The state of the art tion along with example photographs.
is to use negative binomial regression models where crash
frequency is the dependent variable and the independent
Data Used
variables may include site characteristics including major and
minor road AADT. The model coefficients are used to derive Departments of transportation in Nevada, Virginia, and
the CMFs. One problem with using cross-sectional models is North Carolina helped identify treatment sites (i.e., inter-
that the differences in crashes between the sites with treatment sections where DSWF had been installed). They also provided
and without treatment may be due to factors that were meas- geometric, traffic volume, and crash data. For Nevada, data
ured and could not be included in the model, factors which from 1994 to 2008 were available, but only a subset of that data
could not be measured, or even factors that are unknown. was used in order to avoid any major construction activity.
Hence, at this time, the CMFs from cross-sectional models are For Virginia, data from 1998 to 2008 were available, and again
OCR for page 20
20
only a subset was used to avoid construction activity. For DSWF had been installed, and another group consisted of sites
North Carolina, data from 1993 to 2009 were used. in Nevada or Virginia where DSWF had not been installed.
With the North Carolina data, the DSWF were installed at
intersections which were already signalized. Consequently,
Methodology
the problem of separating the effects of signal installation and
With respect to all the treatment sites in Nevada and most DSWF installation was not present, and the state-of-the-art
of the treatment sites in Virginia, it was discovered that the EB before-after method was used. The treatment group con-
traffic signals and DSWF were installed at the same time. This tained 14 sites, 1,000 total crashes in the before period, and
observation had an important implication on the selection of 256 total crashes in the after period. The reference group
an analytical method for this analysis. Since it would be diffi- consisted of 63 signalized intersections in North Carolina
cult to separate the effects of the signal installation from the with 5,948 total crashes.
effects of the DSWF installation, using before-after methods
(e.g., comparison group method or EB method) for those sites
Results
would be problematic. In contrast, all of the treatment sites in
North Carolina were already signalized when the DSWF were The evaluation of DSWF utilized three analysis methods:
installed. Therefore, a before-after method could be employed cross-sectional, before-after with comparison group, and
with the North Carolina data without difficulty. Because of before-after with EB. The cross-sectional analyses for Nevada,
the issue with the timeframe for signal installations, a single Virginia, and the two states combined, show a consistent
method could not be employed for all three states. Instead, reduction in total crashes at intersections that had DSWF. The
three methods were used: cross-sectional analysis, before-after results from the before-after analyses validated these findings.
with comparison group, and EB before-after. The results also suggest that DSWF may help to reduce angle,
With respect to the Nevada data, a cross-sectional method injury, and heavy vehicle crashes, although the sample size was
was employed using two groups of sites: one group consist- limited for many of the individual crash types. It was possible
ing of signalized intersections where DSWF were present and to investigate both angle and injury crashes using all three
another group consisting of signalized intersections where methods and the results consistently indicated a reduction in
DSWF were not present. In all, 261 site-years and 3,224 total expected crashes with the presence of DSWF.
crashes were included in this analysis. The results were less consistent for rear-end crashes. The
With respect to the Virginia data, two analytical methods cross-sectional and comparison group analyses were similar,
were employed. A cross-sectional analysis was conducted with indicating a reduction in expected rear-end crashes with the
the Virginia data using two groups of sites: one group of sites presence of DSWF. However, the EB analysis indicated an
consisting of signalized intersections where DSWF were present increase in rear-end crashes. Note again that the cross-sectional
and another group consisting of signalized intersections where and comparison group analyses were based on data from
DSWF were not present. The Virginia cross-sectional analysis Nevada and Virginia, while the EB analysis was based on data
included 452 site-years and 1,201 total crashes. A before-after from North Carolina.
with comparison group method was also employed with the Multiple methods were used in this analysis of DSWF,
Virginia data with the goal of validating the results of the cross- resulting in multiple sets of CMFs. Of the various sets of
sectional analysis. This analysis was possible because, for a sub- CMFs produced in this analysis, the results of the combined
set of the treatment sites in Virginia, the DSWF installations cross-sectional analysis were ultimately deemed to be the most
occurred after the traffic signal installations. reliable. Table 5.1 presents the CMFs from the combined
Another cross-sectional analysis was performed using a cross-sectional analysis, with the respective standard errors. It
dataset which combined the Nevada and Virginia data. As is important to note that the standard errors shown are `ideal'
with the individual state analyses, two groups were defined. standard errors, and the Highway Safety Manual recommends
One group consisted of sites in Nevada or Virginia where that these standard errors be increased by a factor of 2.0 for
Table 5.1. Crash frequency CMFs (and standard errors) by
crash type for installation of DSWF.
Injury
&
Total Crashes Rear-end Angle Fatal Heavy Vehicle
CMF 0.814# 0.792# 0.745# 0.820# 0.956
Standard Error 0.062 0.079 0.086 0.083 0.177
#
Note: Statistically significant at the 0.05 level (based on the ideal standard errors reported in this table)
OCR for page 21
21
CMFs from cross-sectional regression models to account for Table 5.2. Number of sites for
the fact that results from cross-sectional models are not as treatment group.
reliable as those from well-designed before-after studies for
Location Treatment Sites
estimating CMFs. Colorado 3
The results seem to indicate that the dynamic signal warning Florida 1
Indiana 3
flashers do provide a benefit with the largest percent reduction
Maryland 2
in angle crashes. The relatively large reduction in fatal and Michigan 2
injury crashes is likely the greatest benefit of the dynamic sig- New York 11
North Carolina 2
nal warning flashers in terms of overall safety. Future research
South Carolina 1
could investigate the safety effects of the many variations of Vermont 1
DSWF including roadside and overhead signs. Washington 2
Total 28
Conversion of Signalized Intersections
to Roundabouts in which treatment sites were identified. Reference sites were
identified in Indiana, North Carolina, and New York. Crash,
Description of Treatment and Crash Types of Interest traffic volume, and geometric data were collected for the ref-
erence group. The data from Indiana and North Carolina were
This analysis examined the safety impacts of converting
signalized intersections to roundabouts. Roundabouts have used to directly calibrate SPFs for the two states. For all other
the potential to reduce both the frequency and severity of locations, the SPFs previously used in NCHRP Project 3-65
crashes compared to a similar signalized intersection. The basic were applied.
objective was to estimate the change in crashes. Target crash In order to investigate the effect of approach speed on safety
types considered included: at the roundabouts, the research team attempted to obtain
data from the different States regarding approach speed and/or
· All crashes (all types and severities); speed limits. Data on approach speeds or speed limits were
· Property damage only crashes (all crash types); and not available before the construction of the roundabouts.
· Fatal and injury crashes (all crash types). Speed limit and/or advisory speed data were obtained for the
`after' condition along the major road for each of the study
The change in total crash frequency was analyzed as well as sites. This was called "associated speed" and was based on the
the changes in different crash severities, recognizing that the approach advisory speed when posted, and when it was not
treatment may have a different level of effect on the various posted, based on the nearest upstream posted speed limit.
severities. Appendix B provides the details associated with
this evaluation. Methodology
The primary analysis methodology used was the EB before-
Data Used after analysis as previously described. The evaluation analyzed
Geometric, traffic volume, and crash data for treatment the effects of the treatment on crash frequencies for different
sites were acquired from the States of Indiana (20032008); crash severities before and after the treatment.
New York (3 years before and after treatment); Washington The EB analysis attempted to develop CMFs by severity
(2001March 2009); Michigan (20002009); and North Car- (i.e., PDO vs. fatal/injury vs. total crashes). The reference sites
olina (19992009) to facilitate the analysis. Data were also from Indiana and North Carolina were used to develop SPFs
obtained from NCHRP Project 3-65 which was published for use in the EB before-after analysis. SPFs developed under
as NCHRP Report 572: Roundabouts in the United States a previous effort (NCHRP Project 3-65) were used for the
(Rodegerdts et al., 2007) where signalized intersections were other locations.
replaced with roundabouts. NCHRP Project 3-65 provided In addition to treatment and reference sites, Indiana pro-
data for 1 site in Florida, 3 sites in Colorado, 1 site in South vided data on additional intersections that were newly con-
Carolina, 2 sites in Maryland, and 1 site in Vermont for this structed as roundabouts. It was not possible to include these
analysis. A total of 28 sites were used in the evaluation (see sites in the before-after analysis because there was no before
Table 5.2). period. Instead, these data were used as part of a cross-sectional
Data for reference sites (i.e., signalized intersections similar analysis employed to compare the safety performance of
to those converted to roundabouts) were sought for use similar signalized intersections and roundabouts.
in developing the SPFs required for the EB methodology. The EB analysis was used to investigate the safety effects of
Unfortunately, such data were difficult to obtain for all states converting signals to roundabouts, but the study was based
OCR for page 22
22
on a relatively small sample size. To further investigate the Table 5.3. Crash frequency CMFs (and standard
treatment, a cross-sectional study was employed, using neg- deviations) by crash severity for converting
ative binomial regression models to analyze a larger sample signalized intersections to roundabouts.
of signalized intersections and roundabouts in Indiana and Condition Severity CMF / CMFunction
New York. The cross-sectional analysis was based on a total All 0.792 (0.050)#
All All 0.00004*AADT + 0.303
of 321 site-years, including 42 signalized intersections and Injury and Fatal 0.342 (0.058)#
26 roundabouts. Several potential confounding factors were 2-lane All 0.809 (0.061)#
Injury and Fatal 0.288 (0.065)#
included in the cross-sectional analysis, including traffic 1-lane All 0.735 (0.086)#
volume, area type, number of approaches, and number of Injury and Fatal 0.451 (0.115)#
Suburban All 0.576 (0.053)#
approach/roundabout lanes. Injury and Fatal 0.259 (0.066)#
Urban All 1.150 (0.093)
Injury and Fatal 0.445 (0.100)#
3 approaches All 1.066 (0.163)
Results Injury and Fatal 0.370 (0.172)#
4 approaches All 0.759 (0.052)#
The data collected and analyzed for this study show a general Injury and Fatal 0.338 (0.061) #
#
safety benefit for converting signalized intersections to round- Note: Statistically significant at the 0.05 level
AADT is total intersection AADT
abouts. The EB before-after analysis indicated a significant *represents a product, i.e., 0.0004*AADT is the product of 0.0004 and AADT
reduction in both total and injury crashes. A disaggregate
analysis was also conducted to identify differential effects based
on specific site characteristics (traffic volume, area type, num- as AADT changes. Specifically, with respect to total crashes,
ber of approaches, number of lanes, and associated speed). the safety benefit of roundabouts appears to decrease as traffic
Regarding the effect on total crashes, the safety benefit of volumes increase. The two analysis methods also show a sub-
roundabouts appears to decrease as traffic volumes increase. stantial and sustained reduction in fatal and injury crashes for
The analysis also suggested that the safety benefit is larger for roundabouts across the range of traffic volumes.
suburban than for urban conversions and for intersections Based on the relative rigor of the EB method and the reason-
with four approaches compared to those with three. There ableness of the results, the recommended CMFs were taken
was no clear pattern regarding the effectiveness of the round- from the EB analysis. Table 5.3 shows the CMFs and CMFunc-
about with regard to `associated speed' (as mentioned earlier, tions as applicable. For total crashes, the overall CMF was 0.792,
associated speed is the posted advisory speed or the nearest but the CMF was found to increase (i.e., approach 1.0) with
upstream posted speed limit on the major road during the increasing AADT, and a CMFunction (0.00004*AADT + 0.303)
`after' period). Perhaps the most apparent and telling result was found to be appropriate. The CMFunction is applicable
of the disaggregate analysis is that the reduction in fatal and between a total intersection AADT of 5,300 and 43,000.
injury crashes is substantial and highly significant in all sce-
narios. This is a result of the basic configuration of a round-
about, where crossing-path and left-turn crashes are physically Increasing the Change Interval
eliminated. Description of Treatment and Crash Types of Interest
While the study team employed the EB method to estimate
the safety effects of converting signals to roundabouts, the study This analysis examined the safety impacts of modifying the
was based on a relatively small sample size. A cross-sectional change interval at signalized intersections. The change interval
analysis, employing negative binomial regression, was con- is the time allocated for the yellow and all red phases for a given
ducted to provide support for the EB analysis. Interaction terms approach. The basic objective was to estimate the change in
were explored during the cross-sectional analysis to further crashes. Target crash types considered included:
investigate the relationship between traffic volume and the
effect of installing roundabouts at signalized intersections. · All crashes (all types and severities);
Interaction terms were significant in several of the cross- · Fatal and injury crashes (all crash types);
sectional models for total crashes, indicating differential effects · Angle crashes (all severities); and
for different volumes. The interaction term was insignificant · Rear-end crashes (all severities).
in the injury-related models, confirming the sustained benefit
across the range of traffic volumes. The change in total crash frequency was analyzed as well as
The results of the cross-sectional analysis are relatively con- the changes in different crash types and severities, recognizing
sistent with, and corroborate, the results of the EB analysis. In that the treatment may have a different level of effect on the
particular, both the EB and cross-sectional analyses indicated various types and severities. Appendix C provides the details
that the effects of the treatment on total crashes may change associated with this evaluation.
OCR for page 23
23
Data Used Table 5.4. Number of sites for treatment and
reference groups.
Geometric, traffic volume, signal timing, and crash data for
both treatment and reference sites were acquired from the Location Treatment Sites Reference Sites
Howard County, MD 2 29
States of California (19922002) and Maryland (19922002) Montgomery County, MD 6 38
to facilitate the analysis. Specifically, data were obtained in San Diego, CA 16 36
California from the cities of San Diego and San Francisco and San Francisco, CA 7 32
Total 31 13 5
in Maryland from the counties of Howard and Montgomery.
The sites include data for two types of signalized intersections:
(1) signalized intersections where the change interval was data were only available for a portion of the study period. These
modified during the study period, and (2) signalized inter- data were combined with the reference sites and data from
sections where the change interval was not modified during the the treatment sites in a cross-sectional analysis to investigate
study period. If there were major changes to the geometry or the individual yellow and all red phases with respect to the
operations during the study period, the sites were excluded. ITE recommended practice.
The EB analysis was used to investigate the safety effects of
Methodology modifications to the total change interval with respect to the
ITE recommended practice. Due to a relatively small sample
The primary analysis methodology used was the EB before- size, it was not possible to investigate the individual yellow and
after analysis as previously described. The evaluation analyzed all red phases with respect to the ITE recommended practice,
the effects of the treatment on crash frequencies for different using the EB method. Instead, a cross-sectional study was
crash types and severities before and after the treatment. employed, using negative binomial regression models to ana-
Specifically, the EB analysis was employed to investigate lyze a larger sample of signalized intersections with various
five specific scenarios. Three scenarios were related to various combinations of yellow and all red phases. The cross-sectional
combinations of increasing the yellow and all red time: analysis was based on a total of 916 site-years where the specific
yellow and all red time were known for each year.
· Increasing both the yellow and all red phases,
· Increasing the all red phase only, and
· Increasing the yellow phase only. Results
In discussing the results, it should be noted that the mod-
Two other scenarios were investigated, comparing the ifications to the yellow and all red time were not equivalent
total change interval to the ITE recommended practice (see for all sites. This applies to both the existing conditions and
Appendix C for a description of the ITE recommended prac- the increase in the yellow and/or all red intervals. For example,
tice). In both cases, the before condition was represented by several of the intersections did not include an all red phase in
signalized intersections where the total change interval was the before condition. For sites where both the yellow and all
less than the ITE recommended practice. The after period was red time were increased, the average increases in the yellow
represented by signalized intersections with the following and all red times were 0.8 seconds (minimum of 0.5 seconds
characteristics: and maximum of 1.6 seconds) and 1.2 seconds (minimum of
1.0 second and maximum of 2.0 seconds), respectively. For
· Total change interval remains less than the ITE recom- sites where only the yellow interval was increased, the increase
mended practice and in yellow time was 1.0 second in all the sites. For sites where
· Total change interval is greater than the ITE recommended only the all red interval was increased, the average increase in
practice. the all red time was 1.1 seconds (minimum of 1.0 second and
maximum of 2.0 seconds). For sites where the total change
The analyses attempted to develop CMFs by severity interval was increased, but still less than the ITE recommended
(i.e., fatal/injury vs. total crashes) and by crash types (i.e., total, practice, the average increase was 0.9 seconds (minimum was
angle, and rear-end) for both States. The before-after analysis 0 seconds and the maximum was 1.5 seconds). For sites where
was based on a total of 31 treatment sites as noted in Table 5.4. the total change interval was increased and exceeded the ITE
Reference sites were identified in each jurisdiction to develop recommended practice, the average increase was 1.6 seconds
SPFs for use in the EB before-after analysis. (minimum was 1.0 second and maximum was 3.0 seconds).
In addition to treatment and reference sites, California and Based on the rigor of the EB method, and the generally
Maryland provided data on additional intersections that were insignificant results of the cross-sectional analysis, the rec-
signalized throughout the entire study period, but signal timing ommended CMFs were taken from the EB analysis.
OCR for page 24
24
Table 5.5. Crash frequency CMFs (and standard errors) by crash type for increasing
the yellow and/or all red interval.
Treatment Crash Severity CMF (S.E. of Average Average Number of
(Number of sites) Type CMF) Increase in Increase in All All Red
Yellow Red Interval Intervals =
Interval (min, max) 0 Before
(min, max) Treatment
All All 0.991 (0.146)
Increase Yellow All Injury & 1.020 (0.156)
and All Red (11 Fatal 0.8 (0.5, 1.6) 1.2 (1.0, 2.0) 11
sites) Rear-end All 1.117 (0.288)
Angle All 0.961 (0.217)
All All 1.141 (0.177)
All Injury & 1.073 (0.216)
Increase Yellow
Fatal 1.0 (1.0, 1.0) -- 1
Only (5 sites)
Rear-end All 0.934 (0.237)
Angle All 1.076 (0.297)
All All 0.798 (0.074)#
All Injury & 0.863 (0.114)
Increase All Red
Fatal -- 1.1 (1.0, 2.0) 10
Only (14 sites)
Rear-end All 0.804 (0.135)
Angle All 0.966 (0.164)
#
Note: Statistically significant at the 0.05 level
The EB before-after analyses indicated a significant reduction indicates the average increase in the respective interval, the
in total, injury, and rear-end crashes under various scenarios. applicable range of values, and the number of sites without an
Specifically, the EB analysis indicated a statistically significant all red phase in the before period. It is important to note that
reduction (at the 0.05 level) in total crashes as a result of the number of sites in this evaluation was limited, and hence
(1) increasing the all red phase only, and (2) increasing the the results should be treated with due caution.
total change interval to be less than the ITE recommended
practice. Injury crashes were significantly reduced as a result Change Left-Turn Phasing
of increasing the total change interval to be less than the ITE (From Permissive to Protected-Permissive)
recommended practice. Rear-end crashes were significantly
reduced as a result of increasing the total change interval to Description of Treatment and Crash Types of Interest
be greater than the ITE recommended practice. The change in The objective was to estimate the general safety effects of
angle crashes was statistically insignificant under all scenarios changing from permissive to protected-permissive phasing at
investigated. signalized intersection approaches. Additionally, a particular
Table 5.5 shows the CMFs and standard errors for total, goal was to investigate the effects on non-left-turn related
injury, rear-end, and angle crashes as they relate to increasing crash types and look at the effects of traffic volume, left-turn
the yellow and/or all red intervals. Table 5.6 shows similar volume, and number of opposing lanes on the estimated
results for increasing the total change interval. Each table also change in crashes.
Table 5.6. Crash frequency CMFs (and standard errors) by crash type for
increasing the change interval.
Treatment Crash Severity CMF (S.E. of Average Increase Number of All
Type CMF) in Total Change Red Intervals = 0
Interval Before
(min, max) Treatment
All All 0.728 (0.077)#
All Injury & 0.662 (0.099)#
Increase Change Interval
Fatal 0.9 (0, 1.5) 11
(< ITE) (12 sites)
Rear-end All 0.848 (0.142)
Angle All 0.840 (0.195)
All All 0.922 (0.089)
All Injury & 0.937 (0.114)
Increase Change Interval
Fatal 1.6 (1.0, 3.0) 10
(> ITE) (15 sites)
Rear-end All 0.643 (0.130)#
Angle All 1.068 (0.156)
#
Note: Statistically significant at the 0.05 level
OCR for page 25
25
The site types of interest were signalized intersections with least one leg of the intersection. All these 19 sites were in urban
left-turn lanes in either urban or rural environments, which areas. The change in phasing was one of the following three
have been converted to protected-permissive for at least part categories:
of the daily operation.
The following crash types were of interest: · From Permissive to Protected-Permissive (12 intersections);
· From Permissive or Protected-Permissive to Protected
· Total crashes; (5 intersections); and
· Injury crashes; · From Protected to Permissive or Protected-Permissive on
· Left-turn crashes; at least 2 legs (2 intersections).
· Left-turn opposing through crashes (crashes involving a
left-turn vehicle and a through vehicle from the oppos- Since the number of intersections in the last two categories
ing approach); and is very limited, results are provided here only for the first
· Rear-end crashes.
category of sites, i.e., for intersections where the phasing was
changed from permissive to protected-permissive phasing in
Appendix D provides the details of this evaluation.
at least one leg of the intersection. All the treatment locations
Data were acquired from the City of Toronto, Canada, and
had a left-turn lane on the major legs of the intersection.
urban areas in North Carolina, for both treated and untreated
Unlike Toronto, crash data by approach were not available
signalized intersections.
in North Carolina without a manual review of crash reports.
So, in North Carolina the analysis was focused at the inter-
Data from Toronto section level.
The City maintains a database of signalized intersections
including many variables related to geometry (e.g., number Methodology
of lanes by type by approach), traffic volumes, and crash data.
Volume and crash data from 1999 to 2007 were collected. The methodology applied was the empirical Bayes (EB)
This database was augmented by querying the crash data before-after study, which was described at the beginning of
for specific crash types and adding left-turn AADTs. A separate this chapter. Further details about the methodology are pro-
database of intersection approaches was also created as it was vided in Appendix D.
desired to evaluate left-turn protection improvements at both A number of SPFs were calibrated as follows:
the intersection-level and approach-level. Intersections at
which only one approach had an improvement in left-turn · SPFs were calibrated separately for Total, Injury, Left-turn,
protection were used for the approach-level analysis. Left-turn-opposing through, and Rear-end crashes.
Treated sites were identified in a two-step process. First, an · SPFs at the intersection-level and approach-level were sep-
electronic file of work orders for signalized intersections was arately developed for Toronto. For North Carolina, SPFs
scanned to identify sites where a change in left-turn phasing were estimated at the intersection-level.
was made. Using this list, a subsequent search of hard copy · For the City of Toronto, separate models were also developed
signal timing reports for these sites identified those where the for intersections without and with one-way roads.
left-turn phasing on at least one approach was changed to either
protected-permissive or fully protected at any time of day.
The group of 59 intersection level and 46 approach level Results
treatment sites represented a range of before and after condi- The results are shown in Tables 5.7 and 5.8. Approach
tions with regard to left-turn phasing options. Hence, sites level results are based on data from Toronto. Intersection
were categorized based on the predominant phasing system. level results are based on data from both Toronto and North
A reference group of untreated signalized intersections
was identified to match the treatment sites based on site
characteristics, including number of approaches, presence of Table 5.7. Approach level
left-turn lanes, and traffic volumes. results (Toronto).
Crash Type CMF (s.e.)
#
All 1.077 (0.037)
Data from North Carolina Injury and Fatal 1.150 (0.056) #
#
LTOPP 0.776 (0.098)
In North Carolina, data were available for 19 four-leg inter- Rear end 1.103 (0.118)
sections that experienced a change in left-turn phasing on at #
Note: Statistically significant at the 0.05 level
OCR for page 26
26
Table 5.8. Intersection level results (Toronto and left-turn phase at the same time. There were a few sites in
North Carolina combined). North Carolina where such combined treatments were imple-
mented, but they were not sufficient to conduct an evaluation.
Crash Type Grouping No. Sites CMF (s.e.)
All All sites 71 1.033 (0.023)
1 treated approach 50 1.085 (0.028) #
>1 treated approach 21 0.945 (0.040) Installation of Flashing Yellow Arrow
Injury and Fatal All sites 71 0.958 (0.037) for Permissive Left Turns
1 treated approach 50 1.005 (0.045)
#
>1 treated approach 21 0.878 (0.062) Description of Treatment and Crash Types of Interest
#
LTOPP All sites 71 0.858 (0.056)
1 treated approach 50 0.919 (0.069) The objective was to evaluate the safety impacts due to the
#
>1 treated approach 21 0.762 (0.088) installation of flashing yellow arrow (FYA) for permissive
Rear end All sites 71 1.063 (0.038)
1 treated approach 50 1.091 (0.046)
# left-turn movements. The intent of the flashing yellow arrow
>1 treated approach 21 1.021 (0.062) is to avoid the confusion for drivers turning left on a permis-
#
Note: Statistically significant at the 0.05 level sive circular green signal indication who may assume that the
left turn has the right of way over opposing traffic, especially
Carolina (all intersections were four-leg). Intersection level under some geometric conditions. The following primary tar-
results are provided for two categories of intersections: inter- get crash types were considered:
sections where only 1 approach was treated and intersections
· Total intersection crashes;
where more than 1 approach was treated. Among the 21 inter-
· Total left-turn crashes; and
sections where more than 1 approach was treated, 17 of them
· Total left-turn crashes from the FYA treated approach
had 2 approaches treated, 2 of them had 3 approaches treated,
and 2 of them had 4 approaches treated. (this crash type was examined in Washington and Oregon,
At both intersection and approach levels, the results indicate but not in North Carolina).
substantial benefits for the target crash type, left-turn oppos-
ing involving a left-turn vehicle and a through vehicle from Appendix E provides the details of this evaluation.
the opposing approach (LTOPP). As expected, the benefit at
the intersection level is greater at intersections where more than Data Used
one approach is treated.
One of the fundamental questions the study was expected The data included 5 locations in Kennewick, Washington,
to answer was the extent to which the decrease in target crashes 34 locations from cities in Oregon, and 16 locations from urban
may be offset by a compensating increase in a non-target crash areas in North Carolina. In Kennewick, FYA was introduced
type such as rear-end. At both the intersection and approach in these five locations between 2004 and 2006. Four of these
levels, there were small percentage increases in rear-end crashes. locations had protected-permissive phasing before FYA was
The actual (rather than percentage) increase in rear-end crashes introduced and one location had permissive phasing before
was of the order of 6075% of the decrease in left-turn oppos- FYA was introduced. The City of Kennewick provided many
ing crashes. Disaggregation of the effects by AADT, either variables related to geometry (e.g., number of lanes by type
total entering or left turn, did not reveal any trend. This may and approach), traffic volumes in the form of major and minor
be because the intersections did not have a wide enough dis- road AADTs, peak hour left-turn movements, and crash data.
tribution of these variables. In Oregon, the city of Beaverton provide data for 15 sites, the
In summary, it may be concluded that in estimating the net city of Gresham provided data for 6 sites, the city of Oregon
safety benefit of left-turn protection, consideration must be City provided data for 3 sites, and the city of Portland provided
given to the increase in non-target crashes as well as the decrease data for 10 sites with FYA. Twenty-four of these locations had
in target crashes. It is recommended that the intersection level protected phasing before the FYA was introduced, 3 of them
results for Toronto and North Carolina be used to refine the had permissive phasing, 3 of them had protected-permissive
current CMF for changing from permissive to protected- phasing, and 4 had prohibited left turns. The four cities
permissive in the Highway Safety Manual. Further research in Oregon were able to provide many variables related to
could investigate the specific safety effects of changing left- geometry (e.g., number of lanes by type by approach) and
turn phasing during particular times of day (e.g., peak versus crash data. Left-turn volumes were not available for the Oregon
off-peak) and days of the week (e.g., weekday versus weekend). locations.
Another area of research is to investigate the effect of combined In North Carolina, in all 16 intersections that were evalu-
left-turn treatments: adding a left-turn lane and changing the ated, flashing yellow arrow (FYA) was introduced in two
OCR for page 27
27
out of the four legs. The changes were divided into the fol- in AADT were done by using an SPF calibrated for Kennewich,
lowing three categories: WA, which had sufficient sites for this purpose, and then
dividing the SPF estimate using the after period AADT by the
· Change from protected phasing to FYA protected-permissive SPF estimate using the before period AADT. The adjustment
in 2 legs of the intersection (5 intersections); for time trends was determined using a group of comparison
· Change from doghouse (conventional protected-permissive) sites by dividing the sum of SPF predictions per year for the
to FYA protected-permissive in 1 leg and from permissive after periods by the sum of SPF predictions per year in the
to FYA protected-permissive in another leg (5 intersections); before period for the comparison group.
and In North Carolina, the state of the art EB method could
· Change from doghouse (conventional protected-permissive) be applied. Safety performance functions were estimated using
to FYA protected-permissive in 2 legs of the intersection a reference group of 49 intersections in North Carolina. Further
(6 intersections). detail about the methodology is provided in Appendix E.
In North Carolina, turning volumes were not available at
Results
the treatment or reference sites. However, data on major
and minor road AADT were available for the treatment and Crash Modification Factors are provided in Table 5.9 for
reference sites. total intersection crashes and total intersection left-turn crashes
(the common crash type investigated in the 3 states). Results
are provided for three categories of changes depending on
Methodology
the left-turn phasing of the converted legs before FYA was
For the cities in Washington and Oregon, data on reference introduced:
sites was limited in most of the jurisdictions, and hence the
EB methodology could not be applied with the required rigor. · Intersections where the converted legs had either permis-
The cities did indicate that the sites were not selected based on sive or protected-permissive phasing in the before period,
crash history, but some evidence of an absence of regression- and at least one of the legs had permissive phasing. This
to-the-mean was still desired. The investigation of potential group includes 9 four-leg intersections (total of 36 legs). A
regression-to-the-mean involved aggregating the crash data total of 20 legs were treated with FYA: 15 of the treated legs
over all treatment sites and plotting the totals for each year had permissive phasing in the before period while 5 of the
before treatment (e.g., 1 year before treatment, 2 years before treated legs had protective-permissive phasing in the before
treatment, 3 years before treatment, etc.). This test was period.
conducted for each city separately, and for each, it was con- · Intersections where the converted legs only had protected-
cluded that there was no evidence for regression-to-the-mean permissive phasing in the before period. This group included
notwithstanding the natural randomness of crash counts. 1 3-leg and 12 4-leg intersections (total of 51 legs). A total
The methodology applied combined some aspects of the EB of 27 legs were treated with FYA; all of them had protected-
and Comparison Group approaches. Adjustments for changes permissive phasing in the before period.
Table 5.9. CMFs and standard errors for flashing yellow arrow
installation.
Left-Turn Phasing Before Crash Type CMF (S.E.)
(sites) (legs treated)
Permissive or combination of Total Intersection Crashes 0.753 (0.094) #
permissive and protected-permissive
(at least 1 converted leg was Total Intersection Left-Turn 0.635 (0.126) #
permissive in the before period) (9 Crashes
sites) (20 legs treated)
Protected-Permissive (all converted Total Intersection Crashes 0.922 (0.104)
legs had protected-permissive in the
before period) (13 sites) (27 legs Total Intersection Left-Turn 0.806 (0.146)
treated) Crashes
#
Protected (all converted legs had Total Intersection Crashes 1.338 (0.097)
protected in the before period) (29
sites) (56 legs treated) Total Intersection Left-Turn 2.242 (0.276) #
Crashes
#
Note: Statistically significant at the 0.05 level
OCR for page 28
28
· Intersections where the converted legs only had protected in signal phasing may have had a more significant impact on
only phasing in the before period. This group included 5 3-leg safety than the change to FYA permissive indication. Collec-
intersections and 24 4-leg intersections (total of 111 legs). tively, these results indicate that the largest benefit can be
A total of 56 legs were treated with FYA; all of them had found at sites where at least one of the converted legs had per-
protected only phasing in the before period. missive only operation before the FYA was implemented with
protected-permissive operation. It is important to note that
Intersections in the first group experienced reductions in the number of sites in the first two groups was limited, and
total intersection crashes and total intersection left-turn crashes hence the individual results should be treated with due caution.
that were statistically significant at the 0.05 level. Intersections Most of the sites had 2 legs that were converted (except for
in the second group experienced a smaller reduction that was the few 3-leg intersections in the sample). So, it was not possi-
not statistically significant at the 0.05 level. As expected, on ble to specifically investigate the relationship between the num-
the basis of individual results and those in Noyce et al. (2007), ber of legs that are treated and the associated safety benefits for
intersections in the third group (with protected only phasing left-turn crashes. This could be an area for future research.
in the before period) experienced significant increases in total Another area of future research is an investigation into the
and left-turn crashes. As Noyce et al. commented, the change effect of left-turn volume and opposing through volume.
