Using trip diaries to mitigate route risk and risky driving behavior among older drivers

G Model

ARTICLE IN PRESS

AAP-4292; No. of Pages 12

Accident Analysis and Prevention xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap

Using trip diaries to mitigate route risk and risky driving behavior among older drivers Rashmi P. Payyanadan a,∗ , Adam Maus a , Fabrizzio A. Sanchez b , John D. Lee a , Lillian Miossi a , Amsale Abera a , Jacob Melvin a , Xufan Wang a a b

Department of Industrial and Systems Engineering, University of Wisconsin-Madison, Madison, WI, 53706, USA Department of Statistics, University of Wisconsin-Madison, Madison, WI, 53706, USA

a r t i c l e

i n f o

Article history: Received 30 April 2016 Received in revised form 29 August 2016 Accepted 22 September 2016 Available online xxx Keywords: Older drivers Trip Diary Self-regulation Route risk Retrospective feedback

a b s t r a c t To reduce exposure to risky and challenging driving situations and prolong mobility and independence, older drivers self-regulate their driving behavior. But self-regulation can be challenging because it depends on drivers’ ability to assess their limitations. Studies using self-reports, survey data, and hazard and risk perception tests have shown that driving behavior feedback can help older drivers assess their limitations and adjust their driving behavior. But only limited work has been conducted in developing feedback technology interventions tailored to meet the information needs of older drivers, and the impact these interventions have in helping older drivers self-monitor their driving behavior and risk outcomes. The vehicles of 33 drivers 65 years and older were instrumented with OBD2 devices. Older drivers were provided access to customized web-based Trip Diaries that delivered post-trip feedback of the routes driven, low-risk route alternatives, and frequency of their risky driving behaviors. Data were recorded over four months, with baseline driving behavior collected for one month. Generalized linear mixed effects regression models assessed the effects of post-trip feedback on the route risk and driving behaviors of older drivers. Results showed that post-trip feedback reduced the estimated route risk of older drivers by 2.9% per week, and reduced their speeding frequency on average by 0.9% per week. Overall, the Trip Diary feedback reduced the expected crash rate from 1 in 6172 trips to 1 in 7173 trips, and the expected speeding frequency from 46% to 39%. Thus providing older drivers with tailored feedback of their driving behavior and crash risk could help them appropriately self-regulate their driving behavior, and improve their crash risk outcomes. © 2016 Elsevier Ltd. All rights reserved.

1. Introduction Driving a route whether driving straight, negotiating a turn, crossing an intersection, or changing lanes can result in a potential crash. The National Highway Traffic Safety Administration (NHTSA) examines FARS (Fatality Analysis Reporting) and GES (General Estimates System) to determine the types of driving conditions, driving behaviors, and road infrastructure that contribute to crashes (Stutts et al., 2009). In 2012, the NHTSA report showed that for two-vehicle crashes, older drivers were 75% more likely to be involved in a crash between 2 p.m. and 6 p.m., and during daylight – attributed to their increased driving exposure during the day. Older drivers are also involved in greater number of intersection and crossing-related crashes (Hakamies-Blomqvist, 1993) as a result of their increased exposure to intersections due to choice of road type, such as the

∗ Corresponding author. E-mail address: [email protected] (R.P. Payyanadan).

preference to avoid highways (Langford and Koppel, 2006). The 2012 NHTSA report showed that left turn crashes were particularly high for older drivers, with 20% of drivers 70–79 years, and 25% of drivers 80 years and older involved in a left turn crash. Whereas crashes related to driving straight, passing and overtaking were the only driving maneuvers not associated with an increase in crash risk among older drivers. This is because of their self-regulatory behavior and avoidance of lane change maneuvers unless they were confident, and deemed it necessary and safe (Stutts et al., 2009). These results suggest that the risk of crash among older drivers under certain driving situations is partly influenced by their driving exposure patterns and exposure reduction strategies, which are critical but also challenging to measure. While FARS and GES provide adequate population exposure for assessing the crash risk of older drivers, the ability to estimate and account for self-regulation strategies as a measure of crash risk is limited. Driving strategies and compensatory mechanisms to reduce crash risk depends on the driver’s perception of their driving ability. Early studies commonly used self-ratings to assess driver’s

http://dx.doi.org/10.1016/j.aap.2016.09.023 0001-4575/© 2016 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model AAP-4292; No. of Pages 12 2

ARTICLE IN PRESS R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

self-regulation strategies. These studies showed that older drivers perceived their risk of accident involvement to be significantly lower than other drivers of their same age group when it came to night driving, and driving on wet and snow covered roads (Finn and Bragg, 1986), perceived themselves less likely to be involved in an accident than younger drivers (18–24 years) (Jonah and Dawson, 1982), and less concerned about impaired driving and less likely to think that impaired driving could result in a crash compared to younger drivers (Wilson, 1984). But current studies have shown strong evidence to support that self-enhancement bias is common among drivers of all age groups (Freund et al., 2005; Horswill et al., 2004; White et al., 2011), and has little correlation with actual onroad driving and simulator performance (De Craen et al., 2011; Freund et al., 2005). Thus, while studies using self-ratings have provided important insights into the attitudes and beliefs of driver’s risk perception; using self-ratings to assess self-regulation driving strategies, compensatory mechanisms, and crash risk reduction is often misleading because most drivers, including older drivers overestimate their driving performance (Freund et al., 2005). Recent studies have used drivers’ hazard detection as an indicator of crash risk. A comparison of the driving hazard detection between younger (17–18 years), experienced (22–30 years) and older (65–72 years) drivers showed that experienced and older drivers better anticipated potentially hazardous situations in their driving environment, and continued to scan the environment for potential hazards after a planned event, compared to younger drivers (Borowsky et al., 2010). But hazard perception has been shown to decline with age. Staplin et al. (2012) showed that deficits related to attention, executive function, range of motion, and spatial abilities, increases the likelihood of crashes among older drivers, with reduced contrast sensitivity and Useful Field of View (UFOV) being strong predictors of hazard perception (Horswill et al., 2008). Yet despite age-related declines, older drivers do not have higher crashes than other age-groups (Hakamies-Blomqvist et al., 2002; Langford et al., 2006). This may be due to the fact that when older drivers are aware of their physiological, cognitive, and functional decline, they tend to drive more cautiously, and avoid risky driving situations, restrict their overall driving, and actively regulate their driving behavior (Marottoli and Richardson, 1998; Molnar and Eby, 2008). But evidence to suggest a positive impact of self-regulation on reducing crash risk is not yet clear (Man-Son-Hing et al., 2007). Studies have shown that older drivers who self-regulated their driving behavior reported fewer crashes than those who did not restrict their driving (Holland and Rabbitt, 1992). Whereas others have shown that when at-risk older drivers self-regulated due to poor UFOV scores, they were still twice as likely to incur at-fault crashes over five years compared to low-risk drivers (Ross et al., 2009). These studies have raised three main concerns: a) that older drivers’ do not realize their decline in driving skills to take the necessary self-regulatory action, b) their ability to self-regulate does not match their decline in driving skills, and c) self-regulation is not effective in reducing their crash risk outcomes. The ineffectiveness of self-regulation to reduce crash risk outcomes have been commonly observed in two situations: when cognitive impairments hinder the ability of older drivers to be truly aware of their driving behavior, and when self-regulation strategies do not generate frequent enough behavioral changes to prevent crash involvement (Owsley et al., 2004). But for older drivers who are unaware of their declining ability, or their self-regulation does not compensate for the particular declining ability – providing targeted feedback of their driving behavior may help increase their self-awareness and knowledge of driving safety needs, and improve their driving outcomes. Results from a study by Eby et al. (2003) showed that older drivers who were given the Driving Decisions Workbook to make

them aware of their decline, reported being more aware of their deficits, and began regulating their driving behavior to improve their driving safety. According to the Multifactorial Model for Enabling Driving Safety, matching older driver’s driving ability to their functional capacity can be attained by providing them with the means to accurately assess and evaluate their decline, and accordingly adapt their driving behavior (Anstey et al., 2005). Holland and Rabbitt (1992) showed that two-thirds of the older drivers who were given results of their eyesight and hearing test, made changes to their driving behavior by avoiding driving at night, being cautious when crossing complex or unfamiliar junctions, avoiding rush hour, planning trips in advance, and choosing safer routes. These studies reveal that while older drivers tend to be more cautious and safe, they may be unaware of the risk-reducing alternatives due to a lack of knowledge of the typical crash profile of older drivers, along with the general tendency to stick to familiar routes. With the emergence of in-vehicle data recorders (IVDR), such technologies are making it possible to collect continuous data on true driving behavior, such as risky driving behavior, engagement in secondary tasks, and driver responses (Dingus et al., 2006); to provide more objective feedback to drivers. The goal of this study is to determine whether using personal Trip Diaries to provide older drivers with feedback of their route choice, driving behavior, and low-risk route alternatives (Payyanadan et al., 2016), can result in a reduction of route risk and risky driving behavior. Understanding the influence of the Trip Diary as a feedback tool on driving safety outcomes can aid in the development of more customizable feedback interventions, and targeted driving safety programs to improve the safety and mobility outcomes of older drivers.

1.1. Trip Diary – a web-based feedback for older drivers A web-based Trip Diary was developed for older drivers (Fig. 1). The Trip Diary provided feedback of the driver’s routes, alternate route options with fewer left turns, U-turns, lane closures, and traffic incidents, and a map with directions of the low-risk route alternatives. The Trip Diary also reported the number of left turns, U-turns, speeding, hard braking, harsh cornering, and hard accelerating events by a driver along a driven route. The events were annotated on the map to provide visual feedback of the areas that the driver had a risky driving behavior. To provide the low-risk route alternative, a route risk measure developed by Payyanadan et al. (2016) was implemented in the Trip Diary. The route risk measure compares the crash risk of the routes driven to those suggested by Google and MapQuest. The route with the lowest crash risk was provided as the low-risk route alternative (Fig. 2). The route risk measure considers the number of left turns, U-turns, traffic incidents, and lane closures along a route with minimal cost to distance relative to the route driven. Left turns, Uturns, travel distance, and alternate route options information were retrieved from the recorded driver data, Google and MapQuest APIs, and the traffic incidents and lane closures for a route were retrieved from the 511 Real Time Traffic Map of the Wisconsin Department of Transportation (with permission and licensing agreement from the Wisconsin Department of Transportation).

2. Methods Older drivers 65 years and above were recruited for the study. OBD2 (on-board diagnostic) devices were installed in each of their cars to collect baseline and treatment data. Baseline period was determined based on the power analyses conducted on data collected from a previous study by Payyanadan et al. (2016).

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model AAP-4292; No. of Pages 12

ARTICLE IN PRESS R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

Fig. 1. A web-based Trip Diary developed for older drivers to provide low-risk route alternatives (route with fewer driving challenges), and driving behavior feedback.

Fig. 2. Low-risk route alternative (black route on map) along with the driving directions.

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

3

G Model AAP-4292; No. of Pages 12

ARTICLE IN PRESS R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

4

2.1. Participants A total of 33 drivers 65 and older were recruited from rural (28%), urban (42%), and suburban (30%) settings from a Midwestern state. Participant’s age ranged from 65 to 82 years, and consisted of 15 males and 18 females. County coordinators working at the Aging and Disability Resource Center (ADRC) in each of these regions helped raise awareness about the study. To participate in the study, older adults were required to hold a valid driver’s license, have internet access, and drive at least twice a week. Older adults who provided their contact information to the ADRC county coordinators with interest to participate in the study were contacted by the research team, and participated in the study if they met the inclusion criteria. The vehicles of the 33 participants were instrumented with OBD2 devices for a period of four months, with baseline data collected for a period of one month. During the baseline period, participants were not given access to their driving information. This was to ensure that their baseline route risk and driving behavior was accurately captured. After a month, participants were given access to a customized web-based Trip Diary for a period of three months – where they could log on and access their trip details. Demographic data along with the trip details for baseline and treatment periods are shown in Table 1. 2.2. Device, installation and sensitivity metrics For the study, Geotab GO6 OBD2 devices were purchased from Sprint. The device involved a simple plug-and-play installation – allowing instant trip data access and notification updates to the Trip Diary page. Once the device was installed, participants were asked to go about their normal driving routine. Data collected by the device for analysis and feedback are shown in Table 2, along with the corresponding sensitivity metrics. The sensitivity metrics of the Geotab GO6 OBD2 device were adjusted for passenger vehicle settings recommended by Geotab. 2.3. Trip Diary use and feedback Once baseline period was completed, participants were contacted for a second visit. During the second visit, participants were given access to their Trip Diary page, along with instructions on how to use and understand their trip feedback information on the Trip Diary page. A Trip Diary instructions booklet was provided with steps and explanation on how to use and interpret their trip feedback. Participants were asked to log in to the Trip Diary page daily or at least 2–3 times a week to access feedback about their trips, driving behavior, and alternate route options. During the second visit, 21% of the participants reported limited experience using the internet. To ensure that all participants were able to receive their trip feedback, reports of their Trip Diary feedback information were developed using R (R Development Core Team, 2013) Sweave package and Latex (Leisch, 2002), and mailed to all the participants every month (Fig. 3). Mailed reports consisted of their monthly trip summary report (Fig. 3A), their driving behavior for the current month and past month, driving behavior relative to other participants in the study for the same month (Fig. 3B), map and directions of low-risk route alternatives for trips driven that month (Fig. 3C), and three feedback questions on familiarity with the route driven and low-risk route alternative, near misses along the route driven, and usefulness of the Trip Diary feedback. In this paper, the feedback responses were not analyzed. Participants averaged 6.3 visits to the online Trip Diary during the treatment portion of the study. Combined with the monthly mailed reports, participants were presented with feedback an average of 9.3 times. For each online visit, participants averaged 19 page

views and 40.6 min of activity on the Trip Diary. For the monthly mailed reports, participants were asked to mail back the responses to the three feedback questions (that were also asked on the webbased Trip Diary page). For each monthly report, all participants mailed back their responses. In this paper, the feedback responses were not analyzed. 2.4. Data recording The Geotab SDK (software development kit) link was used to collect data recorded by the OBD2 device (Table 2) and read into the Trip Diary page. For each trip, a CSV file was generated containing information displayed on the Trip Diary page (Table 3). The Trip Diary page displayed information pertaining to seatbelt use, driving with the ‘Check Engine’ light on, map of route driven, left turn and U-turn count, speeding, hard braking, hard cornering, and hard acceleration events along the route driven, and map and directions of the low-risk alternative route. 2.5. Variables, model assumptions and hypotheses Due to a low occurrence rate of crashes, directly measuring significant changes in crash risk requires tens of thousands of drives observed over a very long time frame. In many cases this type of study is impractical to pursue and implement. To address this problem a route crash risk measure was developed (see Payyanadan et al., 2016) to estimate changes in crash risk. This measure uses route characteristics such as left turns, distance, and road closures to determine crash risk. To understand the level of risk associated with route choice, this measure is also applied to relevant route alternatives that could have been driven in lieu of the chosen route, and the ratio of the risk of the chosen route to the risk of the alternatives is determined. Thus if feedback results in changes in route choice, the change in crash risk can be estimated using the new ratio of chosen routes to their alternatives. In this study, to measure the risk associated with a participant’s route choice, for each trip, crash risk of the route driven was determined using the route risk measure developed by Payyanadan et al. (2016) and compared to the crash risk of the low-risk alternative route as in (1). Route Risk Ratio =

Crash Riskroute driven Crash Risklow−risk alternative route

(1)

For simplicity route risk ratio will simply be referred to as route risk except when explicit reference to the ratio is needed for clarity. Participants’ baseline route risk and time under treatment are considered to be independent variables. Baseline route risk is estimated using the median route risk among trips taken during the baseline period of the study. Time under treatment is defined as the number of days since providing access to the Trip Diary. The dependent variables are the driving behaviors (Table 2) and route risk for trips taken while under treatment. To determine the effects of the Trip Diary feedback on route risk and driving behavior, the following assumptions were made, a) Older drivers’ route risk is a function of their inherent route choice and the amount of time under treatment. b) Older drivers’ driving behavior are a function of their inherent driving behavior and the amount of time under treatment. c) The older driver population has a mean route risk and response to treatment, but individuals within the population may differ from either. d) Older drivers’ response to treatment depends on their inherent route choice or driving behavior.

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model AAP-4292; No. of Pages 12

ARTICLE IN PRESS R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

5

Fig. 3. Dynamic report of participant’s Trip Diary mailed every month. (A) Monthly trip summary report. (B) Comparison of their driving behaviors between the present and past months, comparison of present month to other drivers in the same period, and whether their current driving behavior is better or needed improvement. (C) Trip summaries of alternate routes that were of lower risk along with driving directions.

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model

ARTICLE IN PRESS

AAP-4292; No. of Pages 12

R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

6

Table 1 Mean participant ages and within period driving data from the OBD2 device, grouped by gender. Gender

Males Females

Total

15 18

Age

Baseline period (1 month)

74 71

Treatment period (3 months)

Distance (miles)

Time (minutes)

Distance (miles)

Time (minutes)

7.0 7.3

12.8 13.6

7.9 7.2

13.6 13.4

Table 2 OBD2 device settings and sensitivity metrics. Geotab GO6 OBD2 data

Definition

GPS coordinates Trip start/stop

Latitude and longitude data for location retrieval. Event-based A trip starts when the vehicle starts moving. A stop is recorded when the vehicle ignition is turned off, or when the vehicle has a speed of less than 1 km/h for more than 200 s. Distance travelled for each trip from origin to destination. Miles Time taken to travel for each trip from origin to Seconds destination. Records changes in speed during a trip. m/s2 , Event-based 3-axis accelerometer recordings to determine vehicle Threshold change of 300 milli-G in any acceleration. direction Speed is monitored against the posted road speed. If there 5 mph over the posted speed limit was no data on the posted speed limit for a section of a trip, no speed violation was recorded. G-force exertion set at −0.58 A hard braking incident is recorded when it caused a force of 1/2 G to be exerted on the vehicle. A hard cornering incident is recorded when a hard or G-force exertion set at >0.47 and aggressive turn causes a force greater than 2/5 G to be <−0.47 exerted on the vehicle. G-force exertion set at 0.4 A hard acceleration incident is recorded when it causes a force of 1/3 G to be exerted on the vehicle. A seatbelt violation is recorded when the driver is not wearing a seatbelt while the vehicle is moving faster than 6.21 mph. This information is communicated through the ECM (electronic control unit) of the vehicle. But not all vehicles transmit information about the seatbelt, hence reporting depended on the type of vehicle driven. Identifies vehicles driven with the ‘Check Engine’ light on. A possible accident event is recorded when the accelerometer detects a change in speed of more than 15 mph in 1 s in any direction.

Distance Time Speed Acceleration Speed violation

Hard braking Hard cornering

Hard acceleration Seatbelt violation

Engine light Possible accident

Measure and sensitivity settings for passenger vehicle

Table 3 Trip Diary data displayed to participants during the treatment period. Trip Diary data

Definition

Measure/Format of information displayed

Geotab GO6 OBD2 data Google and MapQuest data Alternate low-risk route directions Left turns and U-turns

Table 2

Count

Google and MapQuest API was used to obtain corresponding route alternatives with lower route risk. Google and MapQuest API was used to obtain corresponding lower risk route directions. A left turn algorithm along with manual assessment of routes was used to determine turn count along a driven routes. For route alternatives through Google and MapQuest, text directions were imported into R to conduct a text search of left turns and U-turns count. Direction terms for types of turns was determined by manually comparing turns along a route to corresponding map directions. Distance travelled for each participant trip was obtained from the Geotab GO6 OBD2 data. Distance for the alternate route was obtained from Google and MapQuest API. Travel time for each trip was obtained from the Geotab GO6 OBD2 data. Travel time for the alternate route was obtained from Google and MapQuest API. 511 Wisconsin Department of Transportation (WisDOT) and Mapquest API. 511 Wisconsin Department of Transportation (WisDOT) and Mapquest API. Probability of a crash based on the weighted combination of route characteristics (left turns, U-turns, distance, traffic incidents, and lane closures) from NHTSA crash fatality report.

Latitude and longitude

Distance

Time Traffic incidents Lane closures Route risk

Based on these assumptions, generalized linear mixed effects regression models were used to analyze the treatment effect on route risk and driving behavior. The model assumptions are, a) An older driver’s route risk is a linear function of their baseline route risk XBR , and the amount of time under treatment Xt (Eq.

Count

Miles

Seconds Count Count Ratio (relative to the low-risk alternative route)

(2)). In Eq. (2), T is the effect of time under treatment, BR is the effect of baseline route risk, and TBR is the effect of the interaction between time under treatment and baseline route risk on the dependent variable. to be random   The Zi are considered   effects, where Zi ∼N 0, ␴2i , i ∈ T, BR and ␧j ∼N 0, ␴2 .

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model

ARTICLE IN PRESS

AAP-4292; No. of Pages 12

R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

Route Risk Ratioj ∼␤0 + Z0 + T · Xt + Zt · Xt + BR · XBR +ZBR · XBR + TBR · Xt · XBR + ␧j

(2)

• There is a mean population route risk and effect of time under treatment, but drivers in the population vary around each mean with constant variance. • The effect of treatment time is dependent upon the baseline route risk, i.e. TBR = / 0. • The ␧k ⊥ ␧j for all k = / j 2.5.1. Treatment effects on route risk Using a Gaussian linear mixed effects regression model with an identity link function, four hypotheses were considered to determine the effects of the Trip Diary feedback on route risk (Table 4). It is assumed that the alternate route options generated by Google and MapQuest are a sufficient subset of all possible routes, such that a driver who takes routes with fewer left turns, U-turns, and avoids challenging events (such as construction zones) will have a route risk ratio near 1. It is unlikely that an individual can consistently find a route that has considerably less risk than the group of routes produced by Google and MapQuest, which would result in a mean risk ratio between 0 and 1. Based on this assumption, the expectation is that an effective treatment will drive participants’ route risk to 1. 2.5.2. Treatment effects on driving behavior For analyzing the treatment effects on driving behavior, a binomial family linear mixed effects regression model with logit link function is used since driving behaviors such as speeding, hard braking, cornering, driving with the engine light on, and seatbelt use were measured by their frequency of occurrence. The model assumptions for driving behavior are the same as the model assumptions for route risk. In Eq. (3), XB is the baseline driving behavior, Xt is the amount of time under treatment, T is the effect of treatment time, B is the effect of baseline driving behavior, and TB is the effect of the interaction between treatment time and baseline driving behavior, all on the dependent variable.   The Zi are  con sidered to be random effects, where Zi ∼N 0, ␴2i and i ∈ T, B .



Prob (Driving behavior outcomes) ∼f ˇ0 + Z0 + T · Xt + Zt · Xt +B · XB + ZB · XB + TB · Xt XB )

(3)

Using the generalized linear mixed effects regression model, two hypotheses were considered to determine the effects of the Trip Diary feedback on driving behavior (Table 5). 3. Results The results are organized into three sections. The first section provides a summary of the trip details, route risk, and driving behavior of older drivers in the study. The second section shows the effects of the Trip Diary feedback on the route risk of older drivers. The final section shows the effects of Trip Diary feedback on the driving behavior outcomes of older drivers. 3.1. Route risk and driving behavior of older drivers A total of 33 older adults 65 and above were recruited for the study. Older drivers’ trips were on average 7.4 miles in length, with an average trip time of 13.4 min, and average speed of 24.6 miles/h. Summary statistics of the driving behavior of older drivers for baseline and treatment are shown in Table 6. These summary measures

7

do not account for the sampling differences between baseline and treatment. Hence total trips and average values for baseline and treatment periods should not be used to determine the efficacy of treatment. The remaining variables in Table 6 are reported as the percent of trips, and their odds ratio was used to report the changes in the percentages for meaningful interpretation of the data. Driving with the engine light on was not recorded on all vehicles and only one participant had a possible accident event. Table 7 shows the summary of the route characteristics and route risk of older driver trips for the baseline and treatment periods, with periods representing the first, second, and third month since being given access to the Trip Diary. To minimize differences between trip origins and destinations across periods, the mean difference between participant and low-risk alternative route characteristics were used. Due to extreme outliers in the risk ratio, the median was used as a measure of central tendency. Interpretation of the values in Table 7 shows that participants chose a route with one more left turn than the low-risk alternative, and for approximately every 7–8 trips, they took a U-turn that could have been avoided by choosing the low-risk alternative. In Table 7 none of the factors are considered significant by a traditional one-factor ANOVA. This is true despite treatment 3 showing improvement compared to the baseline for each factor. This result is in part due to two aspects of the data. The first is that traditional ANOVA ignores the ordering of the periods, or in other words, does not estimate the change in each factor with respect to time under treatment. The second is that the one-factor ANOVA does not account for individuals in the study who already demonstrate safe route choice selection or driving behavior. These individuals, whose baseline is near the minimum will not show improvement, potentially drowning out the effect of other individuals showing improvement while under treatment. To address this issue and determine whether there are significant treatment effects over time given an individual’s baseline, generalized linear mixed effects models were applied to the data. 3.2. Influence of Trip Diary feedback on the route risk of older drivers A Gaussian linear mixed effects model with identity link was used to assess the influence of the Trip Diary feedback on the route risk of older drivers. In Table 8 the relative size of the random effects compared to the estimated treatment effects shows that there is a non-negligible amount of variation in the relationship between treatment and risk from participant to participant. Table 8 shows that all of the estimated fixed effects were significantly different from zero at the 95% confidence level (confidence interval calculated using profile likelihood, (Venzon and Moolgavkar, 1988). TBR, the effect of the interaction between time under treatment and baseline route risk was also significantly different from zero, providing evidence to reject the null hypothesis of HA and support the claim that there is an interaction between time under treatment and baseline route risk. Since TBR is estimated to be less than 0, this implies that the higher a participant’s baseline risk ratio, the more effective the treatment. To test the remaining hypothesis B-D (Table 4), the combined effect of T, the effect of time under treatment on the participant’s route risk ratio, and TBR, the effect of the interaction between time under treatment and baseline risk on the route risk was considered (Table 8). The null hypothesis for HB (HC ) can be rejected, if for some baseline risk ratio above (below) 1, the combined effect of T and TBR are statistically different from zero. Fig. 4 shows that for any participant with baseline risk ratio above 1.5, slope (T + TBR*XBR ) was significantly less than zero, and the participant is expected to lower their route risk over time. This provides sufficient evidence to reject the null hypothesis for HB and support the claim that pro-

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model

ARTICLE IN PRESS

AAP-4292; No. of Pages 12

R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

8

Table 4 Hypotheses to determine the effect of the Trip Diary feedback on route risk. HA: There is an interaction between treatment and baseline risk HA0 : TBR = 0, / 0 HA1 : TBR = Evidence to reject TBR = / 0

Not enough evidence to reject TBR = 0 HB

HB

HC

HD

Claim

The effect under treatment is negative T≥0 T<0

Given XBR < 1, the effect under treatment is positive TBR.XBR + T ≤ 0 TBR.XBR + T > 0

Given XBR = 1, the effect under treatment is zero

H0 H1

Given XBR < 1, the effect under treatment is negative TBR.XBR + T ≥ 0 TBR.XBR + T < 0

TBR.XBR + T = 0 TBR.XBR + T = / 0

Table 5 Hypotheses to determine the effect of the Trip Diary feedback on driving behavior. HA: There is an interaction between treatment and baseline driving behavior / 0 HA0 : TB = 0, HA1 : TB =

Claim H0 H1

Not enough evidence to reject TB = 0

Evidence to reject TB = / 0

The effect under treatment is negative T≥0 T<0

Given XB > 0, the effect under treatment is negative TB.XB + T ≥ 0 TB.XB + T < 0

Table 6 Summary of the driving behavior of older drivers for baseline and treatment periods. Percent of trips with

Baseline (1 month) Treatment (3 months) Estimated odds ratio (95% CI)

Total trips

Average speed

Speed violations

Hard braking

Hard cornering

Harsh acceleration

Seatbelt violation

2466.00

23.49

47

0.52

3

88

0.7

5416.00

23.96

44

0.49

5

91

0.4

0.80 [0.70, 0.92]

1.32 [0.634, 2.738]

0.90 [0.64, 1.26]

1.85 [0.84, 4.07]

0.56 [0.37, 0.83]

Table 7 Summary of the route characteristics and route risk of older drivers for baseline and treatment periods. Study period

Mean difference between left turns of route driven and low risk route

Mean difference between U-turns of route driven and low risk route

Mean difference between traffic incidents along route driven and low risk route

Median (route risk of route driven/low route risk)

Proportion of trips the participant drove a safer route than the low risk alternative

Baseline Treatment 1 Treatment 2 Treatment 3 p-value (periods are equal)

0.84 1.04 0.85 0.71 0.74

0.157 0.129 0.134 0.121 0.86

0.0021 0.0014 0.0035 0.0016 0.85

1.45 1.47 1.42 1.32 0.84

0.34 0.31 0.35 0.38 0.76

Table 8 Estimated fixed and random effects on older driver route risk. Fixed Effects

Random Effects

Coefficient

Estimated effect

SE (effect)

2.50%

97.50%

Random component

Estimated standard deviation

␤0 T BR TBR

−0.473 0.143 0.731 −0.119

0.211 0.066 0.156 0.045

−0.902 0.007 0.411 −0.207

−0.049 0.274 1.038 −0.025

Z0 Zt ZBR – ␧j

0.000 0.076 0.166 – 1.559

viding trip feedback with low-risk route alternatives lowered route risk for older drivers who had a baseline risk ratio of at least 1.5. Ten participants (30.3%) had a baseline route risk higher than 1.5. Similarly, there is sufficient evidence to reject HC and support the claim that for older drivers in the study who had a baseline risk less than 0.42, their risk would increase. None of the participants had a baseline route risk lower than 0.42. Lastly, the null hypothesis for HD – the effect of time under treatment is zero, could not

be rejected for participants whose baseline risk was between 0.42 and 1.5; supporting the claim that for these participants there is no significant effect due to time under treatment. Twenty-three of the participants (69.7%) had a baseline route risk between 0.42 and 1.5. Overall, 23 of the 33 participants (69.7%) were estimated to have reduced their route risk ratio, with 10 of the 23 being significant at the 95% confidence level (Fig. 5). These 10 participants reduced their risk by 7.5% per week under treatment (95% CI = (2.3%, 12.2%)).

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model AAP-4292; No. of Pages 12

ARTICLE IN PRESS R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

9

Fig. 4. The treatment effect is determined by the slope = T + TBR*XBR , with the average effect shown by the black curve in the plot. Each light grey line represents the treatment slope for a single participant in the study, and each point represents where a participant fell on their slope based on their baseline route risk (XBR ).

Fig. 5. Distribution of the estimated change in risk ratio per treatment week for 33 participants. On average, the group lowered their route risk by 2.9% per treatment week.

The full group of participants’ mean risk ratio was estimated to reduce by 2.9% per week under treatment (95% CI = (0.4%, 5.3%)). 3.3. Influence of feedback on the driving behavior of older drivers A total of 7 driving behaviors (Table 2) were recorded during baseline and treatment − speeding, hard braking, hard cornering, hard accelerating events, driving without a seatbelt, driving with the engine light on, and possible accident event. Before applying a generalized linear mixed effects regression model to each of the driving behaviors, their baseline and treatment outcomes were assessed for variability. Table 9 shows that frequency of hard braking events and seatbelt use did not have sufficient variability to appropriately model. The remaining behaviors had sufficient variability to model. Driving with the engine light on and a possible

near accident event were not included in the table as there was only 1 event recorded for each of them. A generalized linear mixed effects model was used to assess the influence of the Trip Diary feedback on the speeding behavior of older drivers by comparing their speeding behavior outcomes between baseline and treatment periods. The fixed effects (Table 10) showed that B, the effect of baseline speeding on speeding behavior, was significantly different from zero at the 95% confidence level (confidence interval calculated using profile likelihood). This result suggests that the log odds for speeding under treatment increases by a factor of 0.48 for each percent of baseline speeding. Both T and TB were not significantly different from 0 at the 95% confidence level, but TB was significant at the 90% level. Fig. 6 shows that the combined effect of T and TB is significantly less than zero for participants whose baseline probability of speed-

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model

ARTICLE IN PRESS

AAP-4292; No. of Pages 12

R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

10

Table 9 Assessing variability in driving behavior outcomes. Driving behavior

Median baseline occurrence

Median treatment occurrence

p-value (TB = 0)

p-value (T = 0)

Speeding Hard braking Hard cornering Hard acceleration Seatbelt use

0.447 0.000 0.018 0.948 1.000

0.418 0.002 0.012 0.969 1.000

0.089 NA 0.854 0.405 NA

0.226 NA 0.809 0.444 NA

Table 10 Estimated fixed and random effects on older driver driving behavior. Fixed Effects

Random Effects

Coefficient

Estimated effect

SE (effect)

2.50%

97.50%

Random component

Estimated standard deviation

␤0 T B TB

−2.59 0.048 4.883 −0.122

0.275 0.038 0.542 0.075

−3.09 −0.029 3.75 −0.273

−2.01 0.124 5.87 0.019

Z0 Zt ZB –

0.621 0.016 0.471 –

Fig. 6. The treatment effect is determined by the slope = T + TB*XB with the average effect shown by the black curve in the plot. Each light grey line represents the treatment slope for a single participant, and each point represents where a participant fell on their slope based on their baseline speeding frequency (XB ).

ing exceeded 62.8%. Six of the 33 participants’ baseline speeding probability exceeded 62.8%. Overall 21 of the 33 (63.6%) participants were estimated to have reduced their speeding frequency per week, on average by 0.9% (95% CI = (0.0%, 1.7%)) per treatment week (Fig. 7). 4. Discussion In this study routes driven and the driving behaviors of older drivers were recorded using in-vehicle data recorders, with feedback provided using web-based Trip Diaries. Trip Diary feedback included information about trips driven by the older drivers, lowrisk route alternatives, and frequency of risky driving behaviors along a driven route. Results showed that receiving feedback reduced the route risk of older drivers by 2.9% per week, and reduced their speeding frequency on average by 0.9% per week. Overall, this translated to dropping the expected crash rate from 1 in 6172 trips to 1 in 7173 trips, and the expected speeding frequency from 46% to 39% for the average observed older driver trip. There

has been limited work conducted in understanding the influence of retrospective feedback on driving safety outcomes. A driving simulator study by Donmez et al. (2008) showed that providing retrospective feedback and combined feedback (both retrospective and concurrent) to drivers increased their minimum time to collision when compared to no feedback condition, and improved overall driving performance and engagement in distraction. Results in this study showed that in the presence of feedback, the subject-to-subject variability for the overall effect of treatment time (T) was lower for speeding (Zt = 0.016) than it was for route risk (Zt = 0.079). A possible explanation is that it is simpler to interpret speeding-related feedback as there is less ambiguity in the self-regulation strategy to reduce speeding behavior. Thus when older drivers in the study received their Trip Diary feedback with speeding violations along a route, individuals complied with the feedback. Similar work using OBD2 devices to report driver’s speeding behavior in the form of written reports showed a 75% reduction in speeding violations after receiving feedback (Newnam et al., 2014). Work by Elliott et al. (2003) showed that older drivers are

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model AAP-4292; No. of Pages 12

ARTICLE IN PRESS R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

11

Fig. 7. Distribution of the estimated change in speeding per treatment week for the 33 participants. On average, the group reduced their speeding frequency by 0.9% per treatment week.

more likely to comply with speed limits than younger drivers. Thus a more consistent response to feedback on speeding violations was expected. While there was significant reduction in route risk after receiving feedback, the greater variation observed in the rate of risk reduction can be attributed to the route risk measure being more complex to interpret. Thus some older drivers in the study may have been less willing to trust the alternate low-risk route suggested. The route risk measure involved suggesting alternate routes with fewer left turns, U-turns, traffic incidents, and lane closures. Studies conducted on the willingness of older drivers to change their route choice based on feedback from variable message signs that provide traffic guidance to help choose low traffic routes showed that they were often less willing to alter their route (Yan and Wu, 2014), especially if they were familiar with the chosen route (Dia et al., 2001). It can also be assumed that since the low-risk alternative routes were retrospective, it did not take into account the context, goals and motivations of the trip purpose, which are known to influence route choice. The opportunity to provide older drivers with their driving behavior and route choice feedback may provide older adults with a more comprehensive understanding of their driving behavior, and thereby enable them to appropriately self-regulate their driving without guessing. Having access to their own daily driving records and feedback has the potential to help tailor their mobility needs, address their driving behavior concerns at an early stage, and help customize driving educational and training programs to match their driving safety goals. Additionally, recording continuous driving data of older drivers can help assess their ability and extent to self-regulate their driving behavior. This information would allow for building better personalized driver support system (DSS) features for driver profiles that can be used to automatically configure for the best DSS functioning mode. These systems can then be used to better gauge varying abilities of older driver’s over time to respond to different road and environmental situations. An additional advantage is the recording and analyzing of crash data, which can be used to determine when and why older driver’s exposure to crash risk increases. This information can then be used to assist

them in driving safer or help avoid such driving situations in the future. The opportunity to record and provide feedback of driving behavior and route choice would also allow family members, caregivers and healthcare providers of older drivers to monitor, assess and make more informed decisions about their driving and mobility needs based on empirical information. Interventions such as Trip Diaries that can record and provide feedback are especially transcendental, as driving abilities, mobility needs and even driving challenges vary over time – they can be temporary challenges such as driving after cataract surgery, or more permanent challenges such as decline in driving ability due to arthritis, Alzheimer’s, stroke, etc. For these situations, the ability to record and provide customized feedback based on the changes in driving behavior outcomes can have a great impact on developing better driving safety interventions. At the same time recording and providing feedback in the form of Trip Diaries, customized to fit the needs of older drivers is a medium that can be sustained and used to provide long-term feedback, and providing older drivers more autonomy to oversee their own driving, and take the appropriate self-regulatory measures to improve their driving behavior. This study has a number of limitations that are important to address. The study used the help of the ADRCs to recruit older drivers 65 and older. But the sample may not be representative as it could be argued that those that were interested to volunteer were active drivers, avoided risky driving situations, and were interested in understanding their driving behavior. Although OBD2 devices were installed in the vehicles of participants for a period of 1 month to collect baseline data – knowledge of the device recording their driving behavior might have led participants to drive more carefully and avoid risky behaviors during baseline. Additionally, there are challenges associated with interpreting objective driving data recorded from OBD2 devices as they do not provide context of the driving situation. The analysis conducted did not include the feedback responses from participants, health status of drivers, gender, or location. Future work would include analyzing the driving habits such as time of day, route choice, distance travelled, and the Trip Diary usage data in more detail. Although there were significant reductions in risk and driving behavior, it is assumed

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023

G Model AAP-4292; No. of Pages 12 12

ARTICLE IN PRESS R.P. Payyanadan et al. / Accident Analysis and Prevention xxx (2016) xxx–xxx

that this observed effect is due to the feedback, as the presence of feedback was the only deliberate condition that changed across all participants between baseline and treatment. Other conditions may have varied and had a systematic impact across the participants, however the during- and post-study examinations did not provide evidence that this occurred. The use of Trip Diaries in this study may have helped older drivers reduce their estimated route risk, but it is unclear whether these outcomes translate into reduced near-crashes and crashes. 5. Conclusion Developing web-based Trip Diaries for older drivers is a first step towards customizing their driving information needs and providing feedback in a way that best suits their driving safety and mobility outcomes. Tools such as Trip Diaries serve as a platform that can be incorporated into DSS technologies such as navigation systems. More generally, such tools would also be able to enhance mobility outcomes, by enabling older adults with agerelated declines to take early preventative measures to reduce their crash risk outcomes, allowing them to extend their mobility and length of independent living within the comfort of their homes. Acknowledgements The Agency for Healthcare Research and Quality is the primary funder of the study (5P50HS019917-04). Epic Systems Corporation is a secondary funder. The authors would like to thank the Aging and Disability Resource Center (ADRC), Center for Health Enhancement Systems Studies (CHESS), University of Wisconsin-Madison, Madison, WI 53706, USA, and Dr. David Gustafson, for helping recruit the study participants. The authors would also like to thank Sprint Corporation, especially John A. Menzies for providing the OBD2 devices for the study; the Department of Information Technologies (DoIT), University of Wisconsin-Madison, Madison, WI 53706, USA; members of the CHESS tech team Dave Gustafson Jr., Adam Maus, Matt Wright, Susan Dinauer, Julie Judkins, Gina Landucci, and Patrick Rogne; the undergraduate students from the Industrial and Systems Engineering Department, University of Wisconsin-Madison, Madison, WI 53706, USA; and the older adults who have volunteered their time to participate in our study, and help test and refine the Trip Diary tool. References Anstey, K.J., Wood, J., Lord, S., Walker, J.G., 2005. Cognitive, sensory and physical factors enabling driving safety in older adults. Clin. Psychol. Rev. 25 (1), 45–65, http://dx.doi.org/10.1016/j.cpr.2004.07.008. Borowsky, A., Shinar, D., Oron-Gilad, T., 2010. Age, skill, and hazard perception in driving. Accid. Anal. Prev. 42 (4), 1240–1249, http://dx.doi.org/10.1016/j.aap. 2010.02.001. De Craen, S., Twisk, D.A.M., Hagenzieker, M.P., Elffers, H., Brookhuis, K.A., 2011. Do young novice drivers overestimate their driving skills more than experienced drivers? Different methods lead to different conclusions. Accid. Anal. Prev., http://dx.doi.org/10.1016/j.aap.2011.03.024. Dia, H., Harney, D., Boyle, A., 2001. Dynamics of drivers’ route choice decisions under advanced traveller information systems. Road Transp. Res. 10 (4), 3–13. Dingus, T. A., Klauer, S.G., Neale, V. L., Petersen, A., Lee, S. E.,. Sudweeks, J., Perez, M. A., Hankey, J., Ramsey, D., Gupta, S.,. Bucher, C., Doerzaph, Z. R., Jermeland, J., Knipling, R.R., The 100-car naturalistic driving study, Phase II-results of the 100-car field experiment (No. HS-810 593). Donmez, B., Boyle, L.N., Lee, J.D., 2008. Mitigating driver distraction with retrospective and concurrent feedback. Accid. Anal. Prev. 40 (2), 776–786, http://dx.doi.org/10.1016/j.aap.2007.09.023.

Eby, D.W., Molnar, L.J., Shope, J.T., Vivoda, J.M., Fordyce, T.A., 2003. Improving older driver knowledge and self-awareness through self-assessment: the driving decisions workbook. J. Saf. Res. 34 (4), 371–381, http://dx.doi.org/10.1016/j.jsr. 2003.09.006. Elliott, M.A., Armitage, C.J., Baughan, C.J., 2003. Drivers’ compliance with speed limits: an application of the theory of planned behavior. J. Appl. Psychol. 88 (5), 964–972, http://dx.doi.org/10.1037/0021-9010.88.5.964. Finn, P., Bragg, B.W.E., 1986. Perception of the risk of an accident by young and older drivers. Accid. Anal. Prev. 18 (4), 289–298, http://dx.doi.org/10.1016/ 0001-4575(86)90043-6. Freund, B., Colgrove, L.A., Burke, B.L., McLeod, R., 2005. Self-rated driving performance among elderly drivers referred for driving evaluation. Accid. Anal. Prev. 37 (4), 613–618, http://dx.doi.org/10.1016/j.aap.2005.03.002. Hakamies-Blomqvist, L., Raitanen, T., O’Neill, D., 2002. Driver ageing does not cause higher accident rates per km. Transp. Res. Part F: Traffic Psychol. Behav. 5 (4), 271–274, http://dx.doi.org/10.1016/S1369-8478(03)00005-6. Hakamies-Blomqvist, L.E., 1993. Fatal accidents of older drivers. Accid. Anal. Prev. 25 (1), 19–27, http://dx.doi.org/10.1016/0001-4575(93)90093-C. Holland, C.A., Rabbitt, P.M.A., 1992. People’s awareness of their age-related sensory and cognitive deficits and the implications for road safety. Appl. Cognit. Psychol. 6 (3), 217–231, http://dx.doi.org/10.1002/acp.2350060304. Horswill, M.S., Waylen, A.E., Tofield, M.I., 2004. Drivers’ ratings of different components of their own driving skill: a greater illusion of superiority for skills that relate to accident involvement1. J. Appl. Soc. Psychol. 34 (1), 177–195, http://dx.doi.org/10.1111/j. 1559-1816.2004.tb02543.x. Horswill, M.S., Marrington, S.A., McCullough, C.M., Wood, J., Pachana, N.A., McWilliam, J., Raikos, M.K., 2008. The hazard perception ability of older drivers. J. Gerontol. Ser. B: Psychol. Sci. Soc. Sci. 63 (4), P212–P218. Jonah, B.A., Dawson, N.E., 1982. The National Vehicle Occupant Restraint Survey: Attitudes Toward and Use of Restraints by Canadians (No. TP 3593 E). Langford, J., Koppel, S., 2006. Epidemiology of older driver crashes – identifying older driver risk factors and exposure patterns. Transp. Res. Part F: Traffic Psychol. Behav. 9 (5), 309–321, http://dx.doi.org/10.1016/j.trf.2006.03.005. Langford, J., Methorst, R., Hakamies-Blomqvist, L., 2006. Older drivers do not have a high crash risk – a replication of low mileage bias. Accid. Anal. Prev. 38 (3), 574–578, http://dx.doi.org/10.1016/j.aap.2005.12.002. Leisch, F., 2002. Sweave, part I: mixing r and LaTeX. R News 2 (December), 28–31, http://dx.doi.org/10.1159/000323281. Man-Son-Hing, M., Marshall, S.C., Molnar, F.J., Wilson, K.G., 2007. Systematic review of driving risk and the efficacy of compensatory strategies in persons with dementia. J. Am. Geriatr. Soc. 55 (6), 878–884, http://dx.doi.org/10.1111/j. 1532-5415.2007.01177.x. Marottoli, R.A., Richardson, E.D., 1998. Confidence in, and self-rating of, driving ability among older drivers. Accid. Anal. Prev. 30 (3), 331–336, http://dx.doi. org/10.1016/S0001-4575(97)00100-0. Molnar, L.J., Eby, D.W., 2008. The relationship between self-regulation and driving-related abilities in older drivers: an exploratory study. Traffic Inj. Prev. 9 (4), 314–319. Newnam, S., Lewis, I., Warmerdam, A., 2014. Modifying behaviour to reduce over-speeding in work-related drivers: an objective approach. Accid. Anal. Prev. 64, 23–29, http://dx.doi.org/10.1016/j.aap.2013.10.032. Owsley, C., McGwin, G., Phillips, J.M., McNeal, S.F., Stalvey, B.T., 2004. Impact of an educational program on the safety of high-risk, visually impaired, older drivers. Am. J. Prev. Med. 26 (3), 222–229. Payyanadan, R.P., Sanchez, F.A., Lee, J.D., 2016. Assessing route choice to mitigate older driver risk. IEEE Intelligent Transportation Systems. R Development Core Team, 2013. R: A Language and Environment for Statistical Computing. Development Core Team, Vienna, Austria. Ross, L.A., Clay, O.J., Edwards, J.D., Ball, K.K., Wadley, V.G., Vance, D.E., Joyce, J.J., et al., 2009. Do older drivers at-risk for crashes modify their driving over time? J. Gerontol. Ser. B: Psychol. Sci. Soc. Sci. 64 (2), 163–170, http://dx.doi.org/10. 1093/geronb/gbn034. Staplin, L., Lococo, K.H., Martell, C., Stutts, J., 2012. Taxonomy of Older Driver Behaviors and Crash Risk (No. DOT HS 811 468A). Stutts, J., Martell, C., Staplin, L., 2009. Identifying behaviors and situations associated with increased crash risk for older drivers. Natl. Highw. Traffic Saf. Adm., Retrieved from http://trid.trb.org/view.aspx?id=897748. Venzon, D.J., Moolgavkar, S.H., 1988. A method for computing profile-likelihood-based confidence-intervals. Appl. Stat. J. R. Statist. Soc. Ser. C 37 (1), 87–94. White, M.J., Cunningham, L.C., Titchener, K., 2011. Young drivers’ optimism bias for accident risk and driving skill: accountability and insight experience manipulations. Accid. Anal. Prev. 43 (4), 1309–1315, http://dx.doi.org/10.1016/ j.aap.2011.01.013. Wilson, R.J., 1984. A National Household Survey on Drinking and Driving (No. TP 5865/E). Yan, X., Wu, J., 2014. Effectiveness of variable message signs on driving behavior based on a driving simulation experiment. Discret. Dyn. Nat. Soc. 2014, http:// dx.doi.org/10.1155/2014/206805.

Please cite this article in press as: Payyanadan, R.P., et al., Using trip diaries to mitigate route risk and risky driving behavior among older drivers. Accid. Anal. Prev. (2016), http://dx.doi.org/10.1016/j.aap.2016.09.023