یا در حال حاضر به من اخطار بده و یا بعدا به من اطلاع بده: پذیرش رانندگان از زمان واقعی و سیستم های مقابله با حواس پرتی بعد از رانندگی

Abstract Vehicle crashes caused by driver distraction are of increasing concern. One approach to reduce the number of these crashes mitigates distraction by giving drivers feedback regarding their performance. For these mitigation systems to be effective, drivers must trust and accept them. The objective of this study was to evaluate real-time and post-drive mitigation systems designed to reduce driver distraction. The real-time mitigation system used visual and auditory warnings to alert the driver to distracting behavior. The post-drive mitigation system coached drivers on their performance and encouraged social conformism by comparing their performance to peers. A driving study with 36 participants between the ages of 25 and 50 years old (M=34) was conducted using a high-fidelity driving simulator. An extended Technology Acceptance Model captured drivers' acceptance of mitigation systems using four constructs: perceived ease of use, perceived usefulness, unobtrusiveness, and behavioral intention to use. Perceived ease of use was found to be the primary determinant and perceived usefulness the secondary determinant of behavioral intention to use, while the effect of unobtrusiveness on intention to use was fully mediated by perceived ease of use and perceived usefulness. The real-time system was more obtrusive and less easy to use than the post-drive system. Although this study included a relatively narrow age range (25 to 50 years old), older drivers found both systems more useful. These results suggest that informing drivers with detailed information of their driving performance after driving is more acceptable than warning drivers with auditory and visual alerts while driving.

. Introduction According to the Centers for Disease Control and Prevention, traffic-related crashes are the leading cause of death for those from 8 to 24 years old (Subramanian, 2012). Risky behaviors such as speeding, seat belt disuse, and alcohol consumption are predominant contributors to traffic-related fatalities (Sivak et al., 2007). Adding to these contributors is driver distraction—“the diversion of attention from activities critical for safe driving toward a competing activity” (Regan et al., 2009, p. 7). This is particularly true as ubiquitous computing introduces many new devices into the car, ranging from navigation aids to social networking applications. Even such common tasks as selecting a song from a playlist can lead drivers to look away from the road for a dangerously long time (Lee et al., 2012). Interface modality is also an important consideration because compared to visual interfaces, auditory interfaces can be less distracting and easier to use, but can increase task completion times (Sodnik et al., 2008). As the number of devices carried into the vehicle increases, so does the potential for distraction to contribute to crashes. For example, 40% of drivers report using add-on media devices, 50% use hand-held cell phones, and 60% read text messages while driving (Lansdown, 2012). Even though many drivers believe engaging in these activities is not dangerous (Wogalter and Mayhorn, 2005), using a cell phone results in a fourfold increase in crash likelihood (McEvoy et al., 2005) and texting while driving is associated with a 23-fold increase in crash likelihood (Olson et al., 2009). In 2009 alone, distraction associated with these and other activities contributed to 5,474 deaths and nearly 448,000 injuries in the United States (NHTSA, 2010). To the extent that drivers fail to recognize the risk of distraction, providing drivers with feedback might help mitigate the adverse consequences of distraction. 1.1. Feedback to reduce driver distraction The large number of distraction-related deaths and injuries has prompted many potential solutions, ranging from legislation that outlaws activities to design guidelines that minimize distraction (Regan et al., 2009). One approach to alleviate this issue uses in-vehicle technology to give drivers feedback on their performance. Feedback has substantial promise as a way of shaping safer behavior. According to Schmidt and Bjork (1992), “it [is] understood that any variation of feedback…that makes the information more immediate, more accurate, more frequent, or more useful for modifying behavior will contribute to learning” (p. 212). Hence, more accurate knowledge of skills and capabilities should enable drivers to make accurate assessments of their ability to use a device while driving and adjust behavior accordingly. This suggests that feedback might be a powerful way to reduce distraction-related crashes (Lee, 2009). Evidence from several other driving studies confirms the promise of using feedback to reduce risky driving behaviors. For example, giving feedback about compliance of traffic laws led both young and older drivers to commit 3.5 fewer speed violations per 15–20 min drive and to stop 25% more often at stop signs and signalized intersections (de Waard et al., 1999). For traffic officers, feedback and supervisory inspections reduced the physical injury accident rate from 0.75 accidents to zero accidents per 100,000 miles (Larson et al., 1980). A feedback system warning drivers that they were following too closely reduced the amount of time drivers spent in short headways (<0.8 s) from 20% to 15% and increased the time spent in long headways (>1.2 s) from 57% to 65% (Shinar and Schechtman, 2002). The effect of feedback has also been examined with driver distraction. A series of studies by Donmez et al., 2007 and Donmez et al., 2008 assessed how different types of feedback affected driving performance. Real-time feedback influences immediate performance, but does not provide detailed information that might be needed to affect long-term behavior (Donmez et al., 2008). Post-drive feedback is provided after the driver completes the trip and contains more detailed information that might change long-term behavior by helping drivers to be more aware of dangerous situations (Donmez et al., 2008). Real-time feedback led drivers to glance less toward the distracting device, thereby modulating their engagement in distracting activities (Donmez et al., 2007). When real-time feedback was combined with post-drive feedback, drivers responded to a braking lead vehicle more quickly and made longer glances to the roadway (Donmez et al., 2008). Although real-time and post-drive feedbacks have proven to be beneficial in some instances, giving feedback does not always improve driving performance. For example, some characteristics of a driving coaching system (e.g., negatively framed feedback) led to no change or even undermined driving performance (Arroyo et al., 2006). Hence, only certain types of feedback actually increase safety. One feedback approach that has yet to be tested relies on using social norm conformance to promote safe driving behavior. A powerful way to change behavior is to change attitudes towards the behavior and the subjective norm regarding the behavior (Ajzen, 1991 and Fishbein and Ajzen, 1975). As stated by Bernheim (1994), “individual behavior is motivated by social factors such as the desire for prestige, esteem, popularity, and acceptance. As these factors are so widespread, they also produce conformism and those who do deviate from the norm are often penalized” (p. 842). The notion of modifying one's behavior according to those within the same social network has been seen within the Framingham Heart Study with obesity (Christakis and Fowler, 2007), smoking (Christakis and Fowler, 2008), and happiness (Fowler and Christakis, 2008). Feedback enforcing social norms can also reduce binge drinking among college students (Agostinelli et al., 1995 and Neighbors et al., 2004) and can encourage safe driving behaviors among pizza deliverers (Ludwig et al., 2002). Using social norm conformance to promote a behavior change can prove to be a particularly powerful form of feedback to deter driver distraction. Therefore, one of the goals of this study was to assess how different forms of feedback influence driver's acceptance of the feedback technology. In particular, following the methodology of Donmez et al., 2007 and Donmez et al., 2008, both real-time auditory and visual alerts were to be compared to post-drive reports that use social norm conformance to change behavior. While potentially powerful, such feedback must be accepted by drivers if it is to realize its potential. 1.2. Technology acceptance Feedback effectiveness depends on drivers' acceptance and trust. Several frameworks and methodologies exist that describe how people adopt and accept new technology. Of particular prominence within the driving domain is a simple method that assesses system usefulness and satisfaction (Van Der Laan et al., 1997). Among the more advanced models, the Technology Acceptance Model (TAM) has proven to successfully predict technology use and is broadly used after more than three decades since its introduction. The TAM (Davis et al., 1989), built upon the Theory of Reasoned Action (TRA) of Fishbein and Ajzen (1975), posits that perceived usefulness (PU) and perceived ease of use (PEOU) are the main determinants of attitude toward a technology, which in turn predicts behavioral intention to use (BI) and actual use. Further research has shown that attitude only partially mediates the effect of PU on BI, thus suggesting that attitude could be excluded from the model (Davis and Venkatesh, 1996 and Venkatesh and Davis, 2000). The resulting parsimonious TAM is shown in Fig. 1. The high reliability and validity of the TAM constructs are robust to measurement instrument design (Davis and Venkatesh, 1996). The TAM has been used to assess user acceptance in a variety of domains and has consistently explained a major proportion of variability in use intentions and behavior. These properties make the TAM an apt framework for understanding how people respond to technology innovations such as distraction mitigations systems. Parsimonious Technology Acceptance Model. Fig. 1. Parsimonious Technology Acceptance Model. Figure options The TAM provides a conceptual framework for quantitatively assessing the relationships between external variables (e.g., user characteristics and system features) and user's perceptions, as well as the relationships between these perceptions and user's attitudes and intentions. The TAM's flexibility has enabled researchers to assess the importance of many constructs and relationships not originally part of the model. Examples of such constructs are task-technology fit (Dishaw and Strong, 1999), trust (Gefen et al., 2003), experience and voluntariness (Venkatesh and Davis, 2000), and privacy invasion and fairness (Zweig and Webster, 2002). In another example, Yi and Hwang (2003) added three motivation variables, i.e., self-efficacy, enjoyment, and learning goal orientation, to the TAM framework to predict web-based information systems' use. Another implementation of TAM involved decomposing the TAM and extending it with constructs based on the expectancy disconfirmation theory to assess the continued intention to use e-learning services (Roca et al., 2006). Despite extensive use of the TAM to assess information systems in other fields, only recently has it been applied to driving assistance systems. Xu et al. (2010) used the TAM to assess acceptance of advanced traveler information systems by incorporating four domain-specific constructs (information attributes, trust in travel information, socio-demographics, and cognition of alternate routes). Chen and Chen (2011) also used the TAM for evaluating acceptance of GPS devices, customizing the framework by adding perceived enjoyment and personal innovativeness constructs. Other studies have also used the TAM constructs in their analysis of driving assistance systems (Adell, 2010 and Meschtscherjakov et al., 2009), finding that perceived system disturbance and perceived risk, as well as social factors, strongly influence behavioral intention to use a system. These studies demonstrate the applicability of the TAM framework to in-vehicle systems assessment and provide deeper background and structure of drivers' acceptance than simpler frameworks (Ghazizadeh et al., 2012). 1.3. Purpose of study Distraction mitigation systems are a promising means of addressing driver distraction, but only if drivers accept them. In other domains, the TAM has predicted acceptance and offers a promising framework for assessing acceptance of distraction mitigation systems. Consequently, the goal of this study was to examine how a real-time mitigation system, using auditory and visual alerts, or a post-drive mitigation system, using the concept of social norms, affect driver acceptance.

. Results Data were reduced using MatLab 7.12.0 (MathWorks, 2011). All statistical analyses were done using R 2.14.2 (R Development Core Team, 2011). Descriptive statistics of the four acceptance measures are shown in Table 2. All Cronbach's alphas were near 0.90, indicating acceptable internal reliability. Note that the descriptive statistics in Table 2 (and also the hierarchical linear model that follows) are based on the responses from the real-time and post-drive groups only, because the focus was on those who experienced feedback. What is more, the no-feedback group did not respond to the BI questions for each specific system, which made their set of responses incomparable to the other two groups. Table 2. Means, standard deviations, Cronbach's alphas and correlations between acceptance measures; Cronbach's alphas are on the diagonal. Variable Mean SD PU PEOU Unobtrusiveness BI PU 4.37 1.35 0.93 PEOU 5.26 1.30 0.41⁎ 0.95 Unobtrusiveness 3.46 1.52 0.77⁎⁎ 0.43⁎ 0.88 BI 3.03 1.04 0.57⁎⁎ 0.62⁎⁎ 0.51⁎ 0.88 ⁎ p<0.05. ⁎⁎ p<0.01. Table options Given the high correlations between acceptance measures, it follows that individually, PU, PEOU, and unobtrusiveness all predicted BI (PU: b=0.44, t(22)=3.27, p<0.01; PEOU: b=0.50, t(22)=3.69, p<0.01; unobtrusiveness: b=0.35, t(22)=2.78, p=0.01). However, since PU, PEOU, and unobtrusiveness were also highly correlated with each other, a hierarchical linear model was used to determine how each measure predicted BI after accounting for relationships between the measures. Following the strength of the correlations shown in Table 2, PEOU was entered first, followed by PU, and then unobtrusiveness. Table 3 shows the regression coefficients and significance tests. The model including PEOU and PU (but not unobtrusiveness) best described the variation in BI (Step 2, Table 3). The addition of unobtrusiveness was not justified given the negligible effect of unobtrusiveness and the insignificant improvement in model fit (see Step 3, Table 3). Table 3. Results of hierarchical linear model predicting BI. Variable Step 1 Step 2 Step 3 PEOU 0.50⁎⁎ 0.37⁎ 0.37⁎ PU 0.30⁎ 0.27 Unobtrusiveness 0.03 R2 0.38 0.50 0.50 ΔR2 0.12 0.0008 R2F statistic F(1, 22)=13.64 ⁎⁎ F(2, 21)=10.63 ⁎⁎ F(3, 20)=6.77 ⁎⁎ ΔR2F statistic F(1, 21)=5.09 ⁎ F(1, 20)=0.03 ⁎ p<0.05. ⁎⁎ p<0.01. Table options The relationships between PU, PEOU, and unobtrusiveness were also analyzed. Following the TAM, PEOU predicted PU, b=0.43, t(22)=2.13, p=0.04, and accounted for 17.1% of the variance. Unobtrusiveness predicted both PU (b=0.67, t(22)=5.60, p<0.01), accounting for 59% of the variance, and PEOU (b=0.37, t(22)=2.26, p=0.03), accounting for 19% of the variance. Fig. 8 graphically displays these relationships using standardized regression coefficients. Driver Acceptance Model with associated relationships (standardized regression ... Fig. 8. Driver Acceptance Model with associated relationships (standardized regression coefficients) between variables where solid lines indicate significant relations and dashed lines indicate non-significant relations (⁎p<0.05, ⁎⁎p<0.01); Note that the regression coefficients are standardized, whereas the coefficients presented in the text are unstandardized. Figure options 3.1. Effect of feedback type and age on acceptance It was expected that feedback condition, number of alerts, usefulness of alerts, and distraction engagement, along with age and gender, would be significant predictors of technology acceptance. After initial analysis, number of alerts, distraction engagement, and gender were excluded as they did not account for a significant amount of variance in any of the acceptance measures. In addition, alert usefulness was excluded as it was correlated with feedback condition (Pearson's r=0.62, p<0.01; the real-time group received fewer useful alerts). Linear regressions were fit for each of the three acceptance measures. The model predicting PU accounted for 38% of the overall variance, F(2, 21)=6.35, p<0.01. Age was a significant predictor, b=0.07, t(21)=2.29, p=0.03, indicating that as age increased by one year, usefulness ratings increased by 0.07. Age accounted for 15.6% of the variance. Feedback condition had a nearly significant effect, b=0.84, t(21)=1.74, p=0.10, indicating that the real-time group rated ease of use 0.84 lower than the post-drive group, i.e., the real-time group thought the system was less useful. While feedback condition was not significant, it accounted for 9.0% of the overall 38% model variance. The model predicting perceived ease of use accounted for 44.6% of the variance, F(2, 21)=8.44, p<0.01. Feedback condition was a significant predictor, b=1.70, t(21)=3.85, p<0.01, indicating that the real-time group rated ease of use 1.7 lower than the post-drive group, i.e., the real-time group thought the system was less easy to use. The model predicting unobtrusiveness accounted for 39.8% of the variance, F(2, 21)=6.93, p<0.01. Feedback condition was a significant predictor, b=1.67, t(21)=3.09, p<0.01, indicating that the real-time group rated unobtrusiveness of the system as 1.67 lower than the post-drive group, i.e., the real-time group found the system to be more obtrusive. Fig. 8 shows the relationship between the external variables and the acceptance measures as indicated by standardized regression coefficients. 3.2. Acceptance across feedback and no-feedback conditions The responses from the no-feedback group were compared to the responses of those who received feedback to determine whether experience with the feedback system affected acceptance. It should be noted that for the no-feedback group, the acceptance ratings were based on beliefs about the distraction mitigation systems in the absence of any prior exposure to either system. Previous research has measured pre-usage perceptions of a system and linked them to post-usage perceptions, finding that the pre-usage perceptions are updated sequentially with actual system use (Bhattacherjee and Premkumar, 2004, Böhm et al., 2009, Hsu et al., 2006 and Kim and Malhotra, 2005). As such, the no-feedback group's acceptance ratings can be used as a baseline for gauging the pre-usage expectations and perceptions of the system—to be compared to perceptions of those who have used the system for some time (i.e., received feedback in either real-time or post-drive form). When comparing the real-time versus the no-feedback group, for two of the three acceptance measures (PU and PEOU), feedback condition was a significant predictor (PU: F(1, 22)=10.16, p=0.004; PEOU: F(1, 22)=8.57, p=0.008; unobtrusiveness: F(1, 22)=2.35, p=0.14). These results indicate that the no-feedback group rated these two measures significantly higher than the real-time group when they were asked to imagine their response to real-time feedback ( Table 4). Table 4. Means and standard deviations of acceptance measures for the no-feedback, real-time, and post-drive groups. PU PEOU Unobtrusiveness Real-time system No-feedback 5.23 (0.81) 5.75 (0.95) 3.21 (1.12) Real-time 3.75 (1.39) 4.42 (1.26) 2.54 (1.01) Post-drive system No-feedback 4.39 (1.27) 5.56 (0.89) 4.08 (1.55) Post-drive 4.99 (1.01) 6.11 (0.61) 4.38 (1.42) Table options When comparing the post-drive with the no-feedback group, although the no-feedback group's ratings were higher for all the three measures of acceptance, feedback condition was not a significant predictor of any of the measures, (PU: F(1, 22)=1.62, p=0.22; PEOU: F(1, 22)=3.18, p=0.09; unobtrusiveness: F(1, 22)=0.23, p=0.64). Table 4 shows the mean values across the different groups. The responses were also compared within the no-feedback group to see if drivers rated the two systems differently. The no-feedback group found both the real-time and post-drive systems to be similar on both ease of use and unobtrusiveness, PEOU: F(1, 11)=0.76, p=0.40, unobtrusiveness: F(1, 11)=1.47, p=0.25. However, they thought the post-drive system would be less useful, as indicated by a lower PU rating, PU: F(1, 11)=6.27, p=0.03. Fig. 9 shows the mean acceptance rating with standard error bars for all three acceptance measures for both the real-time and post-drive systems. The no-feedback group is shown in the “not used” category of system usage while both the real-time and post-drive groups are represented in the “used” category of system usage. Overall, use of the system led drivers to see the post-drive feedback more positively and the real-time feedback less positively. Acceptance ratings of the different mitigation systems and experience. Fig. 9. Acceptance ratings of the different mitigation systems and experience.