Statistical Analysis of Vehicle Crash Incidents at the USA Interstate 494 and West using Chi-Square Statistic Goodness-of-Fit Testing

Candace Roberts Department of Resource Analysis, Saint Mary’s University of Minnesota, , MN 55404

Keywords: Vehicle Crash Incidents, Minneapolis I-494 and I-35W , Contributing Factors, Chi-Square Goodness-of-Fit, Contingency Tables

Abstract

The Minneapolis, Minnesota (USA) Interstate 494 and Interstate 35 West cloverleaf interchange experiences a continued overcapacity daily traffic volume, creating the lowest safety ranking of highway interchanges in the metropolitan area. During a 10-year period from 2006 through 2015, an average 8.45 monthly crash incidents were reported. The majority of crashes occurred during good weather conditions and were attributed to driver behavior factors. This study utilized contingency tables and Chi-Square Goodness-of-Fit tests to determine associations and statistical equality of contributing factors that influenced crash incidents. Significant findings point to contributing factors for which improvements, in response to human behaviors, will help mitigate such occurrences at this interchange.

Introduction However, this creates a juxtapositional situation as traffic interweaves lanes while Population growth in the Minneapolis simultaneously adjusting to decreasing (or metro area (USA) has been steadily increasing) speeds depending on exiting increasing over the past years (U.S. (or entering) pathway. Hotchin (2017) Census Bureau, 2018) and this has states, “…weaving is a problem that may increased demand on area infrastructure. lead to breakdown in traffic operation and Growing traffic volume produces stress on [lead to] more accidents.” aging highways and interchanges making The I-494 and I-35W cloverleaf traffic safety a concern. The Interstate 494 interchange was constructed in the 1960’s and Interstate 35 West (I-494 and I-35W) and has a continued overcapacity daily cloverleaf interchange, located in the traffic volume; the daily traffic volume in Minneapolis suburb of Bloomington, is an 2014 was 290,000 vehicles (Minnesota example of this aging interchange under Department of Transportation, 2013). Its stress. safety ranking is among the worst for The cloverleaf structure was interchanges in the Minneapolis metro designed in the early 20th century to keep area (Short Elliott Hendrickson, Inc., traffic flowing without the use of traffic 2010). Congestion and design flaws on I- signals (Hotchin, 2017). The cloverleaf 494 and I-35W arteries impact the design utilizes four loops (ramps) to interchange in a shockwave effect (SEH) transfer traffic flow seamlessly from one creating stress on traffic maneuvering highway to another while maintaining through the interchange. During a 10-year right-lane exit and entrance flow. span, 2006 through 2015, an average 8.45

Roberts, Candace. 2018. Statistical Analysis of Vehicle Crash Incidents at the Minnesota USA Interstate 494 and Interstate 35 West Interchange using Chi-Square Statistic Goodness-of-Fit Testing. Volume 21, Papers in Resource Analysis. 15 pp. Saint Mary’s University of Minnesota University Central Services Press. Winona, MN. Retrieved (date) http://www.gis.smumn.edu monthly crash incidents were reported as significance with the Chi-Square evident in this study’s dataset. Goodness-of-Fit test to determine The state of Minnesota statistical equality. documentation on traffic collision occurrence preferences the term “crash” to Data “accident” since accident suggests “a random, unavoidable quality about the Historical crash data were obtained from events in question” (Minnesota the Minnesota Department of Department of Public Safety, 2016). Transportation in shapefile and CSV When, in fact, the reduction in the number formats. The data incorporated crash and severity of crashes over the past incidents within 1250’ from the center of decades is attributed to advances in the I-494 and I-35W interchange from technology, engineering, public policy, 2006 to 2015. The dataset contained 1014 and driver behavior, which points to crash incidents involving single and preventable instances rather than random multiple vehicle collisions. accounts (MnDPS). Three criteria are The shapefiles revealed the required to define a motor vehicle crash: location for 87% of incidents occurred at occurrence on a public road, minimum the center of the interchange without $1000 damage or a person is injured, dispersion of occurrence elsewhere. Either transport of motor vehicle (MnDPS; this indicated most likely location for a Minnesota Department of Public Safety, crash to occur or data recording n.d.). preference. In either case, location Understanding impacts leading to analysis, such as hot spot clustering, were vehicle crashes by analyzing historical deemed insignificant in determining areas data provides valuable information in of most likely occurrences due to crash efforts to mitigate such occurrences at the location accuracies. Nonetheless, all 1014 I-494 and I-35W cloverleaf interchange. points are shown in Figure 1, of which 881 This study utilizes contingency tables and points are displayed at the same location. Chi-Square Goodness-of-Fit tests to determine significant association and statistical equality using weather conditions, driver age, driver behavior, vehicle safety, time, and collision type of 1014 crash incidents from 2006 to 2015 at the I-494 and I-35W cloverleaf interchange.

Methods

Motor vehicle crash occurrences of the I- 494 and I-35W interchange from 2006 to 2015 were extensively analyzed using contingency tables to calculate the Chi- Square Goodness-of-Fit tests to determine Figure 1. I-494 and I-35W cloverleaf interchange: significant statistical association using 1014 motor vehicle crash occurrences, 2006–2015. single dataset attributes tested for Information analyzed for possible

2 significant statistical association driver behaviors as a contributing factor (indicating influence to crash occurrence) accounted for 43.5% of the crash were month, day, year, time, collision incidents, whereas vehicle safety failures type, weather conditions, driver were accountable for only 3%. Factors contributing factors, and driver age. categorized as ‘not clearly identified as Information analyzed for statistical contributing to crash incident’ were equality testing of occurrence across 2006 44.3%. to 2015 were crash frequencies, two Odds are, if the 44.3% ‘not clearly collision types (rear-end and sideswipe), identified factors’ were countable as and two driver contributing factors (driver vehicle safety failures, the whole 44.3% inattention or distraction, and illegal or would not shift to vehicle safety failures. unsafe driving speeds). At most, 22.2% (or half) would shift and the remaining percentage would shift to Data Preparation driver behavior factors, distributing the 44.3% equally; therefore, distribution of Contributing factors due to driver factors as ‘not clearly identified’ were influence, either human behavior or disregarded. With driver behaviors greater vehicle failure, were recorded in multiple than vehicle safety failures, driver fields as were weather conditions. The behaviors as contributing factors attributed fields were equally weighted. Therefore, at a higher percentage to crash incidents in the fields were analyzed as separate the n=1014 dataset case study. entities, rather than the whole entities as The 2015 Minnesota crash analysis recorded in the crash incident to extract (MnDPS) reports, “…most crashes occur influence of a single entity. in good driving conditions” which In simplifying weather conditions included clear weather conditions. The from the two weather fields, clear was n=1014 dataset case study reflected 655 categorized as clear; cloudy or crashes, or 64.6%, occurred during clear clear/cloudy were categorized as cloudy, weather conditions. The second highest reference to rain was categorized as rain, percentage of crash incidents occurred reference to snow was categorized as during cloudy weather conditions at 231 snow, sleet was categorized as sleet, fog crashes (22.8%). Clear or cloudy weather was categorized as fog, and blowing winds conditions comprised 87.4%. In contrast, were categorized as blowing winds. precipitous conditions – rain, snow, sleet – accounted for 12% with remaining 0.6% Extracting Data Subsets attributed to fog and blowing winds. A subset from the dataset (based According to analysis in the Minnesota on the results of the majority of crash Motor Vehicle Crash Facts 2015 incidents were due to driver behaviors (Minnesota Department of Public Safety, during clear or cloudy weather conditions) 2016), driver behaviors are frequently was extracted. The subset contained 402 cited as contributing factors to crash crash incidents. incidents more often over vehicular safety The distribution of crashes here failures. This was demonstrably accurate from 2006 to 2015 is displayed in scatter in the n=1014 dataset case study for the I- plots with the month (as single-letter 494 and I-35W cloverleaf interchange. initials) along the x-axis and the time In the n=1014 dataset case study, along the y-axis. Figure 2 is the original

3 dataset case study (n=1014); Figure 3 is referenced in Zar (2010). the data subset (n=402). The distributions are similar with saturation of crash Contingency Tables incidents during the morning rush hour (6:00 a.m. to 9:00 a.m.), afternoon (12:00 A contingency table contains enumeration p.m. to 6:00 p.m.), and Thanksgiving data of observed frequencies of two through year end; and absence of categorical variables where the rows distribution around April and October. represent one variable and the columns Based on similar distribution, the data represent the other variable. The number subset (n=402) was representative of the of rows is determined by the number of original dataset case study (n=1014). categories within the row variable and the number of columns is determined by the Crash Occurrences 2006-2015 number of categories in the column 12am variable. In this study (n=402), five 6pm contingency tables were constructed for analysis. The first contingency table 12pm contained two categorical variables, driver behavior and collision types. Driver 6am behaviors were divided into 14 categories (failure to yield, illegal or unsafe driving speed, driving too closely, disregard for J F M A M J J A S O N D Figure 2. All crash occurrences by month and time traffic control device, improper passing or from 2006 – 2015 (n = 1014). Clustered areas of overtaking, improper or unsafe lane use, overlapping, stacking circles are higher densities of improper parking or starting or stopping, crash occurrences. improper turn, over-correcting, impeding traffic, driver inattention or distraction, Crash Occurrences 2006-2015 12am driver inexperience, chemical impairment, and other factor). Collision types were 6pm divided into eight categories (rear-end, sideswipe, off-road left side, off-road right side, right angle, head-on, not specified, 12pm and other type of collision). The second contingency table 6am consisted of driver age and driver behavior. Driver age was divided into nine J F M A M J J A S O N D categories (15-19 year-olds, 20-29 year- Figure 3. Crash occurrences by month and time olds, 30-39 year-olds, 40-49 year-olds, 50- from 2006 – 2015 (n = 402) due to driver behavior 59 year-olds, 60-69 year-olds, 70-79 year- while driving in clear or cloudy conditions. olds, 80-89 year-olds, and not specified). Driver behaviors were divided into 14 Contingency Statistical Analysis categories as in the first contingency table. The third contingency table The data subset (n=402) was analyzed incorporated driver age and collision using contingency tables to calculate Chi- types. Driver age was divided into nine Square Goodness-of-Fit to determine categories as in the second contingency significant statistical association as

4 table. Collision types were divided into the following formula: eight categories as in the first contingency table. (푅풾)(퐶풿) ƒ̂ = , The fourth contingency table 풾풿 푛 comprised of time-periods and driver behavior. Time-periods were divided into where 푅풾 is the total observed frequency eight categories (midnight to 3:00 a.m., of row 풾, 퐶풿 is the total observed 3:00 a.m. to 6:00 a.m., 6:00 a.m. to 9:00 frequency of column 풿, and 푛 is the a.m., 9:00 a.m. to noon, noon to 3:00 p.m., sample size. 3:00 p.m. to 6:00 p.m., 6:00 p.m. to 9:00 Expected frequency calculations p.m., and 9:00 p.m. to midnight). Driver were evaluated using crosstab rules behaviors were divided into 14 categories assuring none were less than or equal to 3, as in the first contingency table. and no more than 20% were less than or The fifth contingency table equal to 5. Columns or rows were contained time-periods and collision types. combined, if possible, or removed until the Time-periods were divided into eight expected frequency calculations adhered categories as in the third contingency to the crosstab rules. table. Collision types were divided into Once the crosstab rules were eight categories as in the first contingency satisfied and the Chi-Square statistic was table. calculated and assessed using the critical The Chi-Square statistics were value of the Chi-Square distribution using derived from the contingency tables and an alpha (훼) level of significance of 0.05 the Goodness-of-Fit tests determined and degrees of freedom calculated as whether to accept or reject the null follows: hypothesis. 퓋 = (푟 − 1)(푐 − 1), Chi-Square Goodness-of-Fit Test where 푟 is the number of rows, and 푐 is the The Chi-Square statistic for evaluating the number of columns. Goodness-of-Fit was performed on the If the Chi-Square statistic was less data subset (n=402) to determine than the critical value, the null hypothesis significant association between categorical was not rejected and that the row and variables. The following Chi-Square column variables were considered statistic formula was performed on data statistically independent (p > 0.05). residing in the contingency tables (Zar, If the calculated Chi-Square 2010): statistic was greater or equal to the critical value, the null hypothesis was rejected (p ̂ 2 (ƒ풾풿 − ƒ풾풿) < 0.05) and the contingency table was 2 = ∑ ∑ , ̂ ƒ풾풿 assessed for possible data reconfiguration – combining rows, columns, or removal – where ƒ is the observed frequency at row if doing so continued to make the Chi- 풾풿 Square test statistic more statistically 풾 and column 풿 of the contingency table, significant when the Chi-Square and ƒ̂ is the expected frequency at 풾, 풿. 풾풿 Goodness-of-Fit test was recalculated. If The expected frequency was data reconfiguration was not successful, derived from the contingency table using the null hypothesis was rejected (p < 0.05)

5 indicating that a dependent association ̂ 2 2 (ƒ − ƒ) between variables existed.  = ∑ , ƒ̂

Equality Statistical Analysis where ƒ is the observed frequency, and ƒ̂ is

the expected frequency. The data subset (n=402) was analyzed The expected frequency was using the Chi-Square Goodness-of-Fit test derived from observed frequencies using to determine significant statistical equality. the following formula: In this study (n=402), collision type, driver behavior, and time were assessed in eleven ∑ ƒ Chi-Square tests. ƒ̂ = , The first test determined the 푛 overall significance by evaluating the where is the observed frequency, and is number of crash occurrences by year. ƒ 푛 The second and third tests turned the sample size. toward evaluating collision types by year. Once the statistic was calculated, The number of rear-end and sideswipe the Goodness-of-Fit test was performed collisions were examined for statistical using the critical value of the Chi-Square significance. Further evaluations distribution using an alpha (훼) level of continued for significance of rear-end and significance of 0.05 and degrees of sideswipe collisions from 6:00 a.m. to freedom calculated as follows: midnight generated the fourth and fifth tests. 퓋 = 푛 − 1, The sixth and seventh tests veered toward evaluating driver behaviors by where 푛 is the sample size. year. The number of collisions attributed If the Chi-Square value was less to driver inattention or distraction, and than the critical value, the null hypothesis illegal or unsafe driving speeds were was not rejected (p > 0.05). This indicated examined for significance. Further that observed frequencies were statistically evaluations continued for significance of equal. rear-end and sideswipe collisions resulting If the calculated Chi-Square from driver inattention or distraction, and statistic was greater or equal to the critical illegal or unsafe driving speeds created value, the null hypothesis was rejected (p eighth through eleventh tests. < 0.05). Observed frequencies were then Chi-Square Goodness-of-Fit tests subdivided, and Chi-Squares were were used to determine whether to accept recalculated to ascertain significant or reject the null hypothesis. differences between observed and expected frequencies. Chi-Square Statistic Goodness-of-Fit Test Results The Chi-Square statistic was performed on the data subset (n=402) to determine Many associations were examined for statistical equality. The following Chi- equality using the Chi-Square Goodness- Square statistic formula was utilized (Zar, of-Fit test to determine influences leading 2010): to vehicle crashes.

6 Testing for Association of Factors attributed to those collisions, illegal or unsafe driving speed was the leading cause Contingency tables were used to calculate for off-road left- and right-side collisions; the Chi-Square Goodness-of-Fit test to driver inattention or distraction was the determine significant association in second leading cause for rear-end, relation to crash occurrences. sideswipe, and off-road left-side collisions. Chi-Square Goodness-of-Fit Test Collision Types Driver behavior and collision types, driver Rear-End 49.5% age and driver behavior, driver age and Sideswipe 17.2% collision types, time-periods and driver Left-Side 12.2% behavior, and time-periods and collision Right-Side 8.5% types were assembled into five Other 7.5% contingency tables and analyzed for Right-Angle 2.5% significant statistical association. Head-On 2.2% Not Spec 0.5% Driver Behavior and Collision Types Figure 5. Percentages of collision types contributing to crash occurrences (n = 402). The data subset (n=402) contained only driver behaviors as contributing factors to The Chi-Square Goodness-of-Fit crash occurrences, which resided in two test between driver behavior and collision fields. Separating and simplifying the types led to data reconfiguration into four fields into single entities for analysis, collision types (rear-end, sideswipe, off- contributing factors totaled 526 driver road left side, and off-road right side) and behavior involvements in n=402 crash four driver behaviors (driver inattention or occurrences. distraction, illegal or unsafe driving speed, The highest percentages of driver following too closely, and improper or behaviors attributed to crashes were driver unsafe lane use). The calculated Chi- inattention or distraction (27.2%), illegal Square statistic was significant indicating or unsafe driving speed (21.5%), following that a dependent association between too closely (20.7%), and improper or driver behavior and collision types existed unsafe lane use (12%) (Figure 4). (p < 0.05) as shown in Table 1.

Driver Behaviors Table 1. Driver behavior and collision types Distract 27.2% contingency table calculating Chi-Square Speed 21.5% Goodness-of-Fit test for significant association (n = 402). Expected frequency noted in parentheses. Close 20.7% Collision Types Lane 12% Behavior Rear-End Sideswipe Left-Side Right-Side Total Figure 4. Highest percentages of driver behaviors Distract 95 24 12 5 136 (84.34) (24.95) (16.17) (10.54) contributing to crash occurrences (n = 402). Speed 28 13 27 20 88 (54.57) (16.14) (10.46) (6.82) Close 101 3 1 2 107 The highest percentages of (66.36) (19.63) (12.72) (8.29) collision type were rear-end (49.5%), Lane 16 31 6 3 56 (34.73) (10.27) (6.66) (4.34) sideswipe (17.2%), off-road left side Total 240 71 46 30 387 2 (12.2%), and off-road right side (8.5%)  0.05,9 = 16.919 (Figure 5). Among driver behaviors 2 = 170.675; therefore, reject null hypothesis (p < 0.05).

7 significant indicating no association Driver Age and Driver Behavior between driver age groups and driver behavior (p > 0.05) as shown in Table 2. In the data subset (n=402), the driver age categories were grouped by decade Table 2. Driver age and behavior contingency table beginning with 20 year-olds; the 15-19 calculating Chi-Square Goodness-of-Fit test for significant association (n = 402). Expected year-olds were grouped as a partial frequency noted in parentheses. decade. Driver Behavior The highest occurrence of crashes Age Distract Speed Close Lane Total 15-19 15 14 14 1 44 by age group was the 20-29 year-olds at (14.81) (11.47) (11.36) (6.36) 36.6% which was twice of the next age 20-29 49 49 37 25 160 (53.84) (41.71) (41.33) (23.13) group (30-39 year-olds) at 18.2% as 30-39 28 16 20 16 80 shown in Figure 6. (26.92) (20.85) (20.66) (11.56) 40-49 21 13 19 7 60 (20.19) (15.64) (15.50) (8.67) Age Groups 50-59 17 12 15 8 52 15 - 19 9.5% (17.50) (6.78) (13.43) (7.52) 60-89 12 6 4 4 26 20 - 29 36.6% (8.75) (6.78) (6.72) (3.76) Total 142 110 109 61 422 30 - 39 18.2% 2  0.05,15 = 24.996 40 - 49 15.2% 2 = 15.310; therefore, accept null hypothesis (p > 0.05). 50 - 59 11.7% 60 - 69 4.2% Driver Age and Collision Types 70 - 79 2.5% 80 - 89 1.0% Across all age groups, rear-end collisions were the leading collision type. Sideswipe Not Spec 1.2% collisions were the second leading Figure 6. Percentages of age groups contributing to crash occurrences (n = 402). collision type, except age groups 15-29 (off-road left side) and age group 60-69 Although population size within (other type of collision). Sideswipe each age group differed, percentage-wise collisions were the third leading collision driver inattention or distraction was the type for age groups 15-29. leading driver behavior cause in all age A significant dependent association groups; and illegal or unsafe driving speed existed between driver age and collision and following too closely were the second types. Passing crosstab rules led to or third leading driver behavior cause for consolidating age groups above 50, age groups 15-59 and third or fourth vehicle-vehicle collision types (rear-end, leading driver behavior cause for age sideswipe, right-angle, and head-on), groups 60-89. vehicle maneuvering off road (off-road left The Chi-Square tests for side and off-road right side), and other Goodness-of-Fit between age groups and collision type with collision not specified. driver behavior led to data reconfiguration The calculated Chi-Square statistic was by combining age groups 60 and above significant indicating that a dependent and four driver behaviors (driver association between driver age and inattention or distraction, illegal or unsafe collision types existed (p < 0.05). driving speed, following too closely, and improper or unsafe lane use). The calculated Chi-Square statistic was not

8 Time-periods and Driver Behavior speed as the first leading cause and chemical impairment as the second leading In the data subset (n=402), the time cause. categories were grouped by three-hour The first Chi-Square Goodness-of- increments. The increment size was Fit test for significant association between chosen to incorporate defined morning and time-periods and driver behavior led to afternoon rush hour time-periods as noted combining time-periods into six-hour in the Minnesota Motor Vehicle Crash increments (midnight to 6:00 a.m., 6:00 Facts 2015 (Minnesota Department of a.m. to noon, noon to 6:00 p.m., and 6:00 Public Safety, 2016). p.m. to midnight) and reducing driver The highest percentage of crashes behaviors to driver inattention or occurred during the 3:00 p.m. to 6:00 p.m. distraction, illegal or unsafe driving speed, (afternoon rush hour) with 22.1%. The following too closely, and improper or next highest was noon to 3:00 p.m. at unsafe lane use. The calculated Chi- 21.1% as shown in Figure 7. Together, Square statistic here was significant noon to 6:00 p.m., accounted for 43.2% indicating that a dependent association which nearly meets the 2015 statistic of between six-hour increment time-periods 45% state-wide in this time-period and driver behavior existed (p < 0.05). (MnDPS). The second Chi-Square Goodness- of-Fit test for significant association Time-periods between time-periods and driver behavior 12am - 3am 4.0% was a continuation of the first test to 3am - 6am 5.7% determine if it was possible to achieve 6am - 9am 15.4% independence of time-periods and driver 9am - 12pm 12.7% behavior. Only when the time-periods 12pm - 3pm 21.1% were consolidated to 12-hour increments 3pm - 6pm 22.1% did the calculated Chi-Square statistic achieve no significance indicating no 6pm - 9pm 9.2% association between time of 12-hour 9pm - 12am 9.7% increments and driver behavior (p > 0.05), Figure 7. Percentages of time-periods contributing anything less was significant (p < 0.05). to crash occurrences (n = 402).

Assessing driver behaviors by Time-periods and Collision Types three-hour increments, driver inattention or distraction was the leading cause for The leading collision type from midnight half of the time increments (6:00 a.m. to to 3:00 a.m. was off-road left side, 3:00 9:00 a.m., 9:00 a.m. to noon, noon to 3:00 a.m. to 6:00 a.m. was sideswipe, and 6:00 p.m., and 6:00 p.m. to 9:00 p.m.). a.m. to midnight was rear-end collisions. Driver inattention or distraction The second leading collision type was the second leading cause from 3:00 midnight to 3:00 a.m. was off-road right p.m. to 6:00 p.m. (the first was following side, 3:00 a.m. to 6:00 a.m. was off-road too closely) and 9:00 p.m. to midnight left side, 6:00 a.m. to 6:00 p.m. was (first was illegal or unsafe driving speed). sideswipe, 6:00 p.m. to 9:00 p.m. was off- Driver inattention or distraction road left side, and 9:00 p.m. to midnight was the third leading cause from midnight was sideswipe collisions. to 6:00 a.m. with illegal or unsafe driving Time-periods and collision types

9 Chi-Square Goodness-of-Fit test were by lines and trend lines were added to shown in two scenario tests. The first test even spikes of extreme variance from year consolidated time-periods to six-hour to year. groups (midnight to 6:00 a.m., 6:00 a.m. to noon, noon to 6:00 p.m., and 6:00 p.m. to Collision Types midnight), vehicle-vehicle collision types (rear-end, sideswipe, right-angle, and The two most occurring collision types head-on), vehicle maneuvering off road were rear-end collisions at 49.5% and (off-road left side and off-road right side), sideswipe collisions at 17.2% (Figure 5). and other collision type with collision not They were analyzed, alongside the total specified. The Chi-Square statistic number of crash occurrences (n = 402), for calculated here was significant indicating similarities of frequencies and trends that a dependent association between six- (Figure 8). hour increment time-periods and collision types existed (p < 0.05). Collision Types by Year (totals) 50 The second test for time-periods All Types and collision type combined time into 12- hour increments and combined other 25 collision type with collision not specified. Rear-End Again, the calculated Chi-Square statistic was significant indicating that a dependent Sideswipe association between time of 12-hour 0 2006 2010 2015 increments and collision types existed (p < Figure 8. Total crash occurrences by year. All 0.05). occurrences, rear-end collision types, sideswipe collision types (n = 402). Testing for Equality of Crashes Total number of crash occurrences Collision type, driver behavior, and time (n=402) Chi-Square Goodness-of-Fit test were used to calculate Chi-Square revealed statistical equality from 2006 to Goodness-of-Fit tests to determine 2015 (p > 0.05), indicating that significant equality by year. statistically significant equal number of crashes occurred from year to year as Chi-Square Goodness-of-Fit Test shown in Table 3. Rear-end collisions (n=402) Chi- Significance of statistical equality was Square test revealed statistical equality determined by year for overall crash from 2006 to 2015 (p > 0.05), indicating frequencies, rear-end collisions, sideswipe again, a statistically equal number of rear- collisions, driver inattention or distraction, end collisions occurred from year to year. and illegal or unsafe driving speeds. These Since all the rear-end collisions occurred particular collision types and driver from 6:00 a.m. to midnight, evaluating the behaviors are of interest because of their Chi-Square Goodness-of-Fit for this high frequency of occurrence as shown in specific time-period was identical to the Figures 4 and 5. prior results. Additionally, collision types and Sideswipe collisions (n=402) Chi- driver behaviors were plotted on line Square test revealed statistical equality graphs where yearly totals were connected from 2006 to 2015 (p > 0.05). A statistically equal number of sideswipe

10 collisions occurred from year to year. similarities of frequencies and trends Since 88% of sideswipe collisions (Figure 10). occurred from 6:00 a.m. to midnight, whereas 100% of rear-end collisions Driver Behavior by Year (totals) 50 occurred during this time-period, All Behaviors evaluating Chi-Square Goodness-of-Fit test revealed statistical equality from 2006 25 to 2015 (p > 0.05), indicating a Distraction statistically equal number of sideswipe collisions from 6:00 a.m. to midnight Speeding occurred from year to year. 0 2006 2010 2015 Figure 9. Total crash occurrences by year. All Table 3. Yearly crash occurrences Chi-Square occurrences, distraction driver behavior, speeding Goodness-of-Fit test for significant equality of driver behavior (n = 402). occurrence (n = 402). Expected frequency noted in parentheses. Distraction Collision Types by Year (totals) All Crash Occurrences Year Frequency 20 2006 48 All Distraction (40.20) 2007 44 (40.20) 10 Rear-End 2008 37 (40.20) Sideswipe 2009 42 (40.20) 0 2010 38 2006 2010 2015 (40.20) 2011 50 Figure 10. Total crash occurrences by year as (40.20) influenced by driver distraction. Rear-end and 2012 36 sideswipe collisions due to distraction (n = 402). (40.20) 2013 30 (40.20) Collisions attributed to driver 2014 35 (40.20) inattention or distraction (n=402) Chi- 2015 42 Square Goodness-of-Fit test revealed (40.20) Total 402 statistical equality from 2006 to 2015 (p > 2  0.05,9 = 16.919 0.05), indicating a statistically equal 2 = 8.50; therefore, accept null hypothesis (p > 0.05). number of collisions occurred from year to year. Similar results were found for Driver Behaviors sideswipe collisions attributed to driver inattention or distraction; however, this The two highest percentages of collisions was not the case for rear-end collisions. due to driver behavior were driver Rear-end collisions attributed to inattention or distraction at 27.2% and driver inattention or distraction revealed illegal or unsafe driving speed at 21.5% that they were not statistically equal from (Figure 4). They were analyzed, alongside 2006 to 2015 (p < 0.05). Subdividing the the total number of crash occurrences (n = observed frequencies and recalculating the 402), for similarities of frequencies and Chi-Square statistic goodness-of-fit test trends (Figure 9). revealed that all years were statistically The 27.2% collisions due to driver equal (p > 0.05), except for year 2010 inattention or distraction, rear-end and (Table 4). sideswipe collisions were analyzed for

11 Regarding the 21.5% collisions due Sideswipe collisions attributed to to illegal or unsafe driving speed, rear-end illegal or unsafe driving speeds revealed and sideswipe collisions were analyzed for statistical equality from 2006 to 2015 (p > similarities of frequencies and trends 0.05), indicating a statistically equal (Figure 11). number of collisions occurred from year to year. Similar results were found for rear- Table 4. Yearly rear-end collisions attributed to end collisions attributed to illegal or driver inattention or distraction Chi-Square unsafe driving speeds. Goodness-of-Fit test for significant equality of occurrence (n = 402), but with 2010 data removed. Expected frequency noted in parentheses. Discussion Driver Inattention or Distraction Year Rear-End Collisions 2006 13 The Chi-Square Goodness-of-Fit test was (9.22) used to determine association and equality 2007 13 (9.22) between collision types, driver behaviors, 2008 7 driver age, and time-periods. (9.22) 2009 5 Significant dependent association (9.22) existed with driver behavior and collision 2011 14 (9.22) types. Rear-end and sideswipe collisions 2012 8 were the top two occurring collision types (9.22) 2013 6 with commonplace driver behaviors of (9.22) driver inattention or distraction, and illegal 2014 13 (9.22) or unsafe driving speed. 2015 4 Assessing significant equality of (9.22) occurrence during 2006 – 2015, sideswipe Total 83 2 collisions in conjunction with driver  0.05,8 = 15.507 2 = 13.83; therefore, accept null hypothesis (p > 0.05). inattention or distraction, or illegal or unsafe driving speed were statistically Collisions attributed to illegal or equal. As well, rear-end collisions were unsafe driving speeds (n=402) were statistically equal, except with driver statistically not equal from 2006 to 2015 inattention or distraction which occurred (p < 0.05). Subdividing the observed statistically higher in 2010. frequencies and recalculating the Chi- A significant association existed Square Goodness-of-Fit test revealed that between driver age and collision types. all years were statistically equal (p > Investigating further, rear-end collisions 0.05), except for year 2006. were the leading collision type across all driver ages; sideswipe collisions were the Speeding Collision Types by Year (totals) second leading collision type for driver 20 ages 30-59 year-olds and 70-89 year-olds – nearly all age groups. All Speeding Significant equality of occurrence 10 Rear-End for rear-end collisions, as well as sideswipe collisions, occurred across all Sideswipe years. Rear-end and sideswipe collisions 0 2006 2010 2015 did not discriminate occurrence by age and Figure 11. Total crash occurrences by year as occurred in statistically equal numbers influenced by driver speeding. Rear-end and from 2006 to 2015. sideswipe collisions due to speeding (n = 402).

12 Significant dependent association collisions were the second leading existed with time-periods and driver collision type as shown in Figure 5; the behavior. The leading driver behaviors flatter trend line is evident in the 32.3% attributed to 75% of the three-hour time- difference between rear-end and sideswipe periods were driver inattention or collision occurrence (Figure 5). distraction, and illegal or unsafe driving Comparing the total number of speed. Regarding significant equality of crash occurrences (n = 402) by year with crash occurrence attributed to driver the crash occurrence frequencies of driver behavior, all years were statistically equal, behaviors (Figure 9) is remarkably similar except illegal or unsafe driving speed to collision types (Figure 8). Trend lines occurred statistically higher in 2006. for collisions attributed to driver A significant dependent association inattention or distraction, and illegal or existed between time-periods and collision unsafe driving speed have similar types. Rear-end and sideswipe collisions downward slope rates as well frequencies accounted for 87.5% of the three-hour by year (Figure 9). Trend line slope time-periods. The time-periods spanning similarities are reflected by the 5.7% 6:00 a.m. to midnight comprised of the difference as first and second highest majority of rear-end and sideswipe contributing factors to occurrences (Figure collisions, and occurred in statistically 4). equal numbers across all years. The total number of crash The total number of crash occurrences resulting from driver occurrences (n = 402) by year showed a inattention or distraction (n = 402) by year downward sloping trend line indicating showed a downward sloping trend line crash occurrence frequency decreasing indicating crash occurrence frequency (Figure 8). Comparing crash occurrences decreasing (Figure 10). Comparing with attributed to rear-end collisions to all types the occurrences resulting in rear-end of collisions, the trend line slopes are collisions from driver inattention or somewhat similar as does the frequencies distraction, the trend line slopes are of occurrence by year (Figure 8). The somewhat similar (Figure 10), indicating similarities indicate possible relationship possible relationship between driver between rear-end collisions and all inattention or distraction and those collisions, and furthermore extrapolate that resulting in rear-end collisions. However, most collisions are rear-end collisions as dissimilar extrapolations made from shown in Figure 5 of percentages of comparing all occurrences from driver collision types contributing to crash inattention or distraction with those occurrences. resulting in sideswipe collisions. Here, the Again, comparing crash trend line slope of sideswipe collisions has occurrences attributed to sideswipe less slope (flatter) indicating less collisions to all types of collisions, the relationship between driver inattention or trend line slopes are somewhat similar, distraction and those resulting in sideswipe although the sideswipe trend line has less collisions. slope (flatter) as shown in Figure 8. The The total number of crash similarities indicate a possible occurrences resulting from illegal or relationship, but not as strong as the unsafe driving speed (n = 402) by year relationship between rear-end collisions showed a downward sloping trend line and all types of collisions. Sideswipe indicating crash occurrence frequency

13 decreasing (Figure 11). Comparing with and I-35W interchange. the occurrences resulting in rear-end collisions from illegal or unsafe driving Conclusion speed, the trend line slopes are somewhat similar (Figure 11), indicating possible The I-494 and I-35W cloverleaf relationship between illegal or unsafe interchange safety ranking is among the driving speed and those resulting in rear- worst for interchanges in the Minneapolis end collisions. The trend line from metro area (Short Elliott Hendrickson, occurrences resulting in sideswipe Inc., 2010). As analysis in this study collisions is flat (nearly a slope rate of validates, the majority of crashes from zero) which is dissimilar to the trend line 2006 through 2015 occurred during good of all occurrence from illegal or unsafe weather conditions which suggests driving speed indicating less relationship modification to driver behavior is highly between from illegal or unsafe driving desirable, especially considering driving speed and those resulting in sideswipe challenges of inclement weather collisions. conditions were shown as noninfluential Statistically higher occurrences of factors to crash occurrence. crash incidents due to driver behavior in Targeting driver behavior 2006 (illegal or unsafe driving speed) and modification efforts to reduce motor 2010 (rear-end collisions caused by driver vehicle crashes in the I-494 and I-35W inattention or distraction) could have been cloverleaf interchange is highly influenced possibly by better weather recommended. Effective educational driving conditions, for which drivers tend objectives addressing the driver behaviors to drive faster (Minnesota Department of of inattention, distraction, and speeding Public Safety, 2016). are suggested for drivers of all ages, Events in the years between 2006 driving during any hour of the day. and 2010 may have decreased influence in 2007 – 2009, resulting in 2006 and 2010 Data Limitations appearing as anomalies of higher influence. The economic downturn from A remarkable 87% of the data points are the Great Recession (2007 – 2009) may located in the center of the interchange. have affected driving behavior as This seemingly indicates the most likely unemployment grew and recovery outlook location for crash occurrence or, perhaps, dimmed (Streff, 2001). Highway a data recording preference. construction in 2008 – 2009 along arteries Cross-referencing data point surrounding the I-494 and I-35W locations with other descriptive attributes interchange may also have affected traffic about crash occurrences (such as roadway flow in the area. direction, character, design; vehicle travel Further research into influences direction; and collision type) did not such as weather, economic, construction, indicate an assurance to center interchange convention and social events, and personal location. Nor did the descriptive attributes technology devices from a historical provide a general sense of occurrence perspective would be beneficial in placement to confidently adjust location. understanding the underlying influences Positional accuracy would have impacting driver behavior leading to crash opened opportunities to spatial hot spot incidents from 2006 to 2015 at the I-494 and regression analysis for areas of most

14 likely occurrence within the interchange. References Positional accuracy also would have led to 2- and 3-dimensional mapping exploration Hotchkin, S. 2017. The amazing world of: for visual analysis and comprehension. Interchange designs (sightseeing series). Retrieved from http://www.sehinc.com/ Statistical Analysis Concerns news/amazing-world-interchange- designs. Results and conclusions were based on the Minnesota Department of Public Safety. data presented. Possible chance of error n.d. In case of a crash: A guide for may occur in analysis at any point which Minnesota motorists. Retrieved from materializes a false conclusion. Additional https://dps.mn.gov/divisions/ots/ information enhancing the data may educational-materials/Documents/In- influence analysis and conclusions Case-of-Crash-Brochure.pdf. likewise. As always, additional studies Minnesota Department of Public Safety. over more years of data along with a study 2016. Minnesota motor vehicle crash of variables and conditions impacting facts 2015. St. Paul, MN: Office of driving would be highly desirable. Traffic Safety. Minnesota Department of Transportation. Acknowledgements 2013. I-494/I-35W interchange: Fact sheet 1 – interchange history & need. I thank Mr. Derek Leuer and Ms Kitty Retrieved from http://www.dot.state.mn. Hurley for access and speedy delivery of us/metro/projects/i494and35winter site I-494 and I-35W crash data for which change/pdf/interchangehistoryneed.pdf. made this study possible. Short Elliott Hendrickson, Inc. 2010. The honor of Dr. David Preliminary design report I-494/I-35W McConville as advisor and professor – interchange preliminary (seh no. mntco always available, realistic, and kindly 107371). Minnetonka, MN: Author. patient – and Mr. John Ebert as co-advisor Streff, E. (Ed.). 2001, June. Connecting and professor – always quick with the Minnesota safety agenda: Towards enthusiasm and encouragement – were zero deaths. Minneapolis, MN: Center appreciated. for Transportation Studies, University of Ms Greta Poser, professor, whose MN. patient expertise assures solid GIS U.S. Census Bureau, Population Division. foundation. Saint Mary’s University of 2018. Annual estimates of the resident Minnesota GIS Department where diverse population: April 1, 2010 to July 1, learning environment built a rich learning 2017. Retrieved from https://factfinder. community was always noted and census.gov/faces/tableservices/jsf/pages/ appreciated, too. productview.xhtml?src=bkmk. Thanks to my husband and son for Zar, J. 2010. Biostatistical analysis, 5th allowing me to cocoon into studies for ed., pp. 466-516. Upper Saddle River, years – life will return to normal…soon! NJ: Prentice Hall, Inc. And, to my parents for instilling the importance of education – thank you.

15