Traffic Records Forum 2011 Understanding Traffic Data: How To Avoid Making the Wrong Turn Presenter: Marc Starnes (202) 366-2186 marc.starnes@dot.gov August 3rd, 2011 1
Summary of Topics Police Crash Reports (PARs) Restraint Use Injury Severity Contributing Factors Distraction Rates vs Counts Exposure 2
Summary of Topics Stratification Belt Use Data in Different Databases Effectiveness Relative Risk & Odds Ratios 3
Please, share your thoughts. That s how we learn the most. 4
Police Crash Reports What exactly is on your State s police crash report (PAR)? When was it last changed? What were the changes? Only a few pages but very important impact on traffic data, and resulting budgets, laws, and more 5
Ch-ch-ch-ch-changes Eating habits before marriage, and five years after marriage 6
Ch-ch-ch-ch-changes Eating habits before marriage, and five years after marriage Social life before and after becoming parents 7
Ch-ch-ch-ch-changes Eating habits before marriage, and five years after marriage Social life before and after becoming parents Sports interests in North Carolina vs sports interests in Texas 8
Ch-ch-ch-ch-changes Eating habits before marriage, and five years later Social life before and after becoming parents Sports interests in North Carolina vs sports interests in Texas Be aware of (data) differences!!! 9
Police Crash Reports How do police crash reports (PARs) vary from State-to-State? Examples - Restraint Use Injury Severity Contributing Factors Distraction 10
Restraint Use What s Needed for Research Child Restraint Use Categories Lap/Shoulder Belt Child Safety Seat (not specific type) Categories on only a few State PARs o Rear-facing CSS o Forward-facing CSS o Booster Seats Due to lack of data, no studies have shown booster seats to have a significant fatality effectiveness, compared to seat belts 11
Injury Severity Include Injured Passengers or All Passengers in PAR Where are people seated in the crash? Are 8-12 year olds in the back seat? Without Seating Position coded for All Passengers, research is limited Front/side airbags, ESC, booster seats 12
Contributing Factors Coding variance between States Select 5 categories for contributing factors, or 3 categories, or 1 category, or as many as officer wants Driver, Vehicle, Environment or all grouped together? 13
Contributing Factors Impact on percentages Driving too fast for conditions Failed to yield right of way Cell phone How many categories can be chosen on the PAR in your State for each type of Contributing Factor? Cell phone gets left out 14
Contributing Factors - Ethics Is cell phone use an option for the officer to select on the PAR? Should it be? How much information on each crash is an officer expected to be able to properly code, even if s/he is doing the best possible job? 15
Contributing Factors What info about cell phones do we need to determine fault What info do we need to determine how unsafe cell phones are? Pretend we know it all 16
Contributing Factors No conclusive research on how much cell phone talk, or texting, impacts safety Was there a cell phone in the car? Was the cell phone in use? By driver? By passenger? Was the person on the cell phone at fault? Harder to estimate than impact of seat belts, air bags, motorcycle helmets 17
Distraction Coding on PARs Some PARs have Distraction as a separate box the police officer must check (yes/no) Distraction, Inattention, Cell phones, eating, text messaging, GPS device Percent fatal crashes involving NHTSA definition of distraction, by State, varies 0.0% to 58.0% in 2009 18
Distraction Coding on PARs State X 2000-05 - below 2 % distraction 2007-09 - 37 to 41 % distraction State Y 2000-03 - above 32 % distraction 2005-09 2 to 8 % distraction State Z 2000-05 3 to 7 % distraction 2007-09 44 to 53 % distraction 19
Rates vs Counts 20
Rate Examples Rates are counts / exposure 55 miles/hr o Counts = miles o Exposure = hours 19 points/game o Counts = points o Exposure = games o 19 points in one game, or 190 points in 10 games 21
Rate Question John drove 30,000 miles and got three speeding tickets Bob drove 5,000 miles and got one speeding ticket Who has the higher rate of speeding tickets, John or Bob? 22
Rate Question John - 30,000 miles and three tickets Bob 5,000 miles and one ticket Different rates John - 1 ticket per 10,000 miles traveled Bob - 2 tickets per 10,000 miles traveled Different exposures John drove 30,000 miles, Bob 5,000 miles 23
2009 Data Rate or Count? 1.14 fatalities per 100 million vehicle miles traveled 33,808 fatalities 11.01 fatalities per 100,000 population 12,233 fatalities where highest BAC in crash was.08+ 24
Rate Example - Percent 77 percent seat belt use o Count = People buckled up in observed vehicles o Exposure = Total people in all vehicles viewed o Rate = Count / Exposure o Rate = 7700 people buckled / 10000 total people =.77 Converted to percent so it is easier to discuss Rate = 77 percent Units cancel (Don t say.77 people per people) All percents are rates!!! o Interest rate of 4.5 percent 25
Exposure Options for Fatality Rates Population Vehicle Miles Traveled (VMT) Registered vehicles Licensed Drivers 26
Population as Exposure Nation/State By: Age, Gender, Race, Hispanic Origin City Counts, County Counts Limitations Shouldn t compare city with metro/subway (NYC, Boston) vs city without (LA, Houston) Not adjusting for VMT 27
VMT as Exposure National: By Vehicle Body Type (i.e. pickup, SUV, van, passenger car) State: by Roadway Function Class (i.e. interstate highway) Limitations VMT not able to be stratified by passenger age, gender, alcohol 28
Registered Vehicles as Exposure National: By Vehicle Body Type State: total only Limitations Not stratified by age of owner, gender of owner, road type 29
Licensed Drivers as Exposure National counts: By Age and Gender State counts: Totals only Limitations Passengers not included Does the minivan have 7 passengers, or no passengers 30
State and County Level Exposure Data Quickfacts.census.gov Demographic Data All States and Counties Cities with population of more than 25,000 Info on age, race, gender, income, population per square mile, and much more 31
Rates With Little Exposure Fatality rate of 16 year-old drivers in Union County, NC in 2009 One multi-vehicle crash can make rate 3 times as high as last year Look at 5 to 10 years combined Expand group All of NC Age 16-20 32
Breaking Data into Two Groups Motorcycle single vehicle fatal crashes 1997 2006, operator age 21+ 81 percent helmeted in law States 24 percent helmeted non-law States Benefit of breaking data into groups BAC levels of weekday / daytime BAC levels of weekend / nighttime 33
Types of Restraint Use Data Comparison of different databases Fatal crash PARS (FARS) Non-fatal crash PARS (NASS-GES) Data from observational study (NOPUS) 34
Fatal Crash Belt Use Data Passenger Vehicle Occupant Fatalities 23,382 in 2009 43 percent restrained 49 percent unrestrained 7 percent unknown Lap/shoulder belts and Child Safety Seats Restrained is including occupants where officer coded improper use 35
Non-Fatal Crash Belt Use Data Restraint use in passenger vehicles among nonfatally injured occupants Around 90 percent restrained for a long time Yes officer, we were all wearing our seat belts! Problems with using restraint use data for nonfatally injured occupants from PARs 36
Observational Belt Use Data National Occupant Protection Use Survey (NOPUS) 85 percent seat belt use in 2010 7 AM to 6 PM Primary law States 88 percent Secondary law States 76 percent Produces national estimates, not State estimates No info on the other 13 hours of the day 37
Effectiveness Lives Saved FAQs, DOT HS 811 105 Effectiveness is a quantitative estimate of how a safety device (e.g., seat belt, air bag, and motorcycle helmet) improves a person s chance of surviving a potentially fatal crash Also effectiveness for minimum drinking age law 38
Effectiveness Lives Saved FAQs, DOT HS 811 105 Example: A driver lap/shoulder belt in a passenger car has an estimated effectiveness of 45 percent. Therefore the passenger car driver in a potentially fatal crash is 45 percent less likely to be fatally injured if they were buckled up 39
Effectiveness Example effectiveness estimates Child safety seat for infants in light trucks / vans 58 percent effective Frontal driver side air bag 14 percent effective Seat belts Depend on vehicle type, seat position, passenger age, belt type 19 percent lap belt in front middle of PC 73 percent - Lap/shoulder belt in rear outboard of LTV 40
Probabilities, Relative Risk, and Odds Ratios 41
Probabilities (hypothetical) Lets say that we did an observational study and found that 2.7 percent of male drivers did not put their infants in CSS, compared to 0.9 percent of female drivers Probability an infant was unrestrained by a male was 0.027 = (2.7 / 100) Probability an infant was unrestrained by a female was 0.009 = (0.9 / 100) 42
Relative Risk The relative risk (RR) of infants being unrestrained by a male (vs female) is: RR = p 1 /p 2 = p male /p female 0.027 / 0.009 = 3 Relative risk is calculated to be 3 o A male is three times as likely as a female to have their infant unrestrained This example is hypothetical 43
Odds The odds of infants being unrestrained by a male is: Odds = [p/(1-p)] = p male /(1-p male ) 0.027/(1-0.027) = 0.02774 Odds are not used much by themselves in research. More often, two odds are compared to get an odds ratio 44
Odds Ratio The odds ratio (OR) of infants being unrestrained by a male (vs female) is: OR = Odds male /Odds female [p male /(1-p male )] / [p female /(1-p female )] = [0.027/(1-0.027)] / [0.009/(1-0.009)] = Odds ratio is calculated to be 3.054 o A male is three times as likely as a female to have their infant unrestrained o OR (3.054) is close to RR (3) when p is small 45
Young Driver Fatalities Factors important to be aware of when looking at young driver fatality trends Graduated driver licensing Age when driver can first get license Restraint use, Vehicle types (i.e. more or less motorcycles), BAC levels, Speeding Much more 46
Young Driver Fatalities Factors in young driver fatality trends Graduated driver licensing Age when driver can first get license Restraint use, Vehicle types (i.e. more or less motorcycles), BAC levels, Speeding Vehicle Miles Traveled o Can families afford more or less cars? o Do young people have more or less jobs to drive to? 47
Push the Boundaries Think big picture What data do we have? Do you know? What other data do we want? Think about what data can be later collected to help answer your questions Adjust PAR Conduct a study 48
Questions? 49