Effectiveness of Anti-Drunken Driving Campaign: Rajasthan Experiment Design By Nina Singh, IPS Inspector General of Police, Rajasthan Rajasthan:Geographical Location Background Road Accidents killed more than 9,100 people and injured more than 31,000 in Rajasthan (2010) Drunken driving one of the major concerns Absence of segregated data about the reasons of these accidents Enforcement by local police stations Low Priority Work 2 MIT Poverty Action Lab Goal: Improve effectiveness of programs by providing policy makers with clear scientific results that help shape successful polices Key Approach: Compare randomly chosen reformed (“treatment”) areas with random un-reformed (“control”) areas and examine difference in outcomes Applies randomized trial approach to a variety of projects in different fields Health Education Governance Reform (such as Police Reforms) Previous collaboration with Rajasthan Police: “Police Performance and Public Perception” (2005-2008) Interventions Use of Breath-analyzers at check points Introduction of dedicated police teams from Reserve Lines for enforcement Use of GPS enabled tracking system for vehicles used by the dedicated teams from the Reserve Police Line 4 Breath Analyzers Device features Provides rapid evidence of blood-alcohol content Automatically maintains a record of the date, time, and alcohol level of each breath test 5 GPS Monitoring Device features Provides up-to-the-second information about vehicle location Maintains a record of vehicle’s travel history Displays GPS information via an online Google Maps portal accessible to J-PAL researchers, District Police, and Jaipur City Control Room 6 Objectives Evaluate the impact of the three interventions: Breath analyzers on reducing road accidents Dedicated police teams on enforcement Technology aided supervision (GPS) on execution of interventions Collect objective evidence of success 9 Pilot Districts Jaipur Rural contributes 7.9% of total deaths in Rajasthan 8.2% of total accidents in Rajasthan Bhilwara contributes 3.9% of the total deaths in Rajasthan 4.0% of the total accidents in Rajasthan Both the districts have long stretches of National Highways 10 Methodology: RCTs 40 police stations in the 2 districts were randomly divided into: How? “Treatment” stations, each holding 2 check points per week between 7 pm and 10 pm “Control” stations, doing no special enforcement Computerized random assignment Designed so treatment and control groups are similar in terms of accident rates, geographic locations, and proximity to national highways Why? With randomization we expect no systematic differences between treatment and control groups Thus, control group can serve as an accurate benchmark for measuring treatment group outcomes 11 Flow Diagram Police Stations Treatment Police Stations Surprise Check Points Police Station Teams Police Lines Teams Control Police Stations Fixed Check Points Police Station Teams Police Lines Teams Fixed/Surprise Check Points Treatment police stations were further randomly divided into: Fixed-Check Point stations: Fixed location and days of checking. Surprise- Check Point stations: Different days and locations of checking, thereby incorporating the element of surprise. Why? Gives objective evidence of whether police should Concentrate enforcement in high-risk areas, or Vary check point locations, to catch offenders offguard. 13 Police Station/Police Line Teams Two types of teams constituted in treatment stations and randomly assigned the duties for conducting the checkings: Local police station teams Conducted 2 checkings per week in the police station area Dedicated teams from the district Police Reserve Lines Conducted 6 checkings per week at 3 different police station areas Assigned dedicated police jeeps, equipped with GPS devices Why? Determine whether the dedicated teams are better while enforcing checkings compared to the police station teams 14 Checking Schedule (30 days of enforcement) Sources of Data Breath analyzer memory GPS database Police logs kept at checking points Accident data from Police Stations Court records Independent surveys by J-PAL Researchers and Surveyors Regarding the traffic flow, police checking pattern and drunk drivers caught at the checking points Regarding the general traffic patterns in absence of checking points 20 Statistical Significance of Pilot Heads Probability that the reported effect is not due to random chance Injuries 66% Deaths 69% Night Accidents 84% Highway Accidents 66% Serious Accidents 34% Total Accidents 19% Future Scale Up Approximately 11 districts Maintain successful practices from pilot Representative sample, based on statistical indicators, accident rates, geography, demographics Proposed district list: Ajmer, Alwar, Banswara, Bharatpur, Bhilwara, Bikaner, Bundi, Jaipur Rural, Jodhpur, Sikar, Udaipur Continued use of dedicated Reserve Line teams Both “Fixed” and “Surprise” checking strategies Use of GPS devices for monitoring Lines teams Comprehensive, objective data, including traffic analysis by JPAL Improve upon pilot design More systematic use of breath analyzers Longer intervention, in order to assess sustainability Introduce variation in number of checkings per week Days/Time of the checkings Scope of improvement: 1 Infrequent use of breath analyzers 14.8% of passing drivers were stopped by police Only 1.2% received breath test More frequent use would send a stronger message, and perhaps help police catch more drunk drivers. 20 Percent of passing drivers stopped by police % 15 10 5 Percent of passing drivers given breath test 0 7:30 to 8:00 8:00 to 8:30 8:30 to 9:00 9:00 to 9:30 9:30 to 10:00 Scope of improvement: 2 Most drunken drivers caught on Tuesdays and Thursdays: 23.8% more than on Saturdays and Sundays. Does that mean more drinking on these nights? 2 Avg. number of drunks caught 1.679 1.5 1.356 1 0.5 0 Tuesday/Thursday Saturday/Sunday Hopefully results from the larger evaluation would help policy planners to make appropriate policy interventions to improve Road Safety. 26 Thank You