Chapter Five. Tool Integration

Introduction

There are several purposes for building an online web tool to assist decision makers in assessing and selecting potential interventions to prevent motor vehicle–related injuries for statewide implementation. First, for a state to implement a new intervention, a decision maker would want to know about the costs and effectiveness of this intervention from states that have implemented and experienced it. Yet, these data are hard to come by or scattered in many places. This tool documentation contains all of this evidence in one place.

Second, in addition to performing a conventional cost-effectiveness analysis, our tool conducts a separate analysis using a portfolio approach to account for the interdependencies among interventions. It is clear that interdependencies exist, and the issue is whether their inclusion makes a difference as to which interventions should be selected to yield the greatest benefit (e.g., the largest reductions in injuries and deaths) for a given intervention implementation budget. In some cases, interdependencies would lead to a different selection, so a methodology that incorporates interdependencies would be important in order to get the greatest benefit for a given budget. Even in cases in which interdependencies make no difference in selection, a tool that can show this is still useful in validating that choices made based on independent cost–benefit ratios are appropriate.

Third, this tool is designed to capture state-specific characteristics, such as demographics and traffic crash patterns. Because the cost and effectiveness data for various interventions are typically collected or estimated across only some states, we have developed a methodology to adjust both effectiveness and cost data to suit a decision maker for implementation in a particular state.

Fourth, this tool is meant to aid both selection and implementation of interventions. Even after interventions are selected, a state decision maker would still need to know what the different implementation activities are and how much each activity would cost. We have classified the cost to implement each intervention into ten components and further classified these into multiple subcomponents. Then, we collected or estimated those cost subcomponents that are relevant to a given intervention. Thus, the costs estimated for each component and subcomponent should be useful not only for intervention selection but also for intervention implementation.

This chapter contains four sections. The first section describes the methodology, assumptions, and model inputs for determining state-specific benefits of interventions in terms of numbers of injuries and deaths reduced. The effectiveness data from Chapter Four are used in this step. The second section discusses how the state-specific projected reductions in injuries and deaths are monetized. The monetary value consists of three parts: physical property damage, injuries, and deaths. The physical property damage is only a small component of the total economic loss from most crashes, but our methodology includes property damage associated with crashes that produce injuries and deaths. The third section of the chapter does the same for determining state-specific costs of implementing interventions. Finally, the fourth section describes how costs and monetized benefits from the preceding sections are used to determine the ranking and selection of interventions employing two different analytical methods. In both this chapter and Chapter Six, we use Ohio as an example to illustrate the various features of the tool.21

Methodology for Estimating State-Specific Injury and Death Reductions of Interventions

In this section, we discuss how we applied the effectiveness estimates to state-specific information on motor vehicle–related injuries and deaths. NHTSA provides annual data on fatal injuries due to motor vehicle crashes through FARS. We use the 2010 FARS data in our analysis to generate the baseline number of deaths in each state. These data include a large set of information about the crash, the people involved, and the circumstances. Using the 2010 FARS data, we calculate the number of deaths in each of the following categories:

  • total deaths in the state
  • deaths that are considered alcohol-related
  • deaths involving drivers with previous DWI convictions
  • deaths involving motorcycles
  • deaths involving bicycles
  • deaths occurring at intersections with traffic lights22
  • deaths of vehicle occupants23
  • deaths caused by drivers over age 70
  • deaths related to speeding.

These categories are necessary because the various estimates from the literature frequently focus on specific types of deaths. These categories help us operationalize the estimates in the literature by providing the correct base with which to work for each state. To estimate the number of deaths that would have been prevented in a state due to a specific intervention, we multiply the number of deaths in the relevant category by the intervention’s effect found in the literature. When a state already has an intervention in place, the effect of the intervention in reducing injuries and deaths would have already been realized and reflected in the FARS data. So a user can assume that any implemented intervention would have already achieved the effectiveness and paid the associated costs. In some cases, however, an intervention may be in place but not fully implemented. In other cases, there may have been changes in the status of the intervention since the data were collected. To address these issues, the tool is designed so that a user can select the set of interventions to be included as candidates for implementation when using the tool.

The deaths by category by state are shown in Table 5.1; note that the sum of the values in each row is higher than the total column because a crash can have multiple causes. Also, some columns include pedestrian deaths, and others do not. If the study on which we relied for a benefit estimate made clear that its fatality reductions included pedestrians, we included pedestrian deaths in that category; if not, we did not include them.

Table 5.1. Deaths per Year, by Category, 2010
State Total Alcohol-Related Previous DWI Motorcycle Bicycle Occurred at Intersection with Light Vehicle Occupants Drivers over 70 Speed-Related
Ala. 862 279 78 86 6 118 702 100 313
Alaska 56 16 1 9 0 6 41 4 23
Ariz. 762 194 31 92 19 128 477 150 228
Ark. 563 173 59 84 1 93 439 64 108
Calif. 2715 791 159 353 99 440 1,531 389 792
Colo. 448 127 25 82 8 155 310 63 159
Conn. 319 121 19 52 7 46 204 38 117
Del. 101 36 13 8 3 16 65 12 41
D.C. 24 5 0 1 2 1 6 1 7
Fla. 2445 660 105 396 83 1,385 1,375 448 425
Ga. 1244 298 87 127 18 255 907 194 212
Hawaii 113 42 11 26 3 11 55 13 46
Idaho 209 71 16 28 4 23 163 24 64
Ill. 927 298 30 131 24 182 631 140 405
Ind. 754 195 42 111 13 206 552 125 181
Iowa 390 90 40 60 8 59 294 56 61
Kan. 431 168 14 40 1 74 371 81 98
Ky. 760 171 95 96 7 209 589 94 151
La. 710 225 47 71 10 242 543 61 230
Maine 161 38 17 19 1 34 128 27 82
Md. 493 154 12 82 8 64 293 72 145
Mass. 314 115 34 56 6 26 185 47 64
Mich. 942 230 85 137 29 1 612 156 229
Minn. 411 127 43 48 9 84 309 66 92
Miss. 641 236 29 42 4 81 541 71 129
Mo. 819 258 39 95 7 107 651 124 317
Mont. 189 73 27 26 0 27 155 24 68
Neb. 190 51 28 14 2 28 163 29 34
Nev. 2 69 12 48 6 45 156 55 73
N.H. 128 44 7 28 0 15 91 21 62
N.J. 556 153 36 71 12 62 322 107 133
N.M. 346 111 5 39 8 114 258 35 129
N.Y. 1200 364 65 184 36 264 631 194 315
N.C. 1319 388 118 191 23 155 909 178 469
N.D. 105 47 16 15 1 7 80 9 40
Ohio 1080 341 97 170 11 154 789 159 293
Okla. 668 220 52 78 9 193 503 94 185
Ore. 317 71 13 38 7 33 203 61 92
Pa. 1324 433 85 223 21 170 908 209 672
R.I. 66 25 4 15 2 7 37 9 26
S.C. 810 3 41 101 14 120 590 79 273
S.D. 140 37 15 27 2 10 100 22 32
Tenn. 1031 283 78 136 4 105 797 134 222
Texas 2998 1259 112 415 42 409 2,141 358 1,149
Utah 236 44 19 20 7 32 176 27 91
Vt. 71 18 3 6 1 5 59 15 27
Va. 740 211 29 86 12 82 553 107 255
Wash. 458 170 14 69 6 55 315 64 171
W.Va. 315 88 4 33 3 18 263 46 130
Wis. 2 205 62 105 9 113 392 97 197
Wyo. 155 54 22 33 0 10 119 20 57
Toral 32,885 10,120 2,095 4,503 618 6,435 22,684 4,773 9,914

SOURCE: NHTSA, undated (c).
NOTE: The “Alcohol-Related,” “Vehicle Occupants,” and “Speed-Related” columns include pedestrian deaths. “Vehicle Occupants” includes people who were occupants of passenger vehicles or large trucks; 98 percent of these were passenger vehicles.
a The total is the total of all vehicle crash deaths in a state; it is not the sum of the columns because fatalities can fall into more than one category.

 

Although the FARS provides a census of motor vehicle–related deaths, we were unable to identify a similar source of comprehensive information on motor vehicle–related injuries. The available data sources provide only a sample of accidents. We chose to use NHTSA’s National Automotive Sampling System (NASS) General Estimates System (GES) for 2010, which provides information on crashes, their circumstances, and resulting injuries. We use these data to understand the number and characteristics of motor vehicle–related injuries. Unfortunately, the GES does not provide data for every state, and, even when crashes in a state are sampled, it is not intended to generate state-specific information on motor vehicle crashes. Consequently, we use the GES to generate ratios of injuries to deaths for specific types of crashes. The GES provides information on injuries and deaths for each sampled motor vehicle crash. We create the following categories:

  • alcohol-related
  • motorcycle
  • bicycle
  • occurred at an intersection with a light
  • vehicle occupants
  • drivers over age 70
  • speed-related.

For each category, we add all of the relevant injuries and deaths. We then create a ratio of the number of injuries per death, as shown in Table 5.2. To generate the number of injuries in that category in a state, we take the number of deaths in that category calculated in the FARS and multiply by the ratio generated in the GES to obtain our estimate of the number of relevant injuries. The result is an estimate of the baseline number of injuries in the state that are potentially affected by the intervention in question.

Unfortunately, the GES does not provide information about whether the driver has a previous DWI conviction, which is provided in the FARS and is necessary for understanding the effects of interventions targeted at repeat offenders. In this case, we use the alcohol-related injury-to-death ratio as an appropriate proxy to generate our injury numbers. To generate the intervention’s estimated state-specific effect on motor vehicle–related injuries, we multiply the number of injuries in the relevant category by the intervention’s effect found in the literature.24

Table 5.2. Injury-to-Death Ratios
Category Ratio
Total 106.5412
Alcohol-related 36.18325
Motorcycle 264.5426
Bike 171.547
Occurred at intersection with light 85.86092
Vehicle occupants 105.5553
Drivers over age 70 90.83724
Speed-related 82.60576

SOURCE: Calculated by RAND researchers based on FARS and GES data.

 

Informed by the available literature, we estimate the reduction in the percentage of injuries and deaths due to the corresponding intervention. Our estimates are shown in Table 5.3. For each intervention, the reduction in deaths does not apply to all motor vehicle deaths, only those with a cause that is affected by the intervention. For example, vehicle impoundment for DWI offenders affects only those deaths caused by drunk drivers. Each study that we have used makes clear the types of deaths that are being affected. If a state has 100 alcohol-related deaths per year and saturation patrols reduce deaths by 17.9 percent (i.e., 0.179 in Table 5.3), then the effectiveness of saturation patrol is the elimination of 18 deaths.

Table 5.3. Injury and Death Reduction Estimates
Baseline Intervention Reduction in Injuries Reduction in Deaths
Occurred at intersection with a light Red-light cameras 0.17 0.17
Speed-related Speed cameras 0.12 0.12
Previous DWI conviction Alcohol interlocks 0.24 0.24
Alcohol-related Sobriety checkpoints 0.2 0.081
Alcohol-related Saturation patrols 0.179 0.179
Bike Bicycle helmet laws 0.15 0.15
Motorcycle Motorcycle helmet laws 0.289 0.289
Vehicle occupants Primary enforcement of seat belt laws 0.07 0.07
Vehicle occupants Seat belt enforcement campaign 0.054 0.054
Previous DWI conviction License plate impoundment 0.27 0.27
Previous DWI conviction Limits on diversion and plea agreements 0.11 0.11
Previous DWI conviction Vehicle impoundment 0.304 0.304
Vehicle occupants Higher seat belt fines 0.072 0.072
Drivers over 70 In-person license renewal 0.090 0.090

SOURCE: Our analysis of literature as discussed in Chapter Four.

 

In many cases, the literature does not assess the reduction in the percentage of injuries due to an intervention. In almost all cases, we assumed that the reduction in the percentage of injuries due to an intervention was the same as the reduction for deaths. The only exception was for sobriety checkpoints, for which the literature did allow us to produce a separate estimate. To find the number of injuries eliminated, we multiply the number of injuries that result from a specific crash cause by the percentage of injuries reduced by an intervention.

To illustrate our method, we use sobriety checkpoints as an example. Fell, Tippetts, and Levy, 2008, studied several NHTSA-funded demonstration projects. They found a range of effects, and we take the average (8.1 percent) as our estimate of the intervention’s effect on deaths. We referenced Elder, Shults, et al., 2002, to obtain estimates of the effect that sobriety checkpoints have on injuries. This study included a review of the literature that looked at U.S. interventions. The authors reported a median finding in the literature of a 20-percent reduction in fatal and nonfatal injury crashes. For both estimates, the estimated effects refer to the intervention’s effect on alcohol-related injuries and deaths.

How does this translate into estimating the effects of implementing sobriety checkpoints in a state where they are not currently used? For example, according to the 2010 FARS, Michigan had 205 alcohol-related motor vehicle deaths. The 2010 GES data tell us that there are 36.2 alcohol-related injuries per alcohol-related death. Consequently, we estimate that there are 7,421 (205 × 36.2) relevant injuries in Michigan. We estimate that implementing sobriety checkpoints would save approximately 16.6 lives and prevent 1,484.2 injuries in Michigan per year.

Methodology for Monetizing Intervention-Related Reductions in Injuries and Deaths at the State Level

Once we estimated the reductions in injuries and deaths, we monetized the effects of the reduction in injuries and deaths associated with a particular intervention. Although this is not a formal cost–benefit analysis, we monetize the effects of the intervention so that we can combine the impact on injuries and deaths and generate a cost-effectiveness estimate that includes both outcomes of interest. We derived the costs per injury and death using a very simple method, relying heavily on estimates already available in the literature. Blincoe, Miller, et al., 2015,25 provided national unit costs by injury severity or death. The costs are separated into different categories: medical, emergency services, market productivity, household productivity, insurance administration, workplace costs, legal costs, travel delays, and property damage.26 These costs are then aggregated, and the resulting values are shown in Table 5.4.

Table 5.4. National per-Injury and per-Death Costs (adjusted to 2012 dollars)
Cost Category Injury Death
Medical 3,624 11,883
Emergency service 88 947
Market productivity 4,4326 979,925
Household productivity 1,501 304,406
Insurance administration 2,662 29,738
Workplace costs 503 12,372
Legal costs 1,225 111,812
Travel delay 1,308 6,006
Property damage 5,044 11,773

SOURCE: Blincoe, Miller, et al., 2014.
NOTE: Blincoe, Miller, et al., 2014, Table 1-2, reported costs by injury severity. We calculate a weighted average using the relative frequencies of each injury severity type (reported in Blincoe, Miller, et al., 2014, Table 5-14) and multiply by 1.05 to adjust for inflation.

 

We incorporate three important changes to these costs:

  • First, we adjust the unit costs for inflation to generate estimates in 2012 dollars.
  • Second, we adjust some of the costs because we believe that there is important variation at the state level.
  • Third, we aggregate the costs by injury severity into one metric.

The literature rarely provides us with any information about interventions’ effects on different types of injuries by severity. Instead, our analysis has considered interventions’ effect on injury and death rates. This is a nice balance between evaluating the heterogeneous impacts of different interventions and using available estimates in the literature.

We believe that state-level cost heterogeneity is especially important for market productivity, household productivity,27 and medical costs, so we adjusted these three categories by state. To do this, we use the state-specific price adjustments employed by CDC’s WISQARS cost-of-injury reports computed using the ACCRA Cost of Living Index data and population data.28 We use the values in the medical column to adjust the national estimates for medical costs in Table 5.4. We multiply the medical national cost by the medical state-specific adjustments for each state to derive a state-adjusted medical cost. Similarly, we use the productivity adjustments in Table 5.5 to adjust the market productivity and household productivity values from Table 5.4.


 

Table 5.5. State-Specific Price Adjusters
State Productivity (Market and Household) Medical
Alabama 0.839 0.898
Alaska 1.045 1.282
Arizona 0.855 0.976
Arkansas 0.779 0.893
California 1.077 1.071
Colorado 1.063 1.010
Connecticut 1.402 1.119
Delaware 1.052 1.160
District of Columbia 1.582 1.058
Florida 0.996 0.979
Georgia 0.867 0.989
Hawaii 1.016 1.061
Idaho 0.808 0.919
Illinois 1.044 0.993
Indiana 0.871 0.890
Iowa 0.907 0.893
Kansas 0.952 0.893
Kentucky 0.806 0.938
Louisiana 0.900 0.923
Maine 0.873 1.019
Maryland 1.192 1.005
Massachusetts 1.271 1.250
Michigan 0.909 0.937
Minnesota 1.063 0.986
Mississippi 0.747 0.978
State Productivity (Market and Household) Medical
Missouri 0.891 0.936
Montana 0.841 0.969
Nebraska 0.945 0.906
Nevada 1.048 1.038
New Hampshire 1.075 1.205
New Jersey 1.274 1.043
New Mexico 0.815 0.963
New York 1.227 1.070
North Carolina 0.871 1.004
North Dakota 0.902 0.926
Ohio 0.903 0.947
Oklahoma 0.885 0.947
Oregon 0.901 1.043
Pennsylvania 1.005 0.974
Rhode Island 1.022 1.114
South Carolina 0.803 1.000
South Dakota 0.878 0.932
Tennessee 0.862 0.912
Texas 0.963 0.950
Utah 0.808 0.918
Vermont 0.950 1.008
Virginia 1.071 0.935
Washington 1.047 1.149
West Virginia 0.765 0.915
Wisconsin 0.934 1.030
Wyoming 1.120 0.954

SOURCE: CDC, 2014b.

 

For the six other factors in Table 5.4, we use the national estimates for all states. We then simply add up the values for the nine cost categories to derive the cost of injuries and deaths in the state.

To arrive at total costs per injury, we need to aggregate the different costs reported by injury severity. Table 1-3 in Blincoe, Miller, et al., 2014, reports the relative frequencies of each injury type (based on severity). We translate these into the fraction of all injuries that are each injury severity type. For example, we calculate the fraction of all injuries with the maximum abbreviated injury score (MAIS) of 2. We add the costs of all injury types, weighted by these fractions, to arrive at the total cost of an average injury.

Finally, we multiply the cost of an injury in a state by the number of injuries that an intervention is estimated to prevent in that state. We do a similar calculation for deaths. We then add these numbers to arrive at the potential monetary savings associated with reducing injuries and deaths by implementing the intervention in each state.

The effectiveness of a particular intervention for preventing motor vehicle crashes is the number of deaths in a state prevented by the intervention multiplied by the state-specific cost per death, then added to the number of injuries prevented by the intervention, then multiplied by the state-specific cost per injury.

Table 5.6 indicates which states had the various interventions in place in most cases as of 2011.29 A cell is coded 1 if the state has the intervention, 0 if not, and 9 if unknown, according to our information on each state. For the two interventions coded as 9 (saturation patrols and seat belt enforcement campaigns), the tool allows a user to place either or both of them as implemented interventions or interventions to be considered for implementation. (See the fact sheets on these interventions in Appendix B for further information on their use.)

Table 5.6. Intervention Status, by State

View Table 5.6

 

To illustrate how we calculate and monetize the effect of a particular motor vehicle intervention, we use vehicle impoundment in Ohio as an example.

DeYoung, 1999, found that first-time DWI offenders with impounded cars had 24.7 percent fewer crashes than a similar group of DWI offenders who did not have their cars impounded. The reduction was 37.6 percent for repeat offenders. Using the relative rates stated in DeYoung’s paper, we calculate that the average reduction is 30.4 percent (Table 5.3). We assume that this reduction leads to proportional reductions in injuries and deaths.

According to the 2010 FARS, Ohio had 97 deaths involving drivers with prior DWI convictions (Table 5.1). We do not have a figure on injuries involving drivers with prior DWIs, so we impute this value using the 2010 GES. Because the GES does not report prior DWI status, only whether a crash involved alcohol, we use the ratio of injuries to deaths in crashes involving alcohol to estimate the ratio for crashes involving prior DWI convictions. For every death involving alcohol, there are 36.18325 injuries (Table 5.2). Consequently, we assume that Ohio had 3,509.5 injuries (97 deaths × 36.18 injuries per death) due to prior-DWI drivers in 2010. Implementation of vehicle impoundment in Ohio, then, is predicted to reduce the number of deaths by 29.49 (i.e., 97 × 30.4 percent) and the number of injuries by 1,066.97 (i.e., 3,509.5 × 30.4 percent) per year.

Then we take the national costs associated with the nine categories (Table 5.4) for injuries and deaths derived from Blincoe, Miller, et al., 2015. As described above, we then adjust three of the nine components by state-specific ratios (Table 5.5). The derived costs per injury and per death for Ohio are $19,794 per injury and $1,343,652 per death.

Finally, we multiply the number of injuries saved by the costs per injury and add the number of deaths saved multiplied by the costs per death. This is (1,066.97 × $19,794) + (29.49 × $1,343,652) = $60,740,834. This is the model’s estimate of the intervention’s monetized annual benefit of implementing vehicle impoundment in the state of Ohio.

Methodology for Estimating State-Specific Costs of Implementing Interventions

To calculate the total costs of developing, implementing, and maintaining the different interventions, we developed a cost-estimating structure and gathered the necessary data from a wide review of literature sources, addressing processes and costs, as well as from the different legal statutes and procedures pertaining to each intervention (e.g., state statutes addressing seat belt laws, fines, and administrative procedures). Data points were normalized so units would be consistent and costs would be expressed in 2012 dollars. Then we gathered relevant statistics for calculating implementation estimates at a state level (e.g., number of licensed drivers in a state).

The costs are broken down into ten components as described in Chapter Three. These ten cost components are broken down into 38 subcomponents and associated with specific interventions, as shown in Table 3.1 in Chapter Three.30

The model combines all of the aforementioned data to estimate the costs for each intervention in each state. To illustrate this, we walk through a sample state, Ohio, and an example intervention, vehicle impoundment. Vehicle impoundment has three basic cost elements: tow staffing, fines, and program management.

First, we calculate the costs associated with the tow portion of the program. As noted above, the most common cost paid by a city or state to tow a vehicle is $637. There were 36,528 DWI arrests in Ohio according to FBI statistics (FBI, 2011a). Under this intervention, we assume that each DWI arrest results in an impounded vehicle, so this amounts to about $23,271,000 in direct costs. Second, each person arrested must pay a fine to retrieve the vehicle or forfeit the car to the impound lot. The average income to the state from fines or forfeitures is about $520 per vehicle. With $520 multiplied by 36,528 arrests, the total potential income after rounding is $18,995,000. Finally, the state needs program oversight by a state employee for coordination and contracting purposes. We assume that the appropriate estimate is 2.5 state employees, with a full-time employee working 2,000 hours per year. Using BLS data for office workers, we selected an hourly cost of $24.59 per hour in 2012 dollars (converted from $23.91 in 2011 dollars in Table 3.3), including benefits, for government employees as representative of Ohio. This hourly wage is similar to the mean for the states ($25.14). This equals a total program staff cost of about $123,000, as shown in Table 5.7.

Table 5.7. Sample Cost Estimate, Ohio Vehicle Impoundment
Cost Element Cost Per Unit ($) Unit Number of Units Cost ($)
Tow staffing 637.06 Arrest 36,528 23,271,000
Program staff 24.59 Hour 5,000 123,000
Cost to state without fines Not applicable Not applicable Not applicable 23,394,000
Fines 519.97 Arrest 36,528 -18,995,000
Cost to state with fines Not applicable Not applicable Not applicable 4,400,000
Cost to offenders Not applicable Not applicable Not applicable 18,995,000

NOTE: Costs have been rounded to the nearest $1,000.

 

The cost of the program varies based on whether the revenues generated by the intervention are included in the calculation. Without fines, the cost to state is almost $23,394,000. When fines are included and can be used to fund the program, the net cost to the state is $4,400,000. We display the fines as a negative cost because they constitute revenue to the state. The tool also uses this convention: Positive costs are those that are actual costs paid by the state, and negative costs are revenues to the state. When the revenues exceed the cost of implementation, the total cost of implementing the intervention will be negative. This indicates that there is a net gain to the state: The revenue received is greater than the cost incurred.

This model provides insight into one possible budget outcome of the vehicle impoundment intervention. The range of costs discovered in our research suggests that some states have budget-neutral vehicle impoundment programs, while others do not. It depends on the combination of costs incurred and revenues generated. For example, if fines are too high for offenders to retrieve their vehicles, then more people will forfeit their vehicles, which are often worth very little at auction. Each state needs to develop its own budget and fine strategy based on its experience.

The annual cost for each intervention in each state for the case in which fines are included is shown in Tables 5.8 and 5.9. The annual cost for each intervention in each state for the case in which fines are excluded is shown in Tables 5.10 and 5.11.

Table 5.8. Annual Cost for Each Intervention for the Case in Which Fines Are Included: Red-Light Cameras, Speed Cameras, Alcohol Interlocks, Sobriety Checkpoints, Saturation Patrols, Bicycle Helmets, and Motorcycle Helmets ($)

View Table 5.8

Table 5.9. Annual Cost for Each Intervention for the Case in Which Fines Are Included: Primary Enforcement of Seat Belt Laws, Seat Belt Enforcement Campaigns, License Plate Impoundments, Limits on Diversion and Plea Agreements, Vehicle Impoundments, In-Person License Renewals, and Higher Seat Belt Fines ($)

View Table 5.9

Table 5.10. Annual Cost for Each Intervention for the Case in Which Fines Are Excluded: Red-Light Cameras, Speed Cameras, Alcohol Interlocks, Sobriety Checkpoints, Saturation Patrols, Bicycle Helmets, and Motorcycle Helmets ($)

View Table 5.10

Table 5.11. Annual Cost for Each Intervention for the Case in Which Fines Are Excluded: Seat Belt Enforcement Campaigns, Primary Enforcement of Seat Belt Laws, License Plate Impoundments, Limits on Diversion and Plea Agreements, Vehicle Impoundments, In-Person License Renewals, and Higher Seat Belt Fines ($)

View Table 5.11

Methodology for Identifying the Optimal Portfolio of Interventions

The portfolio analysis component of the tool helps define the best group of interventions for a state to implement within a given budget. The tool identifies which of the 12 interventions that have not yet been implemented in the state can yield the largest benefit, measured as the greatest reduction in the costs of injuries and deaths. The effectiveness and cost for each intervention estimated in Chapters Three and Four are used as inputs for this final step of optimization.

This optimization can be performed in two ways: the traditional cost-effectiveness analysis and our portfolio analysis. The difference between them is how they treat the interdependencies between interventions. The cost-effectiveness analysis ignores the interdependencies, while portfolio analysis incorporates them. Take the example of primary enforcement of seat belt laws and seat belt enforcement campaigns, both of which encourage occupants in passenger vehicles to wear seat belts. The interdependency between these two interventions is easiest to explain with an extreme hypothetical case, in which each intervention alone can reduce traffic deaths to zero (i.e., a 100-percent reduction). If so, it is clear that implementing both interventions would still cause the reduction to be only 100 percent, not 200 percent. Thus, in the case of two interdependent interventions with the same targeted population and reduction rates in deaths of R1 and R2, the reduction rate for both interventions implemented would be

1-(1-R1) x (1-R2).

Generalizing it to a combination of n such interdependent interventions, we get the combined reduction rate to be

1-(1-R1) x (1-R2) x … x (1-Rn).

On a different issue, there has been controversy about the real purpose of implementing interventions, such as red-light camera enforcement. Is it to generate income to the state? To improve traffic safety at an intersection? Both? To address this issue, we perform an analysis, whether cost-effectiveness or portfolio, under two cases: standard run with fines included and standard run with fines excluded. Fees, charges, and other income generated from interventions under consideration for implementation are also included in the definition of fines here.

Because the types of model runs that can be made and the kinds of screenshots that are displayed are detailed in the user manual (Chapter Six), here we focus on the description of methodology for the two types of analyses.

Cost-Effectiveness Analysis

This analysis is performed under the assumption that the interdependencies among interventions are ignored, as is typically done by traditional analytical methods. The analysis starts by asking for which state the user wants to perform the analysis. We will use Ohio as our example. For each of the 12 candidate interventions, we start with its Ohio-specific effectiveness and cost from Chapters Three and Four and take a ratio of its effectiveness to its cost. The ranking of interventions is based on the ratio; a larger ratio is more attractive and is ranked higher (i.e., more lives saved and more injuries prevented per dollar spent). In cases in which fines are included, the cost can be negative. This signifies that the fines, charges, and other incomes generated by the intervention exceed the costs of implementing the intervention. A negative cost will lead to a negative ratio. Clearly, a negative ratio ranks higher than a positive ratio because the former can generate effectiveness-monetized benefit and net income at the same time, while the latter would cost money to get the benefit.

How do we rank interventions that all have negative ratios? We start with the following observation and assumption. When implementing an intervention generates more in fines than in expenses, we assume that the net cash flow can be used to fund other interventions. Because of this possibility, we add the benefit, which has already been expressed monetarily in dollars, to the absolute value of the cost (i.e., the net cash flow) as a measure of the value generated by the intervention. One with the higher such value is more attractive and ranks higher than one with lower value. It should be emphasized that this ranking scheme applies only to interventions with negative ratios.

After a user identifies a state, the tool displays a screen showing which of the 12 interventions were already implemented in the state around 2010, the year of the traffic fatality data we used. The user can update the list of interventions that have not been implemented and that should stay in the pool for selection by simply clicking all such interventions. Once the user submits an annual implementation budget available and clicks the button to run the model, another screen is displayed. Interventions will be selected for implementation in ranking order according to their cost–benefit ratios, and the cost will be subtracted from the budget until the budget is fully consumed.31

Portfolio Analysis

Portfolio analysis differs from the traditional cost–benefit analysis by including interdependencies among interventions. There are two types of interdependencies: those between implemented and not-yet-implemented interventions and those between not-yet-implemented interventions only. Because the traffic fatality data have already reflected the impacts of interventions that existed at the time the traffic crashes occurred, the first type of interdependencies are ignored, and we focus only on the second type.

Of the 12 interventions, we have identified the following pairs and groups as having interdependencies among them because they target the same population based on crash cause:

  • alcohol interlock, license plate impoundment, limits on diversion and plea agreements, and vehicle impoundment (previous DWI convictions)
  • saturation patrols and sobriety checkpoints (alcohol-related offenses)
  • primary enforcement of seat belt laws, seat belt enforcement campaigns, and higher seat belt fines (vehicle occupants).

The portfolio optimization model is built with mixed-integer linear programming from the SAS software library. The user selects the state where the selected interventions will be implemented. The model’s inputs are the candidate interventions’ annual state-specific benefits and costs, estimated in Chapters Three and Four. The above three sets of interdependencies constitute the constraints in the model. Once a user provides the annual budget available for implementation, the model searches for the optimal combination or portfolio of selected candidate interventions that would produce the greatest effectiveness (i.e., the largest reduction in traffic deaths and injuries as expressed in dollars) for the budget provided. The mixed integer reflects that a candidate intervention is either wholly selected (1) or not selected (0), and the current version of the model does not allow selecting an intervention partially for implementation. The portfolio optimization is a linear-programming model, which has the objective function of maximizing effectiveness subject to constraints in interdependencies. Further, all the decision variables (i.e., which candidate interventions to choose) appear linearly in the objective function and the constraints. The traditional branch-and-bound search algorithm, which is periodically improved by SAS, is used in the model to find the optimal solution or portfolio.

Example of Cost-Effectiveness Analysis and Portfolio Analysis Calculations

To provide a concrete example of the calculations for both types of analyses, we illustrate the calculation of the deaths for interventions in Ohio.

We focus on the three of the four interventions listed above that affect people with previous DWI convictions that are not already implemented in Ohio (Table 5.6). These are alcohol interlock, limits on diversion and plea agreements, and vehicle impoundment. From Table 5.1, we find that, in 2010, 97 people were killed in motor vehicle crashes related to drivers with previous DWI. From Table 4.1 in Chapter Four, we find that alcohol interlocks reduce deaths by 24 percent so that the number of deaths avoided by implementing this intervention alone would be 97 × 0.24 = 23.[1] Limits on diversion and plea agreements would reduce deaths by 11 percent so that the number of deaths avoided by implementing this intervention alone would be 97 × 0.11 = 11. Vehicle impoundment would reduce deaths by 30.4 percent so that the number of deaths avoided by implementing this intervention alone would be 97 × 0.304 = 29.

If all three interventions are implemented together, the cost-effectiveness analysis would simply add these three estimates for reductions in deaths together and estimate the total reduction in deaths to be 63 (23 + 11 + 29). However, the portfolio analysis would take into account the interdependency between these three interventions and use the calculation method shown at the beginning of this section to find the total reduction in deaths to be only 51 (97 × 1 – [1 – 0.24] × [1 – 0.11] × [1 – 0.304]). Thus, ignoring interdependencies would lead to a total reduction in deaths that is 24 percent larger than is accurate. Further, because the injury reduction rates are the same as their corresponding death reduction rates for all three interventions, ignoring interdependencies would also lead to the same 24-percent difference in effectiveness for injuries.

 

 

Footnotes

  1. Ohio was chosen because it is a populous state and it has not implemented eight of the 12 interventions, which allows for a better demonstration of the tool.
  2. Identified in FARS as locations with traffic control devices; the majority of these are stop lights, but the category also includes railroad crossing gates.
  3. FARS categorizes fatalities in three ways: drivers, passengers, and people outside the vehicle (for example, a pedestrian struck by a car). So this category includes both drivers and passengers in vehicles. It does not include people outside the vehicle (pedestrians, bicyclists, and motorcycle riders).
  4. For ease of replication by others for checking and other purposes, numbers in this report are not rounded. 
  5. The revised report by Blincoe, Miller, et al., 2015, fixed errors in the SAS coding; these lowered its estimates of five of the cost categories in Table 5.4 for injuries but not fatalities.
  6. For more information on the cost categories and what they include, please see Blincoe, Miller, et al., 2015.
  7. Market productivity refers to lost earnings, whereas household productivity refers to work done in the home. Essentially, these terms measure the losses of not being able to work in the labor market or perform household duties.
  8. ACCRA previously stood for American Chamber of Commerce Research Association. The organization is now called the Council for Community and Economic Research, and it compiles the ACCRA Cost of Living Index.
  9. The state law data can quickly become out of date as new laws are passed and existing ones repealed. To address this problem, the tool allows users to select which interventions to consider. The information provided in Table 5.6 is the default, but a user can update if he or she knows the status of a particular law in a jurisdiction. 
  10. During the course of the project, we developed additional cost components and subcomponents that may be used if additional interventions are added to the tool.
  11. Should the cost be negative (i.e., a net positive cash flow), the remaining budget will actually increase after selecting the intervention for implementation. Further, should there be insufficient funds to select the next ranked intervention, the model would skip it and check the intervention ranked immediately below until the budget is fully committed or the list of candidate interventions for implementation has been exhausted.
  12. For display purposes, fatality and injury estimates are rounded to the nearest whole number.