sensitivity analysis for cost-benefit studies of justice policies: a primer
DESCRIPTION
TRANSCRIPT
Slide 1
Sensitivity Analysis for Cost-Benefit Studies of Justice Policies: A Primer
September 13, 2012
Ben Bryant, Staff Economist, Millennium Challenge Corpora7on Carl Ma5hies, Senior Policy Analyst, Cost-‐Benefit Analysis Unit, Vera Ins7tute of Jus7ce
Slide 2 • September 13, 2012
Sensitivity Analysis for Cost-Benefit Studies of Justice Policies: A Primer Ben Bryant
Economist Department of Policy and Evaluation Millennium Challenge Corporation
Carl Matthies Senior policy analyst
Cost-Benefit Analysis Unit Vera Institute of Justice
Slide 3 • September 13, 2012
Agenda
Introductions and housekeeping 5 minutes
Sensitivity analysis as part of cost-benefit analysis (CBA) Perspectives included in a CBA Deterministic sensitivity analysis Probabilistic sensitivity analysis Q&A and conclusion
10 minutes 5 minutes 15 minutes 15 minutes 10 minutes
Slide 4 • September 13, 2012
Housekeeping items
• Use the chat feature to send your questions at any time during the webinar.
• If you need webinar support or for troubleshooting: § Type questions into the chat box. § Call 800-843-9166. § Send e-mail to [email protected].
• This webinar is being recorded.
• The recording and PowerPoint slides will be posted on cbkb.org.
Slide 5 • September 13, 2012
Webinar preview
You will learn:
§ Why sensitivity analysis is a critical part of cost-benefit analysis
§ Different techniques for conducting sensitivity analysis
§ How to interpret the sensitivity analysis results
Slide 6 • September 18, 2012
Sensitivity Analysis as Part of Cost-Benefit Analysis (CBA)
Slide 7 • September 13, 2012
Review: Basic steps of CBA
§ Specify the set of alternative projects.
§ Decide whose benefits and costs count.
§ Catalog the impacts and select measurement indicators.
§ Predict the impacts quantitatively over the life of the project.
§ Monetize all impacts.
§ Discount benefits and costs to obtain present values.
§ Compute the net present value (NPV) of each alternative.
§ Perform sensitivity analysis.
§ Recommend/choose a project.
Slide 8 • September 13, 2012
Review: Basic steps of CBA Specify the set of alternative
projects.
Decide whose benefits and costs count.
Catalog the impacts and select measurement indicators.
Predict the impacts quantitatively over the life of the project.
Monetize all impacts.
Discount benefits and costs to obtain present values.
Compute the net present value (NPV) of each alternative.
Perform sensitivity analysis.
Recommend/choose a project.
Can be political
Requires significant topical knowledge
Requires significant quantitative skills and effort
Involves only calculations, in theory
Varies from simple to an entire study
Slide 9 • September 13, 2012
Strengths and weakness of CBA
Some strengths:
§ CBA informs and fosters transparency in policy process by making benefits and costs explicit.
§ It also uses a common unit of measurement for potentially more efficient allocation of resources.
Some weaknesses:
§ Even with common measurement ($), people may disagree on how to value outcomes.
§ Results may vary, subject to choices about who counts and what counts.
Slide 10 • September 13, 2012
State initiatives to legalize marijuana
Decisions on tax rate and regulatory
regime
Remove penalties
for sale and possession
Net impact on state and
local budgets
Marijuana consumption (quantity and
patterns of use)
Source: Kilmer et al. (2010)
Slide 11 • September 13, 2012
A web of uncertain variables
Decisions on tax rate and regulatory
regime
Penalties removed
for sale and possession
Net impact on state and
local budgets
Marijuana consumption (quantity and
patterns of use)
Tax revenues from legal marijuana
sales
Other factors that
influence budgets
(e.g., criminal justice and treatment
costs; federal spending; tourism)
Changes in production and
distribution costs
Tax evasion and tax-induced
shift in mix of marijuana types
Marijuana prices consumers face
Consumer price
sensitivity
Non-price effects on
consumption
Source: Kilmer et al. (2010)
Slide 12 • September 13, 2012
Characterizing uncertainty in a CBA
The way we characterize uncertainty is related to the methods we use:
• Best guess (percent change)
• Plausible bounds (deterministic, break-even)
• Probability distribution (Monte Carlo)
• Distinct stakeholder preferences (scenarios)
Slide 13 • September 13, 2012
Questions?
Use the chat feature to send us your questions.
Slide 14 • September 18, 2012
Perspectives Included in a CBA
Slide 15 • September 13, 2012
Perspectives
• A program or policy’s net benefits can be highly sensitive to the costs and benefits included (who and what “counts”).
• Think of changing the perspective as a form of sensitivity analysis.
Slide 16 • September 13, 2012
Which crime costs are included?
Source: McCollister, French, and Fang (2004)
Offense
Crime Victim Cost
Criminal Justice
System Cost
Crime
Career Cost
Total Tangible
Cost
Murder $744,239 $392,352 $148,555 $1,285,146
Rape/Sexual Assault $5,561 $26,479 $9,212 $41,252
Aggravated Assault $8,635 $8,641 $2,126 $19,472
Robbery $3,274 $13,827 $4,272 $21,373
Arson $11,453 $4,392 $584 $16,429
Motor Vehicle Theft $6,114 $3,867 $553 $10,534
Residential Burglary $1,654 $4,127 $681 $6,169
Larceny $479 $2,879 $163 $523
Slide 17 • September 13, 2012
Which crime costs are included?
Source: McCollister, French, and Fang (2004)
Offense Total Tangible Cost Total Intangible Cost
Murder $1,285,146 $8,442,000
Rape/Sexual Assault $41,252 $199,642
Aggravated Assault $19,472 $95,023
Robbery $21,373 $22,575
Arson $16,429 $5133
Motor Vehicle Theft $10,534 $262
Residential Burglary $6,169 $321
Larceny $3,523 $10
Slide 18 • September 13, 2012
Perspectives may affect project choice
Program
Estimated Impact
Annual Program Cost
Annual Net Benefit
CJS
Annual
Net Benefit
CJS + TV
Annual Net Benefit
CJS + TV + IV
A Prevents 100 motor vehicle thefts per year
$300K
+$87K
+$700K
+$730K
B Prevents 10 sexual assaults per year
$300K
-$35K
+$20K
+$2.0M
CJS: Criminal justice system TV: Tangible victim costs IV: Intangible victim costs
Slide 19 • September 13, 2012
Questions?
Use the chat feature to send us your questions.
Slide 20 • September 18, 2012
Deterministic Sensitivity Analysis
Slide 21 • September 13, 2012
Deterministic sensitivity analysis
This type of analysis examines relationships in specific, carefully chosen ways:
• Partial sensitivity analysis varies model inputs separately to see how sensitive the CBA result is to each input’s value, all else remaining equal.
• Extreme scenario analysis defines worst and best cases.
• Break-even analysis solves for model inputs when the NPV equals zero.
Slide 22 • September 13, 2012
The base case as starting point
• A preliminary result derived from plausible or best-estimate assumptions
• A reference point or benchmark for sensitivity analysis
• A starting point—not necessarily the most likely outcome!
Slide 23 • September 13, 2012
Example: Jail work-release program
• Inmates get three days off sentence for every two days worked.
• The daily marginal cost is slightly higher than for jail overall, but the program potentially reduces sentences and recidivism.
• Assumptions: • A six-year time horizon starts at Year Zero. • No recidivism costs will occur until Year One. • Only first-time offenders participate in the program. • Non-compliers must serve out their entire sentence. • The program has no effect on non-compliers.
Slide 24 • September 13, 2012
The work-release CBA model
Benefits Formula
Cost savings from reduced jail time
npq(SR)(JMC)(0.6)
Cost savings from program effect on recidivism
npq(SR)(BRR)(E)
Costs Formula
Additional staff
s
Program cost for compliers
npq(SR)(PMC)(0.4)
Program cost for non-compliers
np(1-q)(SR)(f) (PMC)+ np(1-q)(JMC)(SR)
Model Inputs Hiring two new corrections staff (s) Number of eligible participants (n) Participation rate (p) Program compliance rate (q) Program marginal cost/day (PMC) Jail marginal cost/day (JMC) Initial sentence range for eligible individuals (ISR) Average program failure point for non-compliers (f) Baseline 5-year recidivism rate (BRR) Change in recidivism rate among participants (E) Sentence range for recidivists (RSR) Discount rate (d)
Slide 25 • September 13, 2012
Example: Jail work-release program
Model Inputs Base-Case Estimate Hiring two new corrections staff (s) $80,000/year
Number of eligible participants (n) 1,000
Participation rate (p) 0.9
Program compliance rate (q) 0.8
Program marginal cost (PMC) $75/day
Jail marginal cost (JMC) $55/day
Initial sentence range for eligible individuals (ISR) 120 Average program failure point for non-compliers (f) 0.1
Baseline 5-year recidivism rate (BRR) 0.5
Change in recidivism rate among participants (E) 0.2
Sentence range for recidivists (RSR) 180
Discount rate (d) 0.03
NPV for work-release program $6,520,950
The base case looks favorable, but what if there are departures from the mean values?
Slide 26 • September 18, 2012
Partial sensitivity analysis
Slide 27 • September 13, 2012
Example: Jail work-release program
Model Inputs Base-Case Estimate Lower Bound Upper Bound Hiring two new corrections staff (s) $80,000/year $ 70,000/year $90,000/year
Number of eligible participants (n) 1,000 700 1300
Participation rate (p) 0.9 0.8 1.0
Program compliance rate (q) 0.8 0.7 1.0
Program marginal cost (PMC) $75/day $65/day $80/day
Jail marginal cost (JMC) $55/day $50/day $60/day
Initial sentence range for eligible individuals (ISR) 120 15 220 Average program failure point for non-compliers (f) 0.1 0.05 0.4
Baseline 5-year recidivism rate (BRR) 0.5 0.3 1.0
Change in recidivism rate among participants (E) 0.2 -0.2 0.6
Sentence range for recidivists (RSR) 180 120 365
Discount rate (d) 0.03 0.03 0.07
NPV for work-release program $6,520,950
Slide 28 • September 18, 2012
Partial sensitivity analysis
Number of eligible participants (n)
Program marginal cost (PMC)
Jail marginal cost (JMC)
Sentence range for recidivists (RSR)
Baseline 5-year recidivism rate (BRR)
Program compliance rate (q)
Participation rate (p)
Change in recidivism rate among participants (E)
Average program failure point in non-compliers (p)
Initial sentence range for eligible individuals (ISR)
Hiring two new corrections staff (s) 90,000 70,000
1,300
Slide 29 • September 13, 2012
Worst-case and best-case scenarios
Model Inputs Base-Case Estimate Worst Case Best Case
Hiring two new corrections staff (s) $80,000/year $90,000/year $70,000/year
Number of eligible participants (n) 1,000 1300 1,300
Participation rate (p) 0.9 1.0 1.0
Program compliance rate (q) 0.8 0.7 1.0
Program marginal cost (PMC) $75/day $80/day $65/day
Jail marginal cost/day (JMC) $55/day $60/day $60/day Initial sentence range for eligible individuals (ISR) 120 15 220 Average program failure point for non-compliers (f) 0.1 0.4 0.05
Baseline 5-year recidivism rate (BRR) 0.5 1.0 1.0 Change in recidivism rate among participants (E) 0.2 -0.2 0.6
Sentence range for recidivists (RSR) 180 120 365
Discount rate (d) 0.03 0.07 0.03
NPV for work-release program $6.5M -$43.3M $104.0M
Slide 30 • September 13, 2012
Break-even analysis
The program would still break even if its marginal cost were $91/day, all else remaining equal.
Model Input Estimate
Hiring two new corrections staff (s) $80,000
Number of eligible participants (n) 1,000
Participation rate (p) 0.9
Program compliance rate (q) 0.8
Program marginal cost (PMC) $91/day
Jail marginal cost/day (JMC) $55/day
Initial sentence range for eligible individuals (SR) 120
Average program failure point for non-compliers (f) 0.1
Baseline 5-year recidivism rate (BRR) 0.5
Change in recidivism rate among participants (E) 0.2
Sentence range for recidivists (SR) 180
Discount rate (d) 0.03
NPV for work-release program $0
Slide 31 • September 13, 2012
Break-even analysis
The program would break even if it had no effect on recidivism, all else remaining equal.
Model Input Estimate
Hiring two new corrections staff (s) $80,000
Number of eligible participants (n) 1,000
Participation rate (p) 0.9
Program compliance rate (q) 0.8
Program marginal cost (PMC) $75/day
Jail marginal cost/day (JMC) $55/day
Initial sentence range for eligible individuals (SR) 120
Average program failure point for non-compliers (f) 0.1
Baseline 5-year recidivism rate (BRR) 0.5
Change in recidivism rate among participants (E) 0.0
Sentence range for recidivists (SR) 180
Discount rate (d) 0.03
NPV for work-release program $0
Slide 32 • September 13, 2012
Questions?
Use the chat feature to send us your questions.
Slide 33 • September 18, 2012
Probabilistic Sensitivity Analysis: Monte Carlo Simulations
Slide 34 • September 13, 2012
Probabilistic sensitivity analysis
This type of analysis changes all model inputs at once.
• The other methods tell us little about the likelihood of various program outcomes.
• Monte Carlo simulation repeatedly draws random values for each model input to create a probability distribution of outcomes.
Slide 35 • September 13, 2012
MS Excel add-ons for Monte Carlo simulation
Slide 36 • September 13, 2012
Step 1: Define assumptions.
Slide 37 • September 13, 2012
Step 1: Define assumptions.
Slide 38 • September 13, 2012
Define assumptions for all uncertain inputs.
Slide 39 • September 13, 2012
Step 2: Set the forecast cell.
Slide 49 • September 18, 2012
MC simulations provide the distribution of outcomes.
Slide 50 • September 18, 2012
Distribution of outcomes
Slide 42 • September 13, 2012
Questions?
Use the chat feature to send us your questions.
Slide 36 • September 18, 2012
Wrap-Up
Slide 44 • September 13, 2012
Review
We covered:
§ Why sensitivity analysis is a critical part of CBA
§ Different techniques for conducting sensitivity analysis
§ How to interpret the sensitivity analysis results
Slide 45 • September 13, 2012
Takeaway points
§ Base-case CBA is just a starting point. § Sensitivity analysis is a way of managing
uncertainty, not eliminating it. § Computation and simulation are
relatively easy. § Defining input boundaries and probability
distributions is harder.
Slide 46 • September 13, 2012
Reminders
§ Please complete the evaluation form before you leave this webinar.
§ To receive information and notifications about upcoming webinars and other events:
• Visit the Cost-Benefit Knowledge Bank for Criminal Justice at cbkb.org.
• Follow us on Twitter at twitter.com/CBKBank.
Slide 47 • September 13, 2012
Contact information
Ben Bryant [email protected]
Carl Matthies [email protected]
213-223-2445
E-mail: [email protected]
Website: cbkb.org
Slide 41
The Cost-Benefit Knowledge Bank for Criminal Justice (CBKB) is a project of the Vera Institute of Justice funded by the U.S. Department of Justice’s Bureau of Justice Assistance.
• Website (cbkb.org) • CBA toolkit • Snapshots of CBA literature • Podcasts, videocasts, and webinars • Roundtable discussions • Community of practice
Slide 42
This project is supported by Grant No. 2009-MU-BX K029 awarded by the Bureau of Justice Assistance. The Bureau of Justice Assistance is a component of the Office of Justice Programs, which also includes the Bureau of Justice Statistics, the National Institute of Justice, the Office of Juvenile Justice and Delinquency Prevention, and the Office of Sex Offender Sentencing, Monitoring, Apprehending, Registering, and Tracking. Points of view or opinions in this presentation are those of the authors and do not represent the official position or policies of the Millennium Challenge Corporation or the United States Department of Justice.
Slide 43 • September 18, 2012
Thank you.