Prioritization using Logic Models and MIRA October 17. 2007 Instituto Nacional de Ecologia Mexico City, Mexico Part I: Logic Models Connecting Program Activities to Environmental Outcomes What is a Logic Model? Tool to help understand how program activities affect environmental outcomes. Kellogg Foundation template to start. Foundation Home Page: http://www.wkkf.org/Default.aspx?LanguageID= 0 Logic Model Guidance Document: http://www.wkkf.org/Pubs/Tools/Evaluation/Pub 3669.pdf Modification for environmental programs. Logic Model Template Baseline What is the condition of the things we care about? Resources In order to accomplish our set of activities we will need the following: Stressors In order of importance what are the stressors and where are they most prevalent? These will be used to target activities. Activities In order to address our problems or asset we will accomplish the following activities. Outputs We expect that once accomplish ed these activities will produce the following evidence or service delivery. (bold = perf stds) Short and Long term outcomes We expect that if accomplish ed the se activities will lead to the following changes in 1-3 then 4-6 years. Impact We expect that if accomplish ed these activities will lead to the following changes in 7-10 years. How to build a logic model Brainstorm activities (“individual logic model”) Program or site activities For each activity, ask: Why do I do this activity? What is(are) the intended outcome(s) of doing that activity? What is(are) the actual outcome(s)? What is the impact (ultimate goal) of this outcome? How to build a logic model – cont’d. Baseline = outcome/impacts Need to measure the same thing at the baseline as at the end. Stressor – Distinguish between: Pollutant stressors E.g., population growth, vehicle emissions Program stressors E.g., conflicting statute, no regulatory authority Different stressors compel different activities/outcomes. Logic Models good for: Linking activities to outcomes/impacts Helps to identify dependent activities. If linking site activities, are different outcomes expected/desired from different sites? Describing indicators needed to measure programs. Defining indicators is necessary for program evaluation; Not always easy. Learning about your programs. Examine why you do your activities. What Logic Models are missing: No indicator data contained in LMs. No way to prioritize program activities. Use MIRA to get these… Part II: MIRA Analyzing Information for Decision Making: Prioritizing Environmental Outcomes and Managing Risk What’s involved in Decision analysis? Criteria/Data Science – exposure, fate/transport models, other Program implementation (logic models). Social science – environmental justice, different demographic impacts. Values Integrative, contextual approach for decision analysis. MIRA Multi-criteria Integrated Resource Assessment MIRA Approach: Multi-criteria Transparent Data driven; relative analysis Iterative/learning-based MIRA Data Collection Manager Geostatistical Indicators Module PRIMARY LEVEL SECONDARY LEVEL THIRD LEVEL Data Fit Region III Data Scatter FOURTH LEVEL Area Wide Population weighted Design Value weighted Attn. Threshold weighted Area Wide Population weighted Design Value weighted Attn. Threshold weighted Worst Outlier Data Fit 1-Hr O3 Non-Attainment Areas Ozone Air Quality Data Scatter Area Wide Population weighted Design Value weighted Attn. Threshold weighted Area Wide Population weighted Design Value weighted Attn. Threshold weighted Worst Outlier Data Fit 1-Hr O3 Attainment Areas Data Scatter Area Wide Population weighted Design Value weighted Attn. Threshold weighted Area Wide Population weighted Design Value weighted Attn. Threshold weighted Worst Outlier Data Fit Class I Areas Data Scatter Worst Outlier Personnel Impact Costs Trends Impact Monitor Servicing Distance Work Load Area Wide Population weighted Design Value weighted Attn. Threshold weighted Area Wide Population weighted Design Value weighted Attn. Threshold weighted Programmatic and Budget Decision Analysis Module Fate and Transport Models What is MIRA designed to do? Policy Development Assist in multi-criteria analyses for the development/implementation of policy. Understanding alternatives Improve understanding of the relationship between the data and the decision alternatives. Address stakeholder concerns Provide an analytical framework for reflecting stakeholder ideas (Inclusive) Steps in the MIRA approach Determine the decision question. Brainstorm initial criteria. Gather data for those criteria. Construct the analytical hierarchy for the decision question. Index data (expert input). Preference criteria (stakeholder value sets). Iterate; Learn. Hazard Ranking System (HRS) Example Possible to use HRS score in different ways with MIRA: Option 1: Use HRS as a decision criterion. Option 2: Use HRS criteria and allow for flexibility for expert input and decision maker judgment. Appropriate when you don’t have or can’t get type of data required by HRS; i.e., need to use surrogate indicators. Option 1: HRS as Criterion Suppose you want to evaluate both the condition of the region and program effectiveness within the region to include: Public health impacts Ecological impacts Balance condition with program (in)effectiveness. Possible to set up a decision hierarchy something like this… Option 1: Sample MIRA Decision Hierarchy Risk Public Health Source Admin. Condition Habitat Condition Ecosystem Health HRS ? Stressors Admin. Risk Public Health Program Source Admin. Habitat Condition Ecosystem Health Stressors Stressors Option 1: Indicator Examples Condition HRS score Economic/social costs Ozone concentration, Nutrient load Cancer risk, Exposure Program # permits/regulations approved; % impaired streams % regulations that include evaluation of alternative control technologies. Amount of time between submittal and approval of…regulation/permit/plan. “x” type of Hazardous Waste implementation plan producing change/improvement in “y” type of risk parameter by “z” amount. Option 1: How to use HRS with other criteria Need to consider the relative environmental significance of HRS with other criteria. Expert discussion What does HRS indicate? Is it a more decision significant indicator than economic cost (for example)? If you believe no other criterion than HRS needs to be considered, you don’t need MIRA. Option 2: Using HRS criteria as the analysis Suppose you only want to consider hazardous waste criteria as currently used in calculating HRS… OR: You are unable to get data required/ expected by HRS and must use surrogate indicators… Possible to set up decision hierarchy as follows… Likelihood of Release SGW Waste Characteristics Observed Release Potential of Release Cancer Toxicity/ Mobility NonCancer Chronic NonCancer Acute Haz. Waste Quantity Source Constituent Nearest Indiv. Waste Stream HRS Targets Population Resources SSW ... SS ... SA ... Sensitive Ecosystems Option 2: MIRA Hierarchy for Hazard Ranking System (HRS) HRS Calculation Example 1 HRS: Likelihood of Release = greater of observed release or potential to release To replicate in MIRA: one of these criterion will have a weight of zero in the calculation (Other = 1.0). MIRA alternative (if not regulatory): weight these criteria in any way that adds up to 1.0 (or 100%). HRS Calculation Example 2 HRS Calculation Methodology Pathway Score, S = (Likelihood of Release x Waste Characteristics x Targets)/82,500 Max values for LR = 550, Waste = 100, Targets = 150. Cont’d Example 2 To replicate in MIRA: Calculate relative weights for each of 3 factors. E.g. LR weight = (550/82,500)/(550/82,500 + 100/82,500 + 150/82,500) = 0.691 (LR) x 0.691 x (waste) x 0.124 x (targets) x 0.185 (Fixed weights via HRS method) Likelihood of Release is designed to be the most important criterion in the HRS calculation method (69% vs. 12% vs. 18%). With MIRA, you can change weights if desired (and allowed by law). HRS Calculation Example 3 HRS = 2 2 S GW S SW S S S A 2 2 4 Max pathway score (S) = 100. HRS equation appears to weight all pathways equally BUT actually weights the pathway score that is highest most heavily (due to squaring). In MIRA: possible to replicate weights via above equation or use other weights. Option 2: HRS Component analysis with MIRA What’s different about using HRS criteria in MIRA vs. just calculating HRS? Allows for transparency in seeing relative importance (weights) of all the criteria composing the HRS. Possible to use additional criteria (economic/ social) if desired. Possible to use surrogate criteria if data required by HRS is not available. If law requires HRS method, using MIRA is not an option. BUT could use MIRA to inform other decisions. MIRA Approach Step 1: Determine the decision question. Step 2: Brainstorm initial criteria. Step 3: Construct the analytical hierarchy for the decision question. Step 4: Address missing data. Step 5: Decide on decision’s unit of measure. Step 6: Index data (expert input). Step 7: Preference criteria (stakeholder value sets). Step 8: Iterate; Learn. Step 1: Formulating the Decision Question Decision makers/stakeholders formulate the question that they want to answer and the criteria they think they need to answer it. What are the problem set elements that you are analyzing/ranking? e.g. watersheds?, counties?, emission control strategies? Step 2: Brainstorm Initial Criteria Are data available for these criteria? Are data available on the scale that you want? States?, Counties?, watersheds?, stream segments? Other? If not: Is another scale possible? Can surrogate data be used? Should this be identified for future data collection? Step 3a: Construct the Decision Hierarchy Provides decision context. Forces stakeholders to assess whether they agree on the decision question that they want to answer. Step 3b: Methodological thinking for constructing the hierarchy Should each criterion currently organized at each level of the hierarchy be directly comparable? E.g., Would you compare Arsenic in ground water with Ozone air quality? OR would a better comparison be Water (with groundwater under it) with Air (with Ozone under it)? Step 4: Determine which criteria have no/missing data Possible alternatives to no data Health impact data – pollutant concentration – source emissions – number of sources? Data collected by volunteers/other organizations. Using similar data (from another program, etc.). Possible alternatives to missing data Statistical analyses – e.g., multivariate analyses Data collected by volunteers/other organizations. Modeling. Note about previously constructed indicators What do these indicators indicate? Is this meaningful in your current analysis? Can better indicators for your analysis be constructed with currently available data? Step 5: Deciding on the Decision’s unit of measure Depends on the decision question What is the condition of the watersheds in the region? Degree of degradation Which watersheds should be restored? Degree of restorability Continued… OR combine questions: Based on the condition of the watersheds and the restorability of the watersheds, which should we restore? Motivation to restore Step 6a: Indexing the data Convert all criteria metrics to the decision unit. Indexing = Relative comparison among the range of metric values on a decision scale; = unit converter (converts units of each criterion metric to the decision unit). Expert Input here What is the decision significance of the indicator values? Same indicator can have different signficance for another decision question. Step 6b: Approach to Indexing the Data Use a decision scale of 1 to 8. Assumption: Each criterion is of equal value or importance. BUT Metrics are not looked at independently. Task in indexing is to define what value of each criterion elicits the same response. Set these values to the same index. E.g., $1 million is a lot of money and 95 ppb of ozone is a high ozone level (on par with $1 mil) (they both elicit a “that’s a lot” response), so set them both to the same index. Step 6c: Thinking about Indexing Range of metric/indicator values? Type of distribution? Double check: Compare values for criteria pairs – same significance? Initialize; Change later if needed. Step 7: Preferencing All criteria are not equally important to the decision makers/stakeholders. Preferencing = Relative comparison of the importance of one criterion to other criteria. Step 7b: Thinking about Preferencing Initialize by setting all criteria preferences to equal weights (i.e., all criteria equally important to the decision question within each level of the hierarchy). = Equal preference value set. Iterate Test different value sets Examine indexing Examine data – including quality assurance of data. Step 8: Iteration Test different value sets Examine indexing Examine data Examine data uncertainty Re-run analysis with different “what if” scenarios. Sulfur Deposition Raw Data Phosphorus Loading Raw Data S_Dep Indexed P_Load Indexed P_Load and S_Dep Combined 80% P_Load, 20% S_Dep (equally important) S_Dep hot spot (NW PA) determined to be more scientifically significant than P_Load hot spot (Delmarva Peninsula). Science significance stays the same. Decision maker judgments alter priorities but decision process is transparent. Role of experts in MIRA Experts in all fields of study to discuss issues: Indicator Types; construction of appropriate indicators? Data* for indicators (existing, new) Missing data issues Scale of indicators/data Combining public health and ecological information Indexing data (determine relative significance of data) Role of decision makers in MIRA Learn the impact of different value sets (i.e., relative preference weights among decision criteria) on the decision options. Science remains constant. Examine/compare the results of different value sets. Make a decision after being informed about the impacts of all the options examined. Build decision confidence. Provide documentation and rationale for decision. MIRA different from other decision support approaches… Hierarchy: represents decision question Indexing: Expert input = relative decision significance of the indicators Preferencing: Decision maker/stakeholder judgments = relative importance of the decision criteria for this decision. Relative contextual analysis. Illustrates what/where the tradeoffs are – as constrained by the data. – Learning. MIRA References http://www.epa.gov/reg3artd/airquality/mira_descr.htm Cimorelli, A. and Stahl, C. (2005), BSTS 25(3): 1, “Tackling the Dilemma of the Science-Policy Interface in Environmental Policy Analysis.” Stahl, C.H. (2003), “Multi-criteria Integrated Resource Assessment (MIRA): A New Decision Analytic Approach to Inform Environmental Policy Analysis.” For the Degree of Doctor of Philosophy, University of Delaware. Stahl, C. H. and Cimorelli, A. J. (2005), Risk Analysis 25(5): 1109, “How Much Uncertainty is Too Much and How Do We Know? A Case Example of Ozone Monitor Network Options.” Part III: Logic Model Outputs as MIRA Inputs Program prioritization What do we get with LMs and MIRA? Integration of Data and Program Activities. Are we doing the right activities? – based on where the “worst” conditions are. Which activities have the greatest effect on the outcomes/impacts we seek? – based on which outcomes/impacts we value most highly AND the condition data. Which activities are dependent on which other activities? Capability to prioritize program outcomes using data. Transparency Learning Example Logic Models Air Quality Monitoring Logic Model Ozone Program Logic Model Trace monitoring activity (certification of ozone air quality data) through Monitoring Logic Model outputs/outcome/impacts See Red text in following figure. Follow black boxes within Monitoring logic model in following figure. Baseline Stressors Activities Outputs O3 SIP Program: O3 SIP Program Stressors O3 Nonattainment Area Designations O3 Nonattainment Area Designations O3 SIP Program Baseline 1= 03 DV weighted by sensitive population (children, elderly, etc.) O3 Monitoring Program: Outcomes Impacts O3 SIP Program O3 SIP Program (Human Health) Impact 1 = Outcome 2= O3 SIP Program Stressor 4 = # 03 DV weighted Human health by sensitive impacts from population O3 pollution (children, elderly, etc.) Based on O3 upwind areas monitor design designated values, concur attainment for with HQ on O3 O3. design monitor for each area. List of DVs for each NA area O3 Monitoring Program Stressors Data Review for Official O3 official O3 monitor monitoring data: DV Output: DV O3 Monitoring program Stressor 1 O3 Monitoring Baseline 1 = need a = regulatory requirement to metric for the certify O3 data and accuracy of the monitoring network calculate O3 DV. Data Review for official O3 DV for O3 monitor: Review states’ certification of O3 data (AQS data prior to official use). MIRA Indicator Complete and certified O3 AQ data (no missing years, etc). Data usable for AQ planning. A) selection of O3 design monitor for R3 areas. Monitored O3 levels accurately represent true O3 levels for AQ planning areas. – Correlation coefficient between monitored and other estimation methods of O3? Cont’d Ozone Logic Model Show dependency of Ozone Program activity on Monitoring certification of data. See red underlined text in previous figure. Trace Ozone program activity through to its outputs/outcomes/impacts. Follow red boxes in previous figure. How does this connect to MIRA? MIRA Indicator Health indicator preferred but currently no data/science. Use Ozone concentration weighted by population as surrogate for now. Summary Logic models improve program understanding. Logic models provide connection between program activities and outcomes/impacts. If prioritization is desired, use as MIRA input. MIRA approach is compatible with use of many environmental, economic and social criteria. Use of HRS criteria possible in 2 different ways. Supports the use of surrogate data (using data that is readily available). MIRA allows transparency, learning, stakeholder inclusiveness.