Promoting a RealWorld and Holistic approach to Impact Evaluation Designing Evaluations under Budget, Time, Data and Political Constraints AEA pre-conference professional development workshop Minneapolis, October 24, 2012 Facilitated by Michael Bamberger and Jim Rugh Note: This PowerPoint presentation and the summary (condensed) chapter of the book are available at: www.RealWorldEvaluation.org 1 Workshop Objectives 1. The basics of the RealWorld Evaluation approach for addressing common issues and constraints faced by evaluators such as: when the evaluator is not called in until the project is nearly completed and there was no baseline nor comparison group; or where the evaluation must be conducted with inadequate budget and insufficient time; and where there are political pressures and expectations for how the evaluation should be conducted and what the conclusions should say; 2 Workshop Objectives 2. 3. 4. 5. 6. 7. Identifying and assessing various design options that could be used in a particular evaluation setting; Some tips on developing logic models; Ways to reconstruct baseline data; How to account for what would have happened without the project’s interventions: alternative counterfactuals; The advantages of using mixed-methods designs; Considerations for more holistic approaches to impact evaluation. 3 Workshop Objectives Note: In this workshop we focus on projectlevel impact evaluations. There are, of course, many other purposes, scopes, evaluands and types of evaluations. Some of these methods may apply to them, but our examples will be based on project impact evaluations, most of them in the context of developing countries. 4 Workshop agenda: morning 8:00-8:30 8:30-9:00 9:00-9:30 9:30-9:50 9:50-10:20 10:20-10:40 10:40-11:00 11:00-12:00 12:00-1:00 Introduction + Brief overview of the RealWorld Evaluation approach Basic evaluation design scenarios Jim 29-48 Small group discussion: personal introductions, including RWE-type constraints you have faced in your own practice break Logic models; reconstructing baselines Jim 51- 77 Mixed Method evaluations Michael 78-102 Alternative counterfactuals Jim 103-115 Small group exercise, Part I: Reading of case studies then preparing an evaluation design when working under budget, time, data or political constraints lunch Workshop agenda: afternoon 1:00-1:45 Impact Evaluations Michael 120-134, Jim 135-155 1:45-2:30 Small group exercise, Part II: 'Clients’ and ‘consultants’ re-negotiate the case study evaluation ToR. Note: In addition to becoming familiar with some of the RWE approaches, the case study exercises will also illustrate the different evaluation agendas and perspectives of evaluation consultants, project implementers and funding agencies. 2:30-2:50 2:50-3:00 Plenary discussion of main learnings from group exercises and the RealWorld Evaluation approach. Jim 156-159 Wrap-up and evaluation of workshop by participants RealWorld Evaluation Designing Evaluations under Budget, Time, Data and Political Constraints OVERVIEW OF THE RWE APPROACH 7 Typical RealWorld Evaluation Scenario Evaluator(s) not brought in until near end of project For political, technical or budget reasons: • There was no life-of-project evaluation plan • There was no baseline survey • Project implementers did not collect adequate data on project participants at the beginning nor during the life of the project • It is difficult to collect data on comparable control groups 8 Reality Check – Real-World Challenges to Evaluation • • • • • • All too often, project designers do not think evaluatively – evaluation not designed until the end There was no baseline – at least not one with data comparable to evaluation There was/can be no control/comparison group. Limited time and resources for evaluation Clients have prior expectations for what they want evaluation findings to say Many stakeholders do not understand evaluation; distrust the process; or even see it as a threat (dislike of being judged) 9 RealWorld Evaluation Quality Control Goals Achieve maximum possible evaluation rigor within the limitations of a given context Identify and control for methodological weaknesses in the evaluation design Negotiate with clients trade-offs between desired rigor and available resources Presentation of findings must acknowledge methodological weaknesses and how they affect generalization to broader populations 10 The Need for the RealWorld Evaluation Approach As a result of these kinds of constraints, many of the basic principles of rigorous impact evaluation design (comparable pre-test -- post test design, control group, adequate instrument development and testing, random sample selection, control for researcher bias, thorough documentation of the evaluation methodology etc.) are often sacrificed. 11 The RealWorld Evaluation Approach An integrated approach to ensure acceptable standards of methodological rigor while operating under real-world budget, time, data and political constraints. This is the cover of the cover of the 1se edition of the RealWorld Evaluation book (2006) 12 EDI T I O N This book addresses the challenges of conducting program evaluations in real-world contexts where evaluators and their clients face budget and time constraints and where critical data may be missing. The book is organized around a seven-step model developed by the authors, which has been tested and refined in workshops and in practice. Vignettes and case studies—representing evaluations from a variety of geographic regions and sectors— demonstrate adaptive possibilities for small projects with budgets of a few thousand dollars to large-scale, longterm evaluations of complex programs. The text incorporates quantitative, qualitative, and mixed-method designs and this Second Edition reflects important developments in the field over the last five years. New to the Second Edition: Adds two new chapters on organizing and managing evaluations, including how to strengthen capacity and promote the institutionalization of evaluation systems Includes a new chapter on the evaluation of complex development interventions, with a number of promising new approaches presented Incorporates new material, including on ethical standards, debates over the “best” evaluation designs and how to assess their validity, and the importance of understanding settings Expands the discussion of program theory, incorporating theory of change, contextual and process analysis, multi-level logic models, using competing theories, and trajectory analysis Provides case studies of each of the 19 evaluation designs, showing how they have been applied in the field “This book represents a significant achievement. The authors have succeeded in creating a book that can be used in a wide variety of locations and by a large community of evaluation practitioners.” —Michael D. Niles, Missouri Western State University RealWorld Evaluation RealWorld Evaluation Bamberger Rugh Mabry 2 “This book is exceptional and unique in the way that it combines foundational knowledge from social sciences with theory and methods that are specific to evaluation.” —Gary Miron, Western Michigan University “The book represents a very good and timely contribution worth having on an evaluator’s shelf, especially if you work in the international development arena.” —Thomaz Chianca, independent evaluation consultant, Rio de Janeiro, Brazil 2 EDITI ON Working Under Budget, Time, Data, and Political Constraints Michael Bamberger Jim Rugh Linda Mabry EDITION The RealWorld Evaluation approach Developed to help evaluation practitioners and clients • managers, funding agencies and external consultants Still a work in progress (we continue to learn more through workshops like this) Originally designed for developing countries, but equally applicable in industrialized nations 14 Most RealWorld Evaluation tools are not new— but promote a holistic, integrated approach Most of the RealWorld Evaluation data collection and analysis tools will be familiar to experienced evaluators. What we emphasize is an integrated approach which combines a wide range of tools adapted to produce the best quality evaluation under RealWorld constraints. 15 What is Special About the RealWorld Evaluation Approach? There is a series of steps, each with checklists for identifying constraints and determining how to address them These steps are summarized on the following slide and in the more detailed flow-chart on page 5 of the Condensed Overview. 16 The Steps of the RealWorld Evaluation Approach Step 1: Planning and scoping the evaluation Step 2: Addressing budget constraints Step 3: Addressing time constraints Step 4: Addressing data constraints Step 5: Addressing political constraints Step 6: Assessing and addressing the strengths and weaknesses of the evaluation design Step 7: Helping clients use the evaluation See page 5 of the Condensed Summary for more details 17 Planning and Scoping the Evaluation Understanding client information needs Defining the program theory model Preliminary identification of constraints to be addressed by the RealWorld Evaluation 18 Understanding client information needs Typical questions clients want answered: Is the project achieving its objectives? Is it having desired impact? Are all sectors of the target population benefiting? Will the results be sustainable? Which contextual factors determine the degree of success or failure? 19 Understanding client information needs A full understanding of client information needs can often reduce the types of information collected and the level of detail and rigor necessary. However, this understanding could also increase the amount of information required! 20 Still part of scoping: Other questions to answer as you customize an evaluation Terms of Reference (ToR): 1. 2. 3. 4. Who asked for the evaluation? (Who are the key stakeholders)? What are the key questions to be answered? Will this be mostly a developmental, formative or summative evaluation? Will there be a next phase, or other projects designed based on the findings of this evaluation? 21 Other questions to answer as you customize an evaluation ToR: 5. 6. 7. 8. 9. What decisions will be made in response to the findings of this evaluation? What is the appropriate level of rigor? What is the scope / scale of the evaluation / evaluand (thing to be evaluated)? How much time will be needed / available? What financial resources are needed / available? 22 Other questions to answer as you customize an evaluation ToR: 10. 11. 12. 13. 14. 15. Should the evaluation rely mainly on quantitative or qualitative methods? Should participatory methods be used? Can / should there be a household survey? Who should be interviewed? Who should be involved in planning / implementing the evaluation? What are the most appropriate forms of communicating the findings to different stakeholder audiences? 23 Before we return to the RealWorld steps, let’s gain a perspective on levels of rigor, and what a life-of-project evaluation plan could look like 24 Different levels of rigor depends on source of evidence; level of confidence; use of information Objective, high precision – but requiring more time & expense Level 5: A very thorough research project is undertaken to conduct indepth analysis of situation; P= +/- 1% Book published! Level 4: Good sampling and data collection methods used to gather data that is representative of target population; P= +/- 5% Decision maker reads full report Level 3: A rapid survey is conducted on a convenient sample of participants; P= +/- 10% Decision maker reads 10-page summary of report Level 2: A fairly good mix of people are asked their perspectives about project; P= +/- 25% Decision maker reads at least executive summary of report Level 1: A few people are asked their perspectives about project; P= +/- 40% Decision made in a few minutes Level 0: Decision-maker’s impressions based on anecdotes and sound bytes heard during brief encounters (hallway gossip), mostly intuition; Level of confidence +/- 50%; Decision made in a few seconds Quick & cheap – but subjective, sloppy 25 CONDUCTING AN EVALUATION IS LIKE LAYING A PIPELINE QUALITY OF INFORMATION GENERATED BY AN EVALUATION DEPENDS UPON LEVEL OF RIGOR OF ALL COMPONENTS AMOUNT OF “FLOW” (QUALITY) OF INFORMATION IS LIMITED TO THE SMALLEST COMPONENT OF THE SURVEY “PIPELINE” Determining appropriate levels of precision for events in a life-of-project evaluation plan High rigor Same level of rigor 4 Final evaluation Baseline study Mid-term evaluation 3 Needs assessment Special Study Annual self-evaluation 2 Low rigor Time during project life cycle 28 RealWorld Evaluation Designing Evaluations under Budget, Time, Data and Political Constraints EVALUATION DESIGNS 29 Some of the purposes for program evaluation Formative: learning and improvement including early identification of possible problems Knowledge generation: identify cause-effect correlations and generic principles about effectiveness. Accountability: to demonstrate that resources are used efficiently to attain desired results Summative judgment: to determine value and future of program Developmental evaluation: adaptation in complex, emergent and dynamic conditions -- Michael Quinn Patton, Utilization-Focused Evaluation, 4th edition, pages 139-140 30 Determining appropriate (and feasible) evaluation design Based on the main purpose for conducting an evaluation, an understanding of client information needs, required level of rigor, and what is possible given the constraints, the evaluator and client need to determine what evaluation design is required and possible under the circumstances. 31 Some of the considerations pertaining to evaluation design 1: When evaluation events take place (baseline, midterm, endline) 2. Review different evaluation designs (experimental, quasi-experimental, other) 3. Determine what needs to be done to “fill in the missing pieces” when the “ideal” evaluation design/scenario is not feasible. 32 An introduction to various evaluation designs Illustrating the need for quasi-experimental longitudinal time series evaluation design Project participants Comparison group baseline scale of major impact indicator end of project evaluation post project evaluation 33 OK, let’s stop the action to identify each of the major types of evaluation (research) design … … one at a time, beginning with the most rigorous design. 34 First of all: the key to the traditional symbols: X = Intervention (treatment), I.e. what the project does in a community O = Observation event (e.g. baseline, mid-term evaluation, end-of-project evaluation) P (top row): Project participants C (bottom row): Comparison (control) group Note: the 7 RWE evaluation designs are laid out on page 8 of the Condensed Overview of the RealWorld Evaluation book 35 Design #1: Longitudinal Quasi-experimental P1 X C1 P2 X C2 P3 P4 C3 C4 Project participants Comparison group baseline midterm end of project evaluation post project evaluation 36 Design #2: Quasi-experimental (pre+post, with comparison) P1 X P2 C1 C2 Project participants Comparison group baseline end of project evaluation 37 Design #2+: Randomized Control Trial P1 X P2 C1 C2 Project participants Research subjects randomly assigned either to project or control group. Control group baseline end of project evaluation 38 Design #3: Truncated Longitudinal X P1 X C1 P2 C2 Project participants Comparison group midterm end of project evaluation 39 Design #4: Pre+post of project; post-only comparison P1 X P2 C Project participants Comparison group baseline end of project evaluation 40 Design #5: Post-test only of project and comparison X P C Project participants Comparison group end of project evaluation 41 Design #6: Pre+post of project; no comparison P1 X P2 Project participants baseline end of project evaluation 42 Design #7: Post-test only of project participants X P Project participants end of project evaluation 43 See Table 2.2 on page 8 of Condensed Overview of RWE D e s i g n T4 cont.) (endline) (ex-post) X P3 C3 P4 C4 X P2 C2 X P2 C2 X X P2 C2 X X P1 C1 X X P2 X X P1 X (baseline) (intervention) 1 P1 C1 X 2 P1 C1 X 3 4 X P1 5 6 7 P1 T2 X T3 T1 (midterm) P2 C2 P1 C1 (intervention, 44 “Non-Experimental” Designs [NEDs] NEDs are impact evaluation designs that do not include a matched comparison group Outcomes and impacts assessed without a conventional statistical counterfactual to address the question • “what would have been the situation of the target population if the project had not taken place?” 45 Situations in which an NED may be the best design option Not possible to define a comparison group When the project involves complex processes of behavioral change Complex, evolving contexts Outcomes not known in advance Many outcomes are qualitative Projects operate in different local settings When it is important to study implementation Project evolves slowly over a long period of time 46 Some potentially strong NEDs A. B. C. D. E. F. G. Interrupted time series Single case evaluation designs Longitudinal designs Mixed method case study designs Analysis of causality through program theory models Concept mapping Contribution Analysis 47 Any questions? 48 TIME FOR SMALL GROUP DISCUSSION 49 1. Self-introductions 2. What constraints of these types have you faced in your evaluation practice? 3. How did you cope with them? 50 RealWorld Evaluation Designing Evaluations under Budget, Time, Data and Political Constraints LOGIC MODELS 51 Defining the program theory model All programs are based on a set of assumptions (hypothesis) about how the project’s interventions should lead to desired outcomes. Sometimes this is clearly spelled out in project documents. Sometimes it is only implicit and the evaluator needs to help stakeholders articulate the hypothesis through a logic model. 52 Defining the program theory model Defining and testing critical assumptions are essential (but often ignored) elements of program theory models. The following is an example of a model to assess the impacts of microcredit on women’s social and economic empowerment 53 Critical logic chain hypothesis for a Gender-Inclusive Micro-Credit Program Sustainability • Structural changes will lead to long-term impacts. Medium/long-term impacts • Increased women’s economic and social empowerment. • Economic and social welfare of women and their families will improve. Short-term outcomes • If women obtain loans they will start income-generating activities. • Women will be able to control the use of loans and reimburse them. Outputs • If credit is available women will be willing and able to obtain loans and technical assistance. 54 What does it take to measure indicators at each level? Impact :Population-based survey (baseline, endline evaluation) Outcome: Change in behavior of participants (can be surveyed annually) Output: Measured and reported by project staff (annually) Activities: On-going (monitoring of interventions) Inputs: On-going (financial accounts) Consequences Consequences Consequences PROBLEM PRIMARY CAUSE 1 Secondary cause 2.1 Tertiary cause 2.2.1 PRIMARY CAUSE 2 Secondary cause 2.2 Tertiary cause 2.2.2 PRIMARY CAUSE 3 Secondary cause 2.3 Tertiary cause 2.2.3 Consequences Consequences Consequences DESIRED IMPACT OUTCOME 1 OUTPUT 2.1 OUTCOME 2 OUTPUT 2.2 OUTCOME 3 OUTPUT 2.3 Intervention Intervention Intervention 2.2.1 2.2.2 2.2.3 Reduction in poverty Women empowered Women in leadership roles Improved educational policies Parents persuaded to send girls to school Young women educated Economic opportunities for women Female enrollment rates increase Curriculum improved Schools built School system hires and pays teachers To have synergy and achieve impact all of these need to address the same target population. Program Goal: Young women educated Advocacy Project Goal: Improved educational policies enacted ASSUMPTION (that others will do this) Construction Project Goal: More classrooms built Teacher Education Project Goal: Improve quality of curriculum OUR project PARTNER will do this Program goal at impact level One form of Program Theory (Logic) Model Economic context in which the project operates Design Inputs Institutional and operational context Political context in which the project operates Implementation Process Outputs Outcomes Impacts Sustainability Socio-economic and cultural characteristics of the affected populations Note: The orange boxes are included in conventional Program Theory Models. The addition of the blue boxes provides the recommended more complete analysis. 60 61 Education Intervention Logic Output Clusters Institutional Management Specific Impact Outcomes Better Allocation of Educational Resources Increased Affordability of Education Quality of Education Economic Growth Skills and Learning Enhancement MDG 2 Equitable Access to Education Improved Participation in Society Education Facilities MDG 3 Poverty Reduction MDG 1 Social Development MDG 2 Health Global Impacts Improved Family Planning & Health Awareness Curricula & Teaching Materials Teacher Recruitment & Training Intermediate Impacts Greater Income Opportunities Optimal Employment Source: OECD/DAC Network on Development Evaluation Expanding the results chain for multi-donor, multicomponent program Impacts Intermediate outcomes Outputs Inputs Increased rural H/H income Increased production Credit for small farmers Donor Increased political participation Access to offfarm employment Rural roads Government Improved education performance Increased school enrolment Improved health Increased use of health services Schools Health services Other donors Attribution gets very difficult! Consider plausible contributions each makes. Contribution Analysis* 1. 2. 3. Develop a program logic (theory of change) Identify existing evidence on the causal links between interventions and possible effects. Identify gaps in information. Assess alternative explanations of possible effects. Assess the possible contributions of different factors to effects. * John Mayne et al in EES’ Evaluation journal, Vol. 18, No. 3, July 2012 64 Contribution Analysis* 4. 5. 6. Create a performance story on the contribution of the intervention to changes in impact variables. Seek out additional evidence to improve the program’s performance story. Revise and strengthen the performance story. * John Mayne et al in EES’ Evaluation journal, Vol. 18, No. 3, July 2012 65 RealWorld Evaluation Designing Evaluations under Budget, Time, Data and Political Constraints Where there was no baseline Ways to reconstruct baseline conditions A. B. C. D. E. Secondary data Project records Recall Key informants Participatory methods 67 Assessing the utility of potential secondary data Reference period Population coverage Inclusion of required indicators Completeness Accuracy Free from bias 68 Assessing the reliability of project records Who collected the data and for what purpose? Were they collected for record-keeping or to influence policymakers or other groups? Do monitoring data only refer to project activities or do they also cover changes in outcomes? Were the data intended exclusively for internal use? For use by a restricted group? Or for public use? 69 Limitations of recall Generally not reliable for precise quantitative data Sample selection bias Deliberate or unintentional distortion Few empirical studies (except on expenditure) to help adjust estimates 70 Sources of bias in recall Who provides the information Under-estimation of small and routine expenditures “Telescoping” of recall concerning major expenditures Distortion to conform to accepted behavior: • • • Intentional or unconscious Romanticizing the past Exaggerating (e.g. “We had nothing before this project came!”) Contextual factors: • • Time intervals used in question Respondents expectations of what interviewer wants to know Implications for the interview protocol 71 Improving the validity of recall Conduct small studies to compare recall with survey or other findings. Ensure all relevant groups interviewed Triangulation Link recall to important reference events • Elections • Drought/flood/tsunami/war/displacement • Construction of road, school etc 72 Key informants Not just officials and high status people Everyone can be a key informant on their own situation: • Single mothers • Factory workers • Users of public transport • Sex workers • Street children 73 Guidelines for key-informant analysis Triangulation greatly enhances validity and understanding Include informants with different experiences and perspectives Understand how each informant fits into the picture Employ multiple rounds if necessary Carefully manage ethical issues 74 PRA and related participatory techniques PRA (Participatory Rapid Appraisal) and PLA (Participatory Learning and Action) techniques collect data at the group or community [rather than individual] level Can either seek to identify consensus or identify different perspectives Risk of bias: • If only certain sectors of the community • participate If certain people dominate the discussion 75 Participatory approaches should be used as much as possible but even they should be used with appropriate rigor: how many (and which) people’s perspectives contributed to the story? 76 Summary of issues in baseline reconstruction Variations in reliability of recall Memory distortion Secondary data not easy to use Secondary data incomplete or unreliable Key informants may distort the past Yet in many cases we need to use one or more of these methods to determine what change occurred during the life of a project. 77 RealWorld Evaluation Designing Evaluations under Budget, Time, Data and Political Constraints Mixed-method (MM) evaluations The Main Messages 1. 2. 3. 4. No single evaluation approach can fully address the complexities of development evaluations MM combines the breadth of quantitative (QUANT) evaluation methods with the depth of qualitative (QUAL) methods MM is an integrated approach to evaluation with specific tools and techniques for each stage of the evaluation cycle MM are used differently by evaluators with a QUANT orientation and those with a QUAL orientation – and offer distinct benefits for each kind of evaluation 79 A. Why mixed methods? No single evaluation methodology can fully explain how development programs operate in the real world This explains the growing interest in mixed methods evaluations 80 Why mixed methods? No single evaluation method can fully explain how development programs operate in the real world 1. 2. 3. 4. 5. Programs operate in complex and changing environments Interventions are affected by historical, cultural, political, economic and other contextual factors Different methodologies are needed to measure different contextual factors, processes and outcomes. Even “simple” interventions often involve complex processes of organizational and behavioral change Programs change depending on how different sectors of the target population respond 81 The benefits of a mixed methods approach QUANTITATIVE breadth How many? How much? How representative of the total population? Are changes statistically significant? QUALITATIVE depth + • • • How were changes experienced by individuals? What actually happened on the ground? The quality of services 82 Four decisions for designing a mixed methods evaluation 1. 2. 3. 4. At which stages of the evaluation are mixed methods used? Is the design sequential or concurrent? Which approach is dominant? Is the design single or multi-level? 83 Decision 1: At which stages of the evaluation are mixed methods used? QUANT QUAL Mixed 1. Formulation of hypotheses 2. Sample design 3. Evaluation design 4. Data collection and recording 5. Triangulation 6. Data analysis and interpretation Mixed methods can be used at any stage of the evaluation. A fully integrated MM design combines QUANT and QUAL methods at all stages of the evaluation 84 Decision 2: Is the design sequential or concurrent? Sequential designs: • QUANT and QUAL approaches are used in sequence Concurrent designs • QUANT and QUAL approaches are both used at the same time 85 Sequential QUAL dominant MM design: Evaluating the adoption of new seed varieties by different types of rural families. quant Rapid QUANT household survey in project villages to estimate HH characteristics, ethnicity, agricultural production and seed adoption QUAL QUAL QUAL data collection using key informants focus groups, observation, and preparation of case studies on households and farming practices. QUAL data analysis using within and between-case analysis and constant comparison. Triangulation among different data sources. 86 A concurrent MM design: Triangulating QUANT and QUAL estimates of household income in project and comparison areas QUANT and QUAL data collection methods are used at the same time Project communities Comparison communities QUANT household surveys QUANT/QUAL Observation of household possessions and construction quality QUAL: Focus groups Triangulation of estimates from the 3 sources – to obtain the most reliable estimate of household income 87 Decision 3: which approach is dominant? QUAL oriented studies gradually incorporating more QUANT focus A B E F QUAL QUANT D G C QUANT oriented studies gradually incorporating more QUAL focus A = completely QUANT design B = dominant QUANT with some QUAL elements C = QUANT oriented design giving equal weight to both approaches D = Study designed as balanced MM E = QUAL oriented design giving equal weight to both approaches. F = dominant QUAL design with some QUANT elements G = completely QUAL design 88 A QUANT dominant evaluation design A B E F QUAL QUANT D G C Example: A rapid qualitative diagnostic study is conducted to help design a quantitative household survey. The data is analyzed using quantitative analysis techniques [e.g. regression analysis] 89 A QUAL dominant evaluation design A B E F QUAL QUANT D G C Example A rapid quantitative sample survey is conducted. This is used to develop a typology of rice production systems. Qualitative case studies are selected to represent each type. The data is analyzed and presented using qualitative methods such as narrative descriptions, photographs and social maps. 90 A Multi-level mixed methods design The effects of a school feeding program on school enrolment C. Using mixed methods to strengthen each stage of the evaluation 1. Hypothesis formulation 2. Sample design 3. Evaluation design 4. Data collection 5. Triangulation 6. Data analysis and interpretation Stage 1. Mixed methods approaches to hypothesis development Combining deductive (QUANT) and inductive (QUAL) hypotheses Basing the evaluation framework on a theory of change Strengthening construct validity by combining different QUANT and QUAL indicators Contextualizing the evaluation 93 Comparing DEDUCTIVE and INDUCTIVE hypotheses Deductive Inductive Mainly used in QUANT research Mainly used in QUAL research Hypotheses test theories based on prior research Hypotheses based on observations in the field Hypotheses defined at start of the evaluation before data collection begins Hypotheses not defined until data collection begins Hypotheses normally do not change Hypotheses evolve as data collection progresses Hypotheses can be tested experimentally Hypotheses are tested using Theory of Change or logically Mixed methods hypotheses combine both deductive and inductive 94 Stage 2. Mixed method sample designs Parallel mixed method sampling • Random (QUANT) and purposive (QUAL) sampling Sequential MM sampling Multi-level MM sampling Strengthening the coverage of the sampling frame Strengthening the matching of the project and control groups 95 Stage 3. Mixed method evaluation design Combining experimental and quasi-experimental; designs with QUAL techniques to explore: • Processes and quality of services • Context • Behavioral change Flexibility to adapt the evaluation to changes in the project design or the project context In-depth analysis of how the project affects different groups Creative identification of comparison groups 96 Stage 4. Strengthening data collection A. B. Integrating survey and QUAL data collection Commonly used mixed method data collection methods for strengthening QUANT evaluations 1. 2. 3. 4. C. D. E. F. Focus groups Observation Secondary data Case studies Reconstructing baseline data Interviewing difficult-to-reach groups Collecting information on sensitive topics Attention to contextual clues 97 Stage 5. Validating findings through triangulation Possible return to field QUANT data collection Household survey data collected on income and expenditures QUANT data analysis: Calculating mean, frequency distributions and standard deviation of income and expenditures QUAL TRIANGULATION PROCESS Findings compared, reconciled and integrated. When different estimates are obtained all of the data is reviewed to understand why differences occur. If necessary teams may return to the field to investigate further Data collection. Sub-sample of household interview families selected. Interviews, key informants and observation. Detailed notes, taped interviews and photos. QUAL Data analysis Review of interview and observation notes, Analysis using constant comparative method 98 Different kinds of triangulation Different data collection methods Different interviewers Collecting information at different times Different locations and contexts 99 Stage 6. Mixed method data analysis and interpretation Parallel MM data analysis Conversion MM data analysis • Converting QUAL data into QUANT indicators and vice versa] Sequential MM data analysis Multi-level MM data analysis Generalizing findings and recommendations to other potential program settings 100 Quantitative data collection methods Strengths and weaknesses Strengths Generalization statistically representative Estimate magnitude and distribution of impacts Clear documentation of methods Standardized approach Statistical control of bias and external factors Weaknesses Surveys cannot capture many types of information Do not work for difficult to reach groups No analysis of context Survey situation may alienate Long delay in obtaining results Reducing data to easy-to-measure indicators loses perspectives on complex information 101 Using Qualitative methods to improve the Evaluation design and results Use recall to reconstruct the pre-test situation Interview key informants to identify other changes in the community or in gender relations Conduct interviews or focus groups with women and men to • • assess the effect of loans on gender relations within the household, such as changes in control of resources and decision-making identify other important results or unintended consequences: • increase in women’s work load, • increase in incidence of gender-based or domestic violence 102 RealWorld Evaluation Designing Evaluations under Budget, Time, Data and Political Constraints Determining Counterfactuals Attribution and counterfactuals How do we know if the observed changes in the project participants or communities • income, health, attitudes, school attendance. etc are due to the implementation of the project • credit, water supply, transport vouchers, school construction, etc or to other unrelated factors? • changes in the economy, demographic movements, other development programs, etc 104 The Counterfactual What change would have occurred in the relevant condition of the target population if there had been no intervention by this project? 105 Where is the counterfactual? After families had been living in a new housing project for 3 years, a study found average household income had increased by an 50% Does this show that housing is an effective way to raise income? 106 I n c o m e Comparing the project with two possible comparison groups Project group. 50% increase 750 Scenario 2. 50% increase in comparison group income. No evidence of project impact 500 Scenario 1. No increase in comparison group income. Potential evidence of project impact 250 2004 2009 Control group and comparison group Control group = randomized allocation of subjects to project and non-treatment group Comparison group = separate procedure for sampling project and non-treatment groups that are as similar as possible in all aspects except the treatment (intervention) 108 IE Designs: Experimental Designs Randomized Control Trials Eligible individuals, communities, schools etc are randomly assigned to either: • The project group (that receives the services) or • The control group (that does not have access to the project services) 109 Primary Outcome A graphical illustration of an ‘ideal’ counterfactual using pre-project trend line then RCT Subjects randomly assigned either to … Intervention IMPACT Impact Treatment group Control group Time 110 There are other methods for assessing the counterfactual Reliable secondary data that depicts relevant trends in the population Longitudinal monitoring data (if it includes non-reached population) Qualitative methods to obtain perspectives of key informants, participants, neighbors, etc. 111 Ways to reconstruct comparison groups Judgmental matching of communities When there is phased introduction of project services beneficiaries entering in later phases can be used as “pipeline” comparison groups Internal controls when different subjects receive different combinations and levels of services 112 Using propensity scores and other methods to strengthen comparison groups Propensity score matching Rapid assessment studies can compare characteristics of project and comparison groups using: • • • • • Observation Key informants Focus groups Secondary data Aerial photos and GIS data 113 Issues in reconstructing comparison groups Project areas often selected purposively and difficult to match Differences between project and comparison groups - difficult to assess whether outcomes were due to project interventions or to these initial differences Lack of good data to select comparison groups Contamination (good ideas tend to spread!) Econometric methods cannot fully adjust for initial differences between the groups [unobservables] 114 What experiences have you had with identifying counterfactual data? 115 Enough of our presentations: it’s time for you (THE RealWorld PEOPLE!) to get involved yourselves. Time for smallgroup work. Read your case studies and begin your discussions. Small group case study work 1. 2. 3. Some of you are playing the role of evaluation consultants, others are clients commissioning the evaluation. Decide what your group will propose to do to address the given constraints/ challenges. Prepare to negotiate the ToR with the other group (later this afternoon). The purpose of this exercise to gain some practical ‘feel’ for applying what we’ve learned about RealWorld Evaluation. Group A (consultants) Evaluation team to consider how they will propose a revised evaluation design and plan that reduces the budget by 25% to 50%, and yet meets the needs of both clients (City Housing Department and the international donor). Group B (clients) will also review the initial proposal in light of what they have learned about RealWorld Evaluation, and prepare to re-negotiate the plans with the consultancy group. Note: there are two types of clients: the Housing Department (project implementers) and the international donor (foundation). Groups are given 45 minutes to prepare their cases. Later the Consultants’ group will meet with Clients’ groups to negotiate their proposed revisions in the plans for this evaluation. 45 minutes will be available for those negotiation sessions. RealWorld Evaluation Designing Evaluations under Budget, Time, Data and Political Constraints More holistic approaches to Impact Evaluation Some recent developments in impact evaluation in development 2003 2006 J-PAL is best understood as a network of affiliated researchers … united by their use of the randomized trial methodology… 2008 Impact Evaluation for Improving Development – 3ie and AfrEA conference in Cairo March 2009 2012 Stern, E., N. Stame, J. Mayne, K. Forss, R. Davies and B. Befani. Broadening the Range of Designs and Methods for Impact Evaluations DFID Working Paper 38 2009 121 Randomized control trials and strong statistical designs When might they be appropriate … and when not? The arguments in favor of RCTs 1. Selection bias has resulted in wrongly attributing changes to the results of an intervention a. b. 2. 3. Selection bias tends to over-estimate program impacts Agencies continue to support their favorite programs – even though they may be producing fewer benefits than claimed A high percentage of evaluations use methodologically unsound methods, but funding agencies are rarely challenged Agencies evaluate themselves and are able to ensure that the findings are never too critical. 123 The arguments in favor of RCTs 4. 5. Isolating defined treatments and testing them in a systematically controlled way is an effective way to improve our understanding of the potential benefits of this specific intervention. It has proved possible to use RCTs in a very wide range of contexts and sectors and in many cases they have produced useful findings at the operational and policy levels 124 The arguments in favor of RCTs 6. RCTs provide a reference point against which to assess the methodological rigor of other evaluation designs. A good way to keep evaluators honest. 125 Practical, political and ethical limitations on the use of RCTs 1. It is only possible to use RCTs in a small proportion of evaluations It is often not possible to identify a comparison group The goal of reaching as many people as possible does not permit the exclusion of certain people Programs are often implemented gradually over a period of time and it is not possible to define a specific starting point. 126 Practical, political and ethical limitations on the use of RCTs 2. Opposition to the random assignment of beneficiaries Politicians do not wish to lose their control over who receives benefits Managers (and others) may wish to target the poorest or most vulnerable sectors and not to give scarce resources to better-off groups Communities may not accept this approach 127 Practical, political and ethical limitations on the use of RCTs 3. Ethical considerations Withholding needed resources from people who need them Giving benefits to less needy groups while depriving the most needy Treating people as guinea pigs in a laboratory 128 Practical, political and ethical limitations on the use of RCTs The ethical arguments are not so clear and one-sided Using weaker evaluation designs may result in poor programs continuing to be funded – thus depriving people of potentially higher benefits from new programs that could use these funds Resources are frequently limited so that some allocation criteria is required – what are the alternatives to randomization? Rigorous evaluations can improve program performance 129 Methodological limitations of RCTs and strong QEDs 1. 2. 3. 4. Can only evaluate a single component (intervention) of a multi-component program Inflexible design – difficult to adjust design to modifications to program design or to changes in the context in which programs are implemented Cannot control what happens during implementation Regression only estimates the average effect size – difficult to generalize the findings to other contexts. 130 Methodological limitations of RCTs and strong QEDs 5. Simple RCTs to not assess the implementation process and when intended impacts are not found, the analysis cannot distinguish between “design failure” and “implementation failure” 131 Implications for RWE 1. RCTs are a potentially useful component of some kinds of evaluation – but should almost never be used in isolation An RCT is strongest when part of a mixed methods design 2. Define whether a program is intended: as an experiment to rigorously assess promising new interventions that might be replicated on a large scale OR To provide services to as many people as possible 132 Implications for RWE 3. 4. If programs are designed as experiments, an RCT can be cheaper than many alternative survey designs. Always ask the question of how nonexperimental and weak QEDs can address the very serious issues of selection bias and the over-estimation of project impacts. Never automatically reject RCTs without assessing the important reasons why these methods were developed But never accept RCTs (or any other evaluation method) as the best or only approach 133 Implications for RWE 5. The real world is too complex for any single evaluation approach to fully address all of the important questions that must be answered. 134 So what should be included in a “rigorous impact evaluation”? 1. Direct cause-effect relationship between one output (or a very limited number of outputs) and an outcome that can be measured by the end of the research project? Pretty clear attribution. … OR … 2. Changes in higher-level indicators of sustainable improvement in the quality of life of people, e.g. the MDGs (Millennium Development Goals)? More significant but much more difficult to assess direct attribution. 135 So what should be included in a “rigorous impact evaluation”? OECD-DAC (2002: 24) defines impact as “the positive and negative, primary and secondary long-term effects produced by a development intervention, directly or indirectly, intended or unintended. These effects can be economic, sociocultural, institutional, environmental, technological or of other types”. Does it mention or imply direct attribution? Or point to the need for counterfactuals or Randomized Control Trials (RCTs)? 136 Different lenses needed for different situations in the RealWorld Simple Complicated Following a recipe Sending a rocket to the Raising a child moon Recipes are tested to assure easy replication Sending one rocket to the moon increases assurance that the next will also be a success The best recipes give There is a high degree good results every time of certainty of outcome Complex Raising one child provides experience but is no guarantee of success with the next Uncertainty of outcome remains Sources: Westley et al (2006) and Stacey (2007), cited in Patton 2008; also presented by Patricia Rodgers at Cairo impact conference 2009. 137 Examples of cause-effect correlations that are generally accepted • Vaccinating young children with a standard set of vaccinations at prescribed ages leads to reduction of childhood diseases (means of verification involves viewing children’s health charts, not just total quantity of vaccines delivered to clinic) • Other examples … ? 138 But look at examples of what kinds of interventions have been “rigorously tested” using RCTs • Conditional cash transfers • The use of visual aids in Kenyan schools • Deworming children (as if that’s all that’s needed to enable them to get a good education) • Note that this kind of research is based on the quest for Silver Bullets – simple, most cost-effective solutions to complex problems. 139 “Far better an approximate answer to the right question, which is often vague, than an exact answer to the wrong question, which can always be made precise.“ J. W. Tukey (1962, page 13), "The future of data analysis". Annals of Mathematical Statistics 33(1), pp. 1-67. Quoted by Patricia Rogers, RMIT University 140 Is that what we call “scientific method”? There is much more to impact, to rigor, and to “the scientific method” than RCTs. Serious impact evaluations require a more holistic approach. 141 Consequences Consequences Consequences DESIRED IMPACT OUTCOME 1 OUTCOME 2 OUTCOME 3 A more comprehensive design OUTPUT 2.1 OUTPUT 2.2 OUTPUT 2.3 A Simple RCT Intervention Intervention Intervention 2.2.1 2.2.2 2.2.3 The limited use of strong evaluation designs In the RealWorld (at least of international development programs) we estimate that: • fewer than 5%-10% of project impact evaluations use a strong experimental or even quasi-experimental designs • significantly less than 5% of impact evaluations use randomized control trials (‘pure’ experimental design) 143 What kinds of evaluation designs are actually used in the real world of international development? Findings from meta-evaluations of 336 evaluation reports of an INGO. Post-test only 59% Before-and-after 25% With-and-without 15% Other counterfactual 1% Rigorous impact evaluation should include (but is not limited to): 1) thorough consultation with and involvement by a variety of stakeholders, 2) articulating a comprehensive logic model that includes relevant external influences, 3) getting agreement on desirable ‘impact level’ goals and indicators, 4) adapting evaluation design as well as data collection and analysis methodologies to respond to the questions being asked, … Rigorous impact evaluation should include (but is not limited to): 5) adequately monitoring and documenting the process throughout the life of the program being evaluated, 6) using an appropriate combination of methods to triangulate evidence being collected, 7) being sufficiently flexible to account for evolving contexts, … Rigorous impact evaluation should include (but is not limited to): 8) using a variety of ways to determine the counterfactual, 9) estimating the potential sustainability of whatever changes have been observed, 10) communicating the findings to different audiences in useful ways, 11) etc. … The point is that the list of what’s required for ‘rigorous’ impact evaluation goes way beyond initial randomization into treatment and ‘control’ groups. We must be careful that in using the “Gold Standard” we do not violate the “Golden Rule”: “Judge not that you not be judged!” In other words: “Evaluate others as you would have them evaluate you.” Caution: Too often what is called Impact Evaluation is based on a “we will examine and judge you” paradigm. When we want our own programs evaluated we prefer a more holistic approach. How much more helpful it is when the approach to evaluation is more like holding up a mirror to help people reflect on their own reality: facilitated self-evaluation. The bottom line is defined by this question: Are our programs making plausible contributions towards positive impact on the quality of life of our intended beneficiaries? Let’s not forget them! 153 153 ANY QUESTIONS? 154 154 Time for consultancy teams to meet with clients to negotiate the revised ToRs for the evaluation of the housing project. 155 What did you learn about RealWorld Evaluation by getting involved in that case study and roleplaying exercise? 156 Next steps: What are some of the practical realities of applying RWE approaches in our own work? 157 Main workshop messages Evaluators must be prepared for RealWorld evaluation challenges. 2. There is considerable experience to learn from. 3. A toolkit of practical “RealWorld” evaluation techniques is available (see www.RealWorldEvaluation.org). 4. Never use time and budget constraints as an excuse for sloppy evaluation methodology. 5. A “threats to validity” checklist helps keep you honest by identifying potential weaknesses in your evaluation design and analysis. 6. The evaluation report should disclose what constraints were faced and what was done to address them. 1. 158 Additional References for IE DFID “Broadening the range of designs and methods for impact evaluations” http://www.oecd.org/dataoecd/0/16/50399683.pdf Robert Picciotto “Experimententalism and development evaluation: Will the bubble burst?” in Evaluation journal (EES) April 2012: http://evi.sagepub.com/ Martin Ravallion “Should the Randomistas Rule?” (Economists’ Voice 2009) http://ideas.repec.org/a/bpj/evoice/v6y2009i2n6.html William Easterly “Measuring How and Why Aid Works—or Doesn't” http://aidwatchers.com/2011/05/controlled-experiments-and-uncontrollablehumans/ Control freaks: Are “randomised evaluations” a better way of doing aid and development policy? The Economist June 12, 2008 http://www.economist.com/node/11535592 Series of guidance notes (and webinars) on impact evaluation produced by InterAction (consortium of US-based INGOs): http://www.interaction.org/impact-evaluation-notes Evaluation (EES journal) Special Issue on Contribution Analysis, Vol. 18, No. 3, July 2012. www.evi.sagepub.com Impact Evaluation In Practice www.worldbank.org/ieinpractice 159 Unanticipated consequences of development interventions: a blind spot for evaluation theory and practice [and if time permits] A few thoughts on evaluating complex development interventions Michael Bamberger 1. Development programs almost never work out as planned The outcomes of most programs are unpredictable • • Programs propose simple, linear solutions to complex problems Even when programs go smoothly and are implemented as planned – human behavior is difficult to predict Programs operate in complex systems with constantly changing economic, political, security, natural environmental and socio-cultural environments Program planners have much less control than they often think 161 The scope of unanticipated outcomes increases as we move from programs to complex development interventions - • such as Country Programs or multi-donor collaborative initiatives 162 2. So why are evaluators still surprised when programs have unanticipated outcomes? Unanticipated outcomes still catch planners and evaluators off-guard even though experience has shown that very few programs work out exactly as planned Many unanticipated outcomes, including serious negative consequences, are not even detected by many evaluations 163 Some serious negative outcomes not captured by the evaluation A food-for-work program in Central America: • • Many women were forbidden by their husbands from participating in the project Some women were seriously beaten by their husbands for attending meetings A slum upgrading project in South East Asia: • Prior to the official start of the project many slum dwellers were forced (sometimes at gunpoint) to sell their houses at a very low price to people with political contacts A road construction project in East Africa: • Prior to the first supervision mission for a road construction project the government destroyed a number of houses to avoid paying compensation to the families as specified in the loan agreement 164 3. Why are evaluations often unable to detect these unanticipated outcomes? Funders only want evaluators to look at whether their programs achieve their intended outcomes/ goals • Don’t rock the boat Real-world evaluation constraints: • Budget, time, data constraints • Political and organizational constraints 165 Many evaluation designs make it difficult to identify unanticipated outcomes Many theories of change and logic models only explain and assess how a program is expected to achieve its objectives • They say very little about other things that might happen Results frameworks only monitor intended outcomes Conventional quantitative evaluation designs only assess whether observed changes can be attributed to the program intervention 166 Many evaluation designs make it difficult to identify unanticipated outcomes, cont. Structured surveys collect information relevant to program objectives but rarely provide an opportunity for respondents to talk about other things on their mind Difficult to find/interview vulnerable groups and those who do not benefit Focus groups often only cover people involved in the project 167 Many development evaluations focus too much on the internal concerns and procedures of their own agency • And do not focus sufficiently on the national development context and the concerns of national stakeholders. 168 Many unanticipated events occur outside the vision field of the evaluation Key events occur before the evaluation begins (e.g. people are forced to sell land or houses at bargain prices) Events occur when the evaluator is not there (at night, during the rainy season, in the privacy of the household) Critical events occur after the evaluation has ended Many quantitative evaluations have no contact with the program between the two applications of the survey instruments (pre-test and post-test) 169 Many unanticipated events occur outside the vision field of the evaluation, cont. Evaluators often ignore context Budget constraints mean that the poorest and most vulnerable groups who live far from the village center, often cannot be interviewed 170 4. Why should we worry? Unanticipated events: • Seriously reduce the efficiency and • • • effectiveness of programs Important lessons are not learned The vulnerable and voiceless tend to be the most affected Some groups or communities can be significantly worse off as a result of development interventions 171 5. What can be done? This section draws on and expands the framework presented in: Jonathan Morell “Evaluation in the Face of Uncertainty: Anticipating Surprise and Responding to the Inevitable” Guilford 2010 172 What can be done? Understanding the nature of the unexpected Types of surprise • • • Surprises we anticipate and address Surprises that could be planned for and detected Surprises that cannot be anticipated The probability and magnitude of surprise is related to: • • • • • • The level of innovation in the program [uncharted territory] The nature of the intervention Fidelity to the design protocol Robustness of the program in different settings Level of understanding of the program context Time between the intervention and the estimation of outcomes 173 What can be done? Ways to foresee and address unanticipated outcomes Use of theory-based evaluation • With focus on mechanisms, linkages and testing assumptions Systematic use of past experience • Participatory consultations with affected populations Break the evaluation down into shorter segments Adapt forecasting methods used by planners Broader definition of the role of monitoring 174 What can be done? Agile evaluation Always have a “Plan B” [see next slide] Flexibility in terms of data collection Flexible designs that can be rapidly adapted to changing circumstances Dynamic monitoring Unpacking the program into separate, but closely linked, components • Each of which is easier to evaluate Constant review of the program theory and the underlying assumptions 175 “Plan B” Planning for the unexpected What to do when: the program begins before the baseline study has been conducted? the comparison groups vanish? the program is restructured? Essential data is not available? • It may be intentionally withheld 176 “Eyes wide open” evaluation Knowing what is really happening on the ground Find out about critical events that took place before the evaluation/ program officially began Develop diverse sources inside and outside the program to make sure you know what everyone else in the country knows Check who you are getting information from and who you are not meeting • Know who participates in focus groups Develop creative ways to get information on, and about, non-beneficiaries and groups who are worse off 177 A few thoughts on the evaluation of complex programs The dimensions of complexity 1. Scope and scale 2. Organizational complexity 3. Contextual factors 4. Behavioral 5. Challenges assessing • • • 6. Attribution Contribution Substitution Some interventions are less complex than claimed 178 1. Scope and scale a. b. c. Number of components and service levels Number and types of geographic areas Larger programs tend to have more features and to be implemented differently in different regions 179 2. Organizational complexity a. • • • b. c. d. Programs co-funded by different agencies tend to be more complex Different implementation strategies and management procedures Different components/services Coordination problems Programs with several coordinating and implementing agencies National and regional level programs have more levels of coordination National agencies may lack implementation capacity 180 3. Contextual factors affecting complexity a. b. c. d. e. f. g. Economic setting Political context Administrative and organizational Climatic and ecological Legal and regulatory Historical Socio-cultural 181 4. Behavioral a. b. c. “Simple” projects may involve complex processes of behavioral change Understanding group behavior Social networks and social communication 182 5. Attribution, contribution and substitution a. b. c. • • • • • • Attribution difficult to assess for complex programs Different donors Target groups exposed to other programs Difficult to define a counterfactual Contribution analysis: Conventional CA only focuses on process but not outcomes Substitution Difficult to track donor funds and other inputs Even more difficult to track substitution of donor for government resources 183 6. Interventions that look “complex” may not be “complex” interventions may comprise a number of “simple” components a. • b. Coordination and management may be the only “complex” dimensions Some agencies may find it convenient to claim their programs are too complex to be rigorously evaluated – to avoid critical scrutiny. 184 STRATEGIES FOR EVALUATING COMPLEX PROGRAMS Counterfactual designs • Attribution analysis • Contribution analysis • Substitution analysis Theorybased approaches Qualitative approaches Quantitative approaches Mixed method designs Rating scales strengthening alternative counterfactuals • “Unpacking complex programs” • Portfolio analysis •Reconstructing baseline data • Creative use of secondary data • Secondary data •Triangulation Estimating impacts The valueadded of agency X Net increase in resources for a program 185