Wednesday, December 19, 2012

What Counts as Credible Evidence in Applied Research and Evaluation Practice?What Counts as Credible Evidence in Applied Research and Evaluation Practice? by Stewart I. Donaldson
My rating: 5 of 5 stars

The evaluation field continues to engage in paradigm wars that involve heated debates over which approaches and methodologies produce the most reliable results to support evidence-based policy-making. Somewhat regretfully, the commendable goal of enhanced rigour in evaluation research has been hijacked by a focus on a narrow set of experimental methods—randomized controlled trials or RCTs—which have been proclaimed as the ‘gold standard’ by their proponents. This trend has been boosted by calls for unambiguous measurements of results and impacts, and cost-efficiency, by policy-makers and bureaucrats struggling with making policy choices and undertaking programs under increasing resource constraints. On the other side, the reaction from the proponents of more qualitative methodologies and participatory approaches to evaluation has been strong, even emotional at times. As a professional evaluator, I’ve witnessed these brawls first hand.

This book makes an excellent contribution to the debate through a balanced presentation of the issues and by letting the different sides to make their respective cases. The authors in the book include a number of leading scholars and practitioners in the field. The perspective is North American (all authors work in the US or Canada) and draws heavily on experiences from education and social services. Although my own work pertains to evaluating international development programs, I found the discussion in the book on what constitutes credible evidence very valuable.

In the two introductory chapters the editors frame the debate in the context of a search for evidence-based society and how this has played out in the use of experimental and non-experimental designs for collecting evidence. Quite didactically, they place the debate within the broader scientific paradigms of social inquiry, including logical positivism, post-positivism, constructivism and related thinking, and pragmatism. They also tentatively place the chapter authors along these paradigmatic axes. The following eight chapters are divided into two sections arguing, respectively, for experimental and non-experimental routes to credible evidence. The trajectory of the chapters moves from the most hard-core case for experimental designs to an argument for the credibility of image-based research in evaluation. The chapters in between tend to take a conciliatory approach to the extent that the last of the chapters in the experimental section could have been moved to the non-experimental section.

Part II entitled ‘Experimental Approaches as a Route to Credible Evidence’ contains four chapters. In the first, Gary Henry argues for high-quality policy and program impact evaluations as a necessity for providing solid evidence for policy choices, linking the matter to democratic theory and the need to detect and debunk bad policies. His view assumes that there actually are ways to objectively define what produces desirable outcomes—and what such outcomes are. He acknowledges types of bias in evaluation research, but sees them in technical, rather than political terms. In the chapter that follows, Leonard Bickman and Stephanie Reich assess the credibility, reliability and validity of RCTs, concluding that while there are threats to the validity (especially external validity) of RCTs, they still can be seen to be amongst the most credible designs available to evaluators. Despite this overall conclusion, Bickman and Reich acknowledge that there are other, non-experimental approaches to establish causality (many natural science disciplines—geology, astronomy, engineering, and subfields of medicine—base their research on non-experimental designs). In social sciences, they particular highlight the program theory, theory-driven or pattern-matching method, recognizing that such other approaches are needed to supplement RCTs that can only answer a very limited number of questions.

The last chapter in this part of the book, by George Julnes and Debra Rog, introduces the concept of actionable evidence. The authors assert that for evidence to be useful, it should not only be credible but also actionable, defined as adequate and appropriate for guiding actions in targeted real-world contexts. The lengthy chapter takes as its starting point the question of relating the choice of methods to the questions that stakeholders want addressed. They proceed to outline a multitude of evaluation tasks (borrowing from Carol      Weiss) and then consider the implications of the tasks for methodology. Another way of framing the evaluation questions relates to the level of conclusion and the different levels of causal questions in impact evaluation, including whether the evaluation seeks to provide an aggregate description, disaggregation for causal analysis, or an inferential analysis of the underlying constructs and causal mechanisms. They state that, “experimental methods are argued as appropriate for strengthening impact-evaluation conclusions, but the value of these methods is dependent on the level of conclusions being addressed” (p. 104). Summarizing the discussion on the relationships between questions and methods, Julnes and Rog express their view that, while particular questions call for quantitative designs, there is substantial territory open to other designs. This summary leads the authors to consider the contextual factors that affect the adequateness and appropriateness of alternative methods, including policy context and the nature of the phenomena studied. They then discuss how to judge the adequacy of methods for providing the evidence that is needed to address the stakeholder questions identified, and when is it appropriate to use particular methods for causal analysis, taking into account constraints posed by factors internal to the program, evaluation capacity and political constraints, as well as ethical considerations. The discussion is nuanced and fair. Julnes and Rog conclude by affirming the primacy of the evaluation stakeholder questions in influencing the types of evidence needed. They caution against “simple frameworks that drive method choice in a somewhat automatic fashion” and instead wish to support “more informed judgments on method choice in political public policy environments” (p. 128). To me, this is one of the strongest, most thoughtful and balanced chapters in the book. Therefore, it also deserves its place in the middle, bridging the quantitative and qualitative parts.

Another chapter in this section of the book, preceding the chapter discussed above, by Russell Gersten and John Hitchcock focuses on the role of the What Works Clearinghouse, established in 2002 by the U.S. Department of Education. The chapter is descriptive and possibly useful to education researchers and evaluators, but did not raise my own somewhat biased interest.

Part III on ‘Nonexperimental Approaches for Building Credible Evidence’ consists of another five chapters, starting with one of the grand old men of evaluation, Michael Scriven. His chapter, ‘Demythologizing Causation and Evidence,’ is written with flair in lively language. Like Julnes and Rog before him, Scriven enlists other sciences—from mathematical physics and geology to anthropology, ethnography and epidemiology—to demonstrate how experimental methods are but one of the many approaches to analyse causation. He writes, “much of the world of science, suffused with causal claims, runs along very well with the usual high standards of evidence, but without RCTs” (p. 136). Breezing through the origins of causal concepts, the cognitive process of causal inference vs. observation, and the level of evidential certainty required for scientific, legal and practical purposes, he then addresses the alleged supremacy of RCTs as well as “other contenders.”  His myth-busting position is that “(i) the attempted takeover of the terms evidence and cause is partly inspired by the false dichotomy between experiment and quasi-experiment, and (ii) the whole effort is closely analogous to the attempted annexation of the concept of significance by statistically significant” (p. 151).

The following chapter by Jennifer Greene discusses evidence as ‘proof’ vs. evidence as ‘inkling.’ Her premise is that evaluation is both influenced by the political, organizational and sociocultural contexts, as well as it serves to shape that context. Consequently, “evaluation is not a bystander or neutral player in the debates that often surround it, but rather an active contributor to those debates and to the institutions that house them” (p. 153). Her chapter attempts to demonstrate how the present discourse assumes that ‘evidence’ can make social systems ‘efficient and effective’ and how these assumptions convey a particular view of human phenomena and responsibilities of government in democratic societies. The argument is a useful antidote to the positivistic view presented by Henry earlier in the book. Her vision of evidence is not providing the truth, or neat and tidy small answers to small questions. Rather, in Greene’s view evidence must provide a “window into the messy complexity of human experience; evidence that accounts for history, culture, and context; evidence that respects difference in perspective and values; evidence about experiences in addition to consequences; evidence about the responsibilities of government, not just the responsibilities of its citizens; evidence with the potential for democratic inclusion and legitimization of multiple voices—evidence not as proof but as inkling” (p. 166).

Sharon Rallis starts her chapter by telling the story of how she first encountered evaluation when she was teaching a federally funded summer program that was subject to an evaluation. The evaluators insisted on holding on to their plan to assess the program against a single outcome, with no regard to the important associated benefits that the program had bolstering the self-esteem of the participating students. Furthermore, the evaluation based on a quasi-experimental design deprived half of the students from participating in an important part of the program, which Rallis and other program colleagues felt was unfair. While the evaluators claimed that their work was scientific and rigorous, Rallis pondered about the missing piece and came to the conclusion that it was ‘probity’—goodness and moral soundness. Consequently, she began to study evaluation with a commitment to make evaluations useful for the program personnel and participants. This chapter elaborates on her vision of evaluation with probity and moral reasoning, grounded in nonconsequentialist theories. She explains: “The evidence we collect looks quite different from that of our colleagues who measure outcomes. Our aim is not to cast judgment … but to discover what happened and what the experience meant to the program participants. We hope that our discoveries can lead to improving the program and thus the well-being of the participants” (pp. 174-175). Rather than RCTs, evaluation done with these principles borrows tools from fields such as ethnography, phenomenology and sociolinguistics/semiotics. She presents a case from an evaluation and needs assessment of an HIV/AIDS education and prevention program that provides some unexpected insights into the participants’ experiences. She asserts that this work is rigorous because “it is grounded in theory and previous research and in moral principles of justice and caring” (p. 178).

Sandra Mathison in her chapter ‘Seeing Is Believing’ explores the credibility of image-based research and evaluation, as one form of evidence to establish and represent truth and value. Like the part III authors before her, she emphasizes how the credibility of evidence and the knowledge thus created is contingent on experiences, perception and social conventions. Image-based research uses images in three ways: (i) as data or evidence; (ii) as an elicitation device to collect other data, and (iii) as a representation of knowledge (p. 184). Mathison posits four considerations for establishing the credibility of image-based research: (1) quality of the research design, (2) attention to context, (3) adequacy of the image from multiple perspectives, and (4) the contribution images make to new knowledge (p. 188).

The last chapter in part III, by Thomas Schwandt, is entitled ‘Toward a Practical Theory of Evidence for Evaluation’ and it functions as a kind of recap of what has come before; as such, it could have equally well been placed in part IV on conclusions. This is another rich chapter that goes to the heart of the debate of what we mean by evidence: “…information helpful in forming a conclusion or judgment. Framed in a more rigorous epistemological perspective, evidence means information bearing on whether a belief or proposition is true or false, valid or invalid, warranted or unsupported. At present, we face some difficulty and confusion with understanding the term evidence in evaluation because it is often taken to be synonymous with the term evidence-based” (p. 199). He then proceeds to problematize the term evidence-based, as being narrowly interpreted to mean only a specific kind of finding regarding causal efficacy. Secondly, Schwandt argues why evidence cannot serve as a secure and infallible base or foundation for action. Furthermore, he emphasizes that, as an aspect of policy making, evaluation must consider ethics. Schwandt concludes that “deciding the question of what constitutes credible evidence is not the same as deciding the question of what constitutes credible evaluation … However necessary, developing credible evidence in evaluation is not sufficient for establishing the credibility of an evaluation” (p. 209). He further asserts that method choice alone does not determine what is credible and convincing evidence. He calls for framing evidence in a practical-theoretical way that that is concerned with the character and ethics of evidence and the contexts in which evidence is used.

In the final part of the book, Melvin Mark summarizes the different perspectives of the book with an aim of changing the terms of the debate. He concludes: “Extensive and continued discussion of the relative merits and credibility of RCTs versus other methods would have limited capacity to move forward our understanding and our practice. … by changing the terms of the debate, we may be able to improve understandings of deeply entrenched disagreements; move toward a common ground where such can be found; better understand the disagreements that remain; allow, at least in select places, credible evidence as part of the conversation; and enhance the capacity of stakeholders to make sensible decisions rather than be bewildered by our disagreement or draw allegiances based on superficial considerations” (pp. 237-8). Certainly a deserving goal.  The book ends with an epilogue by Stewart Donaldson that attempts to provide a practitioners guide for gathering credible evidence in the evidence-based global society.

I thoroughly enjoyed reading this book and although much of the debate around epistemology, approaches and methods is familiar to someone educated and working in evaluation and applied social science research, the way it is framed in this book is truly enlightening. For a thoughtful reader, it becomes evident that the truth—as almost always in such debates—is somewhere in the middle. All of the approaches and methodologies have merit when used appropriately and in appropriate contexts. Both experimental and non-experimental methods can be rigorous, but both can also have serious flaws with regard to internal and external validity, relevance and appropriateness. The old saw about everything looking like a nail when you have only a hammer in your tool box is true here as well. The take-home lesson is that, instead of allowing methods to dictate one’s evaluation questions and designs, one should choose one’s methods according to the questions one wants answered.

View all my reviews

No comments: